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 #include <linux/proc_fs.h>
  82
  83 /* Uncomment to enable debugging */
  84 /* #define TUN_DEBUG 1 */
  85
  86 #ifdef TUN_DEBUG
  87 static int debug;
  88
  89 #define tun_debug(level, tun, fmt, args...)                     \
  90 do {                                                            \
  91         if (tun->debug)                                         \
  92                 netdev_printk(level, tun->dev, fmt, ##args);    \
  93 } while (0)
  94 #define DBG1(level, fmt, args...)                               \
  95 do {                                                            \
  96         if (debug == 2)                                         \
  97                 printk(level fmt, ##args);                      \
  98 } while (0)
  99 #else
 100 #define tun_debug(level, tun, fmt, args...)                     \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 netdev_printk(level, tun->dev, fmt, ##args);    \
 104 } while (0)
 105 #define DBG1(level, fmt, args...)                               \
 106 do {                                                            \
 107         if (0)                                                  \
 108                 printk(level fmt, ##args);                      \
 109 } while (0)
 110 #endif
 111
 112 #define TUN_HEADROOM 256
 113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 127
 128 #define GOODCOPY_LEN 128
 129
 130 #define FLT_EXACT_COUNT 8
 131 struct tap_filter {
 132         unsigned int    count;    /* Number of addrs. Zero means disabled */
 133         u32             mask[2];  /* Mask of the hashed addrs */
 134         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 135 };
 136
 137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 138  * to max number of VCPUs in guest. */
 139 #define MAX_TAP_QUEUES 256
 140 #define MAX_TAP_FLOWS  4096
 141
 142 #define TUN_FLOW_EXPIRE (3 * HZ)
 143
 144 struct tun_pcpu_stats {
 145         u64 rx_packets;
 146         u64 rx_bytes;
 147         u64 tx_packets;
 148         u64 tx_bytes;
 149         struct u64_stats_sync syncp;
 150         u32 rx_dropped;
 151         u32 tx_dropped;
 152         u32 rx_frame_errors;
 153 };
 154
 155 /* A tun_file connects an open character device to a tuntap netdevice. It
 156  * also contains all socket related structures (except sock_fprog and tap_filter)
 157  * to serve as one transmit queue for tuntap device. The sock_fprog and
 158  * tap_filter were kept in tun_struct since they were used for filtering for the
 159  * netdevice not for a specific queue (at least I didn't see the requirement for
 160  * this).
 161  *
 162  * RCU usage:
 163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 164  * other can only be read while rcu_read_lock or rtnl_lock is held.
 165  */
 166 struct tun_file {
 167         struct sock sk;
 168         struct socket socket;
 169         struct socket_wq wq;
 170         struct tun_struct __rcu *tun;
 171         struct fasync_struct *fasync;
 172         /* only used for fasnyc */
 173         unsigned int flags;
 174         union {
 175                 u16 queue_index;
 176                 unsigned int ifindex;
 177         };
 178         struct napi_struct napi;
 179         bool napi_enabled;
 180         struct mutex napi_mutex;        /* Protects access to the above napi */
 181         struct list_head next;
 182         struct tun_struct *detached;
 183         struct ptr_ring tx_ring;
 184         struct xdp_rxq_info xdp_rxq;
 185 };
 186
 187 struct tun_flow_entry {
 188         struct hlist_node hash_link;
 189         struct rcu_head rcu;
 190         struct tun_struct *tun;
 191
 192         u32 rxhash;
 193         u32 rps_rxhash;
 194         int queue_index;
 195         unsigned long updated;
 196 };
 197
 198 #define TUN_NUM_FLOW_ENTRIES 1024
 199
 200 struct tun_prog {
 201         struct rcu_head rcu;
 202         struct bpf_prog *prog;
 203 };
 204
 205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 206  * device, socket filter, sndbuf and vnet header size were restore when the
 207  * file were attached to a persist device.
 208  */
 209 struct tun_struct {
 210         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 211         unsigned int            numqueues;
 212         unsigned int            flags;
 213         kuid_t                  owner;
 214         kgid_t                  group;
 215
 216         struct net_device       *dev;
 217         netdev_features_t       set_features;
 218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 219                           NETIF_F_TSO6)
 220
 221         int                     align;
 222         int                     vnet_hdr_sz;
 223         int                     sndbuf;
 224         struct tap_filter       txflt;
 225         struct sock_fprog       fprog;
 226         /* protected by rtnl lock */
 227         bool                    filter_attached;
 228 #ifdef TUN_DEBUG
 229         int debug;
 230 #endif
 231         spinlock_t lock;
 232         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 233         struct timer_list flow_gc_timer;
 234         unsigned long ageing_time;
 235         unsigned int numdisabled;
 236         struct list_head disabled;
 237         void *security;
 238         u32 flow_count;
 239         u32 rx_batched;
 240         struct tun_pcpu_stats __percpu *pcpu_stats;
 241         struct bpf_prog __rcu *xdp_prog;
 242         struct tun_prog __rcu *steering_prog;
 243         struct tun_prog __rcu *filter_prog;
 244 };
 245
 246 struct veth {
 247         __be16 h_vlan_proto;
 248         __be16 h_vlan_TCI;
 249 };
 250
 251 bool tun_is_xdp_buff(void *ptr)
 252 {
 253         return (unsigned long)ptr & TUN_XDP_FLAG;
 254 }
 255 EXPORT_SYMBOL(tun_is_xdp_buff);
 256
 257 void *tun_xdp_to_ptr(void *ptr)
 258 {
 259         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 260 }
 261 EXPORT_SYMBOL(tun_xdp_to_ptr);
 262
 263 void *tun_ptr_to_xdp(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_ptr_to_xdp);
 268
 269 static int tun_napi_receive(struct napi_struct *napi, int budget)
 270 {
 271         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 272         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 273         struct sk_buff_head process_queue;
 274         struct sk_buff *skb;
 275         int received = 0;
 276
 277         __skb_queue_head_init(&process_queue);
 278
 279         spin_lock(&queue->lock);
 280         skb_queue_splice_tail_init(queue, &process_queue);
 281         spin_unlock(&queue->lock);
 282
 283         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 284                 napi_gro_receive(napi, skb);
 285                 ++received;
 286         }
 287
 288         if (!skb_queue_empty(&process_queue)) {
 289                 spin_lock(&queue->lock);
 290                 skb_queue_splice(&process_queue, queue);
 291                 spin_unlock(&queue->lock);
 292         }
 293
 294         return received;
 295 }
 296
 297 static int tun_napi_poll(struct napi_struct *napi, int budget)
 298 {
 299         unsigned int received;
 300
 301         received = tun_napi_receive(napi, budget);
 302
 303         if (received < budget)
 304                 napi_complete_done(napi, received);
 305
 306         return received;
 307 }
 308
 309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 310                           bool napi_en)
 311 {
 312         tfile->napi_enabled = napi_en;
 313         if (napi_en) {
 314                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 315                                NAPI_POLL_WEIGHT);
 316                 napi_enable(&tfile->napi);
 317                 mutex_init(&tfile->napi_mutex);
 318         }
 319 }
 320
 321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 322 {
 323         if (tfile->napi_enabled)
 324                 napi_disable(&tfile->napi);
 325 }
 326
 327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 netif_napi_del(&tfile->napi);
 331 }
 332
 333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 334 {
 335         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 336 }
 337
 338 #ifdef CONFIG_TUN_VNET_CROSS_LE
 339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 340 {
 341         return tun->flags & TUN_VNET_BE ? false :
 342                 virtio_legacy_is_little_endian();
 343 }
 344
 345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 346 {
 347         int be = !!(tun->flags & TUN_VNET_BE);
 348
 349         if (put_user(be, argp))
 350                 return -EFAULT;
 351
 352         return 0;
 353 }
 354
 355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 356 {
 357         int be;
 358
 359         if (get_user(be, argp))
 360                 return -EFAULT;
 361
 362         if (be)
 363                 tun->flags |= TUN_VNET_BE;
 364         else
 365                 tun->flags &= ~TUN_VNET_BE;
 366
 367         return 0;
 368 }
 369 #else
 370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 371 {
 372         return virtio_legacy_is_little_endian();
 373 }
 374
 375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 376 {
 377         return -EINVAL;
 378 }
 379
 380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 385
 386 static inline bool tun_is_little_endian(struct tun_struct *tun)
 387 {
 388         return tun->flags & TUN_VNET_LE ||
 389                 tun_legacy_is_little_endian(tun);
 390 }
 391
 392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 393 {
 394         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 395 }
 396
 397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 398 {
 399         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline u32 tun_hashfn(u32 rxhash)
 403 {
 404         return rxhash & 0x3ff;
 405 }
 406
 407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 408 {
 409         struct tun_flow_entry *e;
 410
 411         hlist_for_each_entry_rcu(e, head, hash_link) {
 412                 if (e->rxhash == rxhash)
 413                         return e;
 414         }
 415         return NULL;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 419                                               struct hlist_head *head,
 420                                               u32 rxhash, u16 queue_index)
 421 {
 422         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 423
 424         if (e) {
 425                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 426                           rxhash, queue_index);
 427                 e->updated = jiffies;
 428                 e->rxhash = rxhash;
 429                 e->rps_rxhash = 0;
 430                 e->queue_index = queue_index;
 431                 e->tun = tun;
 432                 hlist_add_head_rcu(&e->hash_link, head);
 433                 ++tun->flow_count;
 434         }
 435         return e;
 436 }
 437
 438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 439 {
 440         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 441                   e->rxhash, e->queue_index);
 442         hlist_del_rcu(&e->hash_link);
 443         kfree_rcu(e, rcu);
 444         --tun->flow_count;
 445 }
 446
 447 static void tun_flow_flush(struct tun_struct *tun)
 448 {
 449         int i;
 450
 451         spin_lock_bh(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 457                         tun_flow_delete(tun, e);
 458         }
 459         spin_unlock_bh(&tun->lock);
 460 }
 461
 462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 463 {
 464         int i;
 465
 466         spin_lock_bh(&tun->lock);
 467         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 468                 struct tun_flow_entry *e;
 469                 struct hlist_node *n;
 470
 471                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 472                         if (e->queue_index == queue_index)
 473                                 tun_flow_delete(tun, e);
 474                 }
 475         }
 476         spin_unlock_bh(&tun->lock);
 477 }
 478
 479 static void tun_flow_cleanup(struct timer_list *t)
 480 {
 481         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 482         unsigned long delay = tun->ageing_time;
 483         unsigned long next_timer = jiffies + delay;
 484         unsigned long count = 0;
 485         int i;
 486
 487         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 488
 489         spin_lock(&tun->lock);
 490         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 491                 struct tun_flow_entry *e;
 492                 struct hlist_node *n;
 493
 494                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 495                         unsigned long this_timer;
 496
 497                         this_timer = e->updated + delay;
 498                         if (time_before_eq(this_timer, jiffies)) {
 499                                 tun_flow_delete(tun, e);
 500                                 continue;
 501                         }
 502                         count++;
 503                         if (time_before(this_timer, next_timer))
 504                                 next_timer = this_timer;
 505                 }
 506         }
 507
 508         if (count)
 509                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 510         spin_unlock(&tun->lock);
 511 }
 512
 513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 514                             struct tun_file *tfile)
 515 {
 516         struct hlist_head *head;
 517         struct tun_flow_entry *e;
 518         unsigned long delay = tun->ageing_time;
 519         u16 queue_index = tfile->queue_index;
 520
 521         if (!rxhash)
 522                 return;
 523         else
 524                 head = &tun->flows[tun_hashfn(rxhash)];
 525
 526         rcu_read_lock();
 527
 528         /* We may get a very small possibility of OOO during switching, not
 529          * worth to optimize.*/
 530         if (tun->numqueues == 1 || tfile->detached)
 531                 goto unlock;
 532
 533         e = tun_flow_find(head, rxhash);
 534         if (likely(e)) {
 535                 /* TODO: keep queueing to old queue until it's empty? */
 536                 e->queue_index = queue_index;
 537                 e->updated = jiffies;
 538                 sock_rps_record_flow_hash(e->rps_rxhash);
 539         } else {
 540                 spin_lock_bh(&tun->lock);
 541                 if (!tun_flow_find(head, rxhash) &&
 542                     tun->flow_count < MAX_TAP_FLOWS)
 543                         tun_flow_create(tun, head, rxhash, queue_index);
 544
 545                 if (!timer_pending(&tun->flow_gc_timer))
 546                         mod_timer(&tun->flow_gc_timer,
 547                                   round_jiffies_up(jiffies + delay));
 548                 spin_unlock_bh(&tun->lock);
 549         }
 550
 551 unlock:
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash first. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here. If we could not get rxhash, then we would
 570  * hope the rxq no. may help here.
