drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81
  82 /* Uncomment to enable debugging */
  83 /* #define TUN_DEBUG 1 */
  84
  85 #ifdef TUN_DEBUG
  86 static int debug;
  87
  88 #define tun_debug(level, tun, fmt, args...)                     \
  89 do {                                                            \
  90         if (tun->debug)                                         \
  91                 netdev_printk(level, tun->dev, fmt, ##args);    \
  92 } while (0)
  93 #define DBG1(level, fmt, args...)                               \
  94 do {                                                            \
  95         if (debug == 2)                                         \
  96                 printk(level fmt, ##args);                      \
  97 } while (0)
  98 #else
  99 #define tun_debug(level, tun, fmt, args...)                     \
 100 do {                                                            \
 101         if (0)                                                  \
 102                 netdev_printk(level, tun->dev, fmt, ##args);    \
 103 } while (0)
 104 #define DBG1(level, fmt, args...)                               \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 printk(level fmt, ##args);                      \
 108 } while (0)
 109 #endif
 110
 111 #define TUN_HEADROOM 256
 112 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 113
 114 /* TUN device flags */
 115
 116 /* IFF_ATTACH_QUEUE is never stored in device flags,
 117  * overload it to mean fasync when stored there.
 118  */
 119 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 120 /* High bits in flags field are unused. */
 121 #define TUN_VNET_LE     0x80000000
 122 #define TUN_VNET_BE     0x40000000
 123
 124 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 125                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 126
 127 #define GOODCOPY_LEN 128
 128
 129 #define FLT_EXACT_COUNT 8
 130 struct tap_filter {
 131         unsigned int    count;    /* Number of addrs. Zero means disabled */
 132         u32             mask[2];  /* Mask of the hashed addrs */
 133         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 134 };
 135
 136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 137  * to max number of VCPUs in guest. */
 138 #define MAX_TAP_QUEUES 256
 139 #define MAX_TAP_FLOWS  4096
 140
 141 #define TUN_FLOW_EXPIRE (3 * HZ)
 142
 143 struct tun_pcpu_stats {
 144         u64 rx_packets;
 145         u64 rx_bytes;
 146         u64 tx_packets;
 147         u64 tx_bytes;
 148         struct u64_stats_sync syncp;
 149         u32 rx_dropped;
 150         u32 tx_dropped;
 151         u32 rx_frame_errors;
 152 };
 153
 154 /* A tun_file connects an open character device to a tuntap netdevice. It
 155  * also contains all socket related structures (except sock_fprog and tap_filter)
 156  * to serve as one transmit queue for tuntap device. The sock_fprog and
 157  * tap_filter were kept in tun_struct since they were used for filtering for the
 158  * netdevice not for a specific queue (at least I didn't see the requirement for
 159  * this).
 160  *
 161  * RCU usage:
 162  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 163  * other can only be read while rcu_read_lock or rtnl_lock is held.
 164  */
 165 struct tun_file {
 166         struct sock sk;
 167         struct socket socket;
 168         struct socket_wq wq;
 169         struct tun_struct __rcu *tun;
 170         struct fasync_struct *fasync;
 171         /* only used for fasnyc */
 172         unsigned int flags;
 173         union {
 174                 u16 queue_index;
 175                 unsigned int ifindex;
 176         };
 177         struct napi_struct napi;
 178         bool napi_enabled;
 179         struct mutex napi_mutex;        /* Protects access to the above napi */
 180         struct list_head next;
 181         struct tun_struct *detached;
 182         struct ptr_ring tx_ring;
 183         struct xdp_rxq_info xdp_rxq;
 184 };
 185
 186 struct tun_flow_entry {
 187         struct hlist_node hash_link;
 188         struct rcu_head rcu;
 189         struct tun_struct *tun;
 190
 191         u32 rxhash;
 192         u32 rps_rxhash;
 193         int queue_index;
 194         unsigned long updated;
 195 };
 196
 197 #define TUN_NUM_FLOW_ENTRIES 1024
 198
 199 struct tun_prog {
 200         struct rcu_head rcu;
 201         struct bpf_prog *prog;
 202 };
 203
 204 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 205  * device, socket filter, sndbuf and vnet header size were restore when the
 206  * file were attached to a persist device.
 207  */
 208 struct tun_struct {
 209         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 210         unsigned int            numqueues;
 211         unsigned int            flags;
 212         kuid_t                  owner;
 213         kgid_t                  group;
 214
 215         struct net_device       *dev;
 216         netdev_features_t       set_features;
 217 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 218                           NETIF_F_TSO6)
 219
 220         int                     align;
 221         int                     vnet_hdr_sz;
 222         int                     sndbuf;
 223         struct tap_filter       txflt;
 224         struct sock_fprog       fprog;
 225         /* protected by rtnl lock */
 226         bool                    filter_attached;
 227 #ifdef TUN_DEBUG
 228         int debug;
 229 #endif
 230         spinlock_t lock;
 231         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 232         struct timer_list flow_gc_timer;
 233         unsigned long ageing_time;
 234         unsigned int numdisabled;
 235         struct list_head disabled;
 236         void *security;
 237         u32 flow_count;
 238         u32 rx_batched;
 239         struct tun_pcpu_stats __percpu *pcpu_stats;
 240         struct bpf_prog __rcu *xdp_prog;
 241         struct tun_prog __rcu *steering_prog;
 242         struct tun_prog __rcu *filter_prog;
 243 };
 244
 245 struct veth {
 246         __be16 h_vlan_proto;
 247         __be16 h_vlan_TCI;
 248 };
 249
 250 bool tun_is_xdp_buff(void *ptr)
 251 {
 252         return (unsigned long)ptr & TUN_XDP_FLAG;
 253 }
 254 EXPORT_SYMBOL(tun_is_xdp_buff);
 255
 256 void *tun_xdp_to_ptr(void *ptr)
 257 {
 258         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 259 }
 260 EXPORT_SYMBOL(tun_xdp_to_ptr);
 261
 262 void *tun_ptr_to_xdp(void *ptr)
 263 {
 264         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 265 }
 266 EXPORT_SYMBOL(tun_ptr_to_xdp);
 267
 268 static int tun_napi_receive(struct napi_struct *napi, int budget)
 269 {
 270         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 271         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 272         struct sk_buff_head process_queue;
 273         struct sk_buff *skb;
 274         int received = 0;
 275
 276         __skb_queue_head_init(&process_queue);
 277
 278         spin_lock(&queue->lock);
 279         skb_queue_splice_tail_init(queue, &process_queue);
 280         spin_unlock(&queue->lock);
 281
 282         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 283                 napi_gro_receive(napi, skb);
 284                 ++received;
 285         }
 286
 287         if (!skb_queue_empty(&process_queue)) {
 288                 spin_lock(&queue->lock);
 289                 skb_queue_splice(&process_queue, queue);
 290                 spin_unlock(&queue->lock);
 291         }
 292
 293         return received;
 294 }
 295
 296 static int tun_napi_poll(struct napi_struct *napi, int budget)
 297 {
 298         unsigned int received;
 299
 300         received = tun_napi_receive(napi, budget);
 301
 302         if (received < budget)
 303                 napi_complete_done(napi, received);
 304
 305         return received;
 306 }
 307
 308 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 309                           bool napi_en)
 310 {
 311         tfile->napi_enabled = napi_en;
 312         if (napi_en) {
 313                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 314                                NAPI_POLL_WEIGHT);
 315                 napi_enable(&tfile->napi);
 316                 mutex_init(&tfile->napi_mutex);
 317         }
 318 }
 319
 320 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 321 {
 322         if (tfile->napi_enabled)
 323                 napi_disable(&tfile->napi);
 324 }
 325
 326 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 327 {
 328         if (tfile->napi_enabled)
 329                 netif_napi_del(&tfile->napi);
 330 }
 331
 332 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 333 {
 334         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 335 }
 336
 337 #ifdef CONFIG_TUN_VNET_CROSS_LE
 338 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 339 {
 340         return tun->flags & TUN_VNET_BE ? false :
 341                 virtio_legacy_is_little_endian();
 342 }
 343
 344 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 345 {
 346         int be = !!(tun->flags & TUN_VNET_BE);
 347
 348         if (put_user(be, argp))
 349                 return -EFAULT;
 350
 351         return 0;
 352 }
 353
 354 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 355 {
 356         int be;
 357
 358         if (get_user(be, argp))
 359                 return -EFAULT;
 360
 361         if (be)
 362                 tun->flags |= TUN_VNET_BE;
 363         else
 364                 tun->flags &= ~TUN_VNET_BE;
 365
 366         return 0;
 367 }
 368 #else
 369 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 370 {
 371         return virtio_legacy_is_little_endian();
 372 }
 373
 374 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 375 {
 376         return -EINVAL;
 377 }
 378
 379 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 380 {
 381         return -EINVAL;
 382 }
 383 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 384
 385 static inline bool tun_is_little_endian(struct tun_struct *tun)
 386 {
 387         return tun->flags & TUN_VNET_LE ||
 388                 tun_legacy_is_little_endian(tun);
 389 }
 390
 391 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 392 {
 393         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 394 }
 395
 396 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 397 {
 398         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 399 }
 400
 401 static inline u32 tun_hashfn(u32 rxhash)
 402 {
 403         return rxhash & 0x3ff;
 404 }
 405
 406 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 407 {
 408         struct tun_flow_entry *e;
 409
 410         hlist_for_each_entry_rcu(e, head, hash_link) {
 411                 if (e->rxhash == rxhash)
 412                         return e;
 413         }
 414         return NULL;
 415 }
 416
 417 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 418                                               struct hlist_head *head,
 419                                               u32 rxhash, u16 queue_index)
 420 {
 421         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 422
 423         if (e) {
 424                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 425                           rxhash, queue_index);
 426                 e->updated = jiffies;
 427                 e->rxhash = rxhash;
 428                 e->rps_rxhash = 0;
 429                 e->queue_index = queue_index;
 430                 e->tun = tun;
 431                 hlist_add_head_rcu(&e->hash_link, head);
 432                 ++tun->flow_count;
 433         }
 434         return e;
 435 }
 436
 437 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 438 {
 439         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 440                   e->rxhash, e->queue_index);
 441         hlist_del_rcu(&e->hash_link);
 442         kfree_rcu(e, rcu);
 443         --tun->flow_count;
 444 }
 445
 446 static void tun_flow_flush(struct tun_struct *tun)
 447 {
 448         int i;
 449
 450         spin_lock_bh(&tun->lock);
 451         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 452                 struct tun_flow_entry *e;
 453                 struct hlist_node *n;
 454
 455                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 456                         tun_flow_delete(tun, e);
 457         }
 458         spin_unlock_bh(&tun->lock);
 459 }
 460
 461 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 462 {
 463         int i;
 464
 465         spin_lock_bh(&tun->lock);
 466         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 467                 struct tun_flow_entry *e;
 468                 struct hlist_node *n;
 469
 470                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 471                         if (e->queue_index == queue_index)
 472                                 tun_flow_delete(tun, e);
 473                 }
 474         }
 475         spin_unlock_bh(&tun->lock);
 476 }
 477
 478 static void tun_flow_cleanup(struct timer_list *t)
 479 {
 480         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 481         unsigned long delay = tun->ageing_time;
 482         unsigned long next_timer = jiffies + delay;
 483         unsigned long count = 0;
 484         int i;
 485
 486         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 487
 488         spin_lock(&tun->lock);
 489         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 490                 struct tun_flow_entry *e;
 491                 struct hlist_node *n;
 492
 493                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 494                         unsigned long this_timer;
 495
 496                         this_timer = e->updated + delay;
 497                         if (time_before_eq(this_timer, jiffies)) {
 498                                 tun_flow_delete(tun, e);
 499                                 continue;
 500                         }
 501                         count++;
 502                         if (time_before(this_timer, next_timer))
 503                                 next_timer = this_timer;
 504                 }
 505         }
 506
 507         if (count)
 508                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 509         spin_unlock(&tun->lock);
 510 }
 511
 512 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 513                             struct tun_file *tfile)
 514 {
 515         struct hlist_head *head;
 516         struct tun_flow_entry *e;
 517         unsigned long delay = tun->ageing_time;
 518         u16 queue_index = tfile->queue_index;
 519
 520         if (!rxhash)
 521                 return;
 522         else
 523                 head = &tun->flows[tun_hashfn(rxhash)];
 524
 525         rcu_read_lock();
 526
 527         /* We may get a very small possibility of OOO during switching, not
 528          * worth to optimize.*/
 529         if (tun->numqueues == 1 || tfile->detached)
 530                 goto unlock;
 531
 532         e = tun_flow_find(head, rxhash);
 533         if (likely(e)) {
 534                 /* TODO: keep queueing to old queue until it's empty? */
 535                 e->queue_index = queue_index;
 536                 e->updated = jiffies;
 537                 sock_rps_record_flow_hash(e->rps_rxhash);
 538         } else {
 539                 spin_lock_bh(&tun->lock);
 540                 if (!tun_flow_find(head, rxhash) &&
 541                     tun->flow_count < MAX_TAP_FLOWS)
 542                         tun_flow_create(tun, head, rxhash, queue_index);
 543
 544                 if (!timer_pending(&tun->flow_gc_timer))
 545                         mod_timer(&tun->flow_gc_timer,
 546                                   round_jiffies_up(jiffies + delay));
 547                 spin_unlock_bh(&tun->lock);
 548         }
 549
 550 unlock:
 551         rcu_read_unlock();
 552 }
 553
 554 /**
 555  * Save the hash received in the stack receive path and update the
 556  * flow_hash table accordingly.
