drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81 #include <linux/proc_fs.h>
  82
  83 /* Uncomment to enable debugging */
  84 /* #define TUN_DEBUG 1 */
  85
  86 #ifdef TUN_DEBUG
  87 static int debug;
  88
  89 #define tun_debug(level, tun, fmt, args...)                     \
  90 do {                                                            \
  91         if (tun->debug)                                         \
  92                 netdev_printk(level, tun->dev, fmt, ##args);    \
  93 } while (0)
  94 #define DBG1(level, fmt, args...)                               \
  95 do {                                                            \
  96         if (debug == 2)                                         \
  97                 printk(level fmt, ##args);                      \
  98 } while (0)
  99 #else
 100 #define tun_debug(level, tun, fmt, args...)                     \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 netdev_printk(level, tun->dev, fmt, ##args);    \
 104 } while (0)
 105 #define DBG1(level, fmt, args...)                               \
 106 do {                                                            \
 107         if (0)                                                  \
 108                 printk(level fmt, ##args);                      \
 109 } while (0)
 110 #endif
 111
 112 #define TUN_HEADROOM 256
 113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 127
 128 #define GOODCOPY_LEN 128
 129
 130 #define FLT_EXACT_COUNT 8
 131 struct tap_filter {
 132         unsigned int    count;    /* Number of addrs. Zero means disabled */
 133         u32             mask[2];  /* Mask of the hashed addrs */
 134         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 135 };
 136
 137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 138  * to max number of VCPUs in guest. */
 139 #define MAX_TAP_QUEUES 256
 140 #define MAX_TAP_FLOWS  4096
 141
 142 #define TUN_FLOW_EXPIRE (3 * HZ)
 143
 144 struct tun_pcpu_stats {
 145         u64 rx_packets;
 146         u64 rx_bytes;
 147         u64 tx_packets;
 148         u64 tx_bytes;
 149         struct u64_stats_sync syncp;
 150         u32 rx_dropped;
 151         u32 tx_dropped;
 152         u32 rx_frame_errors;
 153 };
 154
 155 /* A tun_file connects an open character device to a tuntap netdevice. It
 156  * also contains all socket related structures (except sock_fprog and tap_filter)
 157  * to serve as one transmit queue for tuntap device. The sock_fprog and
 158  * tap_filter were kept in tun_struct since they were used for filtering for the
 159  * netdevice not for a specific queue (at least I didn't see the requirement for
 160  * this).
 161  *
 162  * RCU usage:
 163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 164  * other can only be read while rcu_read_lock or rtnl_lock is held.
 165  */
 166 struct tun_file {
 167         struct sock sk;
 168         struct socket socket;
 169         struct socket_wq wq;
 170         struct tun_struct __rcu *tun;
 171         struct fasync_struct *fasync;
 172         /* only used for fasnyc */
 173         unsigned int flags;
 174         union {
 175                 u16 queue_index;
 176                 unsigned int ifindex;
 177         };
 178         struct napi_struct napi;
 179         bool napi_enabled;
 180         struct mutex napi_mutex;        /* Protects access to the above napi */
 181         struct list_head next;
 182         struct tun_struct *detached;
 183         struct ptr_ring tx_ring;
 184         struct xdp_rxq_info xdp_rxq;
 185 };
 186
 187 struct tun_flow_entry {
 188         struct hlist_node hash_link;
 189         struct rcu_head rcu;
 190         struct tun_struct *tun;
 191
 192         u32 rxhash;
 193         u32 rps_rxhash;
 194         int queue_index;
 195         unsigned long updated;
 196 };
 197
 198 #define TUN_NUM_FLOW_ENTRIES 1024
 199
 200 struct tun_prog {
 201         struct rcu_head rcu;
 202         struct bpf_prog *prog;
 203 };
 204
 205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 206  * device, socket filter, sndbuf and vnet header size were restore when the
 207  * file were attached to a persist device.
 208  */
 209 struct tun_struct {
 210         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 211         unsigned int            numqueues;
 212         unsigned int            flags;
 213         kuid_t                  owner;
 214         kgid_t                  group;
 215
 216         struct net_device       *dev;
 217         netdev_features_t       set_features;
 218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 219                           NETIF_F_TSO6)
 220
 221         int                     align;
 222         int                     vnet_hdr_sz;
 223         int                     sndbuf;
 224         struct tap_filter       txflt;
 225         struct sock_fprog       fprog;
 226         /* protected by rtnl lock */
 227         bool                    filter_attached;
 228 #ifdef TUN_DEBUG
 229         int debug;
 230 #endif
 231         spinlock_t lock;
 232         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 233         struct timer_list flow_gc_timer;
 234         unsigned long ageing_time;
 235         unsigned int numdisabled;
 236         struct list_head disabled;
 237         void *security;
 238         u32 flow_count;
 239         u32 rx_batched;
 240         struct tun_pcpu_stats __percpu *pcpu_stats;
 241         struct bpf_prog __rcu *xdp_prog;
 242         struct tun_prog __rcu *steering_prog;
 243         struct tun_prog __rcu *filter_prog;
 244 };
 245
 246 struct veth {
 247         __be16 h_vlan_proto;
 248         __be16 h_vlan_TCI;
 249 };
 250
 251 bool tun_is_xdp_buff(void *ptr)
 252 {
 253         return (unsigned long)ptr & TUN_XDP_FLAG;
 254 }
 255 EXPORT_SYMBOL(tun_is_xdp_buff);
 256
 257 void *tun_xdp_to_ptr(void *ptr)
 258 {
 259         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 260 }
 261 EXPORT_SYMBOL(tun_xdp_to_ptr);
 262
 263 void *tun_ptr_to_xdp(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_ptr_to_xdp);
 268
 269 static int tun_napi_receive(struct napi_struct *napi, int budget)
 270 {
 271         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 272         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 273         struct sk_buff_head process_queue;
 274         struct sk_buff *skb;
 275         int received = 0;
 276
 277         __skb_queue_head_init(&process_queue);
 278
 279         spin_lock(&queue->lock);
 280         skb_queue_splice_tail_init(queue, &process_queue);
 281         spin_unlock(&queue->lock);
 282
 283         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 284                 napi_gro_receive(napi, skb);
 285                 ++received;
 286         }
 287
 288         if (!skb_queue_empty(&process_queue)) {
 289                 spin_lock(&queue->lock);
 290                 skb_queue_splice(&process_queue, queue);
 291                 spin_unlock(&queue->lock);
 292         }
 293
 294         return received;
 295 }
 296
 297 static int tun_napi_poll(struct napi_struct *napi, int budget)
 298 {
 299         unsigned int received;
 300
 301         received = tun_napi_receive(napi, budget);
 302
 303         if (received < budget)
 304                 napi_complete_done(napi, received);
 305
 306         return received;
 307 }
 308
 309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 310                           bool napi_en)
 311 {
 312         tfile->napi_enabled = napi_en;
 313         if (napi_en) {
 314                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 315                                NAPI_POLL_WEIGHT);
 316                 napi_enable(&tfile->napi);
 317                 mutex_init(&tfile->napi_mutex);
 318         }
 319 }
 320
 321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 322 {
 323         if (tfile->napi_enabled)
 324                 napi_disable(&tfile->napi);
 325 }
 326
 327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 netif_napi_del(&tfile->napi);
 331 }
 332
 333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 334 {
 335         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 336 }
 337
 338 #ifdef CONFIG_TUN_VNET_CROSS_LE
 339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 340 {
 341         return tun->flags & TUN_VNET_BE ? false :
 342                 virtio_legacy_is_little_endian();
 343 }
 344
 345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 346 {
 347         int be = !!(tun->flags & TUN_VNET_BE);
 348
 349         if (put_user(be, argp))
 350                 return -EFAULT;
 351
 352         return 0;
 353 }
 354
 355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 356 {
 357         int be;
 358
 359         if (get_user(be, argp))
 360                 return -EFAULT;
 361
 362         if (be)
 363                 tun->flags |= TUN_VNET_BE;
 364         else
 365                 tun->flags &= ~TUN_VNET_BE;
 366
 367         return 0;
 368 }
 369 #else
 370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 371 {
 372         return virtio_legacy_is_little_endian();
 373 }
 374
 375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 376 {
 377         return -EINVAL;
 378 }
 379
 380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 385
 386 static inline bool tun_is_little_endian(struct tun_struct *tun)
 387 {
 388         return tun->flags & TUN_VNET_LE ||
 389                 tun_legacy_is_little_endian(tun);
 390 }
 391
 392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 393 {
 394         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 395 }
 396
 397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 398 {
 399         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline u32 tun_hashfn(u32 rxhash)
 403 {
 404         return rxhash & 0x3ff;
 405 }
 406
 407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 408 {
 409         struct tun_flow_entry *e;
 410
 411         hlist_for_each_entry_rcu(e, head, hash_link) {
 412                 if (e->rxhash == rxhash)
 413                         return e;
 414         }
 415         return NULL;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 419                                               struct hlist_head *head,
 420                                               u32 rxhash, u16 queue_index)
 421 {
 422         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 423
 424         if (e) {
 425                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 426                           rxhash, queue_index);
 427                 e->updated = jiffies;
 428                 e->rxhash = rxhash;
 429                 e->rps_rxhash = 0;
 430                 e->queue_index = queue_index;
 431                 e->tun = tun;
 432                 hlist_add_head_rcu(&e->hash_link, head);
 433                 ++tun->flow_count;
 434         }
 435         return e;
 436 }
 437
 438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 439 {
 440         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 441                   e->rxhash, e->queue_index);
 442         hlist_del_rcu(&e->hash_link);
 443         kfree_rcu(e, rcu);
 444         --tun->flow_count;
 445 }
 446
 447 static void tun_flow_flush(struct tun_struct *tun)
 448 {
 449         int i;
 450
 451         spin_lock_bh(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 457                         tun_flow_delete(tun, e);
 458         }
 459         spin_unlock_bh(&tun->lock);
 460 }
 461
 462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 463 {
 464         int i;
 465
 466         spin_lock_bh(&tun->lock);
 467         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 468                 struct tun_flow_entry *e;
 469                 struct hlist_node *n;
 470
 471                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 472                         if (e->queue_index == queue_index)
 473                                 tun_flow_delete(tun, e);
 474                 }
 475         }
 476         spin_unlock_bh(&tun->lock);
 477 }
 478
 479 static void tun_flow_cleanup(struct timer_list *t)
 480 {
 481         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 482         unsigned long delay = tun->ageing_time;
 483         unsigned long next_timer = jiffies + delay;
 484         unsigned long count = 0;
 485         int i;
 486
 487         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 488
 489         spin_lock(&tun->lock);
 490         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 491                 struct tun_flow_entry *e;
 492                 struct hlist_node *n;
 493
 494                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 495                         unsigned long this_timer;
 496
 497                         this_timer = e->updated + delay;
 498                         if (time_before_eq(this_timer, jiffies)) {
 499                                 tun_flow_delete(tun, e);
 500                                 continue;
 501                         }
 502                         count++;
 503                         if (time_before(this_timer, next_timer))
 504                                 next_timer = this_timer;
 505                 }
 506         }
 507
 508         if (count)
 509                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 510         spin_unlock(&tun->lock);
 511 }
 512
 513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 514                             struct tun_file *tfile)
 515 {
 516         struct hlist_head *head;
 517         struct tun_flow_entry *e;
 518         unsigned long delay = tun->ageing_time;
 519         u16 queue_index = tfile->queue_index;
 520
 521         if (!rxhash)
 522                 return;
 523         else
 524                 head = &tun->flows[tun_hashfn(rxhash)];
 525
 526         rcu_read_lock();
 527
 528         /* We may get a very small possibility of OOO during switching, not
 529          * worth to optimize.*/
 530         if (tun->numqueues == 1 || tfile->detached)
 531                 goto unlock;
 532
 533         e = tun_flow_find(head, rxhash);
 534         if (likely(e)) {
 535                 /* TODO: keep queueing to old queue until it's empty? */
 536                 e->queue_index = queue_index;
 537                 e->updated = jiffies;
 538                 sock_rps_record_flow_hash(e->rps_rxhash);
 539         } else {
 540                 spin_lock_bh(&tun->lock);
 541                 if (!tun_flow_find(head, rxhash) &&
 542                     tun->flow_count < MAX_TAP_FLOWS)
 543                         tun_flow_create(tun, head, rxhash, queue_index);
 544
 545                 if (!timer_pending(&tun->flow_gc_timer))
 546                         mod_timer(&tun->flow_gc_timer,
 547                                   round_jiffies_up(jiffies + delay));
 548                 spin_unlock_bh(&tun->lock);
 549         }
 550
 551 unlock:
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash first. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here. If we could not get rxhash, then we would
 570  * hope the rxq no. may help here.