 571  */
 572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 573 {
 574         struct tun_flow_entry *e;
 575         u32 txq = 0;
 576         u32 numqueues = 0;
 577
 578         numqueues = READ_ONCE(tun->numqueues);
 579
 580         txq = __skb_get_hash_symmetric(skb);
 581         if (txq) {
 582                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 583                 if (e) {
 584                         tun_flow_save_rps_rxhash(e, txq);
 585                         txq = e->queue_index;
 586                 } else
 587                         /* use multiply and shift instead of expensive divide */
 588                         txq = ((u64)txq * numqueues) >> 32;
 589         } else if (likely(skb_rx_queue_recorded(skb))) {
 590                 txq = skb_get_rx_queue(skb);
 591                 while (unlikely(txq >= numqueues))
 592                         txq -= numqueues;
 593         }
 594
 595         return txq;
 596 }
 597
 598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 599 {
 600         struct tun_prog *prog;
 601         u16 ret = 0;
 602
 603         prog = rcu_dereference(tun->steering_prog);
 604         if (prog)
 605                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 606
 607         return ret % tun->numqueues;
 608 }
 609
 610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 611                             void *accel_priv, select_queue_fallback_t fallback)
 612 {
 613         struct tun_struct *tun = netdev_priv(dev);
 614         u16 ret;
 615
 616         rcu_read_lock();
 617         if (rcu_dereference(tun->steering_prog))
 618                 ret = tun_ebpf_select_queue(tun, skb);
 619         else
 620                 ret = tun_automq_select_queue(tun, skb);
 621         rcu_read_unlock();
 622
 623         return ret;
 624 }
 625
 626 static inline bool tun_not_capable(struct tun_struct *tun)
 627 {
 628         const struct cred *cred = current_cred();
 629         struct net *net = dev_net(tun->dev);
 630
 631         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 632                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 633                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 634 }
 635
 636 static void tun_set_real_num_queues(struct tun_struct *tun)
 637 {
 638         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 639         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 640 }
 641
 642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 643 {
 644         tfile->detached = tun;
 645         list_add_tail(&tfile->next, &tun->disabled);
 646         ++tun->numdisabled;
 647 }
 648
 649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 650 {
 651         struct tun_struct *tun = tfile->detached;
 652
 653         tfile->detached = NULL;
 654         list_del_init(&tfile->next);
 655         --tun->numdisabled;
 656         return tun;
 657 }
 658
 659 void tun_ptr_free(void *ptr)
 660 {
 661         if (!ptr)
 662                 return;
 663         if (tun_is_xdp_buff(ptr)) {
 664                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
 665
 666                 put_page(virt_to_head_page(xdp->data));
 667         } else {
 668                 __skb_array_destroy_skb(ptr);
 669         }
 670 }
 671 EXPORT_SYMBOL_GPL(tun_ptr_free);
 672
 673 static void tun_queue_purge(struct tun_file *tfile)
 674 {
 675         void *ptr;
 676
 677         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 678                 tun_ptr_free(ptr);
 679
 680         skb_queue_purge(&tfile->sk.sk_write_queue);
 681         skb_queue_purge(&tfile->sk.sk_error_queue);
 682 }
 683
 684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 685 {
 686         if (tfile->tx_ring.queue) {
 687                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 688                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 690         }
 691 }
 692
 693 static void __tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_file *ntfile;
 696         struct tun_struct *tun;
 697
 698         tun = rtnl_dereference(tfile->tun);
 699
 700         if (tun && clean) {
 701                 tun_napi_disable(tun, tfile);
 702                 tun_napi_del(tun, tfile);
 703         }
 704
 705         if (tun && !tfile->detached) {
 706                 u16 index = tfile->queue_index;
 707                 BUG_ON(index >= tun->numqueues);
 708
 709                 rcu_assign_pointer(tun->tfiles[index],
 710                                    tun->tfiles[tun->numqueues - 1]);
 711                 ntfile = rtnl_dereference(tun->tfiles[index]);
 712                 ntfile->queue_index = index;
 713
 714                 --tun->numqueues;
 715                 if (clean) {
 716                         RCU_INIT_POINTER(tfile->tun, NULL);
 717                         sock_put(&tfile->sk);
 718                 } else
 719                         tun_disable_queue(tun, tfile);
 720
 721                 synchronize_net();
 722                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 723                 /* Drop read queue */
 724                 tun_queue_purge(tfile);
 725                 tun_set_real_num_queues(tun);
 726         } else if (tfile->detached && clean) {
 727                 tun = tun_enable_queue(tfile);
 728                 sock_put(&tfile->sk);
 729         }
 730
 731         if (clean) {
 732                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 733                         netif_carrier_off(tun->dev);
 734
 735                         if (!(tun->flags & IFF_PERSIST) &&
 736                             tun->dev->reg_state == NETREG_REGISTERED)
 737                                 unregister_netdevice(tun->dev);
 738                 }
 739                 tun_cleanup_tx_ring(tfile);
 740                 sock_put(&tfile->sk);
 741         }
 742 }
 743
 744 static void tun_detach(struct tun_file *tfile, bool clean)
 745 {
 746         struct tun_struct *tun;
 747         struct net_device *dev;
 748
 749         rtnl_lock();
 750         tun = rtnl_dereference(tfile->tun);
 751         dev = tun ? tun->dev : NULL;
 752         __tun_detach(tfile, clean);
 753         if (dev)
 754                 netdev_state_change(dev);
 755         rtnl_unlock();
 756 }
 757
 758 static void tun_detach_all(struct net_device *dev)
 759 {
 760         struct tun_struct *tun = netdev_priv(dev);
 761         struct tun_file *tfile, *tmp;
 762         int i, n = tun->numqueues;
 763
 764         for (i = 0; i < n; i++) {
 765                 tfile = rtnl_dereference(tun->tfiles[i]);
 766                 BUG_ON(!tfile);
 767                 tun_napi_disable(tun, tfile);
 768                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 769                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 770                 RCU_INIT_POINTER(tfile->tun, NULL);
 771                 --tun->numqueues;
 772         }
 773         list_for_each_entry(tfile, &tun->disabled, next) {
 774                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 775                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 776                 RCU_INIT_POINTER(tfile->tun, NULL);
 777         }
 778         BUG_ON(tun->numqueues != 0);
 779
 780         synchronize_net();
 781         for (i = 0; i < n; i++) {
 782                 tfile = rtnl_dereference(tun->tfiles[i]);
 783                 tun_napi_del(tun, tfile);
 784                 /* Drop read queue */
 785                 tun_queue_purge(tfile);
 786                 sock_put(&tfile->sk);
 787                 tun_cleanup_tx_ring(tfile);
 788         }
 789         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 790                 tun_enable_queue(tfile);
 791                 tun_queue_purge(tfile);
 792                 sock_put(&tfile->sk);
 793                 tun_cleanup_tx_ring(tfile);
 794         }
 795         BUG_ON(tun->numdisabled != 0);
 796
 797         if (tun->flags & IFF_PERSIST)
 798                 module_put(THIS_MODULE);
 799 }
 800
 801 static int tun_attach(struct tun_struct *tun, struct file *file,
 802                       bool skip_filter, bool napi)
 803 {
 804         struct tun_file *tfile = file->private_data;
 805         struct net_device *dev = tun->dev;
 806         int err;
 807
 808         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 809         if (err < 0)
 810                 goto out;
 811
 812         err = -EINVAL;
 813         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 814                 goto out;
 815
 816         err = -EBUSY;
 817         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 818                 goto out;
 819
 820         err = -E2BIG;
 821         if (!tfile->detached &&
 822             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 823                 goto out;
 824
 825         err = 0;
 826
 827         /* Re-attach the filter to persist device */
 828         if (!skip_filter && (tun->filter_attached == true)) {
 829                 lock_sock(tfile->socket.sk);
 830                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 831                 release_sock(tfile->socket.sk);
 832                 if (!err)
 833                         goto out;
 834         }
 835
 836         if (!tfile->detached &&
 837             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 838                 err = -ENOMEM;
 839                 goto out;
 840         }
 841
 842         tfile->queue_index = tun->numqueues;
 843         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 844
 845         if (tfile->detached) {
 846                 /* Re-attach detached tfile, updating XDP queue_index */
 847                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 848
 849                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 850                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 851         } else {
 852                 /* Setup XDP RX-queue info, for new tfile getting attached */
 853                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 854                                        tun->dev, tfile->queue_index);
 855                 if (err < 0)
 856                         goto out;
 857                 err = 0;
 858         }
 859
 860         rcu_assign_pointer(tfile->tun, tun);
 861         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 862         tun->numqueues++;
 863
 864         if (tfile->detached) {
 865                 tun_enable_queue(tfile);
 866         } else {
 867                 sock_hold(&tfile->sk);
 868                 tun_napi_init(tun, tfile, napi);
 869         }
 870
 871         tun_set_real_num_queues(tun);
 872
 873         /* device is allowed to go away first, so no need to hold extra
 874          * refcnt.
