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_frame(void *ptr)
 252 {
 253         return (unsigned long)ptr & TUN_XDP_FLAG;
 254 }
 255 EXPORT_SYMBOL(tun_is_xdp_frame);
 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         e = tun_flow_find(head, rxhash);
 529         if (likely(e)) {
 530                 /* TODO: keep queueing to old queue until it's empty? */
 531                 e->queue_index = queue_index;
 532                 e->updated = jiffies;
 533                 sock_rps_record_flow_hash(e->rps_rxhash);
 534         } else {
 535                 spin_lock_bh(&tun->lock);
 536                 if (!tun_flow_find(head, rxhash) &&
 537                     tun->flow_count < MAX_TAP_FLOWS)
 538                         tun_flow_create(tun, head, rxhash, queue_index);
 539
 540                 if (!timer_pending(&tun->flow_gc_timer))
 541                         mod_timer(&tun->flow_gc_timer,
 542                                   round_jiffies_up(jiffies + delay));
 543                 spin_unlock_bh(&tun->lock);
 544         }
 545
 546         rcu_read_unlock();
 547 }
 548
 549 /**
 550  * Save the hash received in the stack receive path and update the
 551  * flow_hash table accordingly.
 552  */
 553 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 554 {
 555         if (unlikely(e->rps_rxhash != hash))
 556                 e->rps_rxhash = hash;
 557 }
 558
 559 /* We try to identify a flow through its rxhash first. The reason that
 560  * we do not check rxq no. is because some cards(e.g 82599), chooses
 561  * the rxq based on the txq where the last packet of the flow comes. As
 562  * the userspace application move between processors, we may get a
 563  * different rxq no. here. If we could not get rxhash, then we would
 564  * hope the rxq no. may help here.
 565  */
 566 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 567 {
 568         struct tun_flow_entry *e;
 569         u32 txq = 0;
 570         u32 numqueues = 0;
 571
 572         numqueues = READ_ONCE(tun->numqueues);
 573
 574         txq = __skb_get_hash_symmetric(skb);
 575         if (txq) {
 576                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 577                 if (e) {
 578                         tun_flow_save_rps_rxhash(e, txq);
 579                         txq = e->queue_index;
 580                 } else
 581                         /* use multiply and shift instead of expensive divide */
 582                         txq = ((u64)txq * numqueues) >> 32;
 583         } else if (likely(skb_rx_queue_recorded(skb))) {
 584                 txq = skb_get_rx_queue(skb);
 585                 while (unlikely(txq >= numqueues))
 586                         txq -= numqueues;
 587         }
 588
 589         return txq;
 590 }
 591
 592 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 593 {
 594         struct tun_prog *prog;
 595         u16 ret = 0;
 596
 597         prog = rcu_dereference(tun->steering_prog);
 598         if (prog)
 599                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 600
 601         return ret % tun->numqueues;
 602 }
 603
 604 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 605                             void *accel_priv, select_queue_fallback_t fallback)
 606 {
 607         struct tun_struct *tun = netdev_priv(dev);
 608         u16 ret;
 609
 610         rcu_read_lock();
 611         if (rcu_dereference(tun->steering_prog))
 612                 ret = tun_ebpf_select_queue(tun, skb);
 613         else
 614                 ret = tun_automq_select_queue(tun, skb);
 615         rcu_read_unlock();
 616
 617         return ret;
 618 }
 619
 620 static inline bool tun_not_capable(struct tun_struct *tun)
 621 {
 622         const struct cred *cred = current_cred();
 623         struct net *net = dev_net(tun->dev);
 624
 625         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 626                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 627                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 628 }
 629
 630 static void tun_set_real_num_queues(struct tun_struct *tun)
 631 {
 632         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 633         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 634 }
 635
 636 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 637 {
 638         tfile->detached = tun;
 639         list_add_tail(&tfile->next, &tun->disabled);
 640         ++tun->numdisabled;
 641 }
 642
 643 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 644 {
 645         struct tun_struct *tun = tfile->detached;
 646
 647         tfile->detached = NULL;
 648         list_del_init(&tfile->next);
 649         --tun->numdisabled;
 650         return tun;
 651 }
 652
 653 void tun_ptr_free(void *ptr)
 654 {
 655         if (!ptr)
 656                 return;
 657         if (tun_is_xdp_frame(ptr)) {
 658                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 659
 660                 xdp_return_frame(xdpf);
 661         } else {
 662                 __skb_array_destroy_skb(ptr);
 663         }
 664 }
 665 EXPORT_SYMBOL_GPL(tun_ptr_free);
 666
 667 static void tun_queue_purge(struct tun_file *tfile)
 668 {
 669         void *ptr;
 670
 671         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 672                 tun_ptr_free(ptr);
 673
 674         skb_queue_purge(&tfile->sk.sk_write_queue);
 675         skb_queue_purge(&tfile->sk.sk_error_queue);
 676 }
 677
 678 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 679 {
 680         if (tfile->tx_ring.queue) {
 681                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 682                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 683                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 684         }
 685 }
 686
 687 static void __tun_detach(struct tun_file *tfile, bool clean)
 688 {
 689         struct tun_file *ntfile;
 690         struct tun_struct *tun;
 691
 692         tun = rtnl_dereference(tfile->tun);
 693
 694         if (tun && clean) {
 695                 tun_napi_disable(tun, tfile);
 696                 tun_napi_del(tun, tfile);
 697         }
 698
 699         if (tun && !tfile->detached) {
 700                 u16 index = tfile->queue_index;
 701                 BUG_ON(index >= tun->numqueues);
 702
 703                 rcu_assign_pointer(tun->tfiles[index],
 704                                    tun->tfiles[tun->numqueues - 1]);
 705                 ntfile = rtnl_dereference(tun->tfiles[index]);
 706                 ntfile->queue_index = index;
 707
 708                 --tun->numqueues;
 709                 if (clean) {
 710                         RCU_INIT_POINTER(tfile->tun, NULL);
 711                         sock_put(&tfile->sk);
 712                 } else
 713                         tun_disable_queue(tun, tfile);
 714
 715                 synchronize_net();
 716                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 717                 /* Drop read queue */
 718                 tun_queue_purge(tfile);
 719                 tun_set_real_num_queues(tun);
 720         } else if (tfile->detached && clean) {
 721                 tun = tun_enable_queue(tfile);
 722                 sock_put(&tfile->sk);
 723         }
 724
 725         if (clean) {
 726                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 727                         netif_carrier_off(tun->dev);
 728
 729                         if (!(tun->flags & IFF_PERSIST) &&
 730                             tun->dev->reg_state == NETREG_REGISTERED)
 731                                 unregister_netdevice(tun->dev);
 732                 }
 733                 tun_cleanup_tx_ring(tfile);
 734                 sock_put(&tfile->sk);
 735         }
 736 }
 737
 738 static void tun_detach(struct tun_file *tfile, bool clean)
 739 {
 740         struct tun_struct *tun;
 741         struct net_device *dev;
 742
 743         rtnl_lock();
 744         tun = rtnl_dereference(tfile->tun);
 745         dev = tun ? tun->dev : NULL;
 746         __tun_detach(tfile, clean);
 747         if (dev)
 748                 netdev_state_change(dev);
 749         rtnl_unlock();
 750 }
 751
 752 static void tun_detach_all(struct net_device *dev)
 753 {
 754         struct tun_struct *tun = netdev_priv(dev);
 755         struct tun_file *tfile, *tmp;
 756         int i, n = tun->numqueues;
 757
 758         for (i = 0; i < n; i++) {
 759                 tfile = rtnl_dereference(tun->tfiles[i]);
 760                 BUG_ON(!tfile);
 761                 tun_napi_disable(tun, tfile);
 762                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 763                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 764                 RCU_INIT_POINTER(tfile->tun, NULL);
 765                 --tun->numqueues;
 766         }
 767         list_for_each_entry(tfile, &tun->disabled, next) {
 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         }
 772         BUG_ON(tun->numqueues != 0);
 773
 774         synchronize_net();
 775         for (i = 0; i < n; i++) {
 776                 tfile = rtnl_dereference(tun->tfiles[i]);
 777                 tun_napi_del(tun, tfile);
 778                 /* Drop read queue */
 779                 tun_queue_purge(tfile);
 780                 sock_put(&tfile->sk);
 781                 tun_cleanup_tx_ring(tfile);
 782         }
 783         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 784                 tun_enable_queue(tfile);
 785                 tun_queue_purge(tfile);
 786                 sock_put(&tfile->sk);
 787                 tun_cleanup_tx_ring(tfile);
 788         }
 789         BUG_ON(tun->numdisabled != 0);
 790
 791         if (tun->flags & IFF_PERSIST)
 792                 module_put(THIS_MODULE);
 793 }
 794
 795 static int tun_attach(struct tun_struct *tun, struct file *file,
 796                       bool skip_filter, bool napi)
 797 {
 798         struct tun_file *tfile = file->private_data;
 799         struct net_device *dev = tun->dev;
 800         int err;
 801
 802         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 803         if (err < 0)
 804                 goto out;
 805
 806         err = -EINVAL;
 807         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 808                 goto out;
 809
 810         err = -EBUSY;
 811         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 812                 goto out;
 813
 814         err = -E2BIG;
 815         if (!tfile->detached &&
 816             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 817                 goto out;
 818
 819         err = 0;
 820
 821         /* Re-attach the filter to persist device */
 822         if (!skip_filter && (tun->filter_attached == true)) {
 823                 lock_sock(tfile->socket.sk);
 824                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 825                 release_sock(tfile->socket.sk);
 826                 if (!err)
 827                         goto out;
 828         }
 829
 830         if (!tfile->detached &&
 831             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 832                 err = -ENOMEM;
 833                 goto out;
 834         }
 835
 836         tfile->queue_index = tun->numqueues;
 837         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 838
 839         if (tfile->detached) {
 840                 /* Re-attach detached tfile, updating XDP queue_index */
 841                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 842
 843                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 844                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 845         } else {
 846                 /* Setup XDP RX-queue info, for new tfile getting attached */
 847                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 848                                        tun->dev, tfile->queue_index);
 849                 if (err < 0)
 850                         goto out;
 851                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 852                                                  MEM_TYPE_PAGE_SHARED, NULL);
 853                 if (err < 0) {
 854                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 855                         goto out;
 856                 }
 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_frame *frame)
1294 {
1295         struct tun_struct *tun = netdev_priv(dev);
1296         struct tun_file *tfile;
1297         u32 numqueues;
1298         int ret = 0;
1299
1300         rcu_read_lock();
1301
1302         numqueues = READ_ONCE(tun->numqueues);
1303         if (!numqueues) {
1304                 ret = -ENOSPC;
1305                 goto out;
1306         }
1307
1308         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1309                                             numqueues]);
1310         /* Encode the XDP flag into lowest bit for consumer to differ
1311          * XDP buffer from sk_buff.