 571  */
 572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 573 {
 574         struct tun_flow_entry *e;
 575         u32 txq = 0;
 576         u32 numqueues = 0;
 577
 578         numqueues = READ_ONCE(tun->numqueues);
 579
 580         txq = __skb_get_hash_symmetric(skb);
 581         if (txq) {
 582                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 583                 if (e) {
 584                         tun_flow_save_rps_rxhash(e, txq);
 585                         txq = e->queue_index;
 586                 } else
 587                         /* use multiply and shift instead of expensive divide */
 588                         txq = ((u64)txq * numqueues) >> 32;
 589         } else if (likely(skb_rx_queue_recorded(skb))) {
 590                 txq = skb_get_rx_queue(skb);
 591                 while (unlikely(txq >= numqueues))
 592                         txq -= numqueues;
 593         }
 594
 595         return txq;
 596 }
 597
 598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 599 {
 600         struct tun_prog *prog;
 601         u16 ret = 0;
 602
 603         prog = rcu_dereference(tun->steering_prog);
 604         if (prog)
 605                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 606
 607         return ret % tun->numqueues;
 608 }
 609
 610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 611                             void *accel_priv, select_queue_fallback_t fallback)
 612 {
 613         struct tun_struct *tun = netdev_priv(dev);
 614         u16 ret;
 615
 616         rcu_read_lock();
 617         if (rcu_dereference(tun->steering_prog))
 618                 ret = tun_ebpf_select_queue(tun, skb);
 619         else
 620                 ret = tun_automq_select_queue(tun, skb);
 621         rcu_read_unlock();
 622
 623         return ret;
 624 }
 625
 626 static inline bool tun_not_capable(struct tun_struct *tun)
 627 {
 628         const struct cred *cred = current_cred();
 629         struct net *net = dev_net(tun->dev);
 630
 631         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 632                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 633                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 634 }
 635
 636 static void tun_set_real_num_queues(struct tun_struct *tun)
 637 {
 638         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 639         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 640 }
 641
 642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 643 {
 644         tfile->detached = tun;
 645         list_add_tail(&tfile->next, &tun->disabled);
 646         ++tun->numdisabled;
 647 }
 648
 649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 650 {
 651         struct tun_struct *tun = tfile->detached;
 652
 653         tfile->detached = NULL;
 654         list_del_init(&tfile->next);
 655         --tun->numdisabled;
 656         return tun;
 657 }
 658
 659 void tun_ptr_free(void *ptr)
 660 {
 661         if (!ptr)
 662                 return;
 663         if (tun_is_xdp_buff(ptr)) {
 664                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
 665
 666                 put_page(virt_to_head_page(xdp->data));
 667         } else {
 668                 __skb_array_destroy_skb(ptr);
 669         }
 670 }
 671 EXPORT_SYMBOL_GPL(tun_ptr_free);
 672
 673 static void tun_queue_purge(struct tun_file *tfile)
 674 {
 675         void *ptr;
 676
 677         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 678                 tun_ptr_free(ptr);
 679
 680         skb_queue_purge(&tfile->sk.sk_write_queue);
 681         skb_queue_purge(&tfile->sk.sk_error_queue);
 682 }
 683
 684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 685 {
 686         if (tfile->tx_ring.queue) {
 687                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 688                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 690         }
 691 }
 692
 693 static void __tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_file *ntfile;
 696         struct tun_struct *tun;
 697
 698         tun = rtnl_dereference(tfile->tun);
 699
 700         if (tun && clean) {
 701                 tun_napi_disable(tun, tfile);
 702                 tun_napi_del(tun, tfile);
 703         }
 704
 705         if (tun && !tfile->detached) {
 706                 u16 index = tfile->queue_index;
 707                 BUG_ON(index >= tun->numqueues);
 708
 709                 rcu_assign_pointer(tun->tfiles[index],
 710                                    tun->tfiles[tun->numqueues - 1]);
 711                 ntfile = rtnl_dereference(tun->tfiles[index]);
 712                 ntfile->queue_index = index;
 713
 714                 --tun->numqueues;
 715                 if (clean) {
 716                         RCU_INIT_POINTER(tfile->tun, NULL);
 717                         sock_put(&tfile->sk);
 718                 } else
 719                         tun_disable_queue(tun, tfile);
 720
 721                 synchronize_net();
 722                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 723                 /* Drop read queue */
 724                 tun_queue_purge(tfile);
 725                 tun_set_real_num_queues(tun);
 726         } else if (tfile->detached && clean) {
 727                 tun = tun_enable_queue(tfile);
 728                 sock_put(&tfile->sk);
 729         }
 730
 731         if (clean) {
 732                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 733                         netif_carrier_off(tun->dev);
 734
 735                         if (!(tun->flags & IFF_PERSIST) &&
 736                             tun->dev->reg_state == NETREG_REGISTERED)
 737                                 unregister_netdevice(tun->dev);
 738                 }
 739                 tun_cleanup_tx_ring(tfile);
 740                 sock_put(&tfile->sk);
 741         }
 742 }
 743
 744 static void tun_detach(struct tun_file *tfile, bool clean)
 745 {
 746         rtnl_lock();
 747         __tun_detach(tfile, clean);
 748         rtnl_unlock();
 749 }
 750
 751 static void tun_detach_all(struct net_device *dev)
 752 {
 753         struct tun_struct *tun = netdev_priv(dev);
 754         struct tun_file *tfile, *tmp;
 755         int i, n = tun->numqueues;
 756
 757         for (i = 0; i < n; i++) {
 758                 tfile = rtnl_dereference(tun->tfiles[i]);
 759                 BUG_ON(!tfile);
 760                 tun_napi_disable(tun, tfile);
 761                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 762                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 763                 RCU_INIT_POINTER(tfile->tun, NULL);
 764                 --tun->numqueues;
 765         }
 766         list_for_each_entry(tfile, &tun->disabled, next) {
 767                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 768                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 769                 RCU_INIT_POINTER(tfile->tun, NULL);
 770         }
 771         BUG_ON(tun->numqueues != 0);
 772
 773         synchronize_net();
 774         for (i = 0; i < n; i++) {
 775                 tfile = rtnl_dereference(tun->tfiles[i]);
 776                 tun_napi_del(tun, tfile);
 777                 /* Drop read queue */
 778                 tun_queue_purge(tfile);
 779                 sock_put(&tfile->sk);
 780                 tun_cleanup_tx_ring(tfile);
 781         }
 782         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 783                 tun_enable_queue(tfile);
 784                 tun_queue_purge(tfile);
 785                 sock_put(&tfile->sk);
 786                 tun_cleanup_tx_ring(tfile);
 787         }
 788         BUG_ON(tun->numdisabled != 0);
 789
 790         if (tun->flags & IFF_PERSIST)
 791                 module_put(THIS_MODULE);
 792 }
 793
 794 static int tun_attach(struct tun_struct *tun, struct file *file,
 795                       bool skip_filter, bool napi)
 796 {
 797         struct tun_file *tfile = file->private_data;
 798         struct net_device *dev = tun->dev;
 799         int err;
 800
 801         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 802         if (err < 0)
 803                 goto out;
 804
 805         err = -EINVAL;
 806         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 807                 goto out;
 808
 809         err = -EBUSY;
 810         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 811                 goto out;
 812
 813         err = -E2BIG;
 814         if (!tfile->detached &&
 815             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 816                 goto out;
 817
 818         err = 0;
 819
 820         /* Re-attach the filter to persist device */
 821         if (!skip_filter && (tun->filter_attached == true)) {
 822                 lock_sock(tfile->socket.sk);
 823                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 824                 release_sock(tfile->socket.sk);
 825                 if (!err)
 826                         goto out;
 827         }
 828
 829         if (!tfile->detached &&
 830             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 831                 err = -ENOMEM;
 832                 goto out;
 833         }
 834
 835         tfile->queue_index = tun->numqueues;
 836         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 837
 838         if (tfile->detached) {
 839                 /* Re-attach detached tfile, updating XDP queue_index */
 840                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 841
 842                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 843                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 844         } else {
 845                 /* Setup XDP RX-queue info, for new tfile getting attached */
 846                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 847                                        tun->dev, tfile->queue_index);
 848                 if (err < 0)
 849                         goto out;
 850                 err = 0;
 851         }
 852
 853         rcu_assign_pointer(tfile->tun, tun);
 854         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 855         tun->numqueues++;
 856
 857         if (tfile->detached) {
 858                 tun_enable_queue(tfile);
 859         } else {
 860                 sock_hold(&tfile->sk);
 861                 tun_napi_init(tun, tfile, napi);
 862         }
 863
 864         tun_set_real_num_queues(tun);
 865
 866         /* device is allowed to go away first, so no need to hold extra
 867          * refcnt.
 868          */
 869
 870 out:
 871         return err;
 872 }
 873
 874 static struct tun_struct *tun_get(struct tun_file *tfile)
 875 {
 876         struct tun_struct *tun;
 877
 878         rcu_read_lock();
 879         tun = rcu_dereference(tfile->tun);
 880         if (tun)
 881                 dev_hold(tun->dev);
 882         rcu_read_unlock();
 883
 884         return tun;
 885 }
 886
 887 static void tun_put(struct tun_struct *tun)
 888 {
 889         dev_put(tun->dev);
 890 }
 891
 892 /* TAP filtering */
 893 static void addr_hash_set(u32 *mask, const u8 *addr)
 894 {
 895         int n = ether_crc(ETH_ALEN, addr) >> 26;
 896         mask[n >> 5] |= (1 << (n & 31));
 897 }
 898
 899 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 900 {
 901         int n = ether_crc(ETH_ALEN, addr) >> 26;
 902         return mask[n >> 5] & (1 << (n & 31));
 903 }
 904
 905 static int update_filter(struct tap_filter *filter, void __user *arg)
 906 {
 907         struct { u8 u[ETH_ALEN]; } *addr;
 908         struct tun_filter uf;
 909         int err, alen, n, nexact;
 910
 911         if (copy_from_user(&uf, arg, sizeof(uf)))
 912                 return -EFAULT;
 913
 914         if (!uf.count) {
 915                 /* Disabled */
 916                 filter->count = 0;
 917                 return 0;
 918         }
 919
 920         alen = ETH_ALEN * uf.count;
 921         addr = memdup_user(arg + sizeof(uf), alen);
 922         if (IS_ERR(addr))
 923                 return PTR_ERR(addr);
 924
 925         /* The filter is updated without holding any locks. Which is
 926          * perfectly safe. We disable it first and in the worst
 927          * case we'll accept a few undesired packets. */
 928         filter->count = 0;
 929         wmb();
 930
 931         /* Use first set of addresses as an exact filter */
 932         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 933                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 934
 935         nexact = n;
 936
 937         /* Remaining multicast addresses are hashed,
 938          * unicast will leave the filter disabled. */
 939         memset(filter->mask, 0, sizeof(filter->mask));
 940         for (; n < uf.count; n++) {
 941                 if (!is_multicast_ether_addr(addr[n].u)) {
 942                         err = 0; /* no filter */
 943                         goto free_addr;
 944                 }
 945                 addr_hash_set(filter->mask, addr[n].u);
 946         }
 947
 948         /* For ALLMULTI just set the mask to all ones.
 949          * This overrides the mask populated above. */
 950         if ((uf.flags & TUN_FLT_ALLMULTI))
 951                 memset(filter->mask, ~0, sizeof(filter->mask));
 952
 953         /* Now enable the filter */
 954         wmb();
 955         filter->count = nexact;
 956
 957         /* Return the number of exact filters */
 958         err = nexact;
 959 free_addr:
 960         kfree(addr);
 961         return err;
 962 }
 963
 964 /* Returns: 0 - drop, !=0 - accept */
 965 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 966 {
 967         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 968          * at this point. */
 969         struct ethhdr *eh = (struct ethhdr *) skb->data;
 970         int i;
 971
 972         /* Exact match */
 973         for (i = 0; i < filter->count; i++)
 974                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 975                         return 1;
 976
 977         /* Inexact match (multicast only) */
 978         if (is_multicast_ether_addr(eh->h_dest))
 979                 return addr_hash_test(filter->mask, eh->h_dest);
 980
 981         return 0;
 982 }
 983
 984 /*
 985  * Checks whether the packet is accepted or not.