 875          */
 876
 877 out:
 878         return err;
 879 }
 880
 881 static struct tun_struct *tun_get(struct tun_file *tfile)
 882 {
 883         struct tun_struct *tun;
 884
 885         rcu_read_lock();
 886         tun = rcu_dereference(tfile->tun);
 887         if (tun)
 888                 dev_hold(tun->dev);
 889         rcu_read_unlock();
 890
 891         return tun;
 892 }
 893
 894 static void tun_put(struct tun_struct *tun)
 895 {
 896         dev_put(tun->dev);
 897 }
 898
 899 /* TAP filtering */
 900 static void addr_hash_set(u32 *mask, const u8 *addr)
 901 {
 902         int n = ether_crc(ETH_ALEN, addr) >> 26;
 903         mask[n >> 5] |= (1 << (n & 31));
 904 }
 905
 906 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 907 {
 908         int n = ether_crc(ETH_ALEN, addr) >> 26;
 909         return mask[n >> 5] & (1 << (n & 31));
 910 }
 911
 912 static int update_filter(struct tap_filter *filter, void __user *arg)
 913 {
 914         struct { u8 u[ETH_ALEN]; } *addr;
 915         struct tun_filter uf;
 916         int err, alen, n, nexact;
 917
 918         if (copy_from_user(&uf, arg, sizeof(uf)))
 919                 return -EFAULT;
 920
 921         if (!uf.count) {
 922                 /* Disabled */
 923                 filter->count = 0;
 924                 return 0;
 925         }
 926
 927         alen = ETH_ALEN * uf.count;
 928         addr = memdup_user(arg + sizeof(uf), alen);
 929         if (IS_ERR(addr))
 930                 return PTR_ERR(addr);
 931
 932         /* The filter is updated without holding any locks. Which is
 933          * perfectly safe. We disable it first and in the worst
 934          * case we'll accept a few undesired packets. */
 935         filter->count = 0;
 936         wmb();
 937
 938         /* Use first set of addresses as an exact filter */
 939         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 940                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 941
 942         nexact = n;
 943
 944         /* Remaining multicast addresses are hashed,
 945          * unicast will leave the filter disabled. */
 946         memset(filter->mask, 0, sizeof(filter->mask));
 947         for (; n < uf.count; n++) {
 948                 if (!is_multicast_ether_addr(addr[n].u)) {
 949                         err = 0; /* no filter */
 950                         goto free_addr;
 951                 }
 952                 addr_hash_set(filter->mask, addr[n].u);
 953         }
 954
 955         /* For ALLMULTI just set the mask to all ones.
 956          * This overrides the mask populated above. */
 957         if ((uf.flags & TUN_FLT_ALLMULTI))
 958                 memset(filter->mask, ~0, sizeof(filter->mask));
 959
 960         /* Now enable the filter */
 961         wmb();
 962         filter->count = nexact;
 963
 964         /* Return the number of exact filters */
 965         err = nexact;
 966 free_addr:
 967         kfree(addr);
 968         return err;
 969 }
 970
 971 /* Returns: 0 - drop, !=0 - accept */
 972 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 973 {
 974         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 975          * at this point. */
 976         struct ethhdr *eh = (struct ethhdr *) skb->data;
 977         int i;
 978
 979         /* Exact match */
 980         for (i = 0; i < filter->count; i++)
 981                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 982                         return 1;
 983
 984         /* Inexact match (multicast only) */
 985         if (is_multicast_ether_addr(eh->h_dest))
 986                 return addr_hash_test(filter->mask, eh->h_dest);
 987
 988         return 0;
 989 }
 990
 991 /*
 992  * Checks whether the packet is accepted or not.
 993  * Returns: 0 - drop, !=0 - accept
 994  */
 995 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 996 {
 997         if (!filter->count)
 998                 return 1;
 999
1000         return run_filter(filter, skb);
1001 }
1002
1003 /* Network device part of the driver */
1004
1005 static const struct ethtool_ops tun_ethtool_ops;
1006
1007 /* Net device detach from fd. */
1008 static void tun_net_uninit(struct net_device *dev)
1009 {
1010         tun_detach_all(dev);
1011 }
1012
1013 /* Net device open. */
1014 static int tun_net_open(struct net_device *dev)
1015 {
1016         struct tun_struct *tun = netdev_priv(dev);
1017         int i;
1018
1019         netif_tx_start_all_queues(dev);
1020
1021         for (i = 0; i < tun->numqueues; i++) {
1022                 struct tun_file *tfile;
1023
1024                 tfile = rtnl_dereference(tun->tfiles[i]);
1025                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1026         }
1027
1028         return 0;
1029 }
1030
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1033 {
1034         netif_tx_stop_all_queues(dev);
1035         return 0;
1036 }
1037
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 {
1041 #ifdef CONFIG_RPS
1042         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1043                 /* Select queue was not called for the skbuff, so we extract the
1044                  * RPS hash and save it into the flow_table here.
1045                  */
1046                 __u32 rxhash;
1047
1048                 rxhash = __skb_get_hash_symmetric(skb);
1049                 if (rxhash) {
1050                         struct tun_flow_entry *e;
1051                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1052                                         rxhash);
1053                         if (e)
1054                                 tun_flow_save_rps_rxhash(e, rxhash);
1055                 }
1056         }
1057 #endif
1058 }
1059
1060 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1061                                     struct sk_buff *skb,
1062                                     int len)
1063 {
1064         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1065
1066         if (prog)
1067                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1068
1069         return len;
1070 }
1071
1072 /* Net device start xmit */
1073 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1074 {
1075         struct tun_struct *tun = netdev_priv(dev);
1076         int txq = skb->queue_mapping;
1077         struct tun_file *tfile;
1078         int len = skb->len;
1079
1080         rcu_read_lock();
1081         tfile = rcu_dereference(tun->tfiles[txq]);
1082
1083         /* Drop packet if interface is not attached */
1084         if (txq >= tun->numqueues)
1085                 goto drop;
1086
1087         if (!rcu_dereference(tun->steering_prog))
1088                 tun_automq_xmit(tun, skb);
1089
1090         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1091
1092         BUG_ON(!tfile);
1093
1094         /* Drop if the filter does not like it.
1095          * This is a noop if the filter is disabled.
1096          * Filter can be enabled only for the TAP devices. */
1097         if (!check_filter(&tun->txflt, skb))
1098                 goto drop;
1099
1100         if (tfile->socket.sk->sk_filter &&
1101             sk_filter(tfile->socket.sk, skb))
1102                 goto drop;
1103
1104         len = run_ebpf_filter(tun, skb, len);
1105         if (len == 0 || pskb_trim(skb, len))
1106                 goto drop;
1107
1108         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1109                 goto drop;
1110
1111         skb_tx_timestamp(skb);
1112
1113         /* Orphan the skb - required as we might hang on to it
1114          * for indefinite time.
1115          */
1116         skb_orphan(skb);
1117
1118         nf_reset(skb);
1119
1120         if (ptr_ring_produce(&tfile->tx_ring, skb))
1121                 goto drop;
1122
1123         /* Notify and wake up reader process */
1124         if (tfile->flags & TUN_FASYNC)
1125                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1126         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1127
1128         rcu_read_unlock();
1129         return NETDEV_TX_OK;
1130
1131 drop:
1132         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1133         skb_tx_error(skb);
1134         kfree_skb(skb);
1135         rcu_read_unlock();
1136         return NET_XMIT_DROP;
1137 }
1138
1139 static void tun_net_mclist(struct net_device *dev)
1140 {
1141         /*
1142          * This callback is supposed to deal with mc filter in
1143          * _rx_ path and has nothing to do with the _tx_ path.
1144          * In rx path we always accept everything userspace gives us.
1145          */
1146 }
1147
1148 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1149         netdev_features_t features)
1150 {
1151         struct tun_struct *tun = netdev_priv(dev);
1152
1153         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1154 }
1155 #ifdef CONFIG_NET_POLL_CONTROLLER
1156 static void tun_poll_controller(struct net_device *dev)
1157 {
1158         /*
1159          * Tun only receives frames when:
1160          * 1) the char device endpoint gets data from user space
1161          * 2) the tun socket gets a sendmsg call from user space
1162          * If NAPI is not enabled, since both of those are synchronous
1163          * operations, we are guaranteed never to have pending data when we poll
1164          * for it so there is nothing to do here but return.
1165          * We need this though so netpoll recognizes us as an interface that
1166          * supports polling, which enables bridge devices in virt setups to
1167          * still use netconsole
1168          * If NAPI is enabled, however, we need to schedule polling for all
1169          * queues unless we are using napi_gro_frags(), which we call in
1170          * process context and not in NAPI context.