1312          */
1313         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(frame))) {
1314                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1315                 ret = -ENOSPC;
1316         }
1317
1318 out:
1319         rcu_read_unlock();
1320         return ret;
1321 }
1322
1323 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1324 {
1325         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1326
1327         if (unlikely(!frame))
1328                 return -EOVERFLOW;
1329
1330         return tun_xdp_xmit(dev, frame);
1331 }
1332
1333 static void tun_xdp_flush(struct net_device *dev)
1334 {
1335         struct tun_struct *tun = netdev_priv(dev);
1336         struct tun_file *tfile;
1337         u32 numqueues;
1338
1339         rcu_read_lock();
1340
1341         numqueues = READ_ONCE(tun->numqueues);
1342         if (!numqueues)
1343                 goto out;
1344
1345         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1346                                             numqueues]);
1347         /* Notify and wake up reader process */
1348         if (tfile->flags & TUN_FASYNC)
1349                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1350         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1351
1352 out:
1353         rcu_read_unlock();
1354 }
1355
1356 static const struct net_device_ops tap_netdev_ops = {
1357         .ndo_uninit             = tun_net_uninit,
1358         .ndo_open               = tun_net_open,
1359         .ndo_stop               = tun_net_close,
1360         .ndo_start_xmit         = tun_net_xmit,
1361         .ndo_fix_features       = tun_net_fix_features,
1362         .ndo_set_rx_mode        = tun_net_mclist,
1363         .ndo_set_mac_address    = eth_mac_addr,
1364         .ndo_validate_addr      = eth_validate_addr,
1365         .ndo_select_queue       = tun_select_queue,
1366 #ifdef CONFIG_NET_POLL_CONTROLLER
1367         .ndo_poll_controller    = tun_poll_controller,
1368 #endif
1369         .ndo_features_check     = passthru_features_check,
1370         .ndo_set_rx_headroom    = tun_set_headroom,
1371         .ndo_get_stats64        = tun_net_get_stats64,
1372         .ndo_bpf                = tun_xdp,
1373         .ndo_xdp_xmit           = tun_xdp_xmit,
1374         .ndo_xdp_flush          = tun_xdp_flush,
1375 };
1376
1377 static void tun_flow_init(struct tun_struct *tun)
1378 {
1379         int i;
1380
1381         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1382                 INIT_HLIST_HEAD(&tun->flows[i]);
1383
1384         tun->ageing_time = TUN_FLOW_EXPIRE;
1385         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1386         mod_timer(&tun->flow_gc_timer,
1387                   round_jiffies_up(jiffies + tun->ageing_time));
1388 }
1389
1390 static void tun_flow_uninit(struct tun_struct *tun)
1391 {
1392         del_timer_sync(&tun->flow_gc_timer);
1393         tun_flow_flush(tun);
1394 }
1395
1396 #define MIN_MTU 68
1397 #define MAX_MTU 65535
1398
1399 /* Initialize net device. */
1400 static void tun_net_init(struct net_device *dev)
1401 {
1402         struct tun_struct *tun = netdev_priv(dev);
1403
1404         switch (tun->flags & TUN_TYPE_MASK) {
1405         case IFF_TUN:
1406                 dev->netdev_ops = &tun_netdev_ops;
1407
1408                 /* Point-to-Point TUN Device */
1409                 dev->hard_header_len = 0;
1410                 dev->addr_len = 0;
1411                 dev->mtu = 1500;
1412
1413                 /* Zero header length */
1414                 dev->type = ARPHRD_NONE;
1415                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1416                 break;
1417
1418         case IFF_TAP:
1419                 dev->netdev_ops = &tap_netdev_ops;
1420                 /* Ethernet TAP Device */
1421                 ether_setup(dev);
1422                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1423                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1424
1425                 eth_hw_addr_random(dev);
1426
1427                 break;
1428         }
1429
1430         dev->min_mtu = MIN_MTU;
1431         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1432 }
1433
1434 /* Character device part */
1435
1436 /* Poll */
1437 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1438 {
1439         struct tun_file *tfile = file->private_data;
1440         struct tun_struct *tun = tun_get(tfile);
1441         struct sock *sk;
1442         __poll_t mask = 0;
1443
1444         if (!tun)
1445                 return EPOLLERR;
1446
1447         sk = tfile->socket.sk;
1448
1449         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1450
1451         poll_wait(file, sk_sleep(sk), wait);
1452
1453         if (!ptr_ring_empty(&tfile->tx_ring))
1454                 mask |= EPOLLIN | EPOLLRDNORM;
1455
1456         if (tun->dev->flags & IFF_UP &&
1457             (sock_writeable(sk) ||
1458              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1459               sock_writeable(sk))))
1460                 mask |= EPOLLOUT | EPOLLWRNORM;
1461
1462         if (tun->dev->reg_state != NETREG_REGISTERED)
1463                 mask = EPOLLERR;
1464
1465         tun_put(tun);
1466         return mask;
1467 }
1468
1469 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1470                                             size_t len,
1471                                             const struct iov_iter *it)
1472 {
1473         struct sk_buff *skb;
1474         size_t linear;
1475         int err;
1476         int i;
1477
1478         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1479                 return ERR_PTR(-ENOMEM);
1480
1481         local_bh_disable();
1482         skb = napi_get_frags(&tfile->napi);
1483         local_bh_enable();
1484         if (!skb)
1485                 return ERR_PTR(-ENOMEM);
1486
1487         linear = iov_iter_single_seg_count(it);
1488         err = __skb_grow(skb, linear);
1489         if (err)
1490                 goto free;
1491
1492         skb->len = len;
1493         skb->data_len = len - linear;
1494         skb->truesize += skb->data_len;
1495
1496         for (i = 1; i < it->nr_segs; i++) {
1497                 struct page_frag *pfrag = &current->task_frag;
1498                 size_t fragsz = it->iov[i].iov_len;
1499
1500                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1501                         err = -EINVAL;
1502                         goto free;
1503                 }
1504
1505                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1506                         err = -ENOMEM;
1507                         goto free;
1508                 }
1509
1510                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1511                                    pfrag->offset, fragsz);
1512                 page_ref_inc(pfrag->page);
1513                 pfrag->offset += fragsz;
1514         }
1515
1516         return skb;
1517 free:
1518         /* frees skb and all frags allocated with napi_alloc_frag() */
1519         napi_free_frags(&tfile->napi);
1520         return ERR_PTR(err);
1521 }
1522
1523 /* prepad is the amount to reserve at front.  len is length after that.