 557  */
 558 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 559 {
 560         if (unlikely(e->rps_rxhash != hash))
 561                 e->rps_rxhash = hash;
 562 }
 563
 564 /* We try to identify a flow through its rxhash first. The reason that
 565  * we do not check rxq no. is because some cards(e.g 82599), chooses
 566  * the rxq based on the txq where the last packet of the flow comes. As
 567  * the userspace application move between processors, we may get a
 568  * different rxq no. here. If we could not get rxhash, then we would
 569  * hope the rxq no. may help here.
 570  */
 571 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 572 {
 573         struct tun_flow_entry *e;
 574         u32 txq = 0;
 575         u32 numqueues = 0;
 576
 577         numqueues = READ_ONCE(tun->numqueues);
 578
 579         txq = __skb_get_hash_symmetric(skb);
 580         if (txq) {
 581                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 582                 if (e) {
 583                         tun_flow_save_rps_rxhash(e, txq);
 584                         txq = e->queue_index;
 585                 } else
 586                         /* use multiply and shift instead of expensive divide */
 587                         txq = ((u64)txq * numqueues) >> 32;
 588         } else if (likely(skb_rx_queue_recorded(skb))) {
 589                 txq = skb_get_rx_queue(skb);
 590                 while (unlikely(txq >= numqueues))
 591                         txq -= numqueues;
 592         }
 593
 594         return txq;
 595 }
 596
 597 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 598 {
 599         struct tun_prog *prog;
 600         u16 ret = 0;
 601
 602         prog = rcu_dereference(tun->steering_prog);
 603         if (prog)
 604                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 605
 606         return ret % tun->numqueues;
 607 }
 608
 609 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 610                             void *accel_priv, select_queue_fallback_t fallback)
 611 {
 612         struct tun_struct *tun = netdev_priv(dev);
 613         u16 ret;
 614
 615         rcu_read_lock();
 616         if (rcu_dereference(tun->steering_prog))
 617                 ret = tun_ebpf_select_queue(tun, skb);
 618         else
 619                 ret = tun_automq_select_queue(tun, skb);
 620         rcu_read_unlock();
 621
 622         return ret;
 623 }
 624
 625 static inline bool tun_not_capable(struct tun_struct *tun)
 626 {
 627         const struct cred *cred = current_cred();
 628         struct net *net = dev_net(tun->dev);
 629
 630         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 631                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 632                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 633 }
 634
 635 static void tun_set_real_num_queues(struct tun_struct *tun)
 636 {
 637         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 638         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 639 }
 640
 641 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 642 {
 643         tfile->detached = tun;
 644         list_add_tail(&tfile->next, &tun->disabled);
 645         ++tun->numdisabled;
 646 }
 647
 648 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 649 {
 650         struct tun_struct *tun = tfile->detached;
 651
 652         tfile->detached = NULL;
 653         list_del_init(&tfile->next);
 654         --tun->numdisabled;
 655         return tun;
 656 }
 657
 658 void tun_ptr_free(void *ptr)
 659 {
 660         if (!ptr)
 661                 return;
 662         if (tun_is_xdp_buff(ptr)) {
 663                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
 664
 665                 put_page(virt_to_head_page(xdp->data));
 666         } else {
 667                 __skb_array_destroy_skb(ptr);
 668         }
 669 }
 670 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_cleanup_tx_ring(struct tun_file *tfile)
 684 {
 685         if (tfile->tx_ring.queue) {
 686                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 687                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 688                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 689         }
 690 }
 691
 692 static void __tun_detach(struct tun_file *tfile, bool clean)
 693 {
 694         struct tun_file *ntfile;
 695         struct tun_struct *tun;
 696
 697         tun = rtnl_dereference(tfile->tun);
 698
 699         if (tun && clean) {
 700                 tun_napi_disable(tun, tfile);
 701                 tun_napi_del(tun, tfile);
 702         }
 703
 704         if (tun && !tfile->detached) {
 705                 u16 index = tfile->queue_index;
 706                 BUG_ON(index >= tun->numqueues);
 707
 708                 rcu_assign_pointer(tun->tfiles[index],
 709                                    tun->tfiles[tun->numqueues - 1]);
 710                 ntfile = rtnl_dereference(tun->tfiles[index]);
 711                 ntfile->queue_index = index;
 712
 713                 --tun->numqueues;
 714                 if (clean) {
 715                         RCU_INIT_POINTER(tfile->tun, NULL);
 716                         sock_put(&tfile->sk);
 717                 } else
 718                         tun_disable_queue(tun, tfile);
 719
 720                 synchronize_net();
 721                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 722                 /* Drop read queue */
 723                 tun_queue_purge(tfile);
 724                 tun_set_real_num_queues(tun);
 725         } else if (tfile->detached && clean) {
 726                 tun = tun_enable_queue(tfile);
 727                 sock_put(&tfile->sk);
 728         }
 729
 730         if (clean) {
 731                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 732                         netif_carrier_off(tun->dev);
 733
 734                         if (!(tun->flags & IFF_PERSIST) &&
 735                             tun->dev->reg_state == NETREG_REGISTERED)
 736                                 unregister_netdevice(tun->dev);
 737                 }
 738                 tun_cleanup_tx_ring(tfile);
 739                 sock_put(&tfile->sk);
 740         }
 741 }
 742
 743 static void tun_detach(struct tun_file *tfile, bool clean)
 744 {
 745         rtnl_lock();
 746         __tun_detach(tfile, clean);
 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                 sock_put(&tfile->sk);
 779                 tun_cleanup_tx_ring(tfile);
 780         }
 781         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 782                 tun_enable_queue(tfile);
 783                 tun_queue_purge(tfile);
 784                 sock_put(&tfile->sk);
 785                 tun_cleanup_tx_ring(tfile);
 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_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 830                 err = -ENOMEM;
 831                 goto out;
 832         }
 833
 834         tfile->queue_index = tun->numqueues;
 835         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 836
 837         if (tfile->detached) {
 838                 /* Re-attach detached tfile, updating XDP queue_index */
 839                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 840
 841                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 842                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 843         } else {
 844                 /* Setup XDP RX-queue info, for new tfile getting attached */
 845                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 846                                        tun->dev, tfile->queue_index);
 847                 if (err < 0)
 848                         goto out;
 849                 err = 0;
 850         }
 851
 852         rcu_assign_pointer(tfile->tun, tun);
 853         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 854         tun->numqueues++;
 855
 856         if (tfile->detached) {
 857                 tun_enable_queue(tfile);
 858         } else {
 859                 sock_hold(&tfile->sk);
 860                 tun_napi_init(tun, tfile, napi);
 861         }
 862
 863         tun_set_real_num_queues(tun);
 864
 865         /* device is allowed to go away first, so no need to hold extra
 866          * refcnt.
 867          */
 868
 869 out:
 870         return err;
 871 }
 872
 873 static struct tun_struct *tun_get(struct tun_file *tfile)
 874 {
 875         struct tun_struct *tun;
 876
 877         rcu_read_lock();
 878         tun = rcu_dereference(tfile->tun);
 879         if (tun)
 880                 dev_hold(tun->dev);
 881         rcu_read_unlock();
 882
 883         return tun;
 884 }
 885
 886 static void tun_put(struct tun_struct *tun)
 887 {
 888         dev_put(tun->dev);
 889 }
 890
 891 /* TAP filtering */
 892 static void addr_hash_set(u32 *mask, const u8 *addr)
 893 {
 894         int n = ether_crc(ETH_ALEN, addr) >> 26;
 895         mask[n >> 5] |= (1 << (n & 31));
 896 }
 897
 898 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 899 {
 900         int n = ether_crc(ETH_ALEN, addr) >> 26;
 901         return mask[n >> 5] & (1 << (n & 31));
 902 }
 903
 904 static int update_filter(struct tap_filter *filter, void __user *arg)
 905 {
 906         struct { u8 u[ETH_ALEN]; } *addr;
 907         struct tun_filter uf;
 908         int err, alen, n, nexact;
 909
 910         if (copy_from_user(&uf, arg, sizeof(uf)))
 911                 return -EFAULT;
 912
 913         if (!uf.count) {
 914                 /* Disabled */
 915                 filter->count = 0;
 916                 return 0;
 917         }
 918
 919         alen = ETH_ALEN * uf.count;
 920         addr = memdup_user(arg + sizeof(uf), alen);
 921         if (IS_ERR(addr))
 922                 return PTR_ERR(addr);
 923
 924         /* The filter is updated without holding any locks. Which is
 925          * perfectly safe. We disable it first and in the worst
 926          * case we'll accept a few undesired packets. */
 927         filter->count = 0;
 928         wmb();
 929
 930         /* Use first set of addresses as an exact filter */
 931         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 932                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 933
 934         nexact = n;
 935
 936         /* Remaining multicast addresses are hashed,
 937          * unicast will leave the filter disabled. */
 938         memset(filter->mask, 0, sizeof(filter->mask));
 939         for (; n < uf.count; n++) {
 940                 if (!is_multicast_ether_addr(addr[n].u)) {
 941                         err = 0; /* no filter */
 942                         goto free_addr;
 943                 }
 944                 addr_hash_set(filter->mask, addr[n].u);
 945         }
 946
 947         /* For ALLMULTI just set the mask to all ones.
 948          * This overrides the mask populated above. */
 949         if ((uf.flags & TUN_FLT_ALLMULTI))
 950                 memset(filter->mask, ~0, sizeof(filter->mask));
 951
 952         /* Now enable the filter */
 953         wmb();
 954         filter->count = nexact;
 955
 956         /* Return the number of exact filters */
 957         err = nexact;
 958 free_addr:
 959         kfree(addr);
 960         return err;
 961 }
 962
 963 /* Returns: 0 - drop, !=0 - accept */
 964 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 965 {
 966         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 967          * at this point. */
 968         struct ethhdr *eh = (struct ethhdr *) skb->data;
 969         int i;
 970
 971         /* Exact match */
 972         for (i = 0; i < filter->count; i++)
 973                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 974                         return 1;
 975
 976         /* Inexact match (multicast only) */
 977         if (is_multicast_ether_addr(eh->h_dest))
 978                 return addr_hash_test(filter->mask, eh->h_dest);
 979
 980         return 0;
 981 }
 982
 983 /*
 984  * Checks whether the packet is accepted or not.