 986  * Returns: 0 - drop, !=0 - accept
 987  */
 988 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 989 {
 990         if (!filter->count)
 991                 return 1;
 992
 993         return run_filter(filter, skb);
 994 }
 995
 996 /* Network device part of the driver */
 997
 998 static const struct ethtool_ops tun_ethtool_ops;
 999
1000 /* Net device detach from fd. */
1001 static void tun_net_uninit(struct net_device *dev)
1002 {
1003         tun_detach_all(dev);
1004 }
1005
1006 /* Net device open. */
1007 static int tun_net_open(struct net_device *dev)
1008 {
1009         struct tun_struct *tun = netdev_priv(dev);
1010         int i;
1011
1012         netif_tx_start_all_queues(dev);
1013
1014         for (i = 0; i < tun->numqueues; i++) {
1015                 struct tun_file *tfile;
1016
1017                 tfile = rtnl_dereference(tun->tfiles[i]);
1018                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1019         }
1020
1021         return 0;
1022 }
1023
1024 /* Net device close. */
1025 static int tun_net_close(struct net_device *dev)
1026 {
1027         netif_tx_stop_all_queues(dev);
1028         return 0;
1029 }
1030
1031 /* Net device start xmit */
1032 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1033 {
1034 #ifdef CONFIG_RPS
1035         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1036                 /* Select queue was not called for the skbuff, so we extract the
1037                  * RPS hash and save it into the flow_table here.
1038                  */
1039                 __u32 rxhash;
1040
1041                 rxhash = __skb_get_hash_symmetric(skb);
1042                 if (rxhash) {
1043                         struct tun_flow_entry *e;
1044                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1045                                         rxhash);
1046                         if (e)
1047                                 tun_flow_save_rps_rxhash(e, rxhash);
1048                 }
1049         }
1050 #endif
1051 }
1052
1053 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1054                                     struct sk_buff *skb,
1055                                     int len)
1056 {
1057         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1058
1059         if (prog)
1060                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1061
1062         return len;
1063 }
1064
1065 /* Net device start xmit */
1066 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1067 {
1068         struct tun_struct *tun = netdev_priv(dev);
1069         int txq = skb->queue_mapping;
1070         struct tun_file *tfile;
1071         int len = skb->len;
1072
1073         rcu_read_lock();
1074         tfile = rcu_dereference(tun->tfiles[txq]);
1075
1076         /* Drop packet if interface is not attached */
1077         if (txq >= tun->numqueues)
1078                 goto drop;
1079
1080         if (!rcu_dereference(tun->steering_prog))
1081                 tun_automq_xmit(tun, skb);
1082
1083         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1084
1085         BUG_ON(!tfile);
1086
1087         /* Drop if the filter does not like it.
1088          * This is a noop if the filter is disabled.
1089          * Filter can be enabled only for the TAP devices. */
1090         if (!check_filter(&tun->txflt, skb))
1091                 goto drop;
1092
1093         if (tfile->socket.sk->sk_filter &&
1094             sk_filter(tfile->socket.sk, skb))
1095                 goto drop;
1096
1097         len = run_ebpf_filter(tun, skb, len);
1098
1099         /* Trim extra bytes since we may insert vlan proto & TCI
1100          * in tun_put_user().
1101          */
1102         len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1103         if (len <= 0 || pskb_trim(skb, len))
1104                 goto drop;
1105
1106         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1107                 goto drop;
1108
1109         skb_tx_timestamp(skb);
1110
1111         /* Orphan the skb - required as we might hang on to it
1112          * for indefinite time.
1113          */
1114         skb_orphan(skb);
1115
1116         nf_reset(skb);
1117
1118         if (ptr_ring_produce(&tfile->tx_ring, skb))
1119                 goto drop;
1120
1121         /* Notify and wake up reader process */
1122         if (tfile->flags & TUN_FASYNC)
1123                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1124         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1125
1126         rcu_read_unlock();
1127         return NETDEV_TX_OK;
1128
1129 drop:
1130         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1131         skb_tx_error(skb);
1132         kfree_skb(skb);
1133         rcu_read_unlock();
1134         return NET_XMIT_DROP;
1135 }
1136
1137 static void tun_net_mclist(struct net_device *dev)
1138 {
1139         /*
1140          * This callback is supposed to deal with mc filter in
1141          * _rx_ path and has nothing to do with the _tx_ path.
1142          * In rx path we always accept everything userspace gives us.
1143          */
1144 }
1145
1146 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1147         netdev_features_t features)
1148 {
1149         struct tun_struct *tun = netdev_priv(dev);
1150
1151         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1152 }
1153 #ifdef CONFIG_NET_POLL_CONTROLLER
1154 static void tun_poll_controller(struct net_device *dev)
1155 {
1156         /*
1157          * Tun only receives frames when:
1158          * 1) the char device endpoint gets data from user space
1159          * 2) the tun socket gets a sendmsg call from user space
1160          * If NAPI is not enabled, since both of those are synchronous
1161          * operations, we are guaranteed never to have pending data when we poll
1162          * for it so there is nothing to do here but return.
1163          * We need this though so netpoll recognizes us as an interface that
1164          * supports polling, which enables bridge devices in virt setups to
1165          * still use netconsole
1166          * If NAPI is enabled, however, we need to schedule polling for all
1167          * queues unless we are using napi_gro_frags(), which we call in
1168          * process context and not in NAPI context.
1169          */
1170         struct tun_struct *tun = netdev_priv(dev);
1171
1172         if (tun->flags & IFF_NAPI) {
1173                 struct tun_file *tfile;
1174                 int i;
1175
1176                 if (tun_napi_frags_enabled(tun))
1177                         return;
1178
1179                 rcu_read_lock();
1180                 for (i = 0; i < tun->numqueues; i++) {
1181                         tfile = rcu_dereference(tun->tfiles[i]);
1182                         if (tfile->napi_enabled)
1183                                 napi_schedule(&tfile->napi);
1184                 }
1185                 rcu_read_unlock();
1186         }
1187         return;
1188 }
1189 #endif
1190
1191 static void tun_set_headroom(struct net_device *dev, int new_hr)
1192 {
1193         struct tun_struct *tun = netdev_priv(dev);
1194
1195         if (new_hr < NET_SKB_PAD)
1196                 new_hr = NET_SKB_PAD;
1197
1198         tun->align = new_hr;
1199 }
1200
1201 static void
1202 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1203 {
1204         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1205         struct tun_struct *tun = netdev_priv(dev);
1206         struct tun_pcpu_stats *p;
1207         int i;
1208
1209         for_each_possible_cpu(i) {
1210                 u64 rxpackets, rxbytes, txpackets, txbytes;
1211                 unsigned int start;
1212
1213                 p = per_cpu_ptr(tun->pcpu_stats, i);
1214                 do {
1215                         start = u64_stats_fetch_begin(&p->syncp);
1216                         rxpackets       = p->rx_packets;
1217                         rxbytes         = p->rx_bytes;
1218                         txpackets       = p->tx_packets;
1219                         txbytes         = p->tx_bytes;
1220                 } while (u64_stats_fetch_retry(&p->syncp, start));
1221
1222                 stats->rx_packets       += rxpackets;
1223                 stats->rx_bytes         += rxbytes;
1224                 stats->tx_packets       += txpackets;
1225                 stats->tx_bytes         += txbytes;
1226
1227                 /* u32 counters */
1228                 rx_dropped      += p->rx_dropped;
1229                 rx_frame_errors += p->rx_frame_errors;
1230                 tx_dropped      += p->tx_dropped;
1231         }
1232         stats->rx_dropped  = rx_dropped;
1233         stats->rx_frame_errors = rx_frame_errors;
1234         stats->tx_dropped = tx_dropped;
1235 }
1236
1237 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1238                        struct netlink_ext_ack *extack)
1239 {
1240         struct tun_struct *tun = netdev_priv(dev);
1241         struct bpf_prog *old_prog;
1242
1243         old_prog = rtnl_dereference(tun->xdp_prog);
1244         rcu_assign_pointer(tun->xdp_prog, prog);
1245         if (old_prog)
1246                 bpf_prog_put(old_prog);
1247
1248         return 0;
1249 }
1250
1251 static u32 tun_xdp_query(struct net_device *dev)
1252 {
1253         struct tun_struct *tun = netdev_priv(dev);
1254         const struct bpf_prog *xdp_prog;
1255
1256         xdp_prog = rtnl_dereference(tun->xdp_prog);
1257         if (xdp_prog)
1258                 return xdp_prog->aux->id;
1259
1260         return 0;
1261 }
1262
1263 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1264 {
1265         switch (xdp->command) {
1266         case XDP_SETUP_PROG:
1267                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1268         case XDP_QUERY_PROG:
1269                 xdp->prog_id = tun_xdp_query(dev);
1270                 xdp->prog_attached = !!xdp->prog_id;
1271                 return 0;
1272         default:
1273                 return -EINVAL;
1274         }
1275 }
1276
1277 static const struct net_device_ops tun_netdev_ops = {
1278         .ndo_uninit             = tun_net_uninit,
1279         .ndo_open               = tun_net_open,
1280         .ndo_stop               = tun_net_close,
1281         .ndo_start_xmit         = tun_net_xmit,
1282         .ndo_fix_features       = tun_net_fix_features,
1283         .ndo_select_queue       = tun_select_queue,
1284 #ifdef CONFIG_NET_POLL_CONTROLLER
1285         .ndo_poll_controller    = tun_poll_controller,
1286 #endif
1287         .ndo_set_rx_headroom    = tun_set_headroom,
1288         .ndo_get_stats64        = tun_net_get_stats64,
1289 };
1290
1291 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1292 {
1293         struct tun_struct *tun = netdev_priv(dev);
1294         struct xdp_buff *buff = xdp->data_hard_start;
1295         int headroom = xdp->data - xdp->data_hard_start;
1296         struct tun_file *tfile;
1297         u32 numqueues;
1298         int ret = 0;
1299
1300         /* Assure headroom is available and buff is properly aligned */
1301         if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1302                 return -ENOSPC;
1303
1304         *buff = *xdp;
1305
1306         rcu_read_lock();
1307
1308         numqueues = READ_ONCE(tun->numqueues);
1309         if (!numqueues) {
1310                 ret = -ENOSPC;
1311                 goto out;
1312         }
1313
1314         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1315                                             numqueues]);
1316         /* Encode the XDP flag into lowest bit for consumer to differ
1317          * XDP buffer from sk_buff.