1524  * linear is a hint as to how much to copy (usually headers). */
1525 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1526                                      size_t prepad, size_t len,
1527                                      size_t linear, int noblock)
1528 {
1529         struct sock *sk = tfile->socket.sk;
1530         struct sk_buff *skb;
1531         int err;
1532
1533         /* Under a page?  Don't bother with paged skb. */
1534         if (prepad + len < PAGE_SIZE || !linear)
1535                 linear = len;
1536
1537         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1538                                    &err, 0);
1539         if (!skb)
1540                 return ERR_PTR(err);
1541
1542         skb_reserve(skb, prepad);
1543         skb_put(skb, linear);
1544         skb->data_len = len - linear;
1545         skb->len += len - linear;
1546
1547         return skb;
1548 }
1549
1550 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1551                            struct sk_buff *skb, int more)
1552 {
1553         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1554         struct sk_buff_head process_queue;
1555         u32 rx_batched = tun->rx_batched;
1556         bool rcv = false;
1557
1558         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1559                 local_bh_disable();
1560                 netif_receive_skb(skb);
1561                 local_bh_enable();
1562                 return;
1563         }
1564
1565         spin_lock(&queue->lock);
1566         if (!more || skb_queue_len(queue) == rx_batched) {
1567                 __skb_queue_head_init(&process_queue);
1568                 skb_queue_splice_tail_init(queue, &process_queue);
1569                 rcv = true;
1570         } else {
1571                 __skb_queue_tail(queue, skb);
1572         }
1573         spin_unlock(&queue->lock);
1574
1575         if (rcv) {
1576                 struct sk_buff *nskb;
1577
1578                 local_bh_disable();
1579                 while ((nskb = __skb_dequeue(&process_queue)))
1580                         netif_receive_skb(nskb);
1581                 netif_receive_skb(skb);
1582                 local_bh_enable();
1583         }
1584 }
1585
1586 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1587                               int len, int noblock, bool zerocopy)
1588 {
1589         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1590                 return false;
1591
1592         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1593                 return false;
1594
1595         if (!noblock)
1596                 return false;
1597
1598         if (zerocopy)
1599                 return false;
1600
1601         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1602             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1603                 return false;
1604
1605         return true;
1606 }
1607
1608 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1609                                      struct tun_file *tfile,
1610                                      struct iov_iter *from,
1611                                      struct virtio_net_hdr *hdr,
1612                                      int len, int *skb_xdp)
1613 {
1614         struct page_frag *alloc_frag = &current->task_frag;
1615         struct sk_buff *skb;
1616         struct bpf_prog *xdp_prog;
1617         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1618         unsigned int delta = 0;
1619         char *buf;
1620         size_t copied;
1621         int err, pad = TUN_RX_PAD;
1622
1623         rcu_read_lock();
1624         xdp_prog = rcu_dereference(tun->xdp_prog);
1625         if (xdp_prog)
1626                 pad += TUN_HEADROOM;
1627         buflen += SKB_DATA_ALIGN(len + pad);
1628         rcu_read_unlock();
1629
1630         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1631         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1632                 return ERR_PTR(-ENOMEM);
1633
1634         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1635         copied = copy_page_from_iter(alloc_frag->page,
1636                                      alloc_frag->offset + pad,
1637                                      len, from);
1638         if (copied != len)
1639                 return ERR_PTR(-EFAULT);
1640
1641         /* There's a small window that XDP may be set after the check
1642          * of xdp_prog above, this should be rare and for simplicity
1643          * we do XDP on skb in case the headroom is not enough.
1644          */
1645         if (hdr->gso_type || !xdp_prog)
1646                 *skb_xdp = 1;
1647         else
1648                 *skb_xdp = 0;
1649
1650         preempt_disable();
1651         rcu_read_lock();
1652         xdp_prog = rcu_dereference(tun->xdp_prog);
1653         if (xdp_prog && !*skb_xdp) {
1654                 struct xdp_buff xdp;
1655                 void *orig_data;
1656                 u32 act;
1657
1658                 xdp.data_hard_start = buf;
1659                 xdp.data = buf + pad;
1660                 xdp_set_data_meta_invalid(&xdp);
1661                 xdp.data_end = xdp.data + len;
1662                 xdp.rxq = &tfile->xdp_rxq;
1663                 orig_data = xdp.data;
1664                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1665
1666                 switch (act) {
1667                 case XDP_REDIRECT:
1668                         get_page(alloc_frag->page);
1669                         alloc_frag->offset += buflen;
1670                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1671                         xdp_do_flush_map();
1672                         if (err)
1673                                 goto err_redirect;
1674                         rcu_read_unlock();
1675                         preempt_enable();
1676                         return NULL;
1677                 case XDP_TX:
1678                         get_page(alloc_frag->page);
1679                         alloc_frag->offset += buflen;
1680                         if (tun_xdp_tx(tun->dev, &xdp))
1681                                 goto err_redirect;
1682                         tun_xdp_flush(tun->dev);
1683                         rcu_read_unlock();
1684                         preempt_enable();
1685                         return NULL;
1686                 case XDP_PASS:
1687                         delta = orig_data - xdp.data;
1688                         len = xdp.data_end - xdp.data;
1689                         break;
1690                 default:
1691                         bpf_warn_invalid_xdp_action(act);
1692                         /* fall through */
1693                 case XDP_ABORTED:
1694                         trace_xdp_exception(tun->dev, xdp_prog, act);
1695                         /* fall through */
1696                 case XDP_DROP:
1697                         goto err_xdp;
1698                 }
1699         }
1700
1701         skb = build_skb(buf, buflen);
1702         if (!skb) {
1703                 rcu_read_unlock();
1704                 preempt_enable();
1705                 return ERR_PTR(-ENOMEM);
1706         }
1707
1708         skb_reserve(skb, pad - delta);
1709         skb_put(skb, len);
1710         get_page(alloc_frag->page);
1711         alloc_frag->offset += buflen;
1712
1713         rcu_read_unlock();
1714         preempt_enable();
1715
1716         return skb;
1717
1718 err_redirect:
1719         put_page(alloc_frag->page);
1720 err_xdp:
1721         rcu_read_unlock();
1722         preempt_enable();
1723         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1724         return NULL;
1725 }
1726
1727 /* Get packet from user space buffer */
1728 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1729                             void *msg_control, struct iov_iter *from,
1730                             int noblock, bool more)
1731 {
1732         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1733         struct sk_buff *skb;
1734         size_t total_len = iov_iter_count(from);
1735         size_t len = total_len, align = tun->align, linear;
1736         struct virtio_net_hdr gso = { 0 };
1737         struct tun_pcpu_stats *stats;
1738         int good_linear;
1739         int copylen;
1740         bool zerocopy = false;
1741         int err;
1742         u32 rxhash = 0;
1743         int skb_xdp = 1;
1744         bool frags = tun_napi_frags_enabled(tun);
1745
1746         if (!(tun->dev->flags & IFF_UP))
1747                 return -EIO;
1748
1749         if (!(tun->flags & IFF_NO_PI)) {
1750                 if (len < sizeof(pi))
1751                         return -EINVAL;
1752                 len -= sizeof(pi);
1753
1754                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1755                         return -EFAULT;
1756         }
1757
1758         if (tun->flags & IFF_VNET_HDR) {
1759                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1760
1761                 if (len < vnet_hdr_sz)
1762                         return -EINVAL;
1763                 len -= vnet_hdr_sz;
1764
1765                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1766                         return -EFAULT;
1767
1768                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1769                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1770                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1771
1772                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1773                         return -EINVAL;
1774                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1775         }
1776
1777         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1778                 align += NET_IP_ALIGN;
1779                 if (unlikely(len < ETH_HLEN ||
1780                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1781                         return -EINVAL;
1782         }
1783
1784         good_linear = SKB_MAX_HEAD(align);
1785
1786         if (msg_control) {
1787                 struct iov_iter i = *from;
1788
1789                 /* There are 256 bytes to be copied in skb, so there is
1790                  * enough room for skb expand head in case it is used.
1791                  * The rest of the buffer is mapped from userspace.
1792                  */
1793                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1794                 if (copylen > good_linear)
1795                         copylen = good_linear;
1796                 linear = copylen;
1797                 iov_iter_advance(&i, copylen);
1798                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1799                         zerocopy = true;
1800         }
1801
1802         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1803                 /* For the packet that is not easy to be processed
1804                  * (e.g gso or jumbo packet), we will do it at after
1805                  * skb was created with generic XDP routine.
1806                  */
1807                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1808                 if (IS_ERR(skb)) {
1809                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1810                         return PTR_ERR(skb);
1811                 }
1812                 if (!skb)
1813                         return total_len;
1814         } else {
1815                 if (!zerocopy) {
1816                         copylen = len;
1817                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1818                                 linear = good_linear;
1819                         else
1820                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1821                 }
1822
1823                 if (frags) {
1824                         mutex_lock(&tfile->napi_mutex);
1825                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1826                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1827                          * If zerocopy is enabled, then this layout will be
1828                          * overwritten by zerocopy_sg_from_iter().
1829                          */
1830                         zerocopy = false;
1831                 } else {
1832                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1833                                             noblock);
1834                 }
1835
1836                 if (IS_ERR(skb)) {
1837                         if (PTR_ERR(skb) != -EAGAIN)
1838                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1839                         if (frags)
1840                                 mutex_unlock(&tfile->napi_mutex);
1841                         return PTR_ERR(skb);
1842                 }
1843
1844                 if (zerocopy)
1845                         err = zerocopy_sg_from_iter(skb, from);
1846                 else
1847                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1848
1849                 if (err) {
1850                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1851                         kfree_skb(skb);
1852                         if (frags) {
1853                                 tfile->napi.skb = NULL;
1854                                 mutex_unlock(&tfile->napi_mutex);
1855                         }
1856
1857                         return -EFAULT;
1858                 }
1859         }
1860
1861         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1862                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1863                 kfree_skb(skb);
1864                 if (frags) {
1865                         tfile->napi.skb = NULL;
1866                         mutex_unlock(&tfile->napi_mutex);
1867                 }
1868
1869                 return -EINVAL;
1870         }
1871
1872         switch (tun->flags & TUN_TYPE_MASK) {
1873         case IFF_TUN:
1874                 if (tun->flags & IFF_NO_PI) {
1875                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1876
1877                         switch (ip_version) {
1878                         case 4:
1879                                 pi.proto = htons(ETH_P_IP);
1880                                 break;
1881                         case 6:
1882                                 pi.proto = htons(ETH_P_IPV6);
1883                                 break;
1884                         default:
1885                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1886                                 kfree_skb(skb);
1887                                 return -EINVAL;
1888                         }
1889                 }
1890
1891                 skb_reset_mac_header(skb);
1892                 skb->protocol = pi.proto;
1893                 skb->dev = tun->dev;
1894                 break;
1895         case IFF_TAP:
1896                 if (!frags)
1897                         skb->protocol = eth_type_trans(skb, tun->dev);
1898                 break;
1899         }
1900
1901         /* copy skb_ubuf_info for callback when skb has no error */
1902         if (zerocopy) {
1903                 skb_shinfo(skb)->destructor_arg = msg_control;
1904                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1905                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1906         } else if (msg_control) {
1907                 struct ubuf_info *uarg = msg_control;
1908                 uarg->callback(uarg, false);
1909         }
1910
1911         skb_reset_network_header(skb);
1912         skb_probe_transport_header(skb, 0);
1913
1914         if (skb_xdp) {
1915                 struct bpf_prog *xdp_prog;
1916                 int ret;
1917
1918                 rcu_read_lock();
1919                 xdp_prog = rcu_dereference(tun->xdp_prog);
1920                 if (xdp_prog) {
1921                         ret = do_xdp_generic(xdp_prog, skb);
1922                         if (ret != XDP_PASS) {
1923                                 rcu_read_unlock();
1924                                 return total_len;
1925                         }
1926                 }
1927                 rcu_read_unlock();
1928         }
1929
1930         /* Compute the costly rx hash only if needed for flow updates.