 985  * Returns: 0 - drop, !=0 - accept
 986  */
 987 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 988 {
 989         if (!filter->count)
 990                 return 1;
 991
 992         return run_filter(filter, skb);
 993 }
 994
 995 /* Network device part of the driver */
 996
 997 static const struct ethtool_ops tun_ethtool_ops;
 998
 999 /* Net device detach from fd. */
1000 static void tun_net_uninit(struct net_device *dev)
1001 {
1002         tun_detach_all(dev);
1003 }
1004
1005 /* Net device open. */
1006 static int tun_net_open(struct net_device *dev)
1007 {
1008         struct tun_struct *tun = netdev_priv(dev);
1009         int i;
1010
1011         netif_tx_start_all_queues(dev);
1012
1013         for (i = 0; i < tun->numqueues; i++) {
1014                 struct tun_file *tfile;
1015
1016                 tfile = rtnl_dereference(tun->tfiles[i]);
1017                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1018         }
1019
1020         return 0;
1021 }
1022
1023 /* Net device close. */
1024 static int tun_net_close(struct net_device *dev)
1025 {
1026         netif_tx_stop_all_queues(dev);
1027         return 0;
1028 }
1029
1030 /* Net device start xmit */
1031 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1032 {
1033 #ifdef CONFIG_RPS
1034         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1035                 /* Select queue was not called for the skbuff, so we extract the
1036                  * RPS hash and save it into the flow_table here.
1037                  */
1038                 __u32 rxhash;
1039
1040                 rxhash = __skb_get_hash_symmetric(skb);
1041                 if (rxhash) {
1042                         struct tun_flow_entry *e;
1043                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1044                                         rxhash);
1045                         if (e)
1046                                 tun_flow_save_rps_rxhash(e, rxhash);
1047                 }
1048         }
1049 #endif
1050 }
1051
1052 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1053                                     struct sk_buff *skb,
1054                                     int len)
1055 {
1056         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1057
1058         if (prog)
1059                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1060
1061         return len;
1062 }
1063
1064 /* Net device start xmit */
1065 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1066 {
1067         struct tun_struct *tun = netdev_priv(dev);
1068         int txq = skb->queue_mapping;
1069         struct tun_file *tfile;
1070         int len = skb->len;
1071
1072         rcu_read_lock();
1073         tfile = rcu_dereference(tun->tfiles[txq]);
1074
1075         /* Drop packet if interface is not attached */
1076         if (txq >= tun->numqueues)
1077                 goto drop;
1078
1079         if (!rcu_dereference(tun->steering_prog))
1080                 tun_automq_xmit(tun, skb);
1081
1082         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1083
1084         BUG_ON(!tfile);
1085
1086         /* Drop if the filter does not like it.
1087          * This is a noop if the filter is disabled.
1088          * Filter can be enabled only for the TAP devices. */
1089         if (!check_filter(&tun->txflt, skb))
1090                 goto drop;
1091
1092         if (tfile->socket.sk->sk_filter &&
1093             sk_filter(tfile->socket.sk, skb))
1094                 goto drop;
1095
1096         len = run_ebpf_filter(tun, skb, len);
1097
1098         /* Trim extra bytes since we may insert vlan proto & TCI
1099          * in tun_put_user().
1100          */
1101         len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1102         if (len <= 0 || pskb_trim(skb, len))
1103                 goto drop;
1104
1105         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1106                 goto drop;
1107
1108         skb_tx_timestamp(skb);
1109
1110         /* Orphan the skb - required as we might hang on to it
1111          * for indefinite time.
1112          */
1113         skb_orphan(skb);
1114
1115         nf_reset(skb);
1116
1117         if (ptr_ring_produce(&tfile->tx_ring, skb))
1118                 goto drop;
1119
1120         /* Notify and wake up reader process */
1121         if (tfile->flags & TUN_FASYNC)
1122                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1123         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1124
1125         rcu_read_unlock();
1126         return NETDEV_TX_OK;
1127
1128 drop:
1129         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1130         skb_tx_error(skb);
1131         kfree_skb(skb);
1132         rcu_read_unlock();
1133         return NET_XMIT_DROP;
1134 }
1135
1136 static void tun_net_mclist(struct net_device *dev)
1137 {
1138         /*
1139          * This callback is supposed to deal with mc filter in
1140          * _rx_ path and has nothing to do with the _tx_ path.
1141          * In rx path we always accept everything userspace gives us.
1142          */
1143 }
1144
1145 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1146         netdev_features_t features)
1147 {
1148         struct tun_struct *tun = netdev_priv(dev);
1149
1150         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1151 }
1152 #ifdef CONFIG_NET_POLL_CONTROLLER
1153 static void tun_poll_controller(struct net_device *dev)
1154 {
1155         /*
1156          * Tun only receives frames when:
1157          * 1) the char device endpoint gets data from user space
1158          * 2) the tun socket gets a sendmsg call from user space
1159          * If NAPI is not enabled, since both of those are synchronous
1160          * operations, we are guaranteed never to have pending data when we poll
1161          * for it so there is nothing to do here but return.
1162          * We need this though so netpoll recognizes us as an interface that
1163          * supports polling, which enables bridge devices in virt setups to
1164          * still use netconsole
1165          * If NAPI is enabled, however, we need to schedule polling for all
1166          * queues unless we are using napi_gro_frags(), which we call in
1167          * process context and not in NAPI context.
1168          */
1169         struct tun_struct *tun = netdev_priv(dev);
1170
1171         if (tun->flags & IFF_NAPI) {
1172                 struct tun_file *tfile;
1173                 int i;
1174
1175                 if (tun_napi_frags_enabled(tun))
1176                         return;
1177
1178                 rcu_read_lock();
1179                 for (i = 0; i < tun->numqueues; i++) {
1180                         tfile = rcu_dereference(tun->tfiles[i]);
1181                         if (tfile->napi_enabled)
1182                                 napi_schedule(&tfile->napi);
1183                 }
1184                 rcu_read_unlock();
1185         }
1186         return;
1187 }
1188 #endif
1189
1190 static void tun_set_headroom(struct net_device *dev, int new_hr)
1191 {
1192         struct tun_struct *tun = netdev_priv(dev);
1193
1194         if (new_hr < NET_SKB_PAD)
1195                 new_hr = NET_SKB_PAD;
1196
1197         tun->align = new_hr;
1198 }
1199
1200 static void
1201 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1202 {
1203         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1204         struct tun_struct *tun = netdev_priv(dev);
1205         struct tun_pcpu_stats *p;
1206         int i;
1207
1208         for_each_possible_cpu(i) {
1209                 u64 rxpackets, rxbytes, txpackets, txbytes;
1210                 unsigned int start;
1211
1212                 p = per_cpu_ptr(tun->pcpu_stats, i);
1213                 do {
1214                         start = u64_stats_fetch_begin(&p->syncp);
1215                         rxpackets       = p->rx_packets;
1216                         rxbytes         = p->rx_bytes;
1217                         txpackets       = p->tx_packets;
1218                         txbytes         = p->tx_bytes;
1219                 } while (u64_stats_fetch_retry(&p->syncp, start));
1220
1221                 stats->rx_packets       += rxpackets;
1222                 stats->rx_bytes         += rxbytes;
1223                 stats->tx_packets       += txpackets;
1224                 stats->tx_bytes         += txbytes;
1225
1226                 /* u32 counters */
1227                 rx_dropped      += p->rx_dropped;
1228                 rx_frame_errors += p->rx_frame_errors;
1229                 tx_dropped      += p->tx_dropped;
1230         }
1231         stats->rx_dropped  = rx_dropped;
1232         stats->rx_frame_errors = rx_frame_errors;
1233         stats->tx_dropped = tx_dropped;
1234 }
1235
1236 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1237                        struct netlink_ext_ack *extack)
1238 {
1239         struct tun_struct *tun = netdev_priv(dev);
1240         struct bpf_prog *old_prog;
1241
1242         old_prog = rtnl_dereference(tun->xdp_prog);
1243         rcu_assign_pointer(tun->xdp_prog, prog);
1244         if (old_prog)
1245                 bpf_prog_put(old_prog);
1246
1247         return 0;
1248 }
1249
1250 static u32 tun_xdp_query(struct net_device *dev)
1251 {
1252         struct tun_struct *tun = netdev_priv(dev);
1253         const struct bpf_prog *xdp_prog;
1254
1255         xdp_prog = rtnl_dereference(tun->xdp_prog);
1256         if (xdp_prog)
1257                 return xdp_prog->aux->id;
1258
1259         return 0;
1260 }
1261
1262 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1263 {
1264         switch (xdp->command) {
1265         case XDP_SETUP_PROG:
1266                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1267         case XDP_QUERY_PROG:
1268                 xdp->prog_id = tun_xdp_query(dev);
1269                 xdp->prog_attached = !!xdp->prog_id;
1270                 return 0;
1271         default:
1272                 return -EINVAL;
1273         }
1274 }
1275
1276 static const struct net_device_ops tun_netdev_ops = {
1277         .ndo_uninit             = tun_net_uninit,
1278         .ndo_open               = tun_net_open,
1279         .ndo_stop               = tun_net_close,
1280         .ndo_start_xmit         = tun_net_xmit,
1281         .ndo_fix_features       = tun_net_fix_features,
1282         .ndo_select_queue       = tun_select_queue,
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284         .ndo_poll_controller    = tun_poll_controller,
1285 #endif
1286         .ndo_set_rx_headroom    = tun_set_headroom,
1287         .ndo_get_stats64        = tun_net_get_stats64,
1288 };
1289
1290 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1291 {
1292         struct tun_struct *tun = netdev_priv(dev);
1293         struct xdp_buff *buff = xdp->data_hard_start;
1294         int headroom = xdp->data - xdp->data_hard_start;
1295         struct tun_file *tfile;
1296         u32 numqueues;
1297         int ret = 0;
1298
1299         /* Assure headroom is available and buff is properly aligned */
1300         if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1301                 return -ENOSPC;
1302
1303         *buff = *xdp;
1304
1305         rcu_read_lock();
1306
1307         numqueues = READ_ONCE(tun->numqueues);
1308         if (!numqueues) {
1309                 ret = -ENOSPC;
1310                 goto out;
1311         }
1312
1313         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1314                                             numqueues]);
1315         /* Encode the XDP flag into lowest bit for consumer to differ
1316          * XDP buffer from sk_buff.