1171          */
1172         struct tun_struct *tun = netdev_priv(dev);
1173
1174         if (tun->flags & IFF_NAPI) {
1175                 struct tun_file *tfile;
1176                 int i;
1177
1178                 if (tun_napi_frags_enabled(tun))
1179                         return;
1180
1181                 rcu_read_lock();
1182                 for (i = 0; i < tun->numqueues; i++) {
1183                         tfile = rcu_dereference(tun->tfiles[i]);
1184                         if (tfile->napi_enabled)
1185                                 napi_schedule(&tfile->napi);
1186                 }
1187                 rcu_read_unlock();
1188         }
1189         return;
1190 }
1191 #endif
1192
1193 static void tun_set_headroom(struct net_device *dev, int new_hr)
1194 {
1195         struct tun_struct *tun = netdev_priv(dev);
1196
1197         if (new_hr < NET_SKB_PAD)
1198                 new_hr = NET_SKB_PAD;
1199
1200         tun->align = new_hr;
1201 }
1202
1203 static void
1204 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1205 {
1206         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1207         struct tun_struct *tun = netdev_priv(dev);
1208         struct tun_pcpu_stats *p;
1209         int i;
1210
1211         for_each_possible_cpu(i) {
1212                 u64 rxpackets, rxbytes, txpackets, txbytes;
1213                 unsigned int start;
1214
1215                 p = per_cpu_ptr(tun->pcpu_stats, i);
1216                 do {
1217                         start = u64_stats_fetch_begin(&p->syncp);
1218                         rxpackets       = p->rx_packets;
1219                         rxbytes         = p->rx_bytes;
1220                         txpackets       = p->tx_packets;
1221                         txbytes         = p->tx_bytes;
1222                 } while (u64_stats_fetch_retry(&p->syncp, start));
1223
1224                 stats->rx_packets       += rxpackets;
1225                 stats->rx_bytes         += rxbytes;
1226                 stats->tx_packets       += txpackets;
1227                 stats->tx_bytes         += txbytes;
1228
1229                 /* u32 counters */
1230                 rx_dropped      += p->rx_dropped;
1231                 rx_frame_errors += p->rx_frame_errors;
1232                 tx_dropped      += p->tx_dropped;
1233         }
1234         stats->rx_dropped  = rx_dropped;
1235         stats->rx_frame_errors = rx_frame_errors;
1236         stats->tx_dropped = tx_dropped;
1237 }
1238
1239 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1240                        struct netlink_ext_ack *extack)
1241 {
1242         struct tun_struct *tun = netdev_priv(dev);
1243         struct bpf_prog *old_prog;
1244
1245         old_prog = rtnl_dereference(tun->xdp_prog);
1246         rcu_assign_pointer(tun->xdp_prog, prog);
1247         if (old_prog)
1248                 bpf_prog_put(old_prog);
1249
1250         return 0;
1251 }
1252
1253 static u32 tun_xdp_query(struct net_device *dev)
1254 {
1255         struct tun_struct *tun = netdev_priv(dev);
1256         const struct bpf_prog *xdp_prog;
1257
1258         xdp_prog = rtnl_dereference(tun->xdp_prog);
1259         if (xdp_prog)
1260                 return xdp_prog->aux->id;
1261
1262         return 0;
1263 }
1264
1265 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1266 {
1267         switch (xdp->command) {
1268         case XDP_SETUP_PROG:
1269                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1270         case XDP_QUERY_PROG:
1271                 xdp->prog_id = tun_xdp_query(dev);
1272                 xdp->prog_attached = !!xdp->prog_id;
1273                 return 0;
1274         default:
1275                 return -EINVAL;
1276         }
1277 }
1278
1279 static const struct net_device_ops tun_netdev_ops = {
1280         .ndo_uninit             = tun_net_uninit,
1281         .ndo_open               = tun_net_open,
1282         .ndo_stop               = tun_net_close,
1283         .ndo_start_xmit         = tun_net_xmit,
1284         .ndo_fix_features       = tun_net_fix_features,
1285         .ndo_select_queue       = tun_select_queue,
1286 #ifdef CONFIG_NET_POLL_CONTROLLER
1287         .ndo_poll_controller    = tun_poll_controller,
1288 #endif
1289         .ndo_set_rx_headroom    = tun_set_headroom,
1290         .ndo_get_stats64        = tun_net_get_stats64,
1291 };
1292
1293 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1294 {
1295         struct tun_struct *tun = netdev_priv(dev);
1296         struct xdp_buff *buff = xdp->data_hard_start;
1297         int headroom = xdp->data - xdp->data_hard_start;
1298         struct tun_file *tfile;
1299         u32 numqueues;
1300         int ret = 0;
1301
1302         /* Assure headroom is available and buff is properly aligned */
1303         if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1304                 return -ENOSPC;
1305
1306         *buff = *xdp;
1307
1308         rcu_read_lock();
1309
1310         numqueues = READ_ONCE(tun->numqueues);
1311         if (!numqueues) {
1312                 ret = -ENOSPC;
1313                 goto out;
1314         }
1315
1316         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1317                                             numqueues]);
1318         /* Encode the XDP flag into lowest bit for consumer to differ
1319          * XDP buffer from sk_buff.
1320          */
1321         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1322                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1323                 ret = -ENOSPC;
1324         }
1325
1326 out:
1327         rcu_read_unlock();
1328         return ret;
1329 }
1330
1331 static void tun_xdp_flush(struct net_device *dev)
1332 {
1333         struct tun_struct *tun = netdev_priv(dev);
1334         struct tun_file *tfile;
1335         u32 numqueues;
1336
1337         rcu_read_lock();
1338
1339         numqueues = READ_ONCE(tun->numqueues);
1340         if (!numqueues)
1341                 goto out;
1342
1343         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1344                                             numqueues]);
1345         /* Notify and wake up reader process */
1346         if (tfile->flags & TUN_FASYNC)
1347                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1348         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1349
1350 out:
1351         rcu_read_unlock();
1352 }
1353
1354 static const struct net_device_ops tap_netdev_ops = {
1355         .ndo_uninit             = tun_net_uninit,
1356         .ndo_open               = tun_net_open,
1357         .ndo_stop               = tun_net_close,
1358         .ndo_start_xmit         = tun_net_xmit,
1359         .ndo_fix_features       = tun_net_fix_features,
1360         .ndo_set_rx_mode        = tun_net_mclist,
1361         .ndo_set_mac_address    = eth_mac_addr,
1362         .ndo_validate_addr      = eth_validate_addr,
1363         .ndo_select_queue       = tun_select_queue,
1364 #ifdef CONFIG_NET_POLL_CONTROLLER
1365         .ndo_poll_controller    = tun_poll_controller,
1366 #endif
1367         .ndo_features_check     = passthru_features_check,
1368         .ndo_set_rx_headroom    = tun_set_headroom,
1369         .ndo_get_stats64        = tun_net_get_stats64,
1370         .ndo_bpf                = tun_xdp,
1371         .ndo_xdp_xmit           = tun_xdp_xmit,
1372         .ndo_xdp_flush          = tun_xdp_flush,
1373 };
1374
1375 static void tun_flow_init(struct tun_struct *tun)
1376 {
1377         int i;
1378
1379         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1380                 INIT_HLIST_HEAD(&tun->flows[i]);
1381
1382         tun->ageing_time = TUN_FLOW_EXPIRE;
1383         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1384         mod_timer(&tun->flow_gc_timer,
1385                   round_jiffies_up(jiffies + tun->ageing_time));
1386 }
1387
1388 static void tun_flow_uninit(struct tun_struct *tun)
1389 {
1390         del_timer_sync(&tun->flow_gc_timer);
1391         tun_flow_flush(tun);
1392 }
1393
1394 #define MIN_MTU 68
1395 #define MAX_MTU 65535
1396
1397 /* Initialize net device. */
1398 static void tun_net_init(struct net_device *dev)
1399 {
1400         struct tun_struct *tun = netdev_priv(dev);
1401
1402         switch (tun->flags & TUN_TYPE_MASK) {
1403         case IFF_TUN:
1404                 dev->netdev_ops = &tun_netdev_ops;
1405
1406                 /* Point-to-Point TUN Device */
1407                 dev->hard_header_len = 0;
1408                 dev->addr_len = 0;
1409                 dev->mtu = 1500;
1410
1411                 /* Zero header length */
1412                 dev->type = ARPHRD_NONE;
1413                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1414                 break;
1415
1416         case IFF_TAP:
1417                 dev->netdev_ops = &tap_netdev_ops;
1418                 /* Ethernet TAP Device */
1419                 ether_setup(dev);
1420                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1421                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1422
1423                 eth_hw_addr_random(dev);
1424
1425                 break;
1426         }
1427
1428         dev->min_mtu = MIN_MTU;
1429         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1430 }
1431
1432 /* Character device part */
1433
1434 /* Poll */
1435 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1436 {
1437         struct tun_file *tfile = file->private_data;
1438         struct tun_struct *tun = tun_get(tfile);
1439         struct sock *sk;
1440         __poll_t mask = 0;
1441
1442         if (!tun)
1443                 return EPOLLERR;
1444
1445         sk = tfile->socket.sk;
1446
1447         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1448
1449         poll_wait(file, sk_sleep(sk), wait);
1450
1451         if (!ptr_ring_empty(&tfile->tx_ring))
1452                 mask |= EPOLLIN | EPOLLRDNORM;
1453
1454         if (tun->dev->flags & IFF_UP &&
1455             (sock_writeable(sk) ||
1456              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1457               sock_writeable(sk))))
1458                 mask |= EPOLLOUT | EPOLLWRNORM;
1459
1460         if (tun->dev->reg_state != NETREG_REGISTERED)
1461                 mask = EPOLLERR;
1462
1463         tun_put(tun);
1464         return mask;
1465 }
1466
1467 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1468                                             size_t len,
1469                                             const struct iov_iter *it)
1470 {
1471         struct sk_buff *skb;
1472         size_t linear;
1473         int err;
1474         int i;
1475
1476         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1477                 return ERR_PTR(-ENOMEM);
1478
1479         local_bh_disable();
1480         skb = napi_get_frags(&tfile->napi);
1481         local_bh_enable();
1482         if (!skb)
1483                 return ERR_PTR(-ENOMEM);
1484
1485         linear = iov_iter_single_seg_count(it);
1486         err = __skb_grow(skb, linear);
1487         if (err)
1488                 goto free;
1489
1490         skb->len = len;
1491         skb->data_len = len - linear;
1492         skb->truesize += skb->data_len;
1493
1494         for (i = 1; i < it->nr_segs; i++) {
1495                 struct page_frag *pfrag = &current->task_frag;
1496                 size_t fragsz = it->iov[i].iov_len;
1497
1498                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1499                         err = -EINVAL;
1500                         goto free;
1501                 }
1502
1503                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1504                         err = -ENOMEM;
1505                         goto free;
1506                 }
1507
1508                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1509                                    pfrag->offset, fragsz);
1510                 page_ref_inc(pfrag->page);
1511                 pfrag->offset += fragsz;
1512         }
1513
1514         return skb;
1515 free:
1516         /* frees skb and all frags allocated with napi_alloc_frag() */
1517         napi_free_frags(&tfile->napi);
1518         return ERR_PTR(err);
1519 }
1520
1521 /* prepad is the amount to reserve at front.  len is length after that.