1931          * We may get a very small possibility of OOO during switching, not
1932          * worth to optimize.
1933          */
1934         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1935             !tfile->detached)
1936                 rxhash = __skb_get_hash_symmetric(skb);
1937
1938         if (frags) {
1939                 /* Exercise flow dissector code path. */
1940                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1941
1942                 if (unlikely(headlen > skb_headlen(skb))) {
1943                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1944                         napi_free_frags(&tfile->napi);
1945                         mutex_unlock(&tfile->napi_mutex);
1946                         WARN_ON(1);
1947                         return -ENOMEM;
1948                 }
1949
1950                 local_bh_disable();
1951                 napi_gro_frags(&tfile->napi);
1952                 local_bh_enable();
1953                 mutex_unlock(&tfile->napi_mutex);
1954         } else if (tfile->napi_enabled) {
1955                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1956                 int queue_len;
1957
1958                 spin_lock_bh(&queue->lock);
1959                 __skb_queue_tail(queue, skb);
1960                 queue_len = skb_queue_len(queue);
1961                 spin_unlock(&queue->lock);
1962
1963                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1964                         napi_schedule(&tfile->napi);
1965
1966                 local_bh_enable();
1967         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1968                 tun_rx_batched(tun, tfile, skb, more);
1969         } else {
1970                 netif_rx_ni(skb);
1971         }
1972
1973         stats = get_cpu_ptr(tun->pcpu_stats);
1974         u64_stats_update_begin(&stats->syncp);
1975         stats->rx_packets++;
1976         stats->rx_bytes += len;
1977         u64_stats_update_end(&stats->syncp);
1978         put_cpu_ptr(stats);
1979
1980         if (rxhash)
1981                 tun_flow_update(tun, rxhash, tfile);
1982
1983         return total_len;
1984 }
1985
1986 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1987 {
1988         struct file *file = iocb->ki_filp;
1989         struct tun_file *tfile = file->private_data;
1990         struct tun_struct *tun = tun_get(tfile);
1991         ssize_t result;
1992
1993         if (!tun)
1994                 return -EBADFD;
1995
1996         result = tun_get_user(tun, tfile, NULL, from,
1997                               file->f_flags & O_NONBLOCK, false);
1998
1999         tun_put(tun);
2000         return result;
2001 }
2002
2003 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2004                                 struct tun_file *tfile,
2005                                 struct xdp_frame *xdp_frame,
2006                                 struct iov_iter *iter)
2007 {
2008         int vnet_hdr_sz = 0;
2009         size_t size = xdp_frame->len;
2010         struct tun_pcpu_stats *stats;
2011         size_t ret;
2012
2013         if (tun->flags & IFF_VNET_HDR) {
2014                 struct virtio_net_hdr gso = { 0 };
2015
2016                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2017                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2018                         return -EINVAL;
2019                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2020                              sizeof(gso)))
2021                         return -EFAULT;
2022                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2023         }
2024
2025         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2026
2027         stats = get_cpu_ptr(tun->pcpu_stats);
2028         u64_stats_update_begin(&stats->syncp);
2029         stats->tx_packets++;
2030         stats->tx_bytes += ret;
2031         u64_stats_update_end(&stats->syncp);
2032         put_cpu_ptr(tun->pcpu_stats);
2033
2034         return ret;
2035 }
2036
2037 /* Put packet to the user space buffer */
2038 static ssize_t tun_put_user(struct tun_struct *tun,
2039                             struct tun_file *tfile,
2040                             struct sk_buff *skb,
2041                             struct iov_iter *iter)
2042 {
2043         struct tun_pi pi = { 0, skb->protocol };
2044         struct tun_pcpu_stats *stats;
2045         ssize_t total;
2046         int vlan_offset = 0;
2047         int vlan_hlen = 0;
2048         int vnet_hdr_sz = 0;
2049
2050         if (skb_vlan_tag_present(skb))
2051                 vlan_hlen = VLAN_HLEN;
2052
2053         if (tun->flags & IFF_VNET_HDR)
2054                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2055
2056         total = skb->len + vlan_hlen + vnet_hdr_sz;
2057
2058         if (!(tun->flags & IFF_NO_PI)) {
2059                 if (iov_iter_count(iter) < sizeof(pi))
2060                         return -EINVAL;
2061
2062                 total += sizeof(pi);
2063                 if (iov_iter_count(iter) < total) {
2064                         /* Packet will be striped */
2065                         pi.flags |= TUN_PKT_STRIP;
2066                 }
2067
2068                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2069                         return -EFAULT;
2070         }
2071
2072         if (vnet_hdr_sz) {
2073                 struct virtio_net_hdr gso;
2074
2075                 if (iov_iter_count(iter) < vnet_hdr_sz)
2076                         return -EINVAL;
2077
2078                 if (virtio_net_hdr_from_skb(skb, &gso,
2079                                             tun_is_little_endian(tun), true)) {
2080                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2081                         pr_err("unexpected GSO type: "
2082                                "0x%x, gso_size %d, hdr_len %d\n",
2083                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2084                                tun16_to_cpu(tun, gso.hdr_len));
2085                         print_hex_dump(KERN_ERR, "tun: ",
2086                                        DUMP_PREFIX_NONE,
2087                                        16, 1, skb->head,
2088                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2089                         WARN_ON_ONCE(1);
2090                         return -EINVAL;
2091                 }
2092
2093                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2094                         return -EFAULT;
2095
2096                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2097         }
2098
2099         if (vlan_hlen) {
2100                 int ret;
2101                 struct veth veth;
2102
2103                 veth.h_vlan_proto = skb->vlan_proto;
2104                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2105
2106                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2107
2108                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2109                 if (ret || !iov_iter_count(iter))
2110                         goto done;
2111
2112                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2113                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2114                         goto done;
2115         }
2116
2117         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2118
2119 done:
2120         /* caller is in process context, */
2121         stats = get_cpu_ptr(tun->pcpu_stats);
2122         u64_stats_update_begin(&stats->syncp);
2123         stats->tx_packets++;
2124         stats->tx_bytes += skb->len + vlan_hlen;
2125         u64_stats_update_end(&stats->syncp);
2126         put_cpu_ptr(tun->pcpu_stats);
2127
2128         return total;
2129 }
2130
2131 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2132 {
2133         DECLARE_WAITQUEUE(wait, current);
2134         void *ptr = NULL;
2135         int error = 0;
2136
2137         ptr = ptr_ring_consume(&tfile->tx_ring);
2138         if (ptr)
2139                 goto out;
2140         if (noblock) {
2141                 error = -EAGAIN;
2142                 goto out;
2143         }
2144
2145         add_wait_queue(&tfile->wq.wait, &wait);
2146         current->state = TASK_INTERRUPTIBLE;
2147
2148         while (1) {
2149                 ptr = ptr_ring_consume(&tfile->tx_ring);
2150                 if (ptr)
2151                         break;
2152                 if (signal_pending(current)) {
2153                         error = -ERESTARTSYS;
2154                         break;
2155                 }
2156                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2157                         error = -EFAULT;
2158                         break;
2159                 }
2160
2161                 schedule();
2162         }
2163
2164         current->state = TASK_RUNNING;
2165         remove_wait_queue(&tfile->wq.wait, &wait);
2166
2167 out:
2168         *err = error;
2169         return ptr;
2170 }
2171
2172 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2173                            struct iov_iter *to,
2174                            int noblock, void *ptr)
2175 {
2176         ssize_t ret;
2177         int err;
2178
2179         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2180
2181         if (!iov_iter_count(to)) {
2182                 tun_ptr_free(ptr);
2183                 return 0;
2184         }
2185
2186         if (!ptr) {
2187                 /* Read frames from ring */
2188                 ptr = tun_ring_recv(tfile, noblock, &err);
2189                 if (!ptr)
2190                         return err;
2191         }
2192
2193         if (tun_is_xdp_frame(ptr)) {
2194                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2195
2196                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2197                 xdp_return_frame(xdpf);
2198         } else {
2199                 struct sk_buff *skb = ptr;
2200
2201                 ret = tun_put_user(tun, tfile, skb, to);
2202                 if (unlikely(ret < 0))
2203                         kfree_skb(skb);
2204                 else
2205                         consume_skb(skb);
2206         }
2207
2208         return ret;
2209 }
2210
2211 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2212 {
2213         struct file *file = iocb->ki_filp;
2214         struct tun_file *tfile = file->private_data;
2215         struct tun_struct *tun = tun_get(tfile);
2216         ssize_t len = iov_iter_count(to), ret;
2217
2218         if (!tun)
2219                 return -EBADFD;
2220         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2221         ret = min_t(ssize_t, ret, len);
2222         if (ret > 0)
2223                 iocb->ki_pos = ret;
2224         tun_put(tun);
2225         return ret;
2226 }
2227
2228 static void tun_prog_free(struct rcu_head *rcu)
2229 {
2230         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2231
2232         bpf_prog_destroy(prog->prog);
2233         kfree(prog);
2234 }
2235
2236 static int __tun_set_ebpf(struct tun_struct *tun,
2237                           struct tun_prog __rcu **prog_p,
2238                           struct bpf_prog *prog)
2239 {
2240         struct tun_prog *old, *new = NULL;
2241
2242         if (prog) {
2243                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2244                 if (!new)
2245                         return -ENOMEM;
2246                 new->prog = prog;
2247         }
2248
2249         spin_lock_bh(&tun->lock);
2250         old = rcu_dereference_protected(*prog_p,
2251                                         lockdep_is_held(&tun->lock));
2252         rcu_assign_pointer(*prog_p, new);
2253         spin_unlock_bh(&tun->lock);
2254
2255         if (old)
2256                 call_rcu(&old->rcu, tun_prog_free);
2257
2258         return 0;
2259 }
2260
2261 static void tun_free_netdev(struct net_device *dev)
2262 {
2263         struct tun_struct *tun = netdev_priv(dev);
2264
2265         BUG_ON(!(list_empty(&tun->disabled)));
2266         free_percpu(tun->pcpu_stats);
2267         tun_flow_uninit(tun);
2268         security_tun_dev_free_security(tun->security);
2269         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2270         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2271 }
2272
2273 static void tun_setup(struct net_device *dev)
2274 {
2275         struct tun_struct *tun = netdev_priv(dev);
2276
2277         tun->owner = INVALID_UID;
2278         tun->group = INVALID_GID;
2279
2280         dev->ethtool_ops = &tun_ethtool_ops;
2281         dev->needs_free_netdev = true;
2282         dev->priv_destructor = tun_free_netdev;
2283         /* We prefer our own queue length */
2284         dev->tx_queue_len = TUN_READQ_SIZE;
2285 }
2286
2287 /* Trivial set of netlink ops to allow deleting tun or tap
2288  * device with netlink.