1318          */
1319         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1320                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1321                 ret = -ENOSPC;
1322         }
1323
1324 out:
1325         rcu_read_unlock();
1326         return ret;
1327 }
1328
1329 static void tun_xdp_flush(struct net_device *dev)
1330 {
1331         struct tun_struct *tun = netdev_priv(dev);
1332         struct tun_file *tfile;
1333         u32 numqueues;
1334
1335         rcu_read_lock();
1336
1337         numqueues = READ_ONCE(tun->numqueues);
1338         if (!numqueues)
1339                 goto out;
1340
1341         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1342                                             numqueues]);
1343         /* Notify and wake up reader process */
1344         if (tfile->flags & TUN_FASYNC)
1345                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1346         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1347
1348 out:
1349         rcu_read_unlock();
1350 }
1351
1352 static const struct net_device_ops tap_netdev_ops = {
1353         .ndo_uninit             = tun_net_uninit,
1354         .ndo_open               = tun_net_open,
1355         .ndo_stop               = tun_net_close,
1356         .ndo_start_xmit         = tun_net_xmit,
1357         .ndo_fix_features       = tun_net_fix_features,
1358         .ndo_set_rx_mode        = tun_net_mclist,
1359         .ndo_set_mac_address    = eth_mac_addr,
1360         .ndo_validate_addr      = eth_validate_addr,
1361         .ndo_select_queue       = tun_select_queue,
1362 #ifdef CONFIG_NET_POLL_CONTROLLER
1363         .ndo_poll_controller    = tun_poll_controller,
1364 #endif
1365         .ndo_features_check     = passthru_features_check,
1366         .ndo_set_rx_headroom    = tun_set_headroom,
1367         .ndo_get_stats64        = tun_net_get_stats64,
1368         .ndo_bpf                = tun_xdp,
1369         .ndo_xdp_xmit           = tun_xdp_xmit,
1370         .ndo_xdp_flush          = tun_xdp_flush,
1371 };
1372
1373 static void tun_flow_init(struct tun_struct *tun)
1374 {
1375         int i;
1376
1377         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1378                 INIT_HLIST_HEAD(&tun->flows[i]);
1379
1380         tun->ageing_time = TUN_FLOW_EXPIRE;
1381         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1382         mod_timer(&tun->flow_gc_timer,
1383                   round_jiffies_up(jiffies + tun->ageing_time));
1384 }
1385
1386 static void tun_flow_uninit(struct tun_struct *tun)
1387 {
1388         del_timer_sync(&tun->flow_gc_timer);
1389         tun_flow_flush(tun);
1390 }
1391
1392 #define MIN_MTU 68
1393 #define MAX_MTU 65535
1394
1395 /* Initialize net device. */
1396 static void tun_net_init(struct net_device *dev)
1397 {
1398         struct tun_struct *tun = netdev_priv(dev);
1399
1400         switch (tun->flags & TUN_TYPE_MASK) {
1401         case IFF_TUN:
1402                 dev->netdev_ops = &tun_netdev_ops;
1403
1404                 /* Point-to-Point TUN Device */
1405                 dev->hard_header_len = 0;
1406                 dev->addr_len = 0;
1407                 dev->mtu = 1500;
1408
1409                 /* Zero header length */
1410                 dev->type = ARPHRD_NONE;
1411                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1412                 break;
1413
1414         case IFF_TAP:
1415                 dev->netdev_ops = &tap_netdev_ops;
1416                 /* Ethernet TAP Device */
1417                 ether_setup(dev);
1418                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1420
1421                 eth_hw_addr_random(dev);
1422
1423                 break;
1424         }
1425
1426         dev->min_mtu = MIN_MTU;
1427         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1428 }
1429
1430 /* Character device part */
1431
1432 /* Poll */
1433 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1434 {
1435         struct tun_file *tfile = file->private_data;
1436         struct tun_struct *tun = tun_get(tfile);
1437         struct sock *sk;
1438         __poll_t mask = 0;
1439
1440         if (!tun)
1441                 return EPOLLERR;
1442
1443         sk = tfile->socket.sk;
1444
1445         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1446
1447         poll_wait(file, sk_sleep(sk), wait);
1448
1449         if (!ptr_ring_empty(&tfile->tx_ring))
1450                 mask |= EPOLLIN | EPOLLRDNORM;
1451
1452         if (tun->dev->flags & IFF_UP &&
1453             (sock_writeable(sk) ||
1454              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1455               sock_writeable(sk))))
1456                 mask |= EPOLLOUT | EPOLLWRNORM;
1457
1458         if (tun->dev->reg_state != NETREG_REGISTERED)
1459                 mask = EPOLLERR;
1460
1461         tun_put(tun);
1462         return mask;
1463 }
1464
1465 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1466                                             size_t len,
1467                                             const struct iov_iter *it)
1468 {
1469         struct sk_buff *skb;
1470         size_t linear;
1471         int err;
1472         int i;
1473
1474         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1475                 return ERR_PTR(-ENOMEM);
1476
1477         local_bh_disable();
1478         skb = napi_get_frags(&tfile->napi);
1479         local_bh_enable();
1480         if (!skb)
1481                 return ERR_PTR(-ENOMEM);
1482
1483         linear = iov_iter_single_seg_count(it);
1484         err = __skb_grow(skb, linear);
1485         if (err)
1486                 goto free;
1487
1488         skb->len = len;
1489         skb->data_len = len - linear;
1490         skb->truesize += skb->data_len;
1491
1492         for (i = 1; i < it->nr_segs; i++) {
1493                 struct page_frag *pfrag = &current->task_frag;
1494                 size_t fragsz = it->iov[i].iov_len;
1495
1496                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1497                         err = -EINVAL;
1498                         goto free;
1499                 }
1500
1501                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1502                         err = -ENOMEM;
1503                         goto free;
1504                 }
1505
1506                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1507                                    pfrag->offset, fragsz);
1508                 page_ref_inc(pfrag->page);
1509                 pfrag->offset += fragsz;
1510         }
1511
1512         return skb;
1513 free:
1514         /* frees skb and all frags allocated with napi_alloc_frag() */
1515         napi_free_frags(&tfile->napi);
1516         return ERR_PTR(err);
1517 }
1518
1519 /* prepad is the amount to reserve at front.  len is length after that.
1520  * linear is a hint as to how much to copy (usually headers). */
1521 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1522                                      size_t prepad, size_t len,
1523                                      size_t linear, int noblock)
1524 {
1525         struct sock *sk = tfile->socket.sk;
1526         struct sk_buff *skb;
1527         int err;
1528
1529         /* Under a page?  Don't bother with paged skb. */
1530         if (prepad + len < PAGE_SIZE || !linear)
1531                 linear = len;
1532
1533         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1534                                    &err, 0);
1535         if (!skb)
1536                 return ERR_PTR(err);
1537
1538         skb_reserve(skb, prepad);
1539         skb_put(skb, linear);
1540         skb->data_len = len - linear;
1541         skb->len += len - linear;
1542
1543         return skb;
1544 }
1545
1546 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1547                            struct sk_buff *skb, int more)
1548 {
1549         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1550         struct sk_buff_head process_queue;
1551         u32 rx_batched = tun->rx_batched;
1552         bool rcv = false;
1553
1554         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1555                 local_bh_disable();
1556                 netif_receive_skb(skb);
1557                 local_bh_enable();
1558                 return;
1559         }
1560
1561         spin_lock(&queue->lock);
1562         if (!more || skb_queue_len(queue) == rx_batched) {
1563                 __skb_queue_head_init(&process_queue);
1564                 skb_queue_splice_tail_init(queue, &process_queue);
1565                 rcv = true;
1566         } else {
1567                 __skb_queue_tail(queue, skb);
1568         }
1569         spin_unlock(&queue->lock);
1570
1571         if (rcv) {
1572                 struct sk_buff *nskb;
1573
1574                 local_bh_disable();
1575                 while ((nskb = __skb_dequeue(&process_queue)))
1576                         netif_receive_skb(nskb);
1577                 netif_receive_skb(skb);
1578                 local_bh_enable();
1579         }
1580 }
1581
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583                               int len, int noblock, bool zerocopy)
1584 {
1585         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1586                 return false;
1587
1588         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1589                 return false;
1590
1591         if (!noblock)
1592                 return false;
1593
1594         if (zerocopy)
1595                 return false;
1596
1597         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1598             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1599                 return false;
1600
1601         return true;
1602 }
1603
1604 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1605                                      struct tun_file *tfile,
1606                                      struct iov_iter *from,
1607                                      struct virtio_net_hdr *hdr,
1608                                      int len, int *skb_xdp)
1609 {
1610         struct page_frag *alloc_frag = &current->task_frag;
1611         struct sk_buff *skb;
1612         struct bpf_prog *xdp_prog;
1613         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1614         unsigned int delta = 0;
1615         char *buf;
1616         size_t copied;
1617         int err, pad = TUN_RX_PAD;
1618
1619         rcu_read_lock();
1620         xdp_prog = rcu_dereference(tun->xdp_prog);
1621         if (xdp_prog)
1622                 pad += TUN_HEADROOM;
1623         buflen += SKB_DATA_ALIGN(len + pad);
1624         rcu_read_unlock();
1625
1626         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628                 return ERR_PTR(-ENOMEM);
1629
1630         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631         copied = copy_page_from_iter(alloc_frag->page,
1632                                      alloc_frag->offset + pad,
1633                                      len, from);
1634         if (copied != len)
1635                 return ERR_PTR(-EFAULT);
1636
1637         /* There's a small window that XDP may be set after the check
1638          * of xdp_prog above, this should be rare and for simplicity
1639          * we do XDP on skb in case the headroom is not enough.
1640          */
1641         if (hdr->gso_type || !xdp_prog)
1642                 *skb_xdp = 1;
1643         else
1644                 *skb_xdp = 0;
1645
1646         preempt_disable();
1647         rcu_read_lock();
1648         xdp_prog = rcu_dereference(tun->xdp_prog);
1649         if (xdp_prog && !*skb_xdp) {
1650                 struct xdp_buff xdp;
1651                 void *orig_data;
1652                 u32 act;
1653
1654                 xdp.data_hard_start = buf;
1655                 xdp.data = buf + pad;
1656                 xdp_set_data_meta_invalid(&xdp);
1657                 xdp.data_end = xdp.data + len;
1658                 xdp.rxq = &tfile->xdp_rxq;
1659                 orig_data = xdp.data;
1660                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1661
1662                 switch (act) {
1663                 case XDP_REDIRECT:
1664                         get_page(alloc_frag->page);
1665                         alloc_frag->offset += buflen;
1666                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1667                         xdp_do_flush_map();
1668                         if (err)
1669                                 goto err_redirect;
1670                         rcu_read_unlock();
1671                         preempt_enable();
1672                         return NULL;
1673                 case XDP_TX:
1674                         get_page(alloc_frag->page);
1675                         alloc_frag->offset += buflen;
1676                         if (tun_xdp_xmit(tun->dev, &xdp))
1677                                 goto err_redirect;
1678                         tun_xdp_flush(tun->dev);
1679                         rcu_read_unlock();
1680                         preempt_enable();
1681                         return NULL;
1682                 case XDP_PASS:
1683                         delta = orig_data - xdp.data;
1684                         break;
1685                 default:
1686                         bpf_warn_invalid_xdp_action(act);
1687                         /* fall through */
1688                 case XDP_ABORTED:
1689                         trace_xdp_exception(tun->dev, xdp_prog, act);
1690                         /* fall through */
1691                 case XDP_DROP:
1692                         goto err_xdp;
1693                 }
1694         }
1695
1696         skb = build_skb(buf, buflen);
1697         if (!skb) {
1698                 rcu_read_unlock();
1699                 preempt_enable();
1700                 return ERR_PTR(-ENOMEM);
1701         }
1702
1703         skb_reserve(skb, pad - delta);
1704         skb_put(skb, len + delta);
1705         get_page(alloc_frag->page);
1706         alloc_frag->offset += buflen;
1707
1708         rcu_read_unlock();
1709         preempt_enable();
1710
1711         return skb;
1712
1713 err_redirect:
1714         put_page(alloc_frag->page);
1715 err_xdp:
1716         rcu_read_unlock();
1717         preempt_enable();
1718         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1719         return NULL;
1720 }
1721
1722 /* Get packet from user space buffer */
1723 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1724                             void *msg_control, struct iov_iter *from,
1725                             int noblock, bool more)
1726 {
1727         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1728         struct sk_buff *skb;
1729         size_t total_len = iov_iter_count(from);
1730         size_t len = total_len, align = tun->align, linear;
1731         struct virtio_net_hdr gso = { 0 };
1732         struct tun_pcpu_stats *stats;
1733         int good_linear;
1734         int copylen;
1735         bool zerocopy = false;
1736         int err;
1737         u32 rxhash = 0;
1738         int skb_xdp = 1;
1739         bool frags = tun_napi_frags_enabled(tun);
1740
1741         if (!(tun->dev->flags & IFF_UP))
1742                 return -EIO;
1743
1744         if (!(tun->flags & IFF_NO_PI)) {
1745                 if (len < sizeof(pi))
1746                         return -EINVAL;
1747                 len -= sizeof(pi);
1748
1749                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1750                         return -EFAULT;
1751         }
1752
1753         if (tun->flags & IFF_VNET_HDR) {
1754                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1755
1756                 if (len < vnet_hdr_sz)
1757                         return -EINVAL;
1758                 len -= vnet_hdr_sz;
1759
1760                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1761                         return -EFAULT;
1762
1763                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1764                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1765                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1766
1767                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1768                         return -EINVAL;
1769                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1770         }
1771
1772         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1773                 align += NET_IP_ALIGN;
1774                 if (unlikely(len < ETH_HLEN ||
1775                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1776                         return -EINVAL;
1777         }
1778
1779         good_linear = SKB_MAX_HEAD(align);
1780
1781         if (msg_control) {
1782                 struct iov_iter i = *from;
1783
1784                 /* There are 256 bytes to be copied in skb, so there is
1785                  * enough room for skb expand head in case it is used.
1786                  * The rest of the buffer is mapped from userspace.
1787                  */
1788                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1789                 if (copylen > good_linear)
1790                         copylen = good_linear;
1791                 linear = copylen;
1792                 iov_iter_advance(&i, copylen);
1793                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1794                         zerocopy = true;
1795         }
1796
1797         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1798                 /* For the packet that is not easy to be processed
1799                  * (e.g gso or jumbo packet), we will do it at after
1800                  * skb was created with generic XDP routine.