1317          */
1318         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1319                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1320                 ret = -ENOSPC;
1321         }
1322
1323 out:
1324         rcu_read_unlock();
1325         return ret;
1326 }
1327
1328 static void tun_xdp_flush(struct net_device *dev)
1329 {
1330         struct tun_struct *tun = netdev_priv(dev);
1331         struct tun_file *tfile;
1332         u32 numqueues;
1333
1334         rcu_read_lock();
1335
1336         numqueues = READ_ONCE(tun->numqueues);
1337         if (!numqueues)
1338                 goto out;
1339
1340         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1341                                             numqueues]);
1342         /* Notify and wake up reader process */
1343         if (tfile->flags & TUN_FASYNC)
1344                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1345         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1346
1347 out:
1348         rcu_read_unlock();
1349 }
1350
1351 static const struct net_device_ops tap_netdev_ops = {
1352         .ndo_uninit             = tun_net_uninit,
1353         .ndo_open               = tun_net_open,
1354         .ndo_stop               = tun_net_close,
1355         .ndo_start_xmit         = tun_net_xmit,
1356         .ndo_fix_features       = tun_net_fix_features,
1357         .ndo_set_rx_mode        = tun_net_mclist,
1358         .ndo_set_mac_address    = eth_mac_addr,
1359         .ndo_validate_addr      = eth_validate_addr,
1360         .ndo_select_queue       = tun_select_queue,
1361 #ifdef CONFIG_NET_POLL_CONTROLLER
1362         .ndo_poll_controller    = tun_poll_controller,
1363 #endif
1364         .ndo_features_check     = passthru_features_check,
1365         .ndo_set_rx_headroom    = tun_set_headroom,
1366         .ndo_get_stats64        = tun_net_get_stats64,
1367         .ndo_bpf                = tun_xdp,
1368         .ndo_xdp_xmit           = tun_xdp_xmit,
1369         .ndo_xdp_flush          = tun_xdp_flush,
1370 };
1371
1372 static void tun_flow_init(struct tun_struct *tun)
1373 {
1374         int i;
1375
1376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1377                 INIT_HLIST_HEAD(&tun->flows[i]);
1378
1379         tun->ageing_time = TUN_FLOW_EXPIRE;
1380         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1381         mod_timer(&tun->flow_gc_timer,
1382                   round_jiffies_up(jiffies + tun->ageing_time));
1383 }
1384
1385 static void tun_flow_uninit(struct tun_struct *tun)
1386 {
1387         del_timer_sync(&tun->flow_gc_timer);
1388         tun_flow_flush(tun);
1389 }
1390
1391 #define MIN_MTU 68
1392 #define MAX_MTU 65535
1393
1394 /* Initialize net device. */
1395 static void tun_net_init(struct net_device *dev)
1396 {
1397         struct tun_struct *tun = netdev_priv(dev);
1398
1399         switch (tun->flags & TUN_TYPE_MASK) {
1400         case IFF_TUN:
1401                 dev->netdev_ops = &tun_netdev_ops;
1402
1403                 /* Point-to-Point TUN Device */
1404                 dev->hard_header_len = 0;
1405                 dev->addr_len = 0;
1406                 dev->mtu = 1500;
1407
1408                 /* Zero header length */
1409                 dev->type = ARPHRD_NONE;
1410                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1411                 break;
1412
1413         case IFF_TAP:
1414                 dev->netdev_ops = &tap_netdev_ops;
1415                 /* Ethernet TAP Device */
1416                 ether_setup(dev);
1417                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1418                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1419
1420                 eth_hw_addr_random(dev);
1421
1422                 break;
1423         }
1424
1425         dev->min_mtu = MIN_MTU;
1426         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1427 }
1428
1429 /* Character device part */
1430
1431 /* Poll */
1432 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1433 {
1434         struct tun_file *tfile = file->private_data;
1435         struct tun_struct *tun = tun_get(tfile);
1436         struct sock *sk;
1437         __poll_t mask = 0;
1438
1439         if (!tun)
1440                 return EPOLLERR;
1441
1442         sk = tfile->socket.sk;
1443
1444         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1445
1446         poll_wait(file, sk_sleep(sk), wait);
1447
1448         if (!ptr_ring_empty(&tfile->tx_ring))
1449                 mask |= EPOLLIN | EPOLLRDNORM;
1450
1451         if (tun->dev->flags & IFF_UP &&
1452             (sock_writeable(sk) ||
1453              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1454               sock_writeable(sk))))
1455                 mask |= EPOLLOUT | EPOLLWRNORM;
1456
1457         if (tun->dev->reg_state != NETREG_REGISTERED)
1458                 mask = EPOLLERR;
1459
1460         tun_put(tun);
1461         return mask;
1462 }
1463
1464 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1465                                             size_t len,
1466                                             const struct iov_iter *it)
1467 {
1468         struct sk_buff *skb;
1469         size_t linear;
1470         int err;
1471         int i;
1472
1473         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1474                 return ERR_PTR(-ENOMEM);
1475
1476         local_bh_disable();
1477         skb = napi_get_frags(&tfile->napi);
1478         local_bh_enable();
1479         if (!skb)
1480                 return ERR_PTR(-ENOMEM);
1481
1482         linear = iov_iter_single_seg_count(it);
1483         err = __skb_grow(skb, linear);
1484         if (err)
1485                 goto free;
1486
1487         skb->len = len;
1488         skb->data_len = len - linear;
1489         skb->truesize += skb->data_len;
1490
1491         for (i = 1; i < it->nr_segs; i++) {
1492                 struct page_frag *pfrag = &current->task_frag;
1493                 size_t fragsz = it->iov[i].iov_len;
1494
1495                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1496                         err = -EINVAL;
1497                         goto free;
1498                 }
1499
1500                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1501                         err = -ENOMEM;
1502                         goto free;
1503                 }
1504
1505                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1506                                    pfrag->offset, fragsz);
1507                 page_ref_inc(pfrag->page);
1508                 pfrag->offset += fragsz;
1509         }
1510
1511         return skb;
1512 free:
1513         /* frees skb and all frags allocated with napi_alloc_frag() */
1514         napi_free_frags(&tfile->napi);
1515         return ERR_PTR(err);
1516 }
1517
1518 /* prepad is the amount to reserve at front.  len is length after that.
1519  * linear is a hint as to how much to copy (usually headers). */
1520 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1521                                      size_t prepad, size_t len,
1522                                      size_t linear, int noblock)
1523 {
1524         struct sock *sk = tfile->socket.sk;
1525         struct sk_buff *skb;
1526         int err;
1527
1528         /* Under a page?  Don't bother with paged skb. */
1529         if (prepad + len < PAGE_SIZE || !linear)
1530                 linear = len;
1531
1532         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1533                                    &err, 0);
1534         if (!skb)
1535                 return ERR_PTR(err);
1536
1537         skb_reserve(skb, prepad);
1538         skb_put(skb, linear);
1539         skb->data_len = len - linear;
1540         skb->len += len - linear;
1541
1542         return skb;
1543 }
1544
1545 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1546                            struct sk_buff *skb, int more)
1547 {
1548         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1549         struct sk_buff_head process_queue;
1550         u32 rx_batched = tun->rx_batched;
1551         bool rcv = false;
1552
1553         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1554                 local_bh_disable();
1555                 netif_receive_skb(skb);
1556                 local_bh_enable();
1557                 return;
1558         }
1559
1560         spin_lock(&queue->lock);
1561         if (!more || skb_queue_len(queue) == rx_batched) {
1562                 __skb_queue_head_init(&process_queue);
1563                 skb_queue_splice_tail_init(queue, &process_queue);
1564                 rcv = true;
1565         } else {
1566                 __skb_queue_tail(queue, skb);
1567         }
1568         spin_unlock(&queue->lock);
1569
1570         if (rcv) {
1571                 struct sk_buff *nskb;
1572
1573                 local_bh_disable();
1574                 while ((nskb = __skb_dequeue(&process_queue)))
1575                         netif_receive_skb(nskb);
1576                 netif_receive_skb(skb);
1577                 local_bh_enable();
1578         }
1579 }
1580
1581 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1582                               int len, int noblock, bool zerocopy)
1583 {
1584         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1585                 return false;
1586
1587         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1588                 return false;
1589
1590         if (!noblock)
1591                 return false;
1592
1593         if (zerocopy)
1594                 return false;
1595
1596         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1597             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1598                 return false;
1599
1600         return true;
1601 }
1602
1603 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1604                                      struct tun_file *tfile,
1605                                      struct iov_iter *from,
1606                                      struct virtio_net_hdr *hdr,
1607                                      int len, int *skb_xdp)
1608 {
1609         struct page_frag *alloc_frag = &current->task_frag;
1610         struct sk_buff *skb;
1611         struct bpf_prog *xdp_prog;
1612         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1613         unsigned int delta = 0;
1614         char *buf;
1615         size_t copied;
1616         bool xdp_xmit = false;
1617         int err, pad = TUN_RX_PAD;
1618
1619         rcu_read_lock();
1620         xdp_prog = rcu_dereference(tun->xdp_prog);
1621         if (xdp_prog)
1622                 pad += TUN_HEADROOM;
1623         buflen += SKB_DATA_ALIGN(len + pad);
1624         rcu_read_unlock();
1625
1626         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628                 return ERR_PTR(-ENOMEM);
1629
1630         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631         copied = copy_page_from_iter(alloc_frag->page,
1632                                      alloc_frag->offset + pad,
1633                                      len, from);
1634         if (copied != len)
1635                 return ERR_PTR(-EFAULT);
1636
1637         /* There's a small window that XDP may be set after the check
1638          * of xdp_prog above, this should be rare and for simplicity
1639          * we do XDP on skb in case the headroom is not enough.
1640          */
1641         if (hdr->gso_type || !xdp_prog)
1642                 *skb_xdp = 1;
1643         else
1644                 *skb_xdp = 0;
1645
1646         preempt_disable();
1647         rcu_read_lock();
1648         xdp_prog = rcu_dereference(tun->xdp_prog);
1649         if (xdp_prog && !*skb_xdp) {
1650                 struct xdp_buff xdp;
1651                 void *orig_data;
1652                 u32 act;
1653
1654                 xdp.data_hard_start = buf;
1655                 xdp.data = buf + pad;
1656                 xdp_set_data_meta_invalid(&xdp);
1657                 xdp.data_end = xdp.data + len;
1658                 xdp.rxq = &tfile->xdp_rxq;
1659                 orig_data = xdp.data;
1660                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1661
1662                 switch (act) {
1663                 case XDP_REDIRECT:
1664                         get_page(alloc_frag->page);
1665                         alloc_frag->offset += buflen;
1666                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1667                         xdp_do_flush_map();
1668                         if (err)
1669                                 goto err_redirect;
1670                         rcu_read_unlock();
1671                         preempt_enable();
1672                         return NULL;
1673                 case XDP_TX:
1674                         xdp_xmit = true;
1675                         /* fall through */
1676                 case XDP_PASS:
1677                         delta = orig_data - xdp.data;
1678                         break;
1679                 default:
1680                         bpf_warn_invalid_xdp_action(act);
1681                         /* fall through */
1682                 case XDP_ABORTED:
1683                         trace_xdp_exception(tun->dev, xdp_prog, act);
1684                         /* fall through */
1685                 case XDP_DROP:
1686                         goto err_xdp;
1687                 }
1688         }
1689
1690         skb = build_skb(buf, buflen);
1691         if (!skb) {
1692                 rcu_read_unlock();
1693                 preempt_enable();
1694                 return ERR_PTR(-ENOMEM);
1695         }
1696
1697         skb_reserve(skb, pad - delta);
1698         skb_put(skb, len + delta);
1699         get_page(alloc_frag->page);
1700         alloc_frag->offset += buflen;
1701
1702         if (xdp_xmit) {
1703                 skb->dev = tun->dev;
1704                 generic_xdp_tx(skb, xdp_prog);
1705                 rcu_read_unlock();
1706                 preempt_enable();
1707                 return NULL;
1708         }
1709
1710         rcu_read_unlock();
1711         preempt_enable();
1712
1713         return skb;
1714
1715 err_redirect:
1716         put_page(alloc_frag->page);
1717 err_xdp:
1718         rcu_read_unlock();
1719         preempt_enable();
1720         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1721         return NULL;
1722 }
1723
1724 /* Get packet from user space buffer */
1725 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1726                             void *msg_control, struct iov_iter *from,
1727                             int noblock, bool more)
1728 {
1729         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1730         struct sk_buff *skb;
1731         size_t total_len = iov_iter_count(from);
1732         size_t len = total_len, align = tun->align, linear;
1733         struct virtio_net_hdr gso = { 0 };
1734         struct tun_pcpu_stats *stats;
1735         int good_linear;
1736         int copylen;
1737         bool zerocopy = false;
1738         int err;
1739         u32 rxhash = 0;
1740         int skb_xdp = 1;
1741         bool frags = tun_napi_frags_enabled(tun);
1742
1743         if (!(tun->dev->flags & IFF_UP))
1744                 return -EIO;
1745
1746         if (!(tun->flags & IFF_NO_PI)) {
1747                 if (len < sizeof(pi))
1748                         return -EINVAL;
1749                 len -= sizeof(pi);
1750
1751                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1752                         return -EFAULT;
1753         }
1754
1755         if (tun->flags & IFF_VNET_HDR) {
1756                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1757
1758                 if (len < vnet_hdr_sz)
1759                         return -EINVAL;
1760                 len -= vnet_hdr_sz;
1761
1762                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1763                         return -EFAULT;
1764
1765                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1766                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1767                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1768
1769                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1770                         return -EINVAL;
1771                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1772         }
1773
1774         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1775                 align += NET_IP_ALIGN;
1776                 if (unlikely(len < ETH_HLEN ||
1777                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1778                         return -EINVAL;
1779         }
1780
1781         good_linear = SKB_MAX_HEAD(align);
1782
1783         if (msg_control) {
1784                 struct iov_iter i = *from;
1785
1786                 /* There are 256 bytes to be copied in skb, so there is
1787                  * enough room for skb expand head in case it is used.