2289  */
2290 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2291                         struct netlink_ext_ack *extack)
2292 {
2293         return -EINVAL;
2294 }
2295
2296 static size_t tun_get_size(const struct net_device *dev)
2297 {
2298         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2299         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2300
2301         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2302                nla_total_size(sizeof(gid_t)) + /* GROUP */
2303                nla_total_size(sizeof(u8)) + /* TYPE */
2304                nla_total_size(sizeof(u8)) + /* PI */
2305                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2306                nla_total_size(sizeof(u8)) + /* PERSIST */
2307                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2308                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2309                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2310                0;
2311 }
2312
2313 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2314 {
2315         struct tun_struct *tun = netdev_priv(dev);
2316
2317         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2318                 goto nla_put_failure;
2319         if (uid_valid(tun->owner) &&
2320             nla_put_u32(skb, IFLA_TUN_OWNER,
2321                         from_kuid_munged(current_user_ns(), tun->owner)))
2322                 goto nla_put_failure;
2323         if (gid_valid(tun->group) &&
2324             nla_put_u32(skb, IFLA_TUN_GROUP,
2325                         from_kgid_munged(current_user_ns(), tun->group)))
2326                 goto nla_put_failure;
2327         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2328                 goto nla_put_failure;
2329         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2330                 goto nla_put_failure;
2331         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2332                 goto nla_put_failure;
2333         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2334                        !!(tun->flags & IFF_MULTI_QUEUE)))
2335                 goto nla_put_failure;
2336         if (tun->flags & IFF_MULTI_QUEUE) {
2337                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2338                         goto nla_put_failure;
2339                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2340                                 tun->numdisabled))
2341                         goto nla_put_failure;
2342         }
2343
2344         return 0;
2345
2346 nla_put_failure:
2347         return -EMSGSIZE;
2348 }
2349
2350 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2351         .kind           = DRV_NAME,
2352         .priv_size      = sizeof(struct tun_struct),
2353         .setup          = tun_setup,
2354         .validate       = tun_validate,
2355         .get_size       = tun_get_size,
2356         .fill_info      = tun_fill_info,
2357 };
2358
2359 static void tun_sock_write_space(struct sock *sk)
2360 {
2361         struct tun_file *tfile;
2362         wait_queue_head_t *wqueue;
2363
2364         if (!sock_writeable(sk))
2365                 return;
2366
2367         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2368                 return;
2369
2370         wqueue = sk_sleep(sk);
2371         if (wqueue && waitqueue_active(wqueue))
2372                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2373                                                 EPOLLWRNORM | EPOLLWRBAND);
2374
2375         tfile = container_of(sk, struct tun_file, sk);
2376         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2377 }
2378
2379 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2380 {
2381         int ret;
2382         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2383         struct tun_struct *tun = tun_get(tfile);
2384
2385         if (!tun)
2386                 return -EBADFD;
2387
2388         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2389                            m->msg_flags & MSG_DONTWAIT,
2390                            m->msg_flags & MSG_MORE);
2391         tun_put(tun);
2392         return ret;
2393 }
2394
2395 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2396                        int flags)
2397 {
2398         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2399         struct tun_struct *tun = tun_get(tfile);
2400         void *ptr = m->msg_control;
2401         int ret;
2402
2403         if (!tun) {
2404                 ret = -EBADFD;
2405                 goto out_free;
2406         }
2407
2408         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2409                 ret = -EINVAL;
2410                 goto out_put_tun;
2411         }
2412         if (flags & MSG_ERRQUEUE) {
2413                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2414                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2415                 goto out;
2416         }
2417         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2418         if (ret > (ssize_t)total_len) {
2419                 m->msg_flags |= MSG_TRUNC;
2420                 ret = flags & MSG_TRUNC ? ret : total_len;
2421         }
2422 out:
2423         tun_put(tun);
2424         return ret;
2425
2426 out_put_tun:
2427         tun_put(tun);
2428 out_free:
2429         tun_ptr_free(ptr);
2430         return ret;
2431 }
2432
2433 static int tun_ptr_peek_len(void *ptr)
2434 {
2435         if (likely(ptr)) {
2436                 if (tun_is_xdp_frame(ptr)) {
2437                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2438
2439                         return xdpf->len;
2440                 }
2441                 return __skb_array_len_with_tag(ptr);
2442         } else {
2443                 return 0;
2444         }
2445 }
2446
2447 static int tun_peek_len(struct socket *sock)
2448 {
2449         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2450         struct tun_struct *tun;
2451         int ret = 0;
2452
2453         tun = tun_get(tfile);
2454         if (!tun)
2455                 return 0;
2456
2457         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2458         tun_put(tun);
2459
2460         return ret;
2461 }
2462
2463 /* Ops structure to mimic raw sockets with tun */
2464 static const struct proto_ops tun_socket_ops = {
2465         .peek_len = tun_peek_len,
2466         .sendmsg = tun_sendmsg,
2467         .recvmsg = tun_recvmsg,
2468 };
2469
2470 static struct proto tun_proto = {
2471         .name           = "tun",
2472         .owner          = THIS_MODULE,
2473         .obj_size       = sizeof(struct tun_file),
2474 };
2475
2476 static int tun_flags(struct tun_struct *tun)
2477 {
2478         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2479 }
2480
2481 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2482                               char *buf)
2483 {
2484         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2485         return sprintf(buf, "0x%x\n", tun_flags(tun));
2486 }
2487
2488 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2489                               char *buf)
2490 {
2491         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2492         return uid_valid(tun->owner)?
2493                 sprintf(buf, "%u\n",
2494                         from_kuid_munged(current_user_ns(), tun->owner)):
2495                 sprintf(buf, "-1\n");
2496 }
2497
2498 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2499                               char *buf)
2500 {
2501         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2502         return gid_valid(tun->group) ?
2503                 sprintf(buf, "%u\n",
2504                         from_kgid_munged(current_user_ns(), tun->group)):
2505                 sprintf(buf, "-1\n");
2506 }
2507
2508 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2509 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2510 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2511
2512 static struct attribute *tun_dev_attrs[] = {
2513         &dev_attr_tun_flags.attr,
2514         &dev_attr_owner.attr,
2515         &dev_attr_group.attr,
2516         NULL
2517 };
2518
2519 static const struct attribute_group tun_attr_group = {
2520         .attrs = tun_dev_attrs
2521 };
2522
2523 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2524 {
2525         struct tun_struct *tun;
2526         struct tun_file *tfile = file->private_data;
2527         struct net_device *dev;
2528         int err;
2529
2530         if (tfile->detached)
2531                 return -EINVAL;
2532
2533         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2534                 if (!capable(CAP_NET_ADMIN))
2535                         return -EPERM;
2536
2537                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2538                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2539                         return -EINVAL;
2540         }
2541
2542         dev = __dev_get_by_name(net, ifr->ifr_name);
2543         if (dev) {
2544                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2545                         return -EBUSY;
2546                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2547                         tun = netdev_priv(dev);
2548                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2549                         tun = netdev_priv(dev);
2550                 else
2551                         return -EINVAL;
2552
2553                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2554                     !!(tun->flags & IFF_MULTI_QUEUE))
2555                         return -EINVAL;
2556
2557                 if (tun_not_capable(tun))
2558                         return -EPERM;
2559                 err = security_tun_dev_open(tun->security);
2560                 if (err < 0)
2561                         return err;
2562
2563                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2564                                  ifr->ifr_flags & IFF_NAPI);
2565                 if (err < 0)
2566                         return err;
2567
2568                 if (tun->flags & IFF_MULTI_QUEUE &&
2569                     (tun->numqueues + tun->numdisabled > 1)) {
2570                         /* One or more queue has already been attached, no need
2571                          * to initialize the device again.