1801                  */
1802                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1803                 if (IS_ERR(skb)) {
1804                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1805                         return PTR_ERR(skb);
1806                 }
1807                 if (!skb)
1808                         return total_len;
1809         } else {
1810                 if (!zerocopy) {
1811                         copylen = len;
1812                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1813                                 linear = good_linear;
1814                         else
1815                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1816                 }
1817
1818                 if (frags) {
1819                         mutex_lock(&tfile->napi_mutex);
1820                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1821                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1822                          * If zerocopy is enabled, then this layout will be
1823                          * overwritten by zerocopy_sg_from_iter().
1824                          */
1825                         zerocopy = false;
1826                 } else {
1827                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1828                                             noblock);
1829                 }
1830
1831                 if (IS_ERR(skb)) {
1832                         if (PTR_ERR(skb) != -EAGAIN)
1833                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1834                         if (frags)
1835                                 mutex_unlock(&tfile->napi_mutex);
1836                         return PTR_ERR(skb);
1837                 }
1838
1839                 if (zerocopy)
1840                         err = zerocopy_sg_from_iter(skb, from);
1841                 else
1842                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1843
1844                 if (err) {
1845                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1846                         kfree_skb(skb);
1847                         if (frags) {
1848                                 tfile->napi.skb = NULL;
1849                                 mutex_unlock(&tfile->napi_mutex);
1850                         }
1851
1852                         return -EFAULT;
1853                 }
1854         }
1855
1856         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1857                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1858                 kfree_skb(skb);
1859                 if (frags) {
1860                         tfile->napi.skb = NULL;
1861                         mutex_unlock(&tfile->napi_mutex);
1862                 }
1863
1864                 return -EINVAL;
1865         }
1866
1867         switch (tun->flags & TUN_TYPE_MASK) {
1868         case IFF_TUN:
1869                 if (tun->flags & IFF_NO_PI) {
1870                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1871
1872                         switch (ip_version) {
1873                         case 4:
1874                                 pi.proto = htons(ETH_P_IP);
1875                                 break;
1876                         case 6:
1877                                 pi.proto = htons(ETH_P_IPV6);
1878                                 break;
1879                         default:
1880                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1881                                 kfree_skb(skb);
1882                                 return -EINVAL;
1883                         }
1884                 }
1885
1886                 skb_reset_mac_header(skb);
1887                 skb->protocol = pi.proto;
1888                 skb->dev = tun->dev;
1889                 break;
1890         case IFF_TAP:
1891                 if (!frags)
1892                         skb->protocol = eth_type_trans(skb, tun->dev);
1893                 break;
1894         }
1895
1896         /* copy skb_ubuf_info for callback when skb has no error */
1897         if (zerocopy) {
1898                 skb_shinfo(skb)->destructor_arg = msg_control;
1899                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1900                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1901         } else if (msg_control) {
1902                 struct ubuf_info *uarg = msg_control;
1903                 uarg->callback(uarg, false);
1904         }
1905
1906         skb_reset_network_header(skb);
1907         skb_probe_transport_header(skb, 0);
1908
1909         if (skb_xdp) {
1910                 struct bpf_prog *xdp_prog;
1911                 int ret;
1912
1913                 rcu_read_lock();
1914                 xdp_prog = rcu_dereference(tun->xdp_prog);
1915                 if (xdp_prog) {
1916                         ret = do_xdp_generic(xdp_prog, skb);
1917                         if (ret != XDP_PASS) {
1918                                 rcu_read_unlock();
1919                                 return total_len;
1920                         }
1921                 }
1922                 rcu_read_unlock();
1923         }
1924
1925         rcu_read_lock();
1926         if (!rcu_dereference(tun->steering_prog))
1927                 rxhash = __skb_get_hash_symmetric(skb);
1928         rcu_read_unlock();
1929
1930         if (frags) {
1931                 /* Exercise flow dissector code path. */
1932                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1933
1934                 if (unlikely(headlen > skb_headlen(skb))) {
1935                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1936                         napi_free_frags(&tfile->napi);
1937                         mutex_unlock(&tfile->napi_mutex);
1938                         WARN_ON(1);
1939                         return -ENOMEM;
1940                 }
1941
1942                 local_bh_disable();
1943                 napi_gro_frags(&tfile->napi);
1944                 local_bh_enable();
1945                 mutex_unlock(&tfile->napi_mutex);
1946         } else if (tfile->napi_enabled) {
1947                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1948                 int queue_len;
1949
1950                 spin_lock_bh(&queue->lock);
1951                 __skb_queue_tail(queue, skb);
1952                 queue_len = skb_queue_len(queue);
1953                 spin_unlock(&queue->lock);
1954
1955                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1956                         napi_schedule(&tfile->napi);
1957
1958                 local_bh_enable();
1959         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1960                 tun_rx_batched(tun, tfile, skb, more);
1961         } else {
1962                 netif_rx_ni(skb);
1963         }
1964
1965         stats = get_cpu_ptr(tun->pcpu_stats);
1966         u64_stats_update_begin(&stats->syncp);
1967         stats->rx_packets++;
1968         stats->rx_bytes += len;
1969         u64_stats_update_end(&stats->syncp);
1970         put_cpu_ptr(stats);
1971
1972         if (rxhash)
1973                 tun_flow_update(tun, rxhash, tfile);
1974
1975         return total_len;
1976 }
1977
1978 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1979 {
1980         struct file *file = iocb->ki_filp;
1981         struct tun_file *tfile = file->private_data;
1982         struct tun_struct *tun = tun_get(tfile);
1983         ssize_t result;
1984
1985         if (!tun)
1986                 return -EBADFD;
1987
1988         result = tun_get_user(tun, tfile, NULL, from,
1989                               file->f_flags & O_NONBLOCK, false);
1990
1991         tun_put(tun);
1992         return result;
1993 }
1994
1995 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1996                                 struct tun_file *tfile,
1997                                 struct xdp_buff *xdp,
1998                                 struct iov_iter *iter)
1999 {
2000         int vnet_hdr_sz = 0;
2001         size_t size = xdp->data_end - xdp->data;
2002         struct tun_pcpu_stats *stats;
2003         size_t ret;
2004
2005         if (tun->flags & IFF_VNET_HDR) {
2006                 struct virtio_net_hdr gso = { 0 };
2007
2008                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2009                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2010                         return -EINVAL;
2011                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2012                              sizeof(gso)))
2013                         return -EFAULT;
2014                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2015         }
2016
2017         ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2018
2019         stats = get_cpu_ptr(tun->pcpu_stats);
2020         u64_stats_update_begin(&stats->syncp);
2021         stats->tx_packets++;
2022         stats->tx_bytes += ret;
2023         u64_stats_update_end(&stats->syncp);
2024         put_cpu_ptr(tun->pcpu_stats);
2025
2026         return ret;
2027 }
2028
2029 /* Put packet to the user space buffer */
2030 static ssize_t tun_put_user(struct tun_struct *tun,
2031                             struct tun_file *tfile,
2032                             struct sk_buff *skb,
2033                             struct iov_iter *iter)
2034 {
2035         struct tun_pi pi = { 0, skb->protocol };
2036         struct tun_pcpu_stats *stats;
2037         ssize_t total;
2038         int vlan_offset = 0;
2039         int vlan_hlen = 0;
2040         int vnet_hdr_sz = 0;
2041
2042         if (skb_vlan_tag_present(skb))
2043                 vlan_hlen = VLAN_HLEN;
2044
2045         if (tun->flags & IFF_VNET_HDR)
2046                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2047
2048         total = skb->len + vlan_hlen + vnet_hdr_sz;
2049
2050         if (!(tun->flags & IFF_NO_PI)) {
2051                 if (iov_iter_count(iter) < sizeof(pi))
2052                         return -EINVAL;
2053
2054                 total += sizeof(pi);
2055                 if (iov_iter_count(iter) < total) {
2056                         /* Packet will be striped */
2057                         pi.flags |= TUN_PKT_STRIP;
2058                 }
2059
2060                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2061                         return -EFAULT;
2062         }
2063
2064         if (vnet_hdr_sz) {
2065                 struct virtio_net_hdr gso;
2066
2067                 if (iov_iter_count(iter) < vnet_hdr_sz)
2068                         return -EINVAL;
2069
2070                 if (virtio_net_hdr_from_skb(skb, &gso,
2071                                             tun_is_little_endian(tun), true)) {
2072                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2073                         pr_err("unexpected GSO type: "
2074                                "0x%x, gso_size %d, hdr_len %d\n",
2075                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2076                                tun16_to_cpu(tun, gso.hdr_len));
2077                         print_hex_dump(KERN_ERR, "tun: ",
2078                                        DUMP_PREFIX_NONE,
2079                                        16, 1, skb->head,
2080                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2081                         WARN_ON_ONCE(1);
2082                         return -EINVAL;
2083                 }
2084
2085                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2086                         return -EFAULT;
2087
2088                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2089         }
2090
2091         if (vlan_hlen) {
2092                 int ret;
2093                 struct veth veth;
2094
2095                 veth.h_vlan_proto = skb->vlan_proto;
2096                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2097
2098                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2099
2100                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2101                 if (ret || !iov_iter_count(iter))
2102                         goto done;
2103
2104                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2105                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2106                         goto done;
2107         }
2108
2109         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2110
2111 done:
2112         /* caller is in process context, */
2113         stats = get_cpu_ptr(tun->pcpu_stats);
2114         u64_stats_update_begin(&stats->syncp);
2115         stats->tx_packets++;
2116         stats->tx_bytes += skb->len + vlan_hlen;
2117         u64_stats_update_end(&stats->syncp);
2118         put_cpu_ptr(tun->pcpu_stats);
2119
2120         return total;
2121 }
2122
2123 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2124 {
2125         DECLARE_WAITQUEUE(wait, current);
2126         void *ptr = NULL;
2127         int error = 0;
2128
2129         ptr = ptr_ring_consume(&tfile->tx_ring);
2130         if (ptr)
2131                 goto out;
2132         if (noblock) {
2133                 error = -EAGAIN;
2134                 goto out;
2135         }
2136
2137         add_wait_queue(&tfile->wq.wait, &wait);
2138         current->state = TASK_INTERRUPTIBLE;
2139
2140         while (1) {
2141                 ptr = ptr_ring_consume(&tfile->tx_ring);
2142                 if (ptr)
2143                         break;
2144                 if (signal_pending(current)) {
2145                         error = -ERESTARTSYS;
2146                         break;
2147                 }
2148                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2149                         error = -EFAULT;
2150                         break;
2151                 }
2152
2153                 schedule();
2154         }
2155
2156         current->state = TASK_RUNNING;
2157         remove_wait_queue(&tfile->wq.wait, &wait);
2158
2159 out:
2160         *err = error;
2161         return ptr;
2162 }
2163
2164 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2165                            struct iov_iter *to,
2166                            int noblock, void *ptr)
2167 {
2168         ssize_t ret;
2169         int err;
2170
2171         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2172
2173         if (!iov_iter_count(to)) {
2174                 tun_ptr_free(ptr);
2175                 return 0;
2176         }
2177
2178         if (!ptr) {
2179                 /* Read frames from ring */
2180                 ptr = tun_ring_recv(tfile, noblock, &err);
2181                 if (!ptr)
2182                         return err;
2183         }
2184
2185         if (tun_is_xdp_buff(ptr)) {
2186                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2187
2188                 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2189                 put_page(virt_to_head_page(xdp->data));
2190         } else {
2191                 struct sk_buff *skb = ptr;
2192
2193                 ret = tun_put_user(tun, tfile, skb, to);
2194                 if (unlikely(ret < 0))
2195                         kfree_skb(skb);
2196                 else
2197                         consume_skb(skb);
2198         }
2199
2200         return ret;
2201 }
2202
2203 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2204 {
2205         struct file *file = iocb->ki_filp;
2206         struct tun_file *tfile = file->private_data;
2207         struct tun_struct *tun = tun_get(tfile);
2208         ssize_t len = iov_iter_count(to), ret;
2209
2210         if (!tun)
2211                 return -EBADFD;
2212         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2213         ret = min_t(ssize_t, ret, len);
2214         if (ret > 0)
2215                 iocb->ki_pos = ret;
2216         tun_put(tun);
2217         return ret;
2218 }
2219
2220 static void tun_prog_free(struct rcu_head *rcu)
2221 {
2222         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2223
2224         bpf_prog_destroy(prog->prog);
2225         kfree(prog);
2226 }
2227
2228 static int __tun_set_ebpf(struct tun_struct *tun,
2229                           struct tun_prog __rcu **prog_p,
2230                           struct bpf_prog *prog)
2231 {
2232         struct tun_prog *old, *new = NULL;
2233
2234         if (prog) {
2235                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2236                 if (!new)
2237                         return -ENOMEM;
2238                 new->prog = prog;
2239         }
2240
2241         spin_lock_bh(&tun->lock);
2242         old = rcu_dereference_protected(*prog_p,
2243                                         lockdep_is_held(&tun->lock));
2244         rcu_assign_pointer(*prog_p, new);
2245         spin_unlock_bh(&tun->lock);
2246
2247         if (old)
2248                 call_rcu(&old->rcu, tun_prog_free);
2249
2250         return 0;
2251 }
2252
2253 static void tun_free_netdev(struct net_device *dev)
2254 {
2255         struct tun_struct *tun = netdev_priv(dev);
2256
2257         BUG_ON(!(list_empty(&tun->disabled)));
2258         free_percpu(tun->pcpu_stats);
2259         tun_flow_uninit(tun);
2260         security_tun_dev_free_security(tun->security);
2261         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2262         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2263 }
2264
2265 static void tun_setup(struct net_device *dev)
2266 {
2267         struct tun_struct *tun = netdev_priv(dev);
2268
2269         tun->owner = INVALID_UID;
2270         tun->group = INVALID_GID;
2271
2272         dev->ethtool_ops = &tun_ethtool_ops;
2273         dev->needs_free_netdev = true;
2274         dev->priv_destructor = tun_free_netdev;
2275         /* We prefer our own queue length */
2276         dev->tx_queue_len = TUN_READQ_SIZE;
2277 }
2278
2279 /* Trivial set of netlink ops to allow deleting tun or tap
2280  * device with netlink.