1788                  * The rest of the buffer is mapped from userspace.
1789                  */
1790                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1791                 if (copylen > good_linear)
1792                         copylen = good_linear;
1793                 linear = copylen;
1794                 iov_iter_advance(&i, copylen);
1795                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1796                         zerocopy = true;
1797         }
1798
1799         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1800                 /* For the packet that is not easy to be processed
1801                  * (e.g gso or jumbo packet), we will do it at after
1802                  * skb was created with generic XDP routine.
1803                  */
1804                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1805                 if (IS_ERR(skb)) {
1806                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1807                         return PTR_ERR(skb);
1808                 }
1809                 if (!skb)
1810                         return total_len;
1811         } else {
1812                 if (!zerocopy) {
1813                         copylen = len;
1814                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1815                                 linear = good_linear;
1816                         else
1817                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1818                 }
1819
1820                 if (frags) {
1821                         mutex_lock(&tfile->napi_mutex);
1822                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1823                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1824                          * If zerocopy is enabled, then this layout will be
1825                          * overwritten by zerocopy_sg_from_iter().
1826                          */
1827                         zerocopy = false;
1828                 } else {
1829                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1830                                             noblock);
1831                 }
1832
1833                 if (IS_ERR(skb)) {
1834                         if (PTR_ERR(skb) != -EAGAIN)
1835                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1836                         if (frags)
1837                                 mutex_unlock(&tfile->napi_mutex);
1838                         return PTR_ERR(skb);
1839                 }
1840
1841                 if (zerocopy)
1842                         err = zerocopy_sg_from_iter(skb, from);
1843                 else
1844                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1845
1846                 if (err) {
1847                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1848                         kfree_skb(skb);
1849                         if (frags) {
1850                                 tfile->napi.skb = NULL;
1851                                 mutex_unlock(&tfile->napi_mutex);
1852                         }
1853
1854                         return -EFAULT;
1855                 }
1856         }
1857
1858         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1859                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1860                 kfree_skb(skb);
1861                 if (frags) {
1862                         tfile->napi.skb = NULL;
1863                         mutex_unlock(&tfile->napi_mutex);
1864                 }
1865
1866                 return -EINVAL;
1867         }
1868
1869         switch (tun->flags & TUN_TYPE_MASK) {
1870         case IFF_TUN:
1871                 if (tun->flags & IFF_NO_PI) {
1872                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1873
1874                         switch (ip_version) {
1875                         case 4:
1876                                 pi.proto = htons(ETH_P_IP);
1877                                 break;
1878                         case 6:
1879                                 pi.proto = htons(ETH_P_IPV6);
1880                                 break;
1881                         default:
1882                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1883                                 kfree_skb(skb);
1884                                 return -EINVAL;
1885                         }
1886                 }
1887
1888                 skb_reset_mac_header(skb);
1889                 skb->protocol = pi.proto;
1890                 skb->dev = tun->dev;
1891                 break;
1892         case IFF_TAP:
1893                 if (!frags)
1894                         skb->protocol = eth_type_trans(skb, tun->dev);
1895                 break;
1896         }
1897
1898         /* copy skb_ubuf_info for callback when skb has no error */
1899         if (zerocopy) {
1900                 skb_shinfo(skb)->destructor_arg = msg_control;
1901                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1902                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1903         } else if (msg_control) {
1904                 struct ubuf_info *uarg = msg_control;
1905                 uarg->callback(uarg, false);
1906         }
1907
1908         skb_reset_network_header(skb);
1909         skb_probe_transport_header(skb, 0);
1910
1911         if (skb_xdp) {
1912                 struct bpf_prog *xdp_prog;
1913                 int ret;
1914
1915                 rcu_read_lock();
1916                 xdp_prog = rcu_dereference(tun->xdp_prog);
1917                 if (xdp_prog) {
1918                         ret = do_xdp_generic(xdp_prog, skb);
1919                         if (ret != XDP_PASS) {
1920                                 rcu_read_unlock();
1921                                 return total_len;
1922                         }
1923                 }
1924                 rcu_read_unlock();
1925         }
1926
1927         rcu_read_lock();
1928         if (!rcu_dereference(tun->steering_prog))
1929                 rxhash = __skb_get_hash_symmetric(skb);
1930         rcu_read_unlock();
1931
1932         if (frags) {
1933                 /* Exercise flow dissector code path. */
1934                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1935
1936                 if (unlikely(headlen > skb_headlen(skb))) {
1937                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1938                         napi_free_frags(&tfile->napi);
1939                         mutex_unlock(&tfile->napi_mutex);
1940                         WARN_ON(1);
1941                         return -ENOMEM;
1942                 }
1943
1944                 local_bh_disable();
1945                 napi_gro_frags(&tfile->napi);
1946                 local_bh_enable();
1947                 mutex_unlock(&tfile->napi_mutex);
1948         } else if (tfile->napi_enabled) {
1949                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1950                 int queue_len;
1951
1952                 spin_lock_bh(&queue->lock);
1953                 __skb_queue_tail(queue, skb);
1954                 queue_len = skb_queue_len(queue);
1955                 spin_unlock(&queue->lock);
1956
1957                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1958                         napi_schedule(&tfile->napi);
1959
1960                 local_bh_enable();
1961         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1962                 tun_rx_batched(tun, tfile, skb, more);
1963         } else {
1964                 netif_rx_ni(skb);
1965         }
1966
1967         stats = get_cpu_ptr(tun->pcpu_stats);
1968         u64_stats_update_begin(&stats->syncp);
1969         stats->rx_packets++;
1970         stats->rx_bytes += len;
1971         u64_stats_update_end(&stats->syncp);
1972         put_cpu_ptr(stats);
1973
1974         if (rxhash)
1975                 tun_flow_update(tun, rxhash, tfile);
1976
1977         return total_len;
1978 }
1979
1980 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1981 {
1982         struct file *file = iocb->ki_filp;
1983         struct tun_file *tfile = file->private_data;
1984         struct tun_struct *tun = tun_get(tfile);
1985         ssize_t result;
1986
1987         if (!tun)
1988                 return -EBADFD;
1989
1990         result = tun_get_user(tun, tfile, NULL, from,
1991                               file->f_flags & O_NONBLOCK, false);
1992
1993         tun_put(tun);
1994         return result;
1995 }
1996
1997 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1998                                 struct tun_file *tfile,
1999                                 struct xdp_buff *xdp,
2000                                 struct iov_iter *iter)
2001 {
2002         int vnet_hdr_sz = 0;
2003         size_t size = xdp->data_end - xdp->data;
2004         struct tun_pcpu_stats *stats;
2005         size_t ret;
2006
2007         if (tun->flags & IFF_VNET_HDR) {
2008                 struct virtio_net_hdr gso = { 0 };
2009
2010                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2011                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2012                         return -EINVAL;
2013                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2014                              sizeof(gso)))
2015                         return -EFAULT;
2016                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2017         }
2018
2019         ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2020
2021         stats = get_cpu_ptr(tun->pcpu_stats);
2022         u64_stats_update_begin(&stats->syncp);
2023         stats->tx_packets++;
2024         stats->tx_bytes += ret;
2025         u64_stats_update_end(&stats->syncp);
2026         put_cpu_ptr(tun->pcpu_stats);
2027
2028         return ret;
2029 }
2030
2031 /* Put packet to the user space buffer */
2032 static ssize_t tun_put_user(struct tun_struct *tun,
2033                             struct tun_file *tfile,
2034                             struct sk_buff *skb,
2035                             struct iov_iter *iter)
2036 {
2037         struct tun_pi pi = { 0, skb->protocol };
2038         struct tun_pcpu_stats *stats;
2039         ssize_t total;
2040         int vlan_offset = 0;
2041         int vlan_hlen = 0;
2042         int vnet_hdr_sz = 0;
2043
2044         if (skb_vlan_tag_present(skb))
2045                 vlan_hlen = VLAN_HLEN;
2046
2047         if (tun->flags & IFF_VNET_HDR)
2048                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2049
2050         total = skb->len + vlan_hlen + vnet_hdr_sz;
2051
2052         if (!(tun->flags & IFF_NO_PI)) {
2053                 if (iov_iter_count(iter) < sizeof(pi))
2054                         return -EINVAL;
2055
2056                 total += sizeof(pi);
2057                 if (iov_iter_count(iter) < total) {
2058                         /* Packet will be striped */
2059                         pi.flags |= TUN_PKT_STRIP;
2060                 }
2061
2062                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2063                         return -EFAULT;
2064         }
2065
2066         if (vnet_hdr_sz) {
2067                 struct virtio_net_hdr gso;
2068
2069                 if (iov_iter_count(iter) < vnet_hdr_sz)
2070                         return -EINVAL;
2071
2072                 if (virtio_net_hdr_from_skb(skb, &gso,
2073                                             tun_is_little_endian(tun), true)) {
2074                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2075                         pr_err("unexpected GSO type: "
2076                                "0x%x, gso_size %d, hdr_len %d\n",
2077                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2078                                tun16_to_cpu(tun, gso.hdr_len));
2079                         print_hex_dump(KERN_ERR, "tun: ",
2080                                        DUMP_PREFIX_NONE,
2081                                        16, 1, skb->head,
2082                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2083                         WARN_ON_ONCE(1);
2084                         return -EINVAL;
2085                 }
2086
2087                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2088                         return -EFAULT;
2089
2090                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2091         }
2092
2093         if (vlan_hlen) {
2094                 int ret;
2095                 struct veth veth;
2096
2097                 veth.h_vlan_proto = skb->vlan_proto;
2098                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2099
2100                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2101
2102                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2103                 if (ret || !iov_iter_count(iter))
2104                         goto done;
2105
2106                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2107                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2108                         goto done;
2109         }
2110
2111         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2112
2113 done:
2114         /* caller is in process context, */
2115         stats = get_cpu_ptr(tun->pcpu_stats);
2116         u64_stats_update_begin(&stats->syncp);
2117         stats->tx_packets++;
2118         stats->tx_bytes += skb->len + vlan_hlen;
2119         u64_stats_update_end(&stats->syncp);
2120         put_cpu_ptr(tun->pcpu_stats);
2121
2122         return total;
2123 }
2124
2125 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2126 {
2127         DECLARE_WAITQUEUE(wait, current);
2128         void *ptr = NULL;
2129         int error = 0;
2130
2131         ptr = ptr_ring_consume(&tfile->tx_ring);
2132         if (ptr)
2133                 goto out;
2134         if (noblock) {
2135                 error = -EAGAIN;
2136                 goto out;
2137         }
2138
2139         add_wait_queue(&tfile->wq.wait, &wait);
2140         current->state = TASK_INTERRUPTIBLE;
2141
2142         while (1) {
2143                 ptr = ptr_ring_consume(&tfile->tx_ring);
2144                 if (ptr)
2145                         break;
2146                 if (signal_pending(current)) {
2147                         error = -ERESTARTSYS;
2148                         break;
2149                 }
2150                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2151                         error = -EFAULT;
2152                         break;
2153                 }
2154
2155                 schedule();
2156         }
2157
2158         current->state = TASK_RUNNING;
2159         remove_wait_queue(&tfile->wq.wait, &wait);
2160
2161 out:
2162         *err = error;
2163         return ptr;
2164 }
2165
2166 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2167                            struct iov_iter *to,
2168                            int noblock, void *ptr)
2169 {
2170         ssize_t ret;
2171         int err;
2172
2173         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2174
2175         if (!iov_iter_count(to)) {
2176                 tun_ptr_free(ptr);
2177                 return 0;
2178         }
2179
2180         if (!ptr) {
2181                 /* Read frames from ring */
2182                 ptr = tun_ring_recv(tfile, noblock, &err);
2183                 if (!ptr)
2184                         return err;
2185         }
2186
2187         if (tun_is_xdp_buff(ptr)) {
2188                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2189
2190                 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2191                 put_page(virt_to_head_page(xdp->data));
2192         } else {
2193                 struct sk_buff *skb = ptr;
2194
2195                 ret = tun_put_user(tun, tfile, skb, to);
2196                 if (unlikely(ret < 0))
2197                         kfree_skb(skb);
2198                 else
2199                         consume_skb(skb);
2200         }
2201
2202         return ret;
2203 }
2204
2205 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2206 {
2207         struct file *file = iocb->ki_filp;
2208         struct tun_file *tfile = file->private_data;
2209         struct tun_struct *tun = tun_get(tfile);
2210         ssize_t len = iov_iter_count(to), ret;
2211
2212         if (!tun)
2213                 return -EBADFD;
2214         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2215         ret = min_t(ssize_t, ret, len);
2216         if (ret > 0)
2217                 iocb->ki_pos = ret;
2218         tun_put(tun);
2219         return ret;
2220 }
2221
2222 static void tun_prog_free(struct rcu_head *rcu)
2223 {
2224         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2225
2226         bpf_prog_destroy(prog->prog);
2227         kfree(prog);
2228 }
2229
2230 static int __tun_set_ebpf(struct tun_struct *tun,
2231                           struct tun_prog __rcu **prog_p,
2232                           struct bpf_prog *prog)
2233 {
2234         struct tun_prog *old, *new = NULL;
2235
2236         if (prog) {
2237                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2238                 if (!new)
2239                         return -ENOMEM;
2240                 new->prog = prog;
2241         }
2242
2243         spin_lock_bh(&tun->lock);
2244         old = rcu_dereference_protected(*prog_p,
2245                                         lockdep_is_held(&tun->lock));
2246         rcu_assign_pointer(*prog_p, new);
2247         spin_unlock_bh(&tun->lock);
2248
2249         if (old)
2250                 call_rcu(&old->rcu, tun_prog_free);
2251
2252         return 0;
2253 }
2254
2255 static void tun_free_netdev(struct net_device *dev)
2256 {
2257         struct tun_struct *tun = netdev_priv(dev);
2258
2259         BUG_ON(!(list_empty(&tun->disabled)));
2260         free_percpu(tun->pcpu_stats);
2261         tun_flow_uninit(tun);
2262         security_tun_dev_free_security(tun->security);
2263         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2264         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2265 }
2266
2267 static void tun_setup(struct net_device *dev)
2268 {
2269         struct tun_struct *tun = netdev_priv(dev);
2270
2271         tun->owner = INVALID_UID;
2272         tun->group = INVALID_GID;
2273
2274         dev->ethtool_ops = &tun_ethtool_ops;
2275         dev->needs_free_netdev = true;
2276         dev->priv_destructor = tun_free_netdev;
2277         /* We prefer our own queue length */
2278         dev->tx_queue_len = TUN_READQ_SIZE;
2279 }
2280
2281 /* Trivial set of netlink ops to allow deleting tun or tap
2282  * device with netlink.