2572                          */
2573                         netdev_state_change(dev);
2574                         return 0;
2575                 }
2576
2577                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2578                               (ifr->ifr_flags & TUN_FEATURES);
2579
2580                 netdev_state_change(dev);
2581         } else {
2582                 char *name;
2583                 unsigned long flags = 0;
2584                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2585                              MAX_TAP_QUEUES : 1;
2586
2587                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2588                         return -EPERM;
2589                 err = security_tun_dev_create();
2590                 if (err < 0)
2591                         return err;
2592
2593                 /* Set dev type */
2594                 if (ifr->ifr_flags & IFF_TUN) {
2595                         /* TUN device */
2596                         flags |= IFF_TUN;
2597                         name = "tun%d";
2598                 } else if (ifr->ifr_flags & IFF_TAP) {
2599                         /* TAP device */
2600                         flags |= IFF_TAP;
2601                         name = "tap%d";
2602                 } else
2603                         return -EINVAL;
2604
2605                 if (*ifr->ifr_name)
2606                         name = ifr->ifr_name;
2607
2608                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2609                                        NET_NAME_UNKNOWN, tun_setup, queues,
2610                                        queues);
2611
2612                 if (!dev)
2613                         return -ENOMEM;
2614                 err = dev_get_valid_name(net, dev, name);
2615                 if (err < 0)
2616                         goto err_free_dev;
2617
2618                 dev_net_set(dev, net);
2619                 dev->rtnl_link_ops = &tun_link_ops;
2620                 dev->ifindex = tfile->ifindex;
2621                 dev->sysfs_groups[0] = &tun_attr_group;
2622
2623                 tun = netdev_priv(dev);
2624                 tun->dev = dev;
2625                 tun->flags = flags;
2626                 tun->txflt.count = 0;
2627                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2628
2629                 tun->align = NET_SKB_PAD;
2630                 tun->filter_attached = false;
2631                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2632                 tun->rx_batched = 0;
2633                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2634
2635                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2636                 if (!tun->pcpu_stats) {
2637                         err = -ENOMEM;
2638                         goto err_free_dev;
2639                 }
2640
2641                 spin_lock_init(&tun->lock);
2642
2643                 err = security_tun_dev_alloc_security(&tun->security);
2644                 if (err < 0)
2645                         goto err_free_stat;
2646
2647                 tun_net_init(dev);
2648                 tun_flow_init(tun);
2649
2650                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2651                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2652                                    NETIF_F_HW_VLAN_STAG_TX;
2653                 dev->features = dev->hw_features | NETIF_F_LLTX;
2654                 dev->vlan_features = dev->features &
2655                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2656                                        NETIF_F_HW_VLAN_STAG_TX);
2657
2658                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2659                               (ifr->ifr_flags & TUN_FEATURES);
2660
2661                 INIT_LIST_HEAD(&tun->disabled);
2662                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2663                 if (err < 0)
2664                         goto err_free_flow;
2665
2666                 err = register_netdevice(tun->dev);
2667                 if (err < 0)
2668                         goto err_detach;
2669         }
2670
2671         netif_carrier_on(tun->dev);
2672
2673         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2674
2675         /* Make sure persistent devices do not get stuck in
2676          * xoff state.
2677          */
2678         if (netif_running(tun->dev))
2679                 netif_tx_wake_all_queues(tun->dev);
2680
2681         strcpy(ifr->ifr_name, tun->dev->name);
2682         return 0;
2683
2684 err_detach:
2685         tun_detach_all(dev);
2686         /* register_netdevice() already called tun_free_netdev() */
2687         goto err_free_dev;
2688
2689 err_free_flow:
2690         tun_flow_uninit(tun);
2691         security_tun_dev_free_security(tun->security);
2692 err_free_stat:
2693         free_percpu(tun->pcpu_stats);
2694 err_free_dev:
2695         free_netdev(dev);
2696         return err;
2697 }
2698
2699 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2700                        struct ifreq *ifr)
2701 {
2702         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2703
2704         strcpy(ifr->ifr_name, tun->dev->name);
2705
2706         ifr->ifr_flags = tun_flags(tun);
2707
2708 }
2709
2710 /* This is like a cut-down ethtool ops, except done via tun fd so no
2711  * privs required. */
2712 static int set_offload(struct tun_struct *tun, unsigned long arg)
2713 {
2714         netdev_features_t features = 0;
2715
2716         if (arg & TUN_F_CSUM) {
2717                 features |= NETIF_F_HW_CSUM;
2718                 arg &= ~TUN_F_CSUM;
2719
2720                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2721                         if (arg & TUN_F_TSO_ECN) {
2722                                 features |= NETIF_F_TSO_ECN;
2723                                 arg &= ~TUN_F_TSO_ECN;
2724                         }
2725                         if (arg & TUN_F_TSO4)
2726                                 features |= NETIF_F_TSO;
2727                         if (arg & TUN_F_TSO6)
2728                                 features |= NETIF_F_TSO6;
2729                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2730                 }
2731
2732                 arg &= ~TUN_F_UFO;
2733         }
2734
2735         /* This gives the user a way to test for new features in future by
2736          * trying to set them. */
2737         if (arg)
2738                 return -EINVAL;
2739
2740         tun->set_features = features;
2741         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2742         tun->dev->wanted_features |= features;
2743         netdev_update_features(tun->dev);
2744
2745         return 0;
2746 }
2747
2748 static void tun_detach_filter(struct tun_struct *tun, int n)
2749 {
2750         int i;
2751         struct tun_file *tfile;
2752
2753         for (i = 0; i < n; i++) {
2754                 tfile = rtnl_dereference(tun->tfiles[i]);
2755                 lock_sock(tfile->socket.sk);
2756                 sk_detach_filter(tfile->socket.sk);
2757                 release_sock(tfile->socket.sk);
2758         }
2759
2760         tun->filter_attached = false;
2761 }
2762
2763 static int tun_attach_filter(struct tun_struct *tun)
2764 {
2765         int i, ret = 0;
2766         struct tun_file *tfile;
2767
2768         for (i = 0; i < tun->numqueues; i++) {
2769                 tfile = rtnl_dereference(tun->tfiles[i]);
2770                 lock_sock(tfile->socket.sk);
2771                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2772                 release_sock(tfile->socket.sk);
2773                 if (ret) {
2774                         tun_detach_filter(tun, i);
2775                         return ret;
2776                 }
2777         }
2778
2779         tun->filter_attached = true;
2780         return ret;
2781 }
2782
2783 static void tun_set_sndbuf(struct tun_struct *tun)
2784 {
2785         struct tun_file *tfile;
2786         int i;
2787
2788         for (i = 0; i < tun->numqueues; i++) {
2789                 tfile = rtnl_dereference(tun->tfiles[i]);
2790                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2791         }
2792 }
2793
2794 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2795 {
2796         struct tun_file *tfile = file->private_data;
2797         struct tun_struct *tun;
2798         int ret = 0;
2799
2800         rtnl_lock();
2801
2802         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2803                 tun = tfile->detached;
2804                 if (!tun) {
2805                         ret = -EINVAL;
2806                         goto unlock;
2807                 }
2808                 ret = security_tun_dev_attach_queue(tun->security);
2809                 if (ret < 0)
2810                         goto unlock;
2811                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2812         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2813                 tun = rtnl_dereference(tfile->tun);
2814                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2815                         ret = -EINVAL;
2816                 else
2817                         __tun_detach(tfile, false);
2818         } else
2819                 ret = -EINVAL;
2820
2821         if (ret >= 0)
2822                 netdev_state_change(tun->dev);
2823
2824 unlock:
2825         rtnl_unlock();
2826         return ret;
2827 }
2828
2829 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2830                         void __user *data)
2831 {
2832         struct bpf_prog *prog;
2833         int fd;
2834
2835         if (copy_from_user(&fd, data, sizeof(fd)))
2836                 return -EFAULT;
2837
2838         if (fd == -1) {
2839                 prog = NULL;
2840         } else {
2841                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2842                 if (IS_ERR(prog))
2843                         return PTR_ERR(prog);
2844         }
2845
2846         return __tun_set_ebpf(tun, prog_p, prog);
2847 }
2848
2849 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2850                             unsigned long arg, int ifreq_len)
2851 {
2852         struct tun_file *tfile = file->private_data;
2853         struct net *net = sock_net(&tfile->sk);
2854         struct tun_struct *tun;
2855         void __user* argp = (void __user*)arg;
2856         struct ifreq ifr;
2857         kuid_t owner;
2858         kgid_t group;
2859         int sndbuf;
2860         int vnet_hdr_sz;
2861         unsigned int ifindex;
2862         int le;
2863         int ret;
2864         bool do_notify = false;
2865
2866         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2867             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2868                 if (copy_from_user(&ifr, argp, ifreq_len))
2869                         return -EFAULT;
2870         } else {
2871                 memset(&ifr, 0, sizeof(ifr));
2872         }
2873         if (cmd == TUNGETFEATURES) {
2874                 /* Currently this just means: "what IFF flags are valid?".
2875                  * This is needed because we never checked for invalid flags on
2876                  * TUNSETIFF.