2281  */
2282 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2283                         struct netlink_ext_ack *extack)
2284 {
2285         return -EINVAL;
2286 }
2287
2288 static size_t tun_get_size(const struct net_device *dev)
2289 {
2290         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2291         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2292
2293         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2294                nla_total_size(sizeof(gid_t)) + /* GROUP */
2295                nla_total_size(sizeof(u8)) + /* TYPE */
2296                nla_total_size(sizeof(u8)) + /* PI */
2297                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2298                nla_total_size(sizeof(u8)) + /* PERSIST */
2299                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2300                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2301                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2302                0;
2303 }
2304
2305 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2306 {
2307         struct tun_struct *tun = netdev_priv(dev);
2308
2309         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2310                 goto nla_put_failure;
2311         if (uid_valid(tun->owner) &&
2312             nla_put_u32(skb, IFLA_TUN_OWNER,
2313                         from_kuid_munged(current_user_ns(), tun->owner)))
2314                 goto nla_put_failure;
2315         if (gid_valid(tun->group) &&
2316             nla_put_u32(skb, IFLA_TUN_GROUP,
2317                         from_kgid_munged(current_user_ns(), tun->group)))
2318                 goto nla_put_failure;
2319         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2320                 goto nla_put_failure;
2321         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2322                 goto nla_put_failure;
2323         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2324                 goto nla_put_failure;
2325         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2326                        !!(tun->flags & IFF_MULTI_QUEUE)))
2327                 goto nla_put_failure;
2328         if (tun->flags & IFF_MULTI_QUEUE) {
2329                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2330                         goto nla_put_failure;
2331                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2332                                 tun->numdisabled))
2333                         goto nla_put_failure;
2334         }
2335
2336         return 0;
2337
2338 nla_put_failure:
2339         return -EMSGSIZE;
2340 }
2341
2342 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2343         .kind           = DRV_NAME,
2344         .priv_size      = sizeof(struct tun_struct),
2345         .setup          = tun_setup,
2346         .validate       = tun_validate,
2347         .get_size       = tun_get_size,
2348         .fill_info      = tun_fill_info,
2349 };
2350
2351 static void tun_sock_write_space(struct sock *sk)
2352 {
2353         struct tun_file *tfile;
2354         wait_queue_head_t *wqueue;
2355
2356         if (!sock_writeable(sk))
2357                 return;
2358
2359         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2360                 return;
2361
2362         wqueue = sk_sleep(sk);
2363         if (wqueue && waitqueue_active(wqueue))
2364                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2365                                                 EPOLLWRNORM | EPOLLWRBAND);
2366
2367         tfile = container_of(sk, struct tun_file, sk);
2368         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2369 }
2370
2371 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2372 {
2373         int ret;
2374         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2375         struct tun_struct *tun = tun_get(tfile);
2376
2377         if (!tun)
2378                 return -EBADFD;
2379
2380         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2381                            m->msg_flags & MSG_DONTWAIT,
2382                            m->msg_flags & MSG_MORE);
2383         tun_put(tun);
2384         return ret;
2385 }
2386
2387 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2388                        int flags)
2389 {
2390         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2391         struct tun_struct *tun = tun_get(tfile);
2392         void *ptr = m->msg_control;
2393         int ret;
2394
2395         if (!tun) {
2396                 ret = -EBADFD;
2397                 goto out_free;
2398         }
2399
2400         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2401                 ret = -EINVAL;
2402                 goto out_put_tun;
2403         }
2404         if (flags & MSG_ERRQUEUE) {
2405                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2406                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2407                 goto out;
2408         }
2409         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2410         if (ret > (ssize_t)total_len) {
2411                 m->msg_flags |= MSG_TRUNC;
2412                 ret = flags & MSG_TRUNC ? ret : total_len;
2413         }
2414 out:
2415         tun_put(tun);
2416         return ret;
2417
2418 out_put_tun:
2419         tun_put(tun);
2420 out_free:
2421         tun_ptr_free(ptr);
2422         return ret;
2423 }
2424
2425 static int tun_ptr_peek_len(void *ptr)
2426 {
2427         if (likely(ptr)) {
2428                 if (tun_is_xdp_buff(ptr)) {
2429                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2430
2431                         return xdp->data_end - xdp->data;
2432                 }
2433                 return __skb_array_len_with_tag(ptr);
2434         } else {
2435                 return 0;
2436         }
2437 }
2438
2439 static int tun_peek_len(struct socket *sock)
2440 {
2441         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2442         struct tun_struct *tun;
2443         int ret = 0;
2444
2445         tun = tun_get(tfile);
2446         if (!tun)
2447                 return 0;
2448
2449         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2450         tun_put(tun);
2451
2452         return ret;
2453 }
2454
2455 /* Ops structure to mimic raw sockets with tun */
2456 static const struct proto_ops tun_socket_ops = {
2457         .peek_len = tun_peek_len,
2458         .sendmsg = tun_sendmsg,
2459         .recvmsg = tun_recvmsg,
2460 };
2461
2462 static struct proto tun_proto = {
2463         .name           = "tun",
2464         .owner          = THIS_MODULE,
2465         .obj_size       = sizeof(struct tun_file),
2466 };
2467
2468 static int tun_flags(struct tun_struct *tun)
2469 {
2470         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2471 }
2472
2473 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2474                               char *buf)
2475 {
2476         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2477         return sprintf(buf, "0x%x\n", tun_flags(tun));
2478 }
2479
2480 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2481                               char *buf)
2482 {
2483         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2484         return uid_valid(tun->owner)?
2485                 sprintf(buf, "%u\n",
2486                         from_kuid_munged(current_user_ns(), tun->owner)):
2487                 sprintf(buf, "-1\n");
2488 }
2489
2490 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2491                               char *buf)
2492 {
2493         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2494         return gid_valid(tun->group) ?
2495                 sprintf(buf, "%u\n",
2496                         from_kgid_munged(current_user_ns(), tun->group)):
2497                 sprintf(buf, "-1\n");
2498 }
2499
2500 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2501 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2502 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2503
2504 static struct attribute *tun_dev_attrs[] = {
2505         &dev_attr_tun_flags.attr,
2506         &dev_attr_owner.attr,
2507         &dev_attr_group.attr,
2508         NULL
2509 };
2510
2511 static const struct attribute_group tun_attr_group = {
2512         .attrs = tun_dev_attrs
2513 };
2514
2515 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2516 {
2517         struct tun_struct *tun;
2518         struct tun_file *tfile = file->private_data;
2519         struct net_device *dev;
2520         int err;
2521
2522         if (tfile->detached)
2523                 return -EINVAL;
2524
2525         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2526                 if (!capable(CAP_NET_ADMIN))
2527                         return -EPERM;
2528
2529                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2530                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2531                         return -EINVAL;
2532         }
2533
2534         dev = __dev_get_by_name(net, ifr->ifr_name);
2535         if (dev) {
2536                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2537                         return -EBUSY;
2538                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2539                         tun = netdev_priv(dev);
2540                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2541                         tun = netdev_priv(dev);
2542                 else
2543                         return -EINVAL;
2544
2545                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2546                     !!(tun->flags & IFF_MULTI_QUEUE))
2547                         return -EINVAL;
2548
2549                 if (tun_not_capable(tun))
2550                         return -EPERM;
2551                 err = security_tun_dev_open(tun->security);
2552                 if (err < 0)
2553                         return err;
2554
2555                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2556                                  ifr->ifr_flags & IFF_NAPI);
2557                 if (err < 0)
2558                         return err;
2559
2560                 if (tun->flags & IFF_MULTI_QUEUE &&
2561                     (tun->numqueues + tun->numdisabled > 1)) {
2562                         /* One or more queue has already been attached, no need
2563                          * to initialize the device again.
2564                          */
2565                         return 0;
2566                 }
2567
2568                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2569                               (ifr->ifr_flags & TUN_FEATURES);
2570         }
2571         else {
2572                 char *name;
2573                 unsigned long flags = 0;
2574                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2575                              MAX_TAP_QUEUES : 1;
2576
2577                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2578                         return -EPERM;
2579                 err = security_tun_dev_create();
2580                 if (err < 0)
2581                         return err;
2582
2583                 /* Set dev type */
2584                 if (ifr->ifr_flags & IFF_TUN) {
2585                         /* TUN device */
2586                         flags |= IFF_TUN;
2587                         name = "tun%d";
2588                 } else if (ifr->ifr_flags & IFF_TAP) {
2589                         /* TAP device */
2590                         flags |= IFF_TAP;
2591                         name = "tap%d";
2592                 } else
2593                         return -EINVAL;
2594
2595                 if (*ifr->ifr_name)
2596                         name = ifr->ifr_name;
2597
2598                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2599                                        NET_NAME_UNKNOWN, tun_setup, queues,
2600                                        queues);
2601
2602                 if (!dev)
2603                         return -ENOMEM;
2604                 err = dev_get_valid_name(net, dev, name);
2605                 if (err < 0)
2606                         goto err_free_dev;
2607
2608                 dev_net_set(dev, net);
2609                 dev->rtnl_link_ops = &tun_link_ops;
2610                 dev->ifindex = tfile->ifindex;
2611                 dev->sysfs_groups[0] = &tun_attr_group;
2612
2613                 tun = netdev_priv(dev);
2614                 tun->dev = dev;
2615                 tun->flags = flags;
2616                 tun->txflt.count = 0;
2617                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2618
2619                 tun->align = NET_SKB_PAD;
2620                 tun->filter_attached = false;
2621                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2622                 tun->rx_batched = 0;
2623                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2624
2625                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2626                 if (!tun->pcpu_stats) {
2627                         err = -ENOMEM;
2628                         goto err_free_dev;
2629                 }
2630
2631                 spin_lock_init(&tun->lock);
2632
2633                 err = security_tun_dev_alloc_security(&tun->security);
2634                 if (err < 0)
2635                         goto err_free_stat;
2636
2637                 tun_net_init(dev);
2638                 tun_flow_init(tun);
2639
2640                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2641                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2642                                    NETIF_F_HW_VLAN_STAG_TX;
2643                 dev->features = dev->hw_features | NETIF_F_LLTX;
2644                 dev->vlan_features = dev->features &
2645                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2646                                        NETIF_F_HW_VLAN_STAG_TX);
2647
2648                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2649                               (ifr->ifr_flags & TUN_FEATURES);
2650
2651                 INIT_LIST_HEAD(&tun->disabled);
2652                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2653                 if (err < 0)
2654                         goto err_free_flow;
2655
2656                 err = register_netdevice(tun->dev);
2657                 if (err < 0)
2658                         goto err_detach;
2659         }
2660
2661         netif_carrier_on(tun->dev);
2662
2663         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2664
2665         /* Make sure persistent devices do not get stuck in
2666          * xoff state.