1522  * linear is a hint as to how much to copy (usually headers). */
1523 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1524                                      size_t prepad, size_t len,
1525                                      size_t linear, int noblock)
1526 {
1527         struct sock *sk = tfile->socket.sk;
1528         struct sk_buff *skb;
1529         int err;
1530
1531         /* Under a page?  Don't bother with paged skb. */
1532         if (prepad + len < PAGE_SIZE || !linear)
1533                 linear = len;
1534
1535         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1536                                    &err, 0);
1537         if (!skb)
1538                 return ERR_PTR(err);
1539
1540         skb_reserve(skb, prepad);
1541         skb_put(skb, linear);
1542         skb->data_len = len - linear;
1543         skb->len += len - linear;
1544
1545         return skb;
1546 }
1547
1548 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1549                            struct sk_buff *skb, int more)
1550 {
1551         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1552         struct sk_buff_head process_queue;
1553         u32 rx_batched = tun->rx_batched;
1554         bool rcv = false;
1555
1556         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1557                 local_bh_disable();
1558                 netif_receive_skb(skb);
1559                 local_bh_enable();
1560                 return;
1561         }
1562
1563         spin_lock(&queue->lock);
1564         if (!more || skb_queue_len(queue) == rx_batched) {
1565                 __skb_queue_head_init(&process_queue);
1566                 skb_queue_splice_tail_init(queue, &process_queue);
1567                 rcv = true;
1568         } else {
1569                 __skb_queue_tail(queue, skb);
1570         }
1571         spin_unlock(&queue->lock);
1572
1573         if (rcv) {
1574                 struct sk_buff *nskb;
1575
1576                 local_bh_disable();
1577                 while ((nskb = __skb_dequeue(&process_queue)))
1578                         netif_receive_skb(nskb);
1579                 netif_receive_skb(skb);
1580                 local_bh_enable();
1581         }
1582 }
1583
1584 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1585                               int len, int noblock, bool zerocopy)
1586 {
1587         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1588                 return false;
1589
1590         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1591                 return false;
1592
1593         if (!noblock)
1594                 return false;
1595
1596         if (zerocopy)
1597                 return false;
1598
1599         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1600             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1601                 return false;
1602
1603         return true;
1604 }
1605
1606 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1607                                      struct tun_file *tfile,
1608                                      struct iov_iter *from,
1609                                      struct virtio_net_hdr *hdr,
1610                                      int len, int *skb_xdp)
1611 {
1612         struct page_frag *alloc_frag = &current->task_frag;
1613         struct sk_buff *skb;
1614         struct bpf_prog *xdp_prog;
1615         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1616         unsigned int delta = 0;
1617         char *buf;
1618         size_t copied;
1619         int err, pad = TUN_RX_PAD;
1620
1621         rcu_read_lock();
1622         xdp_prog = rcu_dereference(tun->xdp_prog);
1623         if (xdp_prog)
1624                 pad += TUN_HEADROOM;
1625         buflen += SKB_DATA_ALIGN(len + pad);
1626         rcu_read_unlock();
1627
1628         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1629         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1630                 return ERR_PTR(-ENOMEM);
1631
1632         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1633         copied = copy_page_from_iter(alloc_frag->page,
1634                                      alloc_frag->offset + pad,
1635                                      len, from);
1636         if (copied != len)
1637                 return ERR_PTR(-EFAULT);
1638
1639         /* There's a small window that XDP may be set after the check
1640          * of xdp_prog above, this should be rare and for simplicity
1641          * we do XDP on skb in case the headroom is not enough.
1642          */
1643         if (hdr->gso_type || !xdp_prog)
1644                 *skb_xdp = 1;
1645         else
1646                 *skb_xdp = 0;
1647
1648         preempt_disable();
1649         rcu_read_lock();
1650         xdp_prog = rcu_dereference(tun->xdp_prog);
1651         if (xdp_prog && !*skb_xdp) {
1652                 struct xdp_buff xdp;
1653                 void *orig_data;
1654                 u32 act;
1655
1656                 xdp.data_hard_start = buf;
1657                 xdp.data = buf + pad;
1658                 xdp_set_data_meta_invalid(&xdp);
1659                 xdp.data_end = xdp.data + len;
1660                 xdp.rxq = &tfile->xdp_rxq;
1661                 orig_data = xdp.data;
1662                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1663
1664                 switch (act) {
1665                 case XDP_REDIRECT:
1666                         get_page(alloc_frag->page);
1667                         alloc_frag->offset += buflen;
1668                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1669                         xdp_do_flush_map();
1670                         if (err)
1671                                 goto err_redirect;
1672                         rcu_read_unlock();
1673                         preempt_enable();
1674                         return NULL;
1675                 case XDP_TX:
1676                         get_page(alloc_frag->page);
1677                         alloc_frag->offset += buflen;
1678                         if (tun_xdp_xmit(tun->dev, &xdp))
1679                                 goto err_redirect;
1680                         tun_xdp_flush(tun->dev);
1681                         rcu_read_unlock();
1682                         preempt_enable();
1683                         return NULL;
1684                 case XDP_PASS:
1685                         delta = orig_data - xdp.data;
1686                         break;
1687                 default:
1688                         bpf_warn_invalid_xdp_action(act);
1689                         /* fall through */
1690                 case XDP_ABORTED:
1691                         trace_xdp_exception(tun->dev, xdp_prog, act);
1692                         /* fall through */
1693                 case XDP_DROP:
1694                         goto err_xdp;
1695                 }
1696         }
1697
1698         skb = build_skb(buf, buflen);
1699         if (!skb) {
1700                 rcu_read_unlock();
1701                 preempt_enable();
1702                 return ERR_PTR(-ENOMEM);
1703         }
1704
1705         skb_reserve(skb, pad - delta);
1706         skb_put(skb, len + delta);
1707         get_page(alloc_frag->page);
1708         alloc_frag->offset += buflen;
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 size_t tun_get_size(const struct net_device *dev)
2291 {
2292         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2293         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2294
2295         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2296                nla_total_size(sizeof(gid_t)) + /* GROUP */
2297                nla_total_size(sizeof(u8)) + /* TYPE */
2298                nla_total_size(sizeof(u8)) + /* PI */
2299                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2300                nla_total_size(sizeof(u8)) + /* PERSIST */
2301                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2302                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2303                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2304                0;
2305 }
2306
2307 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2308 {
2309         struct tun_struct *tun = netdev_priv(dev);
2310
2311         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2312                 goto nla_put_failure;
2313         if (uid_valid(tun->owner) &&
2314             nla_put_u32(skb, IFLA_TUN_OWNER,
2315                         from_kuid_munged(current_user_ns(), tun->owner)))
2316                 goto nla_put_failure;
2317         if (gid_valid(tun->group) &&
2318             nla_put_u32(skb, IFLA_TUN_GROUP,
2319                         from_kgid_munged(current_user_ns(), tun->group)))
2320                 goto nla_put_failure;
2321         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2322                 goto nla_put_failure;
2323         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2324                 goto nla_put_failure;
2325         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2326                 goto nla_put_failure;
2327         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2328                        !!(tun->flags & IFF_MULTI_QUEUE)))
2329                 goto nla_put_failure;
2330         if (tun->flags & IFF_MULTI_QUEUE) {
2331                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2332                         goto nla_put_failure;
2333                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2334                                 tun->numdisabled))
2335                         goto nla_put_failure;
2336         }
2337
2338         return 0;
2339
2340 nla_put_failure:
2341         return -EMSGSIZE;
2342 }
2343
2344 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2345         .kind           = DRV_NAME,
2346         .priv_size      = sizeof(struct tun_struct),
2347         .setup          = tun_setup,
2348         .validate       = tun_validate,
2349         .get_size       = tun_get_size,
2350         .fill_info      = tun_fill_info,
2351 };
2352
2353 static void tun_sock_write_space(struct sock *sk)
2354 {
2355         struct tun_file *tfile;
2356         wait_queue_head_t *wqueue;
2357
2358         if (!sock_writeable(sk))
2359                 return;
2360
2361         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2362                 return;
2363
2364         wqueue = sk_sleep(sk);
2365         if (wqueue && waitqueue_active(wqueue))
2366                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2367                                                 EPOLLWRNORM | EPOLLWRBAND);
2368
2369         tfile = container_of(sk, struct tun_file, sk);
2370         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2371 }
2372
2373 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2374 {
2375         int ret;
2376         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2377         struct tun_struct *tun = tun_get(tfile);
2378
2379         if (!tun)
2380                 return -EBADFD;
2381
2382         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2383                            m->msg_flags & MSG_DONTWAIT,
2384                            m->msg_flags & MSG_MORE);
2385         tun_put(tun);
2386         return ret;
2387 }
2388
2389 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2390                        int flags)
2391 {
2392         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2393         struct tun_struct *tun = tun_get(tfile);
2394         void *ptr = m->msg_control;
2395         int ret;
2396
2397         if (!tun) {
2398                 ret = -EBADFD;
2399                 goto out_free;
2400         }
2401
2402         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2403                 ret = -EINVAL;
2404                 goto out_put_tun;
2405         }
2406         if (flags & MSG_ERRQUEUE) {
2407                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2408                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2409                 goto out;
2410         }
2411         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2412         if (ret > (ssize_t)total_len) {
2413                 m->msg_flags |= MSG_TRUNC;
2414                 ret = flags & MSG_TRUNC ? ret : total_len;
2415         }
2416 out:
2417         tun_put(tun);
2418         return ret;
2419
2420 out_put_tun:
2421         tun_put(tun);
2422 out_free:
2423         tun_ptr_free(ptr);
2424         return ret;
2425 }
2426
2427 static int tun_ptr_peek_len(void *ptr)
2428 {
2429         if (likely(ptr)) {
2430                 if (tun_is_xdp_buff(ptr)) {
2431                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2432
2433                         return xdp->data_end - xdp->data;
2434                 }
2435                 return __skb_array_len_with_tag(ptr);
2436         } else {
2437                 return 0;
2438         }
2439 }
2440
2441 static int tun_peek_len(struct socket *sock)
2442 {
2443         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2444         struct tun_struct *tun;
2445         int ret = 0;
2446
2447         tun = tun_get(tfile);
2448         if (!tun)
2449                 return 0;
2450
2451         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2452         tun_put(tun);
2453
2454         return ret;
2455 }
2456
2457 /* Ops structure to mimic raw sockets with tun */
2458 static const struct proto_ops tun_socket_ops = {
2459         .peek_len = tun_peek_len,
2460         .sendmsg = tun_sendmsg,
2461         .recvmsg = tun_recvmsg,
2462 };
2463
2464 static struct proto tun_proto = {
2465         .name           = "tun",
2466         .owner          = THIS_MODULE,
2467         .obj_size       = sizeof(struct tun_file),
2468 };
2469
2470 static int tun_flags(struct tun_struct *tun)
2471 {
2472         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2473 }
2474
2475 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2476                               char *buf)
2477 {
2478         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2479         return sprintf(buf, "0x%x\n", tun_flags(tun));
2480 }
2481
2482 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2483                               char *buf)
2484 {
2485         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2486         return uid_valid(tun->owner)?