2877                  */
2878                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2879                                 (unsigned int __user*)argp);
2880         } else if (cmd == TUNSETQUEUE) {
2881                 return tun_set_queue(file, &ifr);
2882         } else if (cmd == SIOCGSKNS) {
2883                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2884                         return -EPERM;
2885                 return open_related_ns(&net->ns, get_net_ns);
2886         }
2887
2888         ret = 0;
2889         rtnl_lock();
2890
2891         tun = tun_get(tfile);
2892         if (cmd == TUNSETIFF) {
2893                 ret = -EEXIST;
2894                 if (tun)
2895                         goto unlock;
2896
2897                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2898
2899                 ret = tun_set_iff(net, file, &ifr);
2900
2901                 if (ret)
2902                         goto unlock;
2903
2904                 if (copy_to_user(argp, &ifr, ifreq_len))
2905                         ret = -EFAULT;
2906                 goto unlock;
2907         }
2908         if (cmd == TUNSETIFINDEX) {
2909                 ret = -EPERM;
2910                 if (tun)
2911                         goto unlock;
2912
2913                 ret = -EFAULT;
2914                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2915                         goto unlock;
2916
2917                 ret = 0;
2918                 tfile->ifindex = ifindex;
2919                 goto unlock;
2920         }
2921
2922         ret = -EBADFD;
2923         if (!tun)
2924                 goto unlock;
2925
2926         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2927
2928         ret = 0;
2929         switch (cmd) {
2930         case TUNGETIFF:
2931                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2932
2933                 if (tfile->detached)
2934                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2935                 if (!tfile->socket.sk->sk_filter)
2936                         ifr.ifr_flags |= IFF_NOFILTER;
2937
2938                 if (copy_to_user(argp, &ifr, ifreq_len))
2939                         ret = -EFAULT;
2940                 break;
2941
2942         case TUNSETNOCSUM:
2943                 /* Disable/Enable checksum */
2944
2945                 /* [unimplemented] */
2946                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2947                           arg ? "disabled" : "enabled");
2948                 break;
2949
2950         case TUNSETPERSIST:
2951                 /* Disable/Enable persist mode. Keep an extra reference to the
2952                  * module to prevent the module being unprobed.
2953                  */
2954                 if (arg && !(tun->flags & IFF_PERSIST)) {
2955                         tun->flags |= IFF_PERSIST;
2956                         __module_get(THIS_MODULE);
2957                         do_notify = true;
2958                 }
2959                 if (!arg && (tun->flags & IFF_PERSIST)) {
2960                         tun->flags &= ~IFF_PERSIST;
2961                         module_put(THIS_MODULE);
2962                         do_notify = true;
2963                 }
2964
2965                 tun_debug(KERN_INFO, tun, "persist %s\n",
2966                           arg ? "enabled" : "disabled");
2967                 break;
2968
2969         case TUNSETOWNER:
2970                 /* Set owner of the device */
2971                 owner = make_kuid(current_user_ns(), arg);
2972                 if (!uid_valid(owner)) {
2973                         ret = -EINVAL;
2974                         break;
2975                 }
2976                 tun->owner = owner;
2977                 do_notify = true;
2978                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2979                           from_kuid(&init_user_ns, tun->owner));
2980                 break;
2981
2982         case TUNSETGROUP:
2983                 /* Set group of the device */
2984                 group = make_kgid(current_user_ns(), arg);
2985                 if (!gid_valid(group)) {
2986                         ret = -EINVAL;
2987                         break;
2988                 }
2989                 tun->group = group;
2990                 do_notify = true;
2991                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2992                           from_kgid(&init_user_ns, tun->group));
2993                 break;
2994
2995         case TUNSETLINK:
2996                 /* Only allow setting the type when the interface is down */
2997                 if (tun->dev->flags & IFF_UP) {
2998                         tun_debug(KERN_INFO, tun,
2999                                   "Linktype set failed because interface is up\n");
3000                         ret = -EBUSY;
3001                 } else {
3002                         tun->dev->type = (int) arg;
3003                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3004                                   tun->dev->type);
3005                         ret = 0;
3006                 }
3007                 break;
3008
3009 #ifdef TUN_DEBUG
3010         case TUNSETDEBUG:
3011                 tun->debug = arg;
3012                 break;
3013 #endif
3014         case TUNSETOFFLOAD:
3015                 ret = set_offload(tun, arg);
3016                 break;
3017
3018         case TUNSETTXFILTER:
3019                 /* Can be set only for TAPs */
3020                 ret = -EINVAL;
3021                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3022                         break;
3023                 ret = update_filter(&tun->txflt, (void __user *)arg);
3024                 break;
3025
3026         case SIOCGIFHWADDR:
3027                 /* Get hw address */
3028                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3029                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3030                 if (copy_to_user(argp, &ifr, ifreq_len))
3031                         ret = -EFAULT;
3032                 break;
3033
3034         case SIOCSIFHWADDR:
3035                 /* Set hw address */
3036                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3037                           ifr.ifr_hwaddr.sa_data);
3038
3039                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3040                 break;
3041
3042         case TUNGETSNDBUF:
3043                 sndbuf = tfile->socket.sk->sk_sndbuf;
3044                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3045                         ret = -EFAULT;
3046                 break;
3047
3048         case TUNSETSNDBUF:
3049                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3050                         ret = -EFAULT;
3051                         break;
3052                 }
3053                 if (sndbuf <= 0) {
3054                         ret = -EINVAL;
3055                         break;
3056                 }
3057
3058                 tun->sndbuf = sndbuf;
3059                 tun_set_sndbuf(tun);
3060                 break;
3061
3062         case TUNGETVNETHDRSZ:
3063                 vnet_hdr_sz = tun->vnet_hdr_sz;
3064                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3065                         ret = -EFAULT;
3066                 break;
3067
3068         case TUNSETVNETHDRSZ:
3069                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3070                         ret = -EFAULT;
3071                         break;
3072                 }
3073                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3074                         ret = -EINVAL;
3075                         break;
3076                 }
3077
3078                 tun->vnet_hdr_sz = vnet_hdr_sz;
3079                 break;
3080
3081         case TUNGETVNETLE:
3082                 le = !!(tun->flags & TUN_VNET_LE);
3083                 if (put_user(le, (int __user *)argp))
3084                         ret = -EFAULT;
3085                 break;
3086
3087         case TUNSETVNETLE:
3088                 if (get_user(le, (int __user *)argp)) {
3089                         ret = -EFAULT;
3090                         break;
3091                 }
3092                 if (le)
3093                         tun->flags |= TUN_VNET_LE;
3094                 else
3095                         tun->flags &= ~TUN_VNET_LE;
3096                 break;
3097
3098         case TUNGETVNETBE:
3099                 ret = tun_get_vnet_be(tun, argp);
3100                 break;
3101
3102         case TUNSETVNETBE:
3103                 ret = tun_set_vnet_be(tun, argp);
3104                 break;
3105
3106         case TUNATTACHFILTER:
3107                 /* Can be set only for TAPs */
3108                 ret = -EINVAL;
3109                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3110                         break;
3111                 ret = -EFAULT;
3112                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3113                         break;
3114
3115                 ret = tun_attach_filter(tun);
3116                 break;
3117
3118         case TUNDETACHFILTER:
3119                 /* Can be set only for TAPs */
3120                 ret = -EINVAL;
3121                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3122                         break;
3123                 ret = 0;
3124                 tun_detach_filter(tun, tun->numqueues);
3125                 break;
3126
3127         case TUNGETFILTER:
3128                 ret = -EINVAL;
3129                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3130                         break;
3131                 ret = -EFAULT;
3132                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3133                         break;
3134                 ret = 0;
3135                 break;
3136
3137         case TUNSETSTEERINGEBPF:
3138                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3139                 break;
3140
3141         case TUNSETFILTEREBPF:
3142                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3143                 break;
3144
3145         default:
3146                 ret = -EINVAL;
3147                 break;
3148         }
3149
3150         if (do_notify)
3151                 netdev_state_change(tun->dev);
3152
3153 unlock:
3154         rtnl_unlock();
3155         if (tun)
3156                 tun_put(tun);
3157         return ret;
3158 }
3159
3160 static long tun_chr_ioctl(struct file *file,
3161                           unsigned int cmd, unsigned long arg)
3162 {
3163         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3164 }
3165
3166 #ifdef CONFIG_COMPAT
3167 static long tun_chr_compat_ioctl(struct file *file,
3168                          unsigned int cmd, unsigned long arg)
3169 {
3170         switch (cmd) {
3171         case TUNSETIFF:
3172         case TUNGETIFF:
3173         case TUNSETTXFILTER:
3174         case TUNGETSNDBUF:
3175         case TUNSETSNDBUF:
3176         case SIOCGIFHWADDR:
3177         case SIOCSIFHWADDR:
3178                 arg = (unsigned long)compat_ptr(arg);
3179                 break;
3180         default:
3181                 arg = (compat_ulong_t)arg;
3182                 break;
3183         }
3184
3185         /*
3186          * compat_ifreq is shorter than ifreq, so we must not access beyond
3187          * the end of that structure. All fields that are used in this
3188          * driver are compatible though, we don't need to convert the
3189          * contents.