2667          */
2668         if (netif_running(tun->dev))
2669                 netif_tx_wake_all_queues(tun->dev);
2670
2671         strcpy(ifr->ifr_name, tun->dev->name);
2672         return 0;
2673
2674 err_detach:
2675         tun_detach_all(dev);
2676         /* register_netdevice() already called tun_free_netdev() */
2677         goto err_free_dev;
2678
2679 err_free_flow:
2680         tun_flow_uninit(tun);
2681         security_tun_dev_free_security(tun->security);
2682 err_free_stat:
2683         free_percpu(tun->pcpu_stats);
2684 err_free_dev:
2685         free_netdev(dev);
2686         return err;
2687 }
2688
2689 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2690                        struct ifreq *ifr)
2691 {
2692         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2693
2694         strcpy(ifr->ifr_name, tun->dev->name);
2695
2696         ifr->ifr_flags = tun_flags(tun);
2697
2698 }
2699
2700 /* This is like a cut-down ethtool ops, except done via tun fd so no
2701  * privs required. */
2702 static int set_offload(struct tun_struct *tun, unsigned long arg)
2703 {
2704         netdev_features_t features = 0;
2705
2706         if (arg & TUN_F_CSUM) {
2707                 features |= NETIF_F_HW_CSUM;
2708                 arg &= ~TUN_F_CSUM;
2709
2710                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2711                         if (arg & TUN_F_TSO_ECN) {
2712                                 features |= NETIF_F_TSO_ECN;
2713                                 arg &= ~TUN_F_TSO_ECN;
2714                         }
2715                         if (arg & TUN_F_TSO4)
2716                                 features |= NETIF_F_TSO;
2717                         if (arg & TUN_F_TSO6)
2718                                 features |= NETIF_F_TSO6;
2719                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2720                 }
2721
2722                 arg &= ~TUN_F_UFO;
2723         }
2724
2725         /* This gives the user a way to test for new features in future by
2726          * trying to set them. */
2727         if (arg)
2728                 return -EINVAL;
2729
2730         tun->set_features = features;
2731         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2732         tun->dev->wanted_features |= features;
2733         netdev_update_features(tun->dev);
2734
2735         return 0;
2736 }
2737
2738 static void tun_detach_filter(struct tun_struct *tun, int n)
2739 {
2740         int i;
2741         struct tun_file *tfile;
2742
2743         for (i = 0; i < n; i++) {
2744                 tfile = rtnl_dereference(tun->tfiles[i]);
2745                 lock_sock(tfile->socket.sk);
2746                 sk_detach_filter(tfile->socket.sk);
2747                 release_sock(tfile->socket.sk);
2748         }
2749
2750         tun->filter_attached = false;
2751 }
2752
2753 static int tun_attach_filter(struct tun_struct *tun)
2754 {
2755         int i, ret = 0;
2756         struct tun_file *tfile;
2757
2758         for (i = 0; i < tun->numqueues; i++) {
2759                 tfile = rtnl_dereference(tun->tfiles[i]);
2760                 lock_sock(tfile->socket.sk);
2761                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2762                 release_sock(tfile->socket.sk);
2763                 if (ret) {
2764                         tun_detach_filter(tun, i);
2765                         return ret;
2766                 }
2767         }
2768
2769         tun->filter_attached = true;
2770         return ret;
2771 }
2772
2773 static void tun_set_sndbuf(struct tun_struct *tun)
2774 {
2775         struct tun_file *tfile;
2776         int i;
2777
2778         for (i = 0; i < tun->numqueues; i++) {
2779                 tfile = rtnl_dereference(tun->tfiles[i]);
2780                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2781         }
2782 }
2783
2784 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2785 {
2786         struct tun_file *tfile = file->private_data;
2787         struct tun_struct *tun;
2788         int ret = 0;
2789
2790         rtnl_lock();
2791
2792         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2793                 tun = tfile->detached;
2794                 if (!tun) {
2795                         ret = -EINVAL;
2796                         goto unlock;
2797                 }
2798                 ret = security_tun_dev_attach_queue(tun->security);
2799                 if (ret < 0)
2800                         goto unlock;
2801                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2802         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2803                 tun = rtnl_dereference(tfile->tun);
2804                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2805                         ret = -EINVAL;
2806                 else
2807                         __tun_detach(tfile, false);
2808         } else
2809                 ret = -EINVAL;
2810
2811 unlock:
2812         rtnl_unlock();
2813         return ret;
2814 }
2815
2816 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2817                         void __user *data)
2818 {
2819         struct bpf_prog *prog;
2820         int fd;
2821
2822         if (copy_from_user(&fd, data, sizeof(fd)))
2823                 return -EFAULT;
2824
2825         if (fd == -1) {
2826                 prog = NULL;
2827         } else {
2828                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2829                 if (IS_ERR(prog))
2830                         return PTR_ERR(prog);
2831         }
2832
2833         return __tun_set_ebpf(tun, prog_p, prog);
2834 }
2835
2836 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2837                             unsigned long arg, int ifreq_len)
2838 {
2839         struct tun_file *tfile = file->private_data;
2840         struct tun_struct *tun;
2841         void __user* argp = (void __user*)arg;
2842         struct ifreq ifr;
2843         struct net *net;
2844         kuid_t owner;
2845         kgid_t group;
2846         int sndbuf;
2847         int vnet_hdr_sz;
2848         unsigned int ifindex;
2849         int le;
2850         int ret;
2851
2852         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2853             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2854                 if (copy_from_user(&ifr, argp, ifreq_len))
2855                         return -EFAULT;
2856         } else {
2857                 memset(&ifr, 0, sizeof(ifr));
2858         }
2859         if (cmd == TUNGETFEATURES) {
2860                 /* Currently this just means: "what IFF flags are valid?".
2861                  * This is needed because we never checked for invalid flags on
2862                  * TUNSETIFF.
2863                  */
2864                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2865                                 (unsigned int __user*)argp);
2866         } else if (cmd == TUNSETQUEUE)
2867                 return tun_set_queue(file, &ifr);
2868
2869         ret = 0;
2870         rtnl_lock();
2871
2872         tun = tun_get(tfile);
2873         net = sock_net(&tfile->sk);
2874         if (cmd == TUNSETIFF) {
2875                 ret = -EEXIST;
2876                 if (tun)
2877                         goto unlock;
2878
2879                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2880
2881                 ret = tun_set_iff(net, file, &ifr);
2882
2883                 if (ret)
2884                         goto unlock;
2885
2886                 if (copy_to_user(argp, &ifr, ifreq_len))
2887                         ret = -EFAULT;
2888                 goto unlock;
2889         }
2890         if (cmd == TUNSETIFINDEX) {
2891                 ret = -EPERM;
2892                 if (tun)
2893                         goto unlock;
2894
2895                 ret = -EFAULT;
2896                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2897                         goto unlock;
2898
2899                 ret = 0;
2900                 tfile->ifindex = ifindex;
2901                 goto unlock;
2902         }
2903         if (cmd == SIOCGSKNS) {
2904                 ret = -EPERM;
2905                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2906                         goto unlock;
2907
2908                 ret = open_related_ns(&net->ns, get_net_ns);
2909                 goto unlock;
2910         }
2911
2912         ret = -EBADFD;
2913         if (!tun)
2914                 goto unlock;
2915
2916         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2917
2918         ret = 0;
2919         switch (cmd) {
2920         case TUNGETIFF:
2921                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2922
2923                 if (tfile->detached)
2924                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2925                 if (!tfile->socket.sk->sk_filter)
2926                         ifr.ifr_flags |= IFF_NOFILTER;
2927
2928                 if (copy_to_user(argp, &ifr, ifreq_len))
2929                         ret = -EFAULT;
2930                 break;
2931
2932         case TUNSETNOCSUM:
2933                 /* Disable/Enable checksum */
2934
2935                 /* [unimplemented] */
2936                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2937                           arg ? "disabled" : "enabled");
2938                 break;
2939
2940         case TUNSETPERSIST:
2941                 /* Disable/Enable persist mode. Keep an extra reference to the
2942                  * module to prevent the module being unprobed.
2943                  */
2944                 if (arg && !(tun->flags & IFF_PERSIST)) {
2945                         tun->flags |= IFF_PERSIST;
2946                         __module_get(THIS_MODULE);
2947                 }
2948                 if (!arg && (tun->flags & IFF_PERSIST)) {
2949                         tun->flags &= ~IFF_PERSIST;
2950                         module_put(THIS_MODULE);
2951                 }
2952
2953                 tun_debug(KERN_INFO, tun, "persist %s\n",
2954                           arg ? "enabled" : "disabled");
2955                 break;
2956
2957         case TUNSETOWNER:
2958                 /* Set owner of the device */
2959                 owner = make_kuid(current_user_ns(), arg);
2960                 if (!uid_valid(owner)) {
2961                         ret = -EINVAL;
2962                         break;
2963                 }
2964                 tun->owner = owner;
2965                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2966                           from_kuid(&init_user_ns, tun->owner));
2967                 break;
2968
2969         case TUNSETGROUP:
2970                 /* Set group of the device */
2971                 group = make_kgid(current_user_ns(), arg);
2972                 if (!gid_valid(group)) {
2973                         ret = -EINVAL;
2974                         break;
2975                 }
2976                 tun->group = group;
2977                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2978                           from_kgid(&init_user_ns, tun->group));
2979                 break;
2980
2981         case TUNSETLINK:
2982                 /* Only allow setting the type when the interface is down */
2983                 if (tun->dev->flags & IFF_UP) {
2984                         tun_debug(KERN_INFO, tun,
2985                                   "Linktype set failed because interface is up\n");
2986                         ret = -EBUSY;
2987                 } else {
2988                         tun->dev->type = (int) arg;
2989                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2990                                   tun->dev->type);
2991                         ret = 0;
2992                 }
2993                 break;
2994
2995 #ifdef TUN_DEBUG
2996         case TUNSETDEBUG:
2997                 tun->debug = arg;
2998                 break;
2999 #endif
3000         case TUNSETOFFLOAD:
3001                 ret = set_offload(tun, arg);
3002                 break;
3003
3004         case TUNSETTXFILTER:
3005                 /* Can be set only for TAPs */
3006                 ret = -EINVAL;
3007                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3008                         break;
3009                 ret = update_filter(&tun->txflt, (void __user *)arg);
3010                 break;
3011
3012         case SIOCGIFHWADDR:
3013                 /* Get hw address */
3014                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3015                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3016                 if (copy_to_user(argp, &ifr, ifreq_len))
3017                         ret = -EFAULT;
3018                 break;
3019
3020         case SIOCSIFHWADDR:
3021                 /* Set hw address */
3022                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3023                           ifr.ifr_hwaddr.sa_data);
3024
3025                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3026                 break;
3027
3028         case TUNGETSNDBUF:
3029                 sndbuf = tfile->socket.sk->sk_sndbuf;
3030                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3031                         ret = -EFAULT;
3032                 break;
3033
3034         case TUNSETSNDBUF:
3035                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3036                         ret = -EFAULT;
3037                         break;
3038                 }
3039                 if (sndbuf <= 0) {
3040                         ret = -EINVAL;
3041                         break;
3042                 }
3043
3044                 tun->sndbuf = sndbuf;
3045                 tun_set_sndbuf(tun);
3046                 break;
3047
3048         case TUNGETVNETHDRSZ:
3049                 vnet_hdr_sz = tun->vnet_hdr_sz;
3050                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3051                         ret = -EFAULT;
3052                 break;
3053
3054         case TUNSETVNETHDRSZ:
3055                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3056                         ret = -EFAULT;
3057                         break;
3058                 }
3059                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3060                         ret = -EINVAL;
3061                         break;
3062                 }
3063
3064                 tun->vnet_hdr_sz = vnet_hdr_sz;
3065                 break;
3066
3067         case TUNGETVNETLE:
3068                 le = !!(tun->flags & TUN_VNET_LE);
3069                 if (put_user(le, (int __user *)argp))
3070                         ret = -EFAULT;
3071                 break;
3072
3073         case TUNSETVNETLE:
3074                 if (get_user(le, (int __user *)argp)) {
3075                         ret = -EFAULT;
3076                         break;
3077                 }
3078                 if (le)
3079                         tun->flags |= TUN_VNET_LE;
3080                 else
3081                         tun->flags &= ~TUN_VNET_LE;
3082                 break;
3083
3084         case TUNGETVNETBE:
3085                 ret = tun_get_vnet_be(tun, argp);
3086                 break;
3087
3088         case TUNSETVNETBE:
3089                 ret = tun_set_vnet_be(tun, argp);
3090                 break;
3091
3092         case TUNATTACHFILTER:
3093                 /* Can be set only for TAPs */
3094                 ret = -EINVAL;
3095                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3096                         break;
3097                 ret = -EFAULT;
3098                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3099                         break;
3100
3101                 ret = tun_attach_filter(tun);
3102                 break;
3103
3104         case TUNDETACHFILTER:
3105                 /* Can be set only for TAPs */
3106                 ret = -EINVAL;
3107                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3108                         break;
3109                 ret = 0;
3110                 tun_detach_filter(tun, tun->numqueues);
3111                 break;
3112
3113         case TUNGETFILTER:
3114                 ret = -EINVAL;
3115                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3116                         break;
3117                 ret = -EFAULT;
3118                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3119                         break;
3120                 ret = 0;
3121                 break;
3122
3123         case TUNSETSTEERINGEBPF:
3124                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3125                 break;
3126
3127         case TUNSETFILTEREBPF:
3128                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3129                 break;
3130
3131         default:
3132                 ret = -EINVAL;
3133                 break;
3134         }
3135
3136 unlock:
3137         rtnl_unlock();
3138         if (tun)
3139                 tun_put(tun);
3140         return ret;
3141 }
3142
3143 static long tun_chr_ioctl(struct file *file,
3144                           unsigned int cmd, unsigned long arg)
3145 {
3146         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3147 }
3148
3149 #ifdef CONFIG_COMPAT
3150 static long tun_chr_compat_ioctl(struct file *file,
3151                          unsigned int cmd, unsigned long arg)
3152 {
3153         switch (cmd) {
3154         case TUNSETIFF:
3155         case TUNGETIFF:
3156         case TUNSETTXFILTER:
3157         case TUNGETSNDBUF:
3158         case TUNSETSNDBUF:
3159         case SIOCGIFHWADDR:
3160         case SIOCSIFHWADDR:
3161                 arg = (unsigned long)compat_ptr(arg);
3162                 break;
3163         default:
3164                 arg = (compat_ulong_t)arg;
3165                 break;
3166         }
3167
3168         /*
3169          * compat_ifreq is shorter than ifreq, so we must not access beyond
3170          * the end of that structure. All fields that are used in this
3171          * driver are compatible though, we don't need to convert the
3172          * contents.