2283  */
2284 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2285                         struct netlink_ext_ack *extack)
2286 {
2287         return -EINVAL;
2288 }
2289
2290 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2291         .kind           = DRV_NAME,
2292         .priv_size      = sizeof(struct tun_struct),
2293         .setup          = tun_setup,
2294         .validate       = tun_validate,
2295 };
2296
2297 static void tun_sock_write_space(struct sock *sk)
2298 {
2299         struct tun_file *tfile;
2300         wait_queue_head_t *wqueue;
2301
2302         if (!sock_writeable(sk))
2303                 return;
2304
2305         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2306                 return;
2307
2308         wqueue = sk_sleep(sk);
2309         if (wqueue && waitqueue_active(wqueue))
2310                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2311                                                 EPOLLWRNORM | EPOLLWRBAND);
2312
2313         tfile = container_of(sk, struct tun_file, sk);
2314         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2315 }
2316
2317 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2318 {
2319         int ret;
2320         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2321         struct tun_struct *tun = tun_get(tfile);
2322
2323         if (!tun)
2324                 return -EBADFD;
2325
2326         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2327                            m->msg_flags & MSG_DONTWAIT,
2328                            m->msg_flags & MSG_MORE);
2329         tun_put(tun);
2330         return ret;
2331 }
2332
2333 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2334                        int flags)
2335 {
2336         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2337         struct tun_struct *tun = tun_get(tfile);
2338         void *ptr = m->msg_control;
2339         int ret;
2340
2341         if (!tun) {
2342                 ret = -EBADFD;
2343                 goto out_free;
2344         }
2345
2346         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2347                 ret = -EINVAL;
2348                 goto out_put_tun;
2349         }
2350         if (flags & MSG_ERRQUEUE) {
2351                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2352                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2353                 goto out;
2354         }
2355         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2356         if (ret > (ssize_t)total_len) {
2357                 m->msg_flags |= MSG_TRUNC;
2358                 ret = flags & MSG_TRUNC ? ret : total_len;
2359         }
2360 out:
2361         tun_put(tun);
2362         return ret;
2363
2364 out_put_tun:
2365         tun_put(tun);
2366 out_free:
2367         tun_ptr_free(ptr);
2368         return ret;
2369 }
2370
2371 static int tun_ptr_peek_len(void *ptr)
2372 {
2373         if (likely(ptr)) {
2374                 if (tun_is_xdp_buff(ptr)) {
2375                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2376
2377                         return xdp->data_end - xdp->data;
2378                 }
2379                 return __skb_array_len_with_tag(ptr);
2380         } else {
2381                 return 0;
2382         }
2383 }
2384
2385 static int tun_peek_len(struct socket *sock)
2386 {
2387         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2388         struct tun_struct *tun;
2389         int ret = 0;
2390
2391         tun = tun_get(tfile);
2392         if (!tun)
2393                 return 0;
2394
2395         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2396         tun_put(tun);
2397
2398         return ret;
2399 }
2400
2401 /* Ops structure to mimic raw sockets with tun */
2402 static const struct proto_ops tun_socket_ops = {
2403         .peek_len = tun_peek_len,
2404         .sendmsg = tun_sendmsg,
2405         .recvmsg = tun_recvmsg,
2406 };
2407
2408 static struct proto tun_proto = {
2409         .name           = "tun",
2410         .owner          = THIS_MODULE,
2411         .obj_size       = sizeof(struct tun_file),
2412 };
2413
2414 static int tun_flags(struct tun_struct *tun)
2415 {
2416         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2417 }
2418
2419 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2420                               char *buf)
2421 {
2422         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2423         return sprintf(buf, "0x%x\n", tun_flags(tun));
2424 }
2425
2426 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2427                               char *buf)
2428 {
2429         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2430         return uid_valid(tun->owner)?
2431                 sprintf(buf, "%u\n",
2432                         from_kuid_munged(current_user_ns(), tun->owner)):
2433                 sprintf(buf, "-1\n");
2434 }
2435
2436 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2437                               char *buf)
2438 {
2439         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2440         return gid_valid(tun->group) ?
2441                 sprintf(buf, "%u\n",
2442                         from_kgid_munged(current_user_ns(), tun->group)):
2443                 sprintf(buf, "-1\n");
2444 }
2445
2446 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2447 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2448 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2449
2450 static struct attribute *tun_dev_attrs[] = {
2451         &dev_attr_tun_flags.attr,
2452         &dev_attr_owner.attr,
2453         &dev_attr_group.attr,
2454         NULL
2455 };
2456
2457 static const struct attribute_group tun_attr_group = {
2458         .attrs = tun_dev_attrs
2459 };
2460
2461 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2462 {
2463         struct tun_struct *tun;
2464         struct tun_file *tfile = file->private_data;
2465         struct net_device *dev;
2466         int err;
2467
2468         if (tfile->detached)
2469                 return -EINVAL;
2470
2471         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2472                 if (!capable(CAP_NET_ADMIN))
2473                         return -EPERM;
2474
2475                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2476                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2477                         return -EINVAL;
2478         }
2479
2480         dev = __dev_get_by_name(net, ifr->ifr_name);
2481         if (dev) {
2482                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2483                         return -EBUSY;
2484                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2485                         tun = netdev_priv(dev);
2486                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2487                         tun = netdev_priv(dev);
2488                 else
2489                         return -EINVAL;
2490
2491                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2492                     !!(tun->flags & IFF_MULTI_QUEUE))
2493                         return -EINVAL;
2494
2495                 if (tun_not_capable(tun))
2496                         return -EPERM;
2497                 err = security_tun_dev_open(tun->security);
2498                 if (err < 0)
2499                         return err;
2500
2501                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2502                                  ifr->ifr_flags & IFF_NAPI);
2503                 if (err < 0)
2504                         return err;
2505
2506                 if (tun->flags & IFF_MULTI_QUEUE &&
2507                     (tun->numqueues + tun->numdisabled > 1)) {
2508                         /* One or more queue has already been attached, no need
2509                          * to initialize the device again.
2510                          */
2511                         return 0;
2512                 }
2513         }
2514         else {
2515                 char *name;
2516                 unsigned long flags = 0;
2517                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2518                              MAX_TAP_QUEUES : 1;
2519
2520                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2521                         return -EPERM;
2522                 err = security_tun_dev_create();
2523                 if (err < 0)
2524                         return err;
2525
2526                 /* Set dev type */
2527                 if (ifr->ifr_flags & IFF_TUN) {
2528                         /* TUN device */
2529                         flags |= IFF_TUN;
2530                         name = "tun%d";
2531                 } else if (ifr->ifr_flags & IFF_TAP) {
2532                         /* TAP device */
2533                         flags |= IFF_TAP;
2534                         name = "tap%d";
2535                 } else
2536                         return -EINVAL;
2537
2538                 if (*ifr->ifr_name)
2539                         name = ifr->ifr_name;
2540
2541                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2542                                        NET_NAME_UNKNOWN, tun_setup, queues,
2543                                        queues);
2544
2545                 if (!dev)
2546                         return -ENOMEM;
2547                 err = dev_get_valid_name(net, dev, name);
2548                 if (err < 0)
2549                         goto err_free_dev;
2550
2551                 dev_net_set(dev, net);
2552                 dev->rtnl_link_ops = &tun_link_ops;
2553                 dev->ifindex = tfile->ifindex;
2554                 dev->sysfs_groups[0] = &tun_attr_group;
2555
2556                 tun = netdev_priv(dev);
2557                 tun->dev = dev;
2558                 tun->flags = flags;
2559                 tun->txflt.count = 0;
2560                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2561
2562                 tun->align = NET_SKB_PAD;
2563                 tun->filter_attached = false;
2564                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2565                 tun->rx_batched = 0;
2566                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2567
2568                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2569                 if (!tun->pcpu_stats) {
2570                         err = -ENOMEM;
2571                         goto err_free_dev;
2572                 }
2573
2574                 spin_lock_init(&tun->lock);
2575
2576                 err = security_tun_dev_alloc_security(&tun->security);
2577                 if (err < 0)
2578                         goto err_free_stat;
2579
2580                 tun_net_init(dev);
2581                 tun_flow_init(tun);
2582
2583                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2584                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2585                                    NETIF_F_HW_VLAN_STAG_TX;
2586                 dev->features = dev->hw_features | NETIF_F_LLTX;
2587                 dev->vlan_features = dev->features &
2588                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2589                                        NETIF_F_HW_VLAN_STAG_TX);
2590
2591                 INIT_LIST_HEAD(&tun->disabled);
2592                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2593                 if (err < 0)
2594                         goto err_free_flow;
2595
2596                 err = register_netdevice(tun->dev);
2597                 if (err < 0)
2598                         goto err_detach;
2599         }
2600
2601         netif_carrier_on(tun->dev);
2602
2603         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2604
2605         tun->flags = (tun->flags & ~TUN_FEATURES) |
2606                 (ifr->ifr_flags & TUN_FEATURES);
2607
2608         /* Make sure persistent devices do not get stuck in
2609          * xoff state.