2487                 sprintf(buf, "%u\n",
2488                         from_kuid_munged(current_user_ns(), tun->owner)):
2489                 sprintf(buf, "-1\n");
2490 }
2491
2492 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2493                               char *buf)
2494 {
2495         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2496         return gid_valid(tun->group) ?
2497                 sprintf(buf, "%u\n",
2498                         from_kgid_munged(current_user_ns(), tun->group)):
2499                 sprintf(buf, "-1\n");
2500 }
2501
2502 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2503 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2504 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2505
2506 static struct attribute *tun_dev_attrs[] = {
2507         &dev_attr_tun_flags.attr,
2508         &dev_attr_owner.attr,
2509         &dev_attr_group.attr,
2510         NULL
2511 };
2512
2513 static const struct attribute_group tun_attr_group = {
2514         .attrs = tun_dev_attrs
2515 };
2516
2517 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2518 {
2519         struct tun_struct *tun;
2520         struct tun_file *tfile = file->private_data;
2521         struct net_device *dev;
2522         int err;
2523
2524         if (tfile->detached)
2525                 return -EINVAL;
2526
2527         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2528                 if (!capable(CAP_NET_ADMIN))
2529                         return -EPERM;
2530
2531                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2532                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2533                         return -EINVAL;
2534         }
2535
2536         dev = __dev_get_by_name(net, ifr->ifr_name);
2537         if (dev) {
2538                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2539                         return -EBUSY;
2540                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2541                         tun = netdev_priv(dev);
2542                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2543                         tun = netdev_priv(dev);
2544                 else
2545                         return -EINVAL;
2546
2547                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2548                     !!(tun->flags & IFF_MULTI_QUEUE))
2549                         return -EINVAL;
2550
2551                 if (tun_not_capable(tun))
2552                         return -EPERM;
2553                 err = security_tun_dev_open(tun->security);
2554                 if (err < 0)
2555                         return err;
2556
2557                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2558                                  ifr->ifr_flags & IFF_NAPI);
2559                 if (err < 0)
2560                         return err;
2561
2562                 if (tun->flags & IFF_MULTI_QUEUE &&
2563                     (tun->numqueues + tun->numdisabled > 1)) {
2564                         /* One or more queue has already been attached, no need
2565                          * to initialize the device again.
2566                          */
2567                         netdev_state_change(dev);
2568                         return 0;
2569                 }
2570
2571                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2572                               (ifr->ifr_flags & TUN_FEATURES);
2573
2574                 netdev_state_change(dev);
2575         } else {
2576                 char *name;
2577                 unsigned long flags = 0;
2578                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2579                              MAX_TAP_QUEUES : 1;
2580
2581                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2582                         return -EPERM;
2583                 err = security_tun_dev_create();
2584                 if (err < 0)
2585                         return err;
2586
2587                 /* Set dev type */
2588                 if (ifr->ifr_flags & IFF_TUN) {
2589                         /* TUN device */
2590                         flags |= IFF_TUN;
2591                         name = "tun%d";
2592                 } else if (ifr->ifr_flags & IFF_TAP) {
2593                         /* TAP device */
2594                         flags |= IFF_TAP;
2595                         name = "tap%d";
2596                 } else
2597                         return -EINVAL;
2598
2599                 if (*ifr->ifr_name)
2600                         name = ifr->ifr_name;
2601
2602                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2603                                        NET_NAME_UNKNOWN, tun_setup, queues,
2604                                        queues);
2605
2606                 if (!dev)
2607                         return -ENOMEM;
2608                 err = dev_get_valid_name(net, dev, name);
2609                 if (err < 0)
2610                         goto err_free_dev;
2611
2612                 dev_net_set(dev, net);
2613                 dev->rtnl_link_ops = &tun_link_ops;
2614                 dev->ifindex = tfile->ifindex;
2615                 dev->sysfs_groups[0] = &tun_attr_group;
2616
2617                 tun = netdev_priv(dev);
2618                 tun->dev = dev;
2619                 tun->flags = flags;
2620                 tun->txflt.count = 0;
2621                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2622
2623                 tun->align = NET_SKB_PAD;
2624                 tun->filter_attached = false;
2625                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2626                 tun->rx_batched = 0;
2627                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2628
2629                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2630                 if (!tun->pcpu_stats) {
2631                         err = -ENOMEM;
2632                         goto err_free_dev;
2633                 }
2634
2635                 spin_lock_init(&tun->lock);
2636
2637                 err = security_tun_dev_alloc_security(&tun->security);
2638                 if (err < 0)
2639                         goto err_free_stat;
2640
2641                 tun_net_init(dev);
2642                 tun_flow_init(tun);
2643
2644                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2645                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2646                                    NETIF_F_HW_VLAN_STAG_TX;
2647                 dev->features = dev->hw_features | NETIF_F_LLTX;
2648                 dev->vlan_features = dev->features &
2649                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2650                                        NETIF_F_HW_VLAN_STAG_TX);
2651
2652                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2653                               (ifr->ifr_flags & TUN_FEATURES);
2654
2655                 INIT_LIST_HEAD(&tun->disabled);
2656                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2657                 if (err < 0)
2658                         goto err_free_flow;
2659
2660                 err = register_netdevice(tun->dev);
2661                 if (err < 0)
2662                         goto err_detach;
2663         }
2664
2665         netif_carrier_on(tun->dev);
2666
2667         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2668
2669         /* Make sure persistent devices do not get stuck in
2670          * xoff state.
2671          */
2672         if (netif_running(tun->dev))
2673                 netif_tx_wake_all_queues(tun->dev);
2674
2675         strcpy(ifr->ifr_name, tun->dev->name);
2676         return 0;
2677
2678 err_detach:
2679         tun_detach_all(dev);
2680         /* register_netdevice() already called tun_free_netdev() */
2681         goto err_free_dev;
2682
2683 err_free_flow:
2684         tun_flow_uninit(tun);
2685         security_tun_dev_free_security(tun->security);
2686 err_free_stat:
2687         free_percpu(tun->pcpu_stats);
2688 err_free_dev:
2689         free_netdev(dev);
2690         return err;
2691 }
2692
2693 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2694                        struct ifreq *ifr)
2695 {
2696         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2697
2698         strcpy(ifr->ifr_name, tun->dev->name);
2699
2700         ifr->ifr_flags = tun_flags(tun);
2701
2702 }
2703
2704 /* This is like a cut-down ethtool ops, except done via tun fd so no
2705  * privs required. */
2706 static int set_offload(struct tun_struct *tun, unsigned long arg)
2707 {
2708         netdev_features_t features = 0;
2709
2710         if (arg & TUN_F_CSUM) {
2711                 features |= NETIF_F_HW_CSUM;
2712                 arg &= ~TUN_F_CSUM;
2713
2714                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2715                         if (arg & TUN_F_TSO_ECN) {
2716                                 features |= NETIF_F_TSO_ECN;
2717                                 arg &= ~TUN_F_TSO_ECN;
2718                         }
2719                         if (arg & TUN_F_TSO4)
2720                                 features |= NETIF_F_TSO;
2721                         if (arg & TUN_F_TSO6)
2722                                 features |= NETIF_F_TSO6;
2723                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2724                 }
2725
2726                 arg &= ~TUN_F_UFO;
2727         }
2728
2729         /* This gives the user a way to test for new features in future by
2730          * trying to set them. */
2731         if (arg)
2732                 return -EINVAL;
2733
2734         tun->set_features = features;
2735         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2736         tun->dev->wanted_features |= features;
2737         netdev_update_features(tun->dev);
2738
2739         return 0;
2740 }
2741
2742 static void tun_detach_filter(struct tun_struct *tun, int n)
2743 {
2744         int i;
2745         struct tun_file *tfile;
2746
2747         for (i = 0; i < n; i++) {
2748                 tfile = rtnl_dereference(tun->tfiles[i]);
2749                 lock_sock(tfile->socket.sk);
2750                 sk_detach_filter(tfile->socket.sk);
2751                 release_sock(tfile->socket.sk);
2752         }
2753
2754         tun->filter_attached = false;
2755 }
2756
2757 static int tun_attach_filter(struct tun_struct *tun)
2758 {
2759         int i, ret = 0;
2760         struct tun_file *tfile;
2761
2762         for (i = 0; i < tun->numqueues; i++) {
2763                 tfile = rtnl_dereference(tun->tfiles[i]);
2764                 lock_sock(tfile->socket.sk);
2765                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2766                 release_sock(tfile->socket.sk);
2767                 if (ret) {
2768                         tun_detach_filter(tun, i);
2769                         return ret;
2770                 }
2771         }
2772
2773         tun->filter_attached = true;
2774         return ret;
2775 }
2776
2777 static void tun_set_sndbuf(struct tun_struct *tun)
2778 {
2779         struct tun_file *tfile;
2780         int i;
2781
2782         for (i = 0; i < tun->numqueues; i++) {
2783                 tfile = rtnl_dereference(tun->tfiles[i]);
2784                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2785         }
2786 }
2787
2788 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2789 {
2790         struct tun_file *tfile = file->private_data;
2791         struct tun_struct *tun;
2792         int ret = 0;
2793
2794         rtnl_lock();
2795
2796         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2797                 tun = tfile->detached;
2798                 if (!tun) {
2799                         ret = -EINVAL;
2800                         goto unlock;
2801                 }
2802                 ret = security_tun_dev_attach_queue(tun->security);
2803                 if (ret < 0)
2804                         goto unlock;
2805                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2806         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2807                 tun = rtnl_dereference(tfile->tun);
2808                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2809                         ret = -EINVAL;
2810                 else
2811                         __tun_detach(tfile, false);
2812         } else
2813                 ret = -EINVAL;
2814
2815         if (ret >= 0)
2816                 netdev_state_change(tun->dev);
2817
2818 unlock:
2819         rtnl_unlock();
2820         return ret;
2821 }
2822
2823 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2824                         void __user *data)
2825 {
2826         struct bpf_prog *prog;
2827         int fd;
2828
2829         if (copy_from_user(&fd, data, sizeof(fd)))
2830                 return -EFAULT;
2831
2832         if (fd == -1) {
2833                 prog = NULL;
2834         } else {
2835                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2836                 if (IS_ERR(prog))
2837                         return PTR_ERR(prog);
2838         }
2839
2840         return __tun_set_ebpf(tun, prog_p, prog);
2841 }
2842
2843 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2844                             unsigned long arg, int ifreq_len)
2845 {
2846         struct tun_file *tfile = file->private_data;
2847         struct tun_struct *tun;
2848         void __user* argp = (void __user*)arg;
2849         struct ifreq ifr;
2850         struct net *net;
2851         kuid_t owner;
2852         kgid_t group;
2853         int sndbuf;
2854         int vnet_hdr_sz;
2855         unsigned int ifindex;
2856         int le;
2857         int ret;
2858         bool do_notify = false;
2859
2860         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2861             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2862                 if (copy_from_user(&ifr, argp, ifreq_len))
2863                         return -EFAULT;
2864         } else {
2865                 memset(&ifr, 0, sizeof(ifr));
2866         }
2867         if (cmd == TUNGETFEATURES) {
2868                 /* Currently this just means: "what IFF flags are valid?".