3173          */
3174         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3175 }
3176 #endif /* CONFIG_COMPAT */
3177
3178 static int tun_chr_fasync(int fd, struct file *file, int on)
3179 {
3180         struct tun_file *tfile = file->private_data;
3181         int ret;
3182
3183         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3184                 goto out;
3185
3186         if (on) {
3187                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3188                 tfile->flags |= TUN_FASYNC;
3189         } else
3190                 tfile->flags &= ~TUN_FASYNC;
3191         ret = 0;
3192 out:
3193         return ret;
3194 }
3195
3196 static int tun_chr_open(struct inode *inode, struct file * file)
3197 {
3198         struct net *net = current->nsproxy->net_ns;
3199         struct tun_file *tfile;
3200
3201         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3202
3203         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3204                                             &tun_proto, 0);
3205         if (!tfile)
3206                 return -ENOMEM;
3207         RCU_INIT_POINTER(tfile->tun, NULL);
3208         tfile->flags = 0;
3209         tfile->ifindex = 0;
3210
3211         init_waitqueue_head(&tfile->wq.wait);
3212         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3213
3214         tfile->socket.file = file;
3215         tfile->socket.ops = &tun_socket_ops;
3216
3217         sock_init_data(&tfile->socket, &tfile->sk);
3218
3219         tfile->sk.sk_write_space = tun_sock_write_space;
3220         tfile->sk.sk_sndbuf = INT_MAX;
3221
3222         file->private_data = tfile;
3223         INIT_LIST_HEAD(&tfile->next);
3224
3225         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3226
3227         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3228
3229         return 0;
3230 }
3231
3232 static int tun_chr_close(struct inode *inode, struct file *file)
3233 {
3234         struct tun_file *tfile = file->private_data;
3235
3236         tun_detach(tfile, true);
3237
3238         return 0;
3239 }
3240
3241 #ifdef CONFIG_PROC_FS
3242 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3243 {
3244         struct tun_file *tfile = file->private_data;
3245         struct tun_struct *tun;
3246         struct ifreq ifr;
3247
3248         memset(&ifr, 0, sizeof(ifr));
3249
3250         rtnl_lock();
3251         tun = tun_get(tfile);
3252         if (tun)
3253                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3254         rtnl_unlock();
3255
3256         if (tun)
3257                 tun_put(tun);
3258
3259         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3260 }
3261 #endif
3262
3263 static const struct file_operations tun_fops = {
3264         .owner  = THIS_MODULE,
3265         .llseek = no_llseek,
3266         .read_iter  = tun_chr_read_iter,
3267         .write_iter = tun_chr_write_iter,
3268         .poll   = tun_chr_poll,
3269         .unlocked_ioctl = tun_chr_ioctl,
3270 #ifdef CONFIG_COMPAT
3271         .compat_ioctl = tun_chr_compat_ioctl,
3272 #endif
3273         .open   = tun_chr_open,
3274         .release = tun_chr_close,
3275         .fasync = tun_chr_fasync,
3276 #ifdef CONFIG_PROC_FS
3277         .show_fdinfo = tun_chr_show_fdinfo,
3278 #endif
3279 };
3280
3281 static struct miscdevice tun_miscdev = {
3282         .minor = TUN_MINOR,
3283         .name = "tun",
3284         .nodename = "net/tun",
3285         .fops = &tun_fops,
3286 };
3287
3288 /* ethtool interface */
3289
3290 static int tun_get_link_ksettings(struct net_device *dev,
3291                                   struct ethtool_link_ksettings *cmd)
3292 {
3293         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3294         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3295         cmd->base.speed         = SPEED_10;
3296         cmd->base.duplex        = DUPLEX_FULL;
3297         cmd->base.port          = PORT_TP;
3298         cmd->base.phy_address   = 0;
3299         cmd->base.autoneg       = AUTONEG_DISABLE;
3300         return 0;
3301 }
3302
3303 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3304 {
3305         struct tun_struct *tun = netdev_priv(dev);
3306
3307         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3308         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3309
3310         switch (tun->flags & TUN_TYPE_MASK) {
3311         case IFF_TUN:
3312                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3313                 break;
3314         case IFF_TAP:
3315                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3316                 break;
3317         }
3318 }
3319
3320 static u32 tun_get_msglevel(struct net_device *dev)
3321 {
3322 #ifdef TUN_DEBUG
3323         struct tun_struct *tun = netdev_priv(dev);
3324         return tun->debug;
3325 #else
3326         return -EOPNOTSUPP;
3327 #endif
3328 }
3329
3330 static void tun_set_msglevel(struct net_device *dev, u32 value)
3331 {
3332 #ifdef TUN_DEBUG
3333         struct tun_struct *tun = netdev_priv(dev);
3334         tun->debug = value;
3335 #endif
3336 }
3337
3338 static int tun_get_coalesce(struct net_device *dev,
3339                             struct ethtool_coalesce *ec)
3340 {
3341         struct tun_struct *tun = netdev_priv(dev);
3342
3343         ec->rx_max_coalesced_frames = tun->rx_batched;
3344
3345         return 0;
3346 }
3347
3348 static int tun_set_coalesce(struct net_device *dev,
3349                             struct ethtool_coalesce *ec)
3350 {
3351         struct tun_struct *tun = netdev_priv(dev);
3352
3353         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3354                 tun->rx_batched = NAPI_POLL_WEIGHT;
3355         else
3356                 tun->rx_batched = ec->rx_max_coalesced_frames;
3357
3358         return 0;
3359 }
3360
3361 static const struct ethtool_ops tun_ethtool_ops = {
3362         .get_drvinfo    = tun_get_drvinfo,
3363         .get_msglevel   = tun_get_msglevel,
3364         .set_msglevel   = tun_set_msglevel,
3365         .get_link       = ethtool_op_get_link,
3366         .get_ts_info    = ethtool_op_get_ts_info,
3367         .get_coalesce   = tun_get_coalesce,
3368         .set_coalesce   = tun_set_coalesce,
3369         .get_link_ksettings = tun_get_link_ksettings,
3370 };
3371
3372 static int tun_queue_resize(struct tun_struct *tun)
3373 {
3374         struct net_device *dev = tun->dev;
3375         struct tun_file *tfile;
3376         struct ptr_ring **rings;
3377         int n = tun->numqueues + tun->numdisabled;
3378         int ret, i;
3379
3380         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3381         if (!rings)
3382                 return -ENOMEM;
3383
3384         for (i = 0; i < tun->numqueues; i++) {
3385                 tfile = rtnl_dereference(tun->tfiles[i]);
3386                 rings[i] = &tfile->tx_ring;
3387         }
3388         list_for_each_entry(tfile, &tun->disabled, next)
3389                 rings[i++] = &tfile->tx_ring;
3390
3391         ret = ptr_ring_resize_multiple(rings, n,
3392                                        dev->tx_queue_len, GFP_KERNEL,
3393                                        tun_ptr_free);
3394
3395         kfree(rings);
3396         return ret;
3397 }
3398
3399 static int tun_device_event(struct notifier_block *unused,
3400                             unsigned long event, void *ptr)
3401 {
3402         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3403         struct tun_struct *tun = netdev_priv(dev);
3404
3405         if (dev->rtnl_link_ops != &tun_link_ops)
3406                 return NOTIFY_DONE;
3407
3408         switch (event) {
3409         case NETDEV_CHANGE_TX_QUEUE_LEN:
3410                 if (tun_queue_resize(tun))
3411                         return NOTIFY_BAD;
3412                 break;
3413         default:
3414                 break;
3415         }
3416
3417         return NOTIFY_DONE;
3418 }
3419
3420 static struct notifier_block tun_notifier_block __read_mostly = {
3421         .notifier_call  = tun_device_event,
3422 };
3423
3424 static int __init tun_init(void)
3425 {
3426         int ret = 0;
3427
3428         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3429
3430         ret = rtnl_link_register(&tun_link_ops);
3431         if (ret) {
3432                 pr_err("Can't register link_ops\n");
3433                 goto err_linkops;
3434         }
3435
3436         ret = misc_register(&tun_miscdev);
3437         if (ret) {
3438                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3439                 goto err_misc;
3440         }
3441
3442         ret = register_netdevice_notifier(&tun_notifier_block);
3443         if (ret) {
3444                 pr_err("Can't register netdevice notifier\n");
3445                 goto err_notifier;
3446         }
3447
3448         return  0;
3449
3450 err_notifier:
3451         misc_deregister(&tun_miscdev);
3452 err_misc:
3453         rtnl_link_unregister(&tun_link_ops);
3454 err_linkops:
3455         return ret;
3456 }
3457
3458 static void tun_cleanup(void)
3459 {
3460         misc_deregister(&tun_miscdev);
3461         rtnl_link_unregister(&tun_link_ops);
3462         unregister_netdevice_notifier(&tun_notifier_block);
3463 }
3464
3465 /* Get an underlying socket object from tun file.  Returns error unless file is
3466  * attached to a device.  The returned object works like a packet socket, it
3467  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3468  * holding a reference to the file for as long as the socket is in use. */
3469 struct socket *tun_get_socket(struct file *file)
3470 {
3471         struct tun_file *tfile;
3472         if (file->f_op != &tun_fops)
3473                 return ERR_PTR(-EINVAL);
3474         tfile = file->private_data;
3475         if (!tfile)
3476                 return ERR_PTR(-EBADFD);
3477         return &tfile->socket;
3478 }
3479 EXPORT_SYMBOL_GPL(tun_get_socket);
3480
3481 struct ptr_ring *tun_get_tx_ring(struct file *file)
3482 {
3483         struct tun_file *tfile;
3484
3485         if (file->f_op != &tun_fops)
3486                 return ERR_PTR(-EINVAL);
3487         tfile = file->private_data;
3488         if (!tfile)
3489                 return ERR_PTR(-EBADFD);
3490         return &tfile->tx_ring;
3491 }
3492 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3493
3494 module_init(tun_init);
3495 module_exit(tun_cleanup);
3496 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3497 MODULE_AUTHOR(DRV_COPYRIGHT);
3498 MODULE_LICENSE("GPL");
3499 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3500 MODULE_ALIAS("devname:net/tun");