3190          */
3191         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3192 }
3193 #endif /* CONFIG_COMPAT */
3194
3195 static int tun_chr_fasync(int fd, struct file *file, int on)
3196 {
3197         struct tun_file *tfile = file->private_data;
3198         int ret;
3199
3200         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3201                 goto out;
3202
3203         if (on) {
3204                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3205                 tfile->flags |= TUN_FASYNC;
3206         } else
3207                 tfile->flags &= ~TUN_FASYNC;
3208         ret = 0;
3209 out:
3210         return ret;
3211 }
3212
3213 static int tun_chr_open(struct inode *inode, struct file * file)
3214 {
3215         struct net *net = current->nsproxy->net_ns;
3216         struct tun_file *tfile;
3217
3218         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3219
3220         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3221                                             &tun_proto, 0);
3222         if (!tfile)
3223                 return -ENOMEM;
3224         RCU_INIT_POINTER(tfile->tun, NULL);
3225         tfile->flags = 0;
3226         tfile->ifindex = 0;
3227
3228         init_waitqueue_head(&tfile->wq.wait);
3229         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3230
3231         tfile->socket.file = file;
3232         tfile->socket.ops = &tun_socket_ops;
3233
3234         sock_init_data(&tfile->socket, &tfile->sk);
3235
3236         tfile->sk.sk_write_space = tun_sock_write_space;
3237         tfile->sk.sk_sndbuf = INT_MAX;
3238
3239         file->private_data = tfile;
3240         INIT_LIST_HEAD(&tfile->next);
3241
3242         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3243
3244         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3245
3246         return 0;
3247 }
3248
3249 static int tun_chr_close(struct inode *inode, struct file *file)
3250 {
3251         struct tun_file *tfile = file->private_data;
3252
3253         tun_detach(tfile, true);
3254
3255         return 0;
3256 }
3257
3258 #ifdef CONFIG_PROC_FS
3259 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3260 {
3261         struct tun_file *tfile = file->private_data;
3262         struct tun_struct *tun;
3263         struct ifreq ifr;
3264
3265         memset(&ifr, 0, sizeof(ifr));
3266
3267         rtnl_lock();
3268         tun = tun_get(tfile);
3269         if (tun)
3270                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3271         rtnl_unlock();
3272
3273         if (tun)
3274                 tun_put(tun);
3275
3276         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3277 }
3278 #endif
3279
3280 static const struct file_operations tun_fops = {
3281         .owner  = THIS_MODULE,
3282         .llseek = no_llseek,
3283         .read_iter  = tun_chr_read_iter,
3284         .write_iter = tun_chr_write_iter,
3285         .poll   = tun_chr_poll,
3286         .unlocked_ioctl = tun_chr_ioctl,
3287 #ifdef CONFIG_COMPAT
3288         .compat_ioctl = tun_chr_compat_ioctl,
3289 #endif
3290         .open   = tun_chr_open,
3291         .release = tun_chr_close,
3292         .fasync = tun_chr_fasync,
3293 #ifdef CONFIG_PROC_FS
3294         .show_fdinfo = tun_chr_show_fdinfo,
3295 #endif
3296 };
3297
3298 static struct miscdevice tun_miscdev = {
3299         .minor = TUN_MINOR,
3300         .name = "tun",
3301         .nodename = "net/tun",
3302         .fops = &tun_fops,
3303 };
3304
3305 /* ethtool interface */
3306
3307 static int tun_get_link_ksettings(struct net_device *dev,
3308                                   struct ethtool_link_ksettings *cmd)
3309 {
3310         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3311         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3312         cmd->base.speed         = SPEED_10;
3313         cmd->base.duplex        = DUPLEX_FULL;
3314         cmd->base.port          = PORT_TP;
3315         cmd->base.phy_address   = 0;
3316         cmd->base.autoneg       = AUTONEG_DISABLE;
3317         return 0;
3318 }
3319
3320 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3321 {
3322         struct tun_struct *tun = netdev_priv(dev);
3323
3324         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3325         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3326
3327         switch (tun->flags & TUN_TYPE_MASK) {
3328         case IFF_TUN:
3329                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3330                 break;
3331         case IFF_TAP:
3332                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3333                 break;
3334         }
3335 }
3336
3337 static u32 tun_get_msglevel(struct net_device *dev)
3338 {
3339 #ifdef TUN_DEBUG
3340         struct tun_struct *tun = netdev_priv(dev);
3341         return tun->debug;
3342 #else
3343         return -EOPNOTSUPP;
3344 #endif
3345 }
3346
3347 static void tun_set_msglevel(struct net_device *dev, u32 value)
3348 {
3349 #ifdef TUN_DEBUG
3350         struct tun_struct *tun = netdev_priv(dev);
3351         tun->debug = value;
3352 #endif
3353 }
3354
3355 static int tun_get_coalesce(struct net_device *dev,
3356                             struct ethtool_coalesce *ec)
3357 {
3358         struct tun_struct *tun = netdev_priv(dev);
3359
3360         ec->rx_max_coalesced_frames = tun->rx_batched;
3361
3362         return 0;
3363 }
3364
3365 static int tun_set_coalesce(struct net_device *dev,
3366                             struct ethtool_coalesce *ec)
3367 {
3368         struct tun_struct *tun = netdev_priv(dev);
3369
3370         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3371                 tun->rx_batched = NAPI_POLL_WEIGHT;
3372         else
3373                 tun->rx_batched = ec->rx_max_coalesced_frames;
3374
3375         return 0;
3376 }
3377
3378 static const struct ethtool_ops tun_ethtool_ops = {
3379         .get_drvinfo    = tun_get_drvinfo,
3380         .get_msglevel   = tun_get_msglevel,
3381         .set_msglevel   = tun_set_msglevel,
3382         .get_link       = ethtool_op_get_link,
3383         .get_ts_info    = ethtool_op_get_ts_info,
3384         .get_coalesce   = tun_get_coalesce,
3385         .set_coalesce   = tun_set_coalesce,
3386         .get_link_ksettings = tun_get_link_ksettings,
3387 };
3388
3389 static int tun_queue_resize(struct tun_struct *tun)
3390 {
3391         struct net_device *dev = tun->dev;
3392         struct tun_file *tfile;
3393         struct ptr_ring **rings;
3394         int n = tun->numqueues + tun->numdisabled;
3395         int ret, i;
3396
3397         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3398         if (!rings)
3399                 return -ENOMEM;
3400
3401         for (i = 0; i < tun->numqueues; i++) {
3402                 tfile = rtnl_dereference(tun->tfiles[i]);
3403                 rings[i] = &tfile->tx_ring;
3404         }
3405         list_for_each_entry(tfile, &tun->disabled, next)
3406                 rings[i++] = &tfile->tx_ring;
3407
3408         ret = ptr_ring_resize_multiple(rings, n,
3409                                        dev->tx_queue_len, GFP_KERNEL,
3410                                        tun_ptr_free);
3411
3412         kfree(rings);
3413         return ret;
3414 }
3415
3416 static int tun_device_event(struct notifier_block *unused,
3417                             unsigned long event, void *ptr)
3418 {
3419         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3420         struct tun_struct *tun = netdev_priv(dev);
3421
3422         if (dev->rtnl_link_ops != &tun_link_ops)
3423                 return NOTIFY_DONE;
3424
3425         switch (event) {
3426         case NETDEV_CHANGE_TX_QUEUE_LEN:
3427                 if (tun_queue_resize(tun))
3428                         return NOTIFY_BAD;
3429                 break;
3430         default:
3431                 break;
3432         }
3433
3434         return NOTIFY_DONE;
3435 }
3436
3437 static struct notifier_block tun_notifier_block __read_mostly = {
3438         .notifier_call  = tun_device_event,
3439 };
3440
3441 static int __init tun_init(void)
3442 {
3443         int ret = 0;
3444
3445         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3446
3447         ret = rtnl_link_register(&tun_link_ops);
3448         if (ret) {
3449                 pr_err("Can't register link_ops\n");
3450                 goto err_linkops;
3451         }
3452
3453         ret = misc_register(&tun_miscdev);
3454         if (ret) {
3455                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3456                 goto err_misc;
3457         }
3458
3459         ret = register_netdevice_notifier(&tun_notifier_block);
3460         if (ret) {
3461                 pr_err("Can't register netdevice notifier\n");
3462                 goto err_notifier;
3463         }
3464
3465         return  0;
3466
3467 err_notifier:
3468         misc_deregister(&tun_miscdev);
3469 err_misc:
3470         rtnl_link_unregister(&tun_link_ops);
3471 err_linkops:
3472         return ret;
3473 }
3474
3475 static void tun_cleanup(void)
3476 {
3477         misc_deregister(&tun_miscdev);
3478         rtnl_link_unregister(&tun_link_ops);
3479         unregister_netdevice_notifier(&tun_notifier_block);
3480 }
3481
3482 /* Get an underlying socket object from tun file.  Returns error unless file is
3483  * attached to a device.  The returned object works like a packet socket, it
3484  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3485  * holding a reference to the file for as long as the socket is in use. */
3486 struct socket *tun_get_socket(struct file *file)
3487 {
3488         struct tun_file *tfile;
3489         if (file->f_op != &tun_fops)
3490                 return ERR_PTR(-EINVAL);
3491         tfile = file->private_data;
3492         if (!tfile)
3493                 return ERR_PTR(-EBADFD);
3494         return &tfile->socket;
3495 }
3496 EXPORT_SYMBOL_GPL(tun_get_socket);
3497
3498 struct ptr_ring *tun_get_tx_ring(struct file *file)
3499 {
3500         struct tun_file *tfile;
3501
3502         if (file->f_op != &tun_fops)
3503                 return ERR_PTR(-EINVAL);
3504         tfile = file->private_data;
3505         if (!tfile)
3506                 return ERR_PTR(-EBADFD);
3507         return &tfile->tx_ring;
3508 }
3509 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3510
3511 module_init(tun_init);
3512 module_exit(tun_cleanup);
3513 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3514 MODULE_AUTHOR(DRV_COPYRIGHT);
3515 MODULE_LICENSE("GPL");
3516 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3517 MODULE_ALIAS("devname:net/tun");