2869                  * This is needed because we never checked for invalid flags on
2870                  * TUNSETIFF.
2871                  */
2872                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2873                                 (unsigned int __user*)argp);
2874         } else if (cmd == TUNSETQUEUE)
2875                 return tun_set_queue(file, &ifr);
2876
2877         ret = 0;
2878         rtnl_lock();
2879
2880         tun = tun_get(tfile);
2881         net = sock_net(&tfile->sk);
2882         if (cmd == TUNSETIFF) {
2883                 ret = -EEXIST;
2884                 if (tun)
2885                         goto unlock;
2886
2887                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2888
2889                 ret = tun_set_iff(net, file, &ifr);
2890
2891                 if (ret)
2892                         goto unlock;
2893
2894                 if (copy_to_user(argp, &ifr, ifreq_len))
2895                         ret = -EFAULT;
2896                 goto unlock;
2897         }
2898         if (cmd == TUNSETIFINDEX) {
2899                 ret = -EPERM;
2900                 if (tun)
2901                         goto unlock;
2902
2903                 ret = -EFAULT;
2904                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2905                         goto unlock;
2906
2907                 ret = 0;
2908                 tfile->ifindex = ifindex;
2909                 goto unlock;
2910         }
2911         if (cmd == SIOCGSKNS) {
2912                 ret = -EPERM;
2913                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2914                         goto unlock;
2915
2916                 ret = open_related_ns(&net->ns, get_net_ns);
2917                 goto unlock;
2918         }
2919
2920         ret = -EBADFD;
2921         if (!tun)
2922                 goto unlock;
2923
2924         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2925
2926         ret = 0;
2927         switch (cmd) {
2928         case TUNGETIFF:
2929                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2930
2931                 if (tfile->detached)
2932                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2933                 if (!tfile->socket.sk->sk_filter)
2934                         ifr.ifr_flags |= IFF_NOFILTER;
2935
2936                 if (copy_to_user(argp, &ifr, ifreq_len))
2937                         ret = -EFAULT;
2938                 break;
2939
2940         case TUNSETNOCSUM:
2941                 /* Disable/Enable checksum */
2942
2943                 /* [unimplemented] */
2944                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2945                           arg ? "disabled" : "enabled");
2946                 break;
2947
2948         case TUNSETPERSIST:
2949                 /* Disable/Enable persist mode. Keep an extra reference to the
2950                  * module to prevent the module being unprobed.
2951                  */
2952                 if (arg && !(tun->flags & IFF_PERSIST)) {
2953                         tun->flags |= IFF_PERSIST;
2954                         __module_get(THIS_MODULE);
2955                         do_notify = true;
2956                 }
2957                 if (!arg && (tun->flags & IFF_PERSIST)) {
2958                         tun->flags &= ~IFF_PERSIST;
2959                         module_put(THIS_MODULE);
2960                         do_notify = true;
2961                 }
2962
2963                 tun_debug(KERN_INFO, tun, "persist %s\n",
2964                           arg ? "enabled" : "disabled");
2965                 break;
2966
2967         case TUNSETOWNER:
2968                 /* Set owner of the device */
2969                 owner = make_kuid(current_user_ns(), arg);
2970                 if (!uid_valid(owner)) {
2971                         ret = -EINVAL;
2972                         break;
2973                 }
2974                 tun->owner = owner;
2975                 do_notify = true;
2976                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2977                           from_kuid(&init_user_ns, tun->owner));
2978                 break;
2979
2980         case TUNSETGROUP:
2981                 /* Set group of the device */
2982                 group = make_kgid(current_user_ns(), arg);
2983                 if (!gid_valid(group)) {
2984                         ret = -EINVAL;
2985                         break;
2986                 }
2987                 tun->group = group;
2988                 do_notify = true;
2989                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2990                           from_kgid(&init_user_ns, tun->group));
2991                 break;
2992
2993         case TUNSETLINK:
2994                 /* Only allow setting the type when the interface is down */
2995                 if (tun->dev->flags & IFF_UP) {
2996                         tun_debug(KERN_INFO, tun,
2997                                   "Linktype set failed because interface is up\n");
2998                         ret = -EBUSY;
2999                 } else {
3000                         tun->dev->type = (int) arg;
3001                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3002                                   tun->dev->type);
3003                         ret = 0;
3004                 }
3005                 break;
3006
3007 #ifdef TUN_DEBUG
3008         case TUNSETDEBUG:
3009                 tun->debug = arg;
3010                 break;
3011 #endif
3012         case TUNSETOFFLOAD:
3013                 ret = set_offload(tun, arg);
3014                 break;
3015
3016         case TUNSETTXFILTER:
3017                 /* Can be set only for TAPs */
3018                 ret = -EINVAL;
3019                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3020                         break;
3021                 ret = update_filter(&tun->txflt, (void __user *)arg);
3022                 break;
3023
3024         case SIOCGIFHWADDR:
3025                 /* Get hw address */
3026                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3027                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3028                 if (copy_to_user(argp, &ifr, ifreq_len))
3029                         ret = -EFAULT;
3030                 break;
3031
3032         case SIOCSIFHWADDR:
3033                 /* Set hw address */
3034                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3035                           ifr.ifr_hwaddr.sa_data);
3036
3037                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3038                 break;
3039
3040         case TUNGETSNDBUF:
3041                 sndbuf = tfile->socket.sk->sk_sndbuf;
3042                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3043                         ret = -EFAULT;
3044                 break;
3045
3046         case TUNSETSNDBUF:
3047                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3048                         ret = -EFAULT;
3049                         break;
3050                 }
3051                 if (sndbuf <= 0) {
3052                         ret = -EINVAL;
3053                         break;
3054                 }
3055
3056                 tun->sndbuf = sndbuf;
3057                 tun_set_sndbuf(tun);
3058                 break;
3059
3060         case TUNGETVNETHDRSZ:
3061                 vnet_hdr_sz = tun->vnet_hdr_sz;
3062                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3063                         ret = -EFAULT;
3064                 break;
3065
3066         case TUNSETVNETHDRSZ:
3067                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3068                         ret = -EFAULT;
3069                         break;
3070                 }
3071                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3072                         ret = -EINVAL;
3073                         break;
3074                 }
3075
3076                 tun->vnet_hdr_sz = vnet_hdr_sz;
3077                 break;
3078
3079         case TUNGETVNETLE:
3080                 le = !!(tun->flags & TUN_VNET_LE);
3081                 if (put_user(le, (int __user *)argp))
3082                         ret = -EFAULT;
3083                 break;
3084
3085         case TUNSETVNETLE:
3086                 if (get_user(le, (int __user *)argp)) {
3087                         ret = -EFAULT;
3088                         break;
3089                 }
3090                 if (le)
3091                         tun->flags |= TUN_VNET_LE;
3092                 else
3093                         tun->flags &= ~TUN_VNET_LE;
3094                 break;
3095
3096         case TUNGETVNETBE:
3097                 ret = tun_get_vnet_be(tun, argp);
3098                 break;
3099
3100         case TUNSETVNETBE:
3101                 ret = tun_set_vnet_be(tun, argp);
3102                 break;
3103
3104         case TUNATTACHFILTER:
3105                 /* Can be set only for TAPs */
3106                 ret = -EINVAL;
3107                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3108                         break;
3109                 ret = -EFAULT;
3110                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3111                         break;
3112
3113                 ret = tun_attach_filter(tun);
3114                 break;
3115
3116         case TUNDETACHFILTER:
3117                 /* Can be set only for TAPs */
3118                 ret = -EINVAL;
3119                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3120                         break;
3121                 ret = 0;
3122                 tun_detach_filter(tun, tun->numqueues);
3123                 break;
3124
3125         case TUNGETFILTER:
3126                 ret = -EINVAL;
3127                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3128                         break;
3129                 ret = -EFAULT;
3130                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3131                         break;
3132                 ret = 0;
3133                 break;
3134
3135         case TUNSETSTEERINGEBPF:
3136                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3137                 break;
3138
3139         case TUNSETFILTEREBPF:
3140                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3141                 break;
3142
3143         default:
3144                 ret = -EINVAL;
3145                 break;
3146         }
3147
3148         if (do_notify)
3149                 netdev_state_change(tun->dev);
3150
3151 unlock:
3152         rtnl_unlock();
3153         if (tun)
3154                 tun_put(tun);
3155         return ret;
3156 }
3157
3158 static long tun_chr_ioctl(struct file *file,
3159                           unsigned int cmd, unsigned long arg)
3160 {
3161         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3162 }
3163
3164 #ifdef CONFIG_COMPAT
3165 static long tun_chr_compat_ioctl(struct file *file,
3166                          unsigned int cmd, unsigned long arg)
3167 {
3168         switch (cmd) {
3169         case TUNSETIFF:
3170         case TUNGETIFF:
3171         case TUNSETTXFILTER:
3172         case TUNGETSNDBUF:
3173         case TUNSETSNDBUF:
3174         case SIOCGIFHWADDR:
3175         case SIOCSIFHWADDR:
3176                 arg = (unsigned long)compat_ptr(arg);
3177                 break;
3178         default:
3179                 arg = (compat_ulong_t)arg;
3180                 break;
3181         }
3182
3183         /*
3184          * compat_ifreq is shorter than ifreq, so we must not access beyond
3185          * the end of that structure. All fields that are used in this
3186          * driver are compatible though, we don't need to convert the
3187          * contents.