2610          */
2611         if (netif_running(tun->dev))
2612                 netif_tx_wake_all_queues(tun->dev);
2613
2614         strcpy(ifr->ifr_name, tun->dev->name);
2615         return 0;
2616
2617 err_detach:
2618         tun_detach_all(dev);
2619         /* register_netdevice() already called tun_free_netdev() */
2620         goto err_free_dev;
2621
2622 err_free_flow:
2623         tun_flow_uninit(tun);
2624         security_tun_dev_free_security(tun->security);
2625 err_free_stat:
2626         free_percpu(tun->pcpu_stats);
2627 err_free_dev:
2628         free_netdev(dev);
2629         return err;
2630 }
2631
2632 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2633                        struct ifreq *ifr)
2634 {
2635         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2636
2637         strcpy(ifr->ifr_name, tun->dev->name);
2638
2639         ifr->ifr_flags = tun_flags(tun);
2640
2641 }
2642
2643 /* This is like a cut-down ethtool ops, except done via tun fd so no
2644  * privs required. */
2645 static int set_offload(struct tun_struct *tun, unsigned long arg)
2646 {
2647         netdev_features_t features = 0;
2648
2649         if (arg & TUN_F_CSUM) {
2650                 features |= NETIF_F_HW_CSUM;
2651                 arg &= ~TUN_F_CSUM;
2652
2653                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2654                         if (arg & TUN_F_TSO_ECN) {
2655                                 features |= NETIF_F_TSO_ECN;
2656                                 arg &= ~TUN_F_TSO_ECN;
2657                         }
2658                         if (arg & TUN_F_TSO4)
2659                                 features |= NETIF_F_TSO;
2660                         if (arg & TUN_F_TSO6)
2661                                 features |= NETIF_F_TSO6;
2662                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2663                 }
2664
2665                 arg &= ~TUN_F_UFO;
2666         }
2667
2668         /* This gives the user a way to test for new features in future by
2669          * trying to set them. */
2670         if (arg)
2671                 return -EINVAL;
2672
2673         tun->set_features = features;
2674         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2675         tun->dev->wanted_features |= features;
2676         netdev_update_features(tun->dev);
2677
2678         return 0;
2679 }
2680
2681 static void tun_detach_filter(struct tun_struct *tun, int n)
2682 {
2683         int i;
2684         struct tun_file *tfile;
2685
2686         for (i = 0; i < n; i++) {
2687                 tfile = rtnl_dereference(tun->tfiles[i]);
2688                 lock_sock(tfile->socket.sk);
2689                 sk_detach_filter(tfile->socket.sk);
2690                 release_sock(tfile->socket.sk);
2691         }
2692
2693         tun->filter_attached = false;
2694 }
2695
2696 static int tun_attach_filter(struct tun_struct *tun)
2697 {
2698         int i, ret = 0;
2699         struct tun_file *tfile;
2700
2701         for (i = 0; i < tun->numqueues; i++) {
2702                 tfile = rtnl_dereference(tun->tfiles[i]);
2703                 lock_sock(tfile->socket.sk);
2704                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2705                 release_sock(tfile->socket.sk);
2706                 if (ret) {
2707                         tun_detach_filter(tun, i);
2708                         return ret;
2709                 }
2710         }
2711
2712         tun->filter_attached = true;
2713         return ret;
2714 }
2715
2716 static void tun_set_sndbuf(struct tun_struct *tun)
2717 {
2718         struct tun_file *tfile;
2719         int i;
2720
2721         for (i = 0; i < tun->numqueues; i++) {
2722                 tfile = rtnl_dereference(tun->tfiles[i]);
2723                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2724         }
2725 }
2726
2727 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2728 {
2729         struct tun_file *tfile = file->private_data;
2730         struct tun_struct *tun;
2731         int ret = 0;
2732
2733         rtnl_lock();
2734
2735         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2736                 tun = tfile->detached;
2737                 if (!tun) {
2738                         ret = -EINVAL;
2739                         goto unlock;
2740                 }
2741                 ret = security_tun_dev_attach_queue(tun->security);
2742                 if (ret < 0)
2743                         goto unlock;
2744                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2745         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2746                 tun = rtnl_dereference(tfile->tun);
2747                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2748                         ret = -EINVAL;
2749                 else
2750                         __tun_detach(tfile, false);
2751         } else
2752                 ret = -EINVAL;
2753
2754 unlock:
2755         rtnl_unlock();
2756         return ret;
2757 }
2758
2759 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2760                         void __user *data)
2761 {
2762         struct bpf_prog *prog;
2763         int fd;
2764
2765         if (copy_from_user(&fd, data, sizeof(fd)))
2766                 return -EFAULT;
2767
2768         if (fd == -1) {
2769                 prog = NULL;
2770         } else {
2771                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2772                 if (IS_ERR(prog))
2773                         return PTR_ERR(prog);
2774         }
2775
2776         return __tun_set_ebpf(tun, prog_p, prog);
2777 }
2778
2779 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2780                             unsigned long arg, int ifreq_len)
2781 {
2782         struct tun_file *tfile = file->private_data;
2783         struct tun_struct *tun;
2784         void __user* argp = (void __user*)arg;
2785         struct ifreq ifr;
2786         kuid_t owner;
2787         kgid_t group;
2788         int sndbuf;
2789         int vnet_hdr_sz;
2790         unsigned int ifindex;
2791         int le;
2792         int ret;
2793
2794         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2795                 if (copy_from_user(&ifr, argp, ifreq_len))
2796                         return -EFAULT;
2797         } else {
2798                 memset(&ifr, 0, sizeof(ifr));
2799         }
2800         if (cmd == TUNGETFEATURES) {
2801                 /* Currently this just means: "what IFF flags are valid?".
2802                  * This is needed because we never checked for invalid flags on
2803                  * TUNSETIFF.
2804                  */
2805                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2806                                 (unsigned int __user*)argp);
2807         } else if (cmd == TUNSETQUEUE)
2808                 return tun_set_queue(file, &ifr);
2809
2810         ret = 0;
2811         rtnl_lock();
2812
2813         tun = tun_get(tfile);
2814         if (cmd == TUNSETIFF) {
2815                 ret = -EEXIST;
2816                 if (tun)
2817                         goto unlock;
2818
2819                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2820
2821                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2822
2823                 if (ret)
2824                         goto unlock;
2825
2826                 if (copy_to_user(argp, &ifr, ifreq_len))
2827                         ret = -EFAULT;
2828                 goto unlock;
2829         }
2830         if (cmd == TUNSETIFINDEX) {
2831                 ret = -EPERM;
2832                 if (tun)
2833                         goto unlock;
2834
2835                 ret = -EFAULT;
2836                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2837                         goto unlock;
2838
2839                 ret = 0;
2840                 tfile->ifindex = ifindex;
2841                 goto unlock;
2842         }
2843
2844         ret = -EBADFD;
2845         if (!tun)
2846                 goto unlock;
2847
2848         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2849
2850         ret = 0;
2851         switch (cmd) {
2852         case TUNGETIFF:
2853                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2854
2855                 if (tfile->detached)
2856                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2857                 if (!tfile->socket.sk->sk_filter)
2858                         ifr.ifr_flags |= IFF_NOFILTER;
2859
2860                 if (copy_to_user(argp, &ifr, ifreq_len))
2861                         ret = -EFAULT;
2862                 break;
2863
2864         case TUNSETNOCSUM:
2865                 /* Disable/Enable checksum */
2866
2867                 /* [unimplemented] */
2868                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2869                           arg ? "disabled" : "enabled");
2870                 break;
2871
2872         case TUNSETPERSIST:
2873                 /* Disable/Enable persist mode. Keep an extra reference to the
2874                  * module to prevent the module being unprobed.
2875                  */
2876                 if (arg && !(tun->flags & IFF_PERSIST)) {
2877                         tun->flags |= IFF_PERSIST;
2878                         __module_get(THIS_MODULE);
2879                 }
2880                 if (!arg && (tun->flags & IFF_PERSIST)) {
2881                         tun->flags &= ~IFF_PERSIST;
2882                         module_put(THIS_MODULE);
2883                 }
2884
2885                 tun_debug(KERN_INFO, tun, "persist %s\n",
2886                           arg ? "enabled" : "disabled");
2887                 break;
2888
2889         case TUNSETOWNER:
2890                 /* Set owner of the device */
2891                 owner = make_kuid(current_user_ns(), arg);
2892                 if (!uid_valid(owner)) {
2893                         ret = -EINVAL;
2894                         break;
2895                 }
2896                 tun->owner = owner;
2897                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2898                           from_kuid(&init_user_ns, tun->owner));
2899                 break;
2900
2901         case TUNSETGROUP:
2902                 /* Set group of the device */
2903                 group = make_kgid(current_user_ns(), arg);
2904                 if (!gid_valid(group)) {
2905                         ret = -EINVAL;
2906                         break;
2907                 }
2908                 tun->group = group;
2909                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2910                           from_kgid(&init_user_ns, tun->group));
2911                 break;
2912
2913         case TUNSETLINK:
2914                 /* Only allow setting the type when the interface is down */
2915                 if (tun->dev->flags & IFF_UP) {
2916                         tun_debug(KERN_INFO, tun,
2917                                   "Linktype set failed because interface is up\n");
2918                         ret = -EBUSY;
2919                 } else {
2920                         tun->dev->type = (int) arg;
2921                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2922                                   tun->dev->type);
2923                         ret = 0;
2924                 }
2925                 break;
2926
2927 #ifdef TUN_DEBUG
2928         case TUNSETDEBUG:
2929                 tun->debug = arg;
2930                 break;
2931 #endif
2932         case TUNSETOFFLOAD:
2933                 ret = set_offload(tun, arg);
2934                 break;
2935
2936         case TUNSETTXFILTER:
2937                 /* Can be set only for TAPs */
2938                 ret = -EINVAL;
2939                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2940                         break;
2941                 ret = update_filter(&tun->txflt, (void __user *)arg);
2942                 break;
2943
2944         case SIOCGIFHWADDR:
2945                 /* Get hw address */
2946                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2947                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2948                 if (copy_to_user(argp, &ifr, ifreq_len))
2949                         ret = -EFAULT;
2950                 break;
2951
2952         case SIOCSIFHWADDR:
2953                 /* Set hw address */
2954                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2955                           ifr.ifr_hwaddr.sa_data);
2956
2957                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2958                 break;
2959
2960         case TUNGETSNDBUF:
2961                 sndbuf = tfile->socket.sk->sk_sndbuf;
2962                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2963                         ret = -EFAULT;
2964                 break;
2965
2966         case TUNSETSNDBUF:
2967                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2968                         ret = -EFAULT;
2969                         break;
2970                 }
2971                 if (sndbuf <= 0) {
2972                         ret = -EINVAL;
2973                         break;
2974                 }
2975
2976                 tun->sndbuf = sndbuf;
2977                 tun_set_sndbuf(tun);
2978                 break;
2979
2980         case TUNGETVNETHDRSZ:
2981                 vnet_hdr_sz = tun->vnet_hdr_sz;
2982                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2983                         ret = -EFAULT;
2984                 break;
2985
2986         case TUNSETVNETHDRSZ:
2987                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2988                         ret = -EFAULT;
2989                         break;
2990                 }
2991                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2992                         ret = -EINVAL;
2993                         break;
2994                 }
2995
2996                 tun->vnet_hdr_sz = vnet_hdr_sz;
2997                 break;
2998
2999         case TUNGETVNETLE:
3000                 le = !!(tun->flags & TUN_VNET_LE);
3001                 if (put_user(le, (int __user *)argp))
3002                         ret = -EFAULT;
3003                 break;
3004
3005         case TUNSETVNETLE:
3006                 if (get_user(le, (int __user *)argp)) {
3007                         ret = -EFAULT;
3008                         break;
3009                 }
3010                 if (le)
3011                         tun->flags |= TUN_VNET_LE;
3012                 else
3013                         tun->flags &= ~TUN_VNET_LE;
3014                 break;
3015
3016         case TUNGETVNETBE:
3017                 ret = tun_get_vnet_be(tun, argp);
3018                 break;
3019
3020         case TUNSETVNETBE:
3021                 ret = tun_set_vnet_be(tun, argp);
3022                 break;
3023
3024         case TUNATTACHFILTER:
3025                 /* Can be set only for TAPs */
3026                 ret = -EINVAL;
3027                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3028                         break;
3029                 ret = -EFAULT;
3030                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3031                         break;
3032
3033                 ret = tun_attach_filter(tun);
3034                 break;
3035
3036         case TUNDETACHFILTER:
3037                 /* Can be set only for TAPs */
3038                 ret = -EINVAL;
3039                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3040                         break;
3041                 ret = 0;
3042                 tun_detach_filter(tun, tun->numqueues);
3043                 break;
3044
3045         case TUNGETFILTER:
3046                 ret = -EINVAL;
3047                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3048                         break;
3049                 ret = -EFAULT;
3050                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3051                         break;
3052                 ret = 0;
3053                 break;
3054
3055         case TUNSETSTEERINGEBPF:
3056                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3057                 break;
3058
3059         case TUNSETFILTEREBPF:
3060                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3061                 break;
3062
3063         default:
3064                 ret = -EINVAL;
3065                 break;
3066         }
3067
3068 unlock:
3069         rtnl_unlock();
3070         if (tun)
3071                 tun_put(tun);
3072         return ret;
3073 }
3074
3075 static long tun_chr_ioctl(struct file *file,
3076                           unsigned int cmd, unsigned long arg)
3077 {
3078         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3079 }
3080
3081 #ifdef CONFIG_COMPAT
3082 static long tun_chr_compat_ioctl(struct file *file,
3083                          unsigned int cmd, unsigned long arg)
3084 {
3085         switch (cmd) {
3086         case TUNSETIFF:
3087         case TUNGETIFF:
3088         case TUNSETTXFILTER:
3089         case TUNGETSNDBUF:
3090         case TUNSETSNDBUF:
3091         case SIOCGIFHWADDR:
3092         case SIOCSIFHWADDR:
3093                 arg = (unsigned long)compat_ptr(arg);
3094                 break;
3095         default:
3096                 arg = (compat_ulong_t)arg;
3097                 break;
3098         }
3099
3100         /*
3101          * compat_ifreq is shorter than ifreq, so we must not access beyond
3102          * the end of that structure. All fields that are used in this
3103          * driver are compatible though, we don't need to convert the
3104          * contents.