3188          */
3189         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3190 }
3191 #endif /* CONFIG_COMPAT */
3192
3193 static int tun_chr_fasync(int fd, struct file *file, int on)
3194 {
3195         struct tun_file *tfile = file->private_data;
3196         int ret;
3197
3198         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3199                 goto out;
3200
3201         if (on) {
3202                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3203                 tfile->flags |= TUN_FASYNC;
3204         } else
3205                 tfile->flags &= ~TUN_FASYNC;
3206         ret = 0;
3207 out:
3208         return ret;
3209 }
3210
3211 static int tun_chr_open(struct inode *inode, struct file * file)
3212 {
3213         struct net *net = current->nsproxy->net_ns;
3214         struct tun_file *tfile;
3215
3216         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3217
3218         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3219                                             &tun_proto, 0);
3220         if (!tfile)
3221                 return -ENOMEM;
3222         RCU_INIT_POINTER(tfile->tun, NULL);
3223         tfile->flags = 0;
3224         tfile->ifindex = 0;
3225
3226         init_waitqueue_head(&tfile->wq.wait);
3227         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3228
3229         tfile->socket.file = file;
3230         tfile->socket.ops = &tun_socket_ops;
3231
3232         sock_init_data(&tfile->socket, &tfile->sk);
3233
3234         tfile->sk.sk_write_space = tun_sock_write_space;
3235         tfile->sk.sk_sndbuf = INT_MAX;
3236
3237         file->private_data = tfile;
3238         INIT_LIST_HEAD(&tfile->next);
3239
3240         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3241
3242         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3243
3244         return 0;
3245 }
3246
3247 static int tun_chr_close(struct inode *inode, struct file *file)
3248 {
3249         struct tun_file *tfile = file->private_data;
3250
3251         tun_detach(tfile, true);
3252
3253         return 0;
3254 }
3255
3256 #ifdef CONFIG_PROC_FS
3257 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3258 {
3259         struct tun_file *tfile = file->private_data;
3260         struct tun_struct *tun;
3261         struct ifreq ifr;
3262
3263         memset(&ifr, 0, sizeof(ifr));
3264
3265         rtnl_lock();
3266         tun = tun_get(tfile);
3267         if (tun)
3268                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3269         rtnl_unlock();
3270
3271         if (tun)
3272                 tun_put(tun);
3273
3274         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3275 }
3276 #endif
3277
3278 static const struct file_operations tun_fops = {
3279         .owner  = THIS_MODULE,
3280         .llseek = no_llseek,
3281         .read_iter  = tun_chr_read_iter,
3282         .write_iter = tun_chr_write_iter,
3283         .poll   = tun_chr_poll,
3284         .unlocked_ioctl = tun_chr_ioctl,
3285 #ifdef CONFIG_COMPAT
3286         .compat_ioctl = tun_chr_compat_ioctl,
3287 #endif
3288         .open   = tun_chr_open,
3289         .release = tun_chr_close,
3290         .fasync = tun_chr_fasync,
3291 #ifdef CONFIG_PROC_FS
3292         .show_fdinfo = tun_chr_show_fdinfo,
3293 #endif
3294 };
3295
3296 static struct miscdevice tun_miscdev = {
3297         .minor = TUN_MINOR,
3298         .name = "tun",
3299         .nodename = "net/tun",
3300         .fops = &tun_fops,
3301 };
3302
3303 /* ethtool interface */
3304
3305 static int tun_get_link_ksettings(struct net_device *dev,
3306                                   struct ethtool_link_ksettings *cmd)
3307 {
3308         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3309         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3310         cmd->base.speed         = SPEED_10;
3311         cmd->base.duplex        = DUPLEX_FULL;
3312         cmd->base.port          = PORT_TP;
3313         cmd->base.phy_address   = 0;
3314         cmd->base.autoneg       = AUTONEG_DISABLE;
3315         return 0;
3316 }
3317
3318 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3319 {
3320         struct tun_struct *tun = netdev_priv(dev);
3321
3322         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3323         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3324
3325         switch (tun->flags & TUN_TYPE_MASK) {
3326         case IFF_TUN:
3327                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3328                 break;
3329         case IFF_TAP:
3330                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3331                 break;
3332         }
3333 }
3334
3335 static u32 tun_get_msglevel(struct net_device *dev)
3336 {
3337 #ifdef TUN_DEBUG
3338         struct tun_struct *tun = netdev_priv(dev);
3339         return tun->debug;
3340 #else
3341         return -EOPNOTSUPP;
3342 #endif
3343 }
3344
3345 static void tun_set_msglevel(struct net_device *dev, u32 value)
3346 {
3347 #ifdef TUN_DEBUG
3348         struct tun_struct *tun = netdev_priv(dev);
3349         tun->debug = value;
3350 #endif
3351 }
3352
3353 static int tun_get_coalesce(struct net_device *dev,
3354                             struct ethtool_coalesce *ec)
3355 {
3356         struct tun_struct *tun = netdev_priv(dev);
3357
3358         ec->rx_max_coalesced_frames = tun->rx_batched;
3359
3360         return 0;
3361 }
3362
3363 static int tun_set_coalesce(struct net_device *dev,
3364                             struct ethtool_coalesce *ec)
3365 {
3366         struct tun_struct *tun = netdev_priv(dev);
3367
3368         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3369                 tun->rx_batched = NAPI_POLL_WEIGHT;
3370         else
3371                 tun->rx_batched = ec->rx_max_coalesced_frames;
3372
3373         return 0;
3374 }
3375
3376 static const struct ethtool_ops tun_ethtool_ops = {
3377         .get_drvinfo    = tun_get_drvinfo,
3378         .get_msglevel   = tun_get_msglevel,
3379         .set_msglevel   = tun_set_msglevel,
3380         .get_link       = ethtool_op_get_link,
3381         .get_ts_info    = ethtool_op_get_ts_info,
3382         .get_coalesce   = tun_get_coalesce,
3383         .set_coalesce   = tun_set_coalesce,
3384         .get_link_ksettings = tun_get_link_ksettings,
3385 };
3386
3387 static int tun_queue_resize(struct tun_struct *tun)
3388 {
3389         struct net_device *dev = tun->dev;
3390         struct tun_file *tfile;
3391         struct ptr_ring **rings;
3392         int n = tun->numqueues + tun->numdisabled;
3393         int ret, i;
3394
3395         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3396         if (!rings)
3397                 return -ENOMEM;
3398
3399         for (i = 0; i < tun->numqueues; i++) {
3400                 tfile = rtnl_dereference(tun->tfiles[i]);
3401                 rings[i] = &tfile->tx_ring;
3402         }
3403         list_for_each_entry(tfile, &tun->disabled, next)
3404                 rings[i++] = &tfile->tx_ring;
3405
3406         ret = ptr_ring_resize_multiple(rings, n,
3407                                        dev->tx_queue_len, GFP_KERNEL,
3408                                        tun_ptr_free);
3409
3410         kfree(rings);
3411         return ret;
3412 }
3413
3414 static int tun_device_event(struct notifier_block *unused,
3415                             unsigned long event, void *ptr)
3416 {
3417         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3418         struct tun_struct *tun = netdev_priv(dev);
3419
3420         if (dev->rtnl_link_ops != &tun_link_ops)
3421                 return NOTIFY_DONE;
3422
3423         switch (event) {
3424         case NETDEV_CHANGE_TX_QUEUE_LEN:
3425                 if (tun_queue_resize(tun))
3426                         return NOTIFY_BAD;
3427                 break;
3428         default:
3429                 break;
3430         }
3431
3432         return NOTIFY_DONE;
3433 }
3434
3435 static struct notifier_block tun_notifier_block __read_mostly = {
3436         .notifier_call  = tun_device_event,
3437 };
3438
3439 static int __init tun_init(void)
3440 {
3441         int ret = 0;
3442
3443         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3444
3445         ret = rtnl_link_register(&tun_link_ops);
3446         if (ret) {
3447                 pr_err("Can't register link_ops\n");
3448                 goto err_linkops;
3449         }
3450
3451         ret = misc_register(&tun_miscdev);
3452         if (ret) {
3453                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3454                 goto err_misc;
3455         }
3456
3457         ret = register_netdevice_notifier(&tun_notifier_block);
3458         if (ret) {
3459                 pr_err("Can't register netdevice notifier\n");
3460                 goto err_notifier;
3461         }
3462
3463         return  0;
3464
3465 err_notifier:
3466         misc_deregister(&tun_miscdev);
3467 err_misc:
3468         rtnl_link_unregister(&tun_link_ops);
3469 err_linkops:
3470         return ret;
3471 }
3472
3473 static void tun_cleanup(void)
3474 {
3475         misc_deregister(&tun_miscdev);
3476         rtnl_link_unregister(&tun_link_ops);
3477         unregister_netdevice_notifier(&tun_notifier_block);
3478 }
3479
3480 /* Get an underlying socket object from tun file.  Returns error unless file is
3481  * attached to a device.  The returned object works like a packet socket, it
3482  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3483  * holding a reference to the file for as long as the socket is in use. */
3484 struct socket *tun_get_socket(struct file *file)
3485 {
3486         struct tun_file *tfile;
3487         if (file->f_op != &tun_fops)
3488                 return ERR_PTR(-EINVAL);
3489         tfile = file->private_data;
3490         if (!tfile)
3491                 return ERR_PTR(-EBADFD);
3492         return &tfile->socket;
3493 }
3494 EXPORT_SYMBOL_GPL(tun_get_socket);
3495
3496 struct ptr_ring *tun_get_tx_ring(struct file *file)
3497 {
3498         struct tun_file *tfile;
3499
3500         if (file->f_op != &tun_fops)
3501                 return ERR_PTR(-EINVAL);
3502         tfile = file->private_data;
3503         if (!tfile)
3504                 return ERR_PTR(-EBADFD);
3505         return &tfile->tx_ring;
3506 }
3507 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3508
3509 module_init(tun_init);
3510 module_exit(tun_cleanup);
3511 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3512 MODULE_AUTHOR(DRV_COPYRIGHT);
3513 MODULE_LICENSE("GPL");
3514 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3515 MODULE_ALIAS("devname:net/tun");