3105          */
3106         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3107 }
3108 #endif /* CONFIG_COMPAT */
3109
3110 static int tun_chr_fasync(int fd, struct file *file, int on)
3111 {
3112         struct tun_file *tfile = file->private_data;
3113         int ret;
3114
3115         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3116                 goto out;
3117
3118         if (on) {
3119                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3120                 tfile->flags |= TUN_FASYNC;
3121         } else
3122                 tfile->flags &= ~TUN_FASYNC;
3123         ret = 0;
3124 out:
3125         return ret;
3126 }
3127
3128 static int tun_chr_open(struct inode *inode, struct file * file)
3129 {
3130         struct net *net = current->nsproxy->net_ns;
3131         struct tun_file *tfile;
3132
3133         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3134
3135         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3136                                             &tun_proto, 0);
3137         if (!tfile)
3138                 return -ENOMEM;
3139         RCU_INIT_POINTER(tfile->tun, NULL);
3140         tfile->flags = 0;
3141         tfile->ifindex = 0;
3142
3143         init_waitqueue_head(&tfile->wq.wait);
3144         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3145
3146         tfile->socket.file = file;
3147         tfile->socket.ops = &tun_socket_ops;
3148
3149         sock_init_data(&tfile->socket, &tfile->sk);
3150
3151         tfile->sk.sk_write_space = tun_sock_write_space;
3152         tfile->sk.sk_sndbuf = INT_MAX;
3153
3154         file->private_data = tfile;
3155         INIT_LIST_HEAD(&tfile->next);
3156
3157         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3158
3159         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3160
3161         return 0;
3162 }
3163
3164 static int tun_chr_close(struct inode *inode, struct file *file)
3165 {
3166         struct tun_file *tfile = file->private_data;
3167
3168         tun_detach(tfile, true);
3169
3170         return 0;
3171 }
3172
3173 #ifdef CONFIG_PROC_FS
3174 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3175 {
3176         struct tun_file *tfile = file->private_data;
3177         struct tun_struct *tun;
3178         struct ifreq ifr;
3179
3180         memset(&ifr, 0, sizeof(ifr));
3181
3182         rtnl_lock();
3183         tun = tun_get(tfile);
3184         if (tun)
3185                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3186         rtnl_unlock();
3187
3188         if (tun)
3189                 tun_put(tun);
3190
3191         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3192 }
3193 #endif
3194
3195 static const struct file_operations tun_fops = {
3196         .owner  = THIS_MODULE,
3197         .llseek = no_llseek,
3198         .read_iter  = tun_chr_read_iter,
3199         .write_iter = tun_chr_write_iter,
3200         .poll   = tun_chr_poll,
3201         .unlocked_ioctl = tun_chr_ioctl,
3202 #ifdef CONFIG_COMPAT
3203         .compat_ioctl = tun_chr_compat_ioctl,
3204 #endif
3205         .open   = tun_chr_open,
3206         .release = tun_chr_close,
3207         .fasync = tun_chr_fasync,
3208 #ifdef CONFIG_PROC_FS
3209         .show_fdinfo = tun_chr_show_fdinfo,
3210 #endif
3211 };
3212
3213 static struct miscdevice tun_miscdev = {
3214         .minor = TUN_MINOR,
3215         .name = "tun",
3216         .nodename = "net/tun",
3217         .fops = &tun_fops,
3218 };
3219
3220 /* ethtool interface */
3221
3222 static int tun_get_link_ksettings(struct net_device *dev,
3223                                   struct ethtool_link_ksettings *cmd)
3224 {
3225         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3226         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3227         cmd->base.speed         = SPEED_10;
3228         cmd->base.duplex        = DUPLEX_FULL;
3229         cmd->base.port          = PORT_TP;
3230         cmd->base.phy_address   = 0;
3231         cmd->base.autoneg       = AUTONEG_DISABLE;
3232         return 0;
3233 }
3234
3235 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3236 {
3237         struct tun_struct *tun = netdev_priv(dev);
3238
3239         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3240         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3241
3242         switch (tun->flags & TUN_TYPE_MASK) {
3243         case IFF_TUN:
3244                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3245                 break;
3246         case IFF_TAP:
3247                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3248                 break;
3249         }
3250 }
3251
3252 static u32 tun_get_msglevel(struct net_device *dev)
3253 {
3254 #ifdef TUN_DEBUG
3255         struct tun_struct *tun = netdev_priv(dev);
3256         return tun->debug;
3257 #else
3258         return -EOPNOTSUPP;
3259 #endif
3260 }
3261
3262 static void tun_set_msglevel(struct net_device *dev, u32 value)
3263 {
3264 #ifdef TUN_DEBUG
3265         struct tun_struct *tun = netdev_priv(dev);
3266         tun->debug = value;
3267 #endif
3268 }
3269
3270 static int tun_get_coalesce(struct net_device *dev,
3271                             struct ethtool_coalesce *ec)
3272 {
3273         struct tun_struct *tun = netdev_priv(dev);
3274
3275         ec->rx_max_coalesced_frames = tun->rx_batched;
3276
3277         return 0;
3278 }
3279
3280 static int tun_set_coalesce(struct net_device *dev,
3281                             struct ethtool_coalesce *ec)
3282 {
3283         struct tun_struct *tun = netdev_priv(dev);
3284
3285         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3286                 tun->rx_batched = NAPI_POLL_WEIGHT;
3287         else
3288                 tun->rx_batched = ec->rx_max_coalesced_frames;
3289
3290         return 0;
3291 }
3292
3293 static const struct ethtool_ops tun_ethtool_ops = {
3294         .get_drvinfo    = tun_get_drvinfo,
3295         .get_msglevel   = tun_get_msglevel,
3296         .set_msglevel   = tun_set_msglevel,
3297         .get_link       = ethtool_op_get_link,
3298         .get_ts_info    = ethtool_op_get_ts_info,
3299         .get_coalesce   = tun_get_coalesce,
3300         .set_coalesce   = tun_set_coalesce,
3301         .get_link_ksettings = tun_get_link_ksettings,
3302 };
3303
3304 static int tun_queue_resize(struct tun_struct *tun)
3305 {
3306         struct net_device *dev = tun->dev;
3307         struct tun_file *tfile;
3308         struct ptr_ring **rings;
3309         int n = tun->numqueues + tun->numdisabled;
3310         int ret, i;
3311
3312         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3313         if (!rings)
3314                 return -ENOMEM;
3315
3316         for (i = 0; i < tun->numqueues; i++) {
3317                 tfile = rtnl_dereference(tun->tfiles[i]);
3318                 rings[i] = &tfile->tx_ring;
3319         }
3320         list_for_each_entry(tfile, &tun->disabled, next)
3321                 rings[i++] = &tfile->tx_ring;
3322
3323         ret = ptr_ring_resize_multiple(rings, n,
3324                                        dev->tx_queue_len, GFP_KERNEL,
3325                                        tun_ptr_free);
3326
3327         kfree(rings);
3328         return ret;
3329 }
3330
3331 static int tun_device_event(struct notifier_block *unused,
3332                             unsigned long event, void *ptr)
3333 {
3334         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3335         struct tun_struct *tun = netdev_priv(dev);
3336
3337         if (dev->rtnl_link_ops != &tun_link_ops)
3338                 return NOTIFY_DONE;
3339
3340         switch (event) {
3341         case NETDEV_CHANGE_TX_QUEUE_LEN:
3342                 if (tun_queue_resize(tun))
3343                         return NOTIFY_BAD;
3344                 break;
3345         default:
3346                 break;
3347         }
3348
3349         return NOTIFY_DONE;
3350 }
3351
3352 static struct notifier_block tun_notifier_block __read_mostly = {
3353         .notifier_call  = tun_device_event,
3354 };
3355
3356 static int __init tun_init(void)
3357 {
3358         int ret = 0;
3359
3360         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3361
3362         ret = rtnl_link_register(&tun_link_ops);
3363         if (ret) {
3364                 pr_err("Can't register link_ops\n");
3365                 goto err_linkops;
3366         }
3367
3368         ret = misc_register(&tun_miscdev);
3369         if (ret) {
3370                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3371                 goto err_misc;
3372         }
3373
3374         ret = register_netdevice_notifier(&tun_notifier_block);
3375         if (ret) {
3376                 pr_err("Can't register netdevice notifier\n");
3377                 goto err_notifier;
3378         }
3379
3380         return  0;
3381
3382 err_notifier:
3383         misc_deregister(&tun_miscdev);
3384 err_misc:
3385         rtnl_link_unregister(&tun_link_ops);
3386 err_linkops:
3387         return ret;
3388 }
3389
3390 static void tun_cleanup(void)
3391 {
3392         misc_deregister(&tun_miscdev);
3393         rtnl_link_unregister(&tun_link_ops);
3394         unregister_netdevice_notifier(&tun_notifier_block);
3395 }
3396
3397 /* Get an underlying socket object from tun file.  Returns error unless file is
3398  * attached to a device.  The returned object works like a packet socket, it
3399  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3400  * holding a reference to the file for as long as the socket is in use. */
3401 struct socket *tun_get_socket(struct file *file)
3402 {
3403         struct tun_file *tfile;
3404         if (file->f_op != &tun_fops)
3405                 return ERR_PTR(-EINVAL);
3406         tfile = file->private_data;
3407         if (!tfile)
3408                 return ERR_PTR(-EBADFD);
3409         return &tfile->socket;
3410 }
3411 EXPORT_SYMBOL_GPL(tun_get_socket);
3412
3413 struct ptr_ring *tun_get_tx_ring(struct file *file)
3414 {
3415         struct tun_file *tfile;
3416
3417         if (file->f_op != &tun_fops)
3418                 return ERR_PTR(-EINVAL);
3419         tfile = file->private_data;
3420         if (!tfile)
3421                 return ERR_PTR(-EBADFD);
3422         return &tfile->tx_ring;
3423 }
3424 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3425
3426 module_init(tun_init);
3427 module_exit(tun_cleanup);
3428 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3429 MODULE_AUTHOR(DRV_COPYRIGHT);
3430 MODULE_LICENSE("GPL");
3431 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3432 MODULE_ALIAS("devname:net/tun");