drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <[email protected]>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <[email protected]>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81 #include <linux/proc_fs.h>
  82
  83 /* Uncomment to enable debugging */
  84 /* #define TUN_DEBUG 1 */
  85
  86 #ifdef TUN_DEBUG
  87 static int debug;
  88
  89 #define tun_debug(level, tun, fmt, args...)                     \
  90 do {                                                            \
  91         if (tun->debug)                                         \
  92                 netdev_printk(level, tun->dev, fmt, ##args);    \
  93 } while (0)
  94 #define DBG1(level, fmt, args...)                               \
  95 do {                                                            \
  96         if (debug == 2)                                         \
  97                 printk(level fmt, ##args);                      \
  98 } while (0)
  99 #else
 100 #define tun_debug(level, tun, fmt, args...)                     \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 netdev_printk(level, tun->dev, fmt, ##args);    \
 104 } while (0)
 105 #define DBG1(level, fmt, args...)                               \
 106 do {                                                            \
 107         if (0)                                                  \
 108                 printk(level fmt, ##args);                      \
 109 } while (0)
 110 #endif
 111
 112 #define TUN_HEADROOM 256
 113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 127
 128 #define GOODCOPY_LEN 128
 129
 130 #define FLT_EXACT_COUNT 8
 131 struct tap_filter {
 132         unsigned int    count;    /* Number of addrs. Zero means disabled */
 133         u32             mask[2];  /* Mask of the hashed addrs */
 134         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 135 };
 136
 137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 138  * to max number of VCPUs in guest. */
 139 #define MAX_TAP_QUEUES 256
 140 #define MAX_TAP_FLOWS  4096
 141
 142 #define TUN_FLOW_EXPIRE (3 * HZ)
 143
 144 struct tun_pcpu_stats {
 145         u64 rx_packets;
 146         u64 rx_bytes;
 147         u64 tx_packets;
 148         u64 tx_bytes;
 149         struct u64_stats_sync syncp;
 150         u32 rx_dropped;
 151         u32 tx_dropped;
 152         u32 rx_frame_errors;
 153 };
 154
 155 /* A tun_file connects an open character device to a tuntap netdevice. It
 156  * also contains all socket related structures (except sock_fprog and tap_filter)
 157  * to serve as one transmit queue for tuntap device. The sock_fprog and
 158  * tap_filter were kept in tun_struct since they were used for filtering for the
 159  * netdevice not for a specific queue (at least I didn't see the requirement for
 160  * this).
 161  *
 162  * RCU usage:
 163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 164  * other can only be read while rcu_read_lock or rtnl_lock is held.
 165  */
 166 struct tun_file {
 167         struct sock sk;
 168         struct socket socket;
 169         struct socket_wq wq;
 170         struct tun_struct __rcu *tun;
 171         struct fasync_struct *fasync;
 172         /* only used for fasnyc */
 173         unsigned int flags;
 174         union {
 175                 u16 queue_index;
 176                 unsigned int ifindex;
 177         };
 178         struct napi_struct napi;
 179         bool napi_enabled;
 180         struct mutex napi_mutex;        /* Protects access to the above napi */
 181         struct list_head next;
 182         struct tun_struct *detached;
 183         struct ptr_ring tx_ring;
 184         struct xdp_rxq_info xdp_rxq;
 185 };
 186
 187 struct tun_flow_entry {
 188         struct hlist_node hash_link;
 189         struct rcu_head rcu;
 190         struct tun_struct *tun;
 191
 192         u32 rxhash;
 193         u32 rps_rxhash;
 194         int queue_index;
 195         unsigned long updated;
 196 };
 197
 198 #define TUN_NUM_FLOW_ENTRIES 1024
 199
 200 struct tun_prog {
 201         struct rcu_head rcu;
 202         struct bpf_prog *prog;
 203 };
 204
 205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 206  * device, socket filter, sndbuf and vnet header size were restore when the
 207  * file were attached to a persist device.
 208  */
 209 struct tun_struct {
 210         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 211         unsigned int            numqueues;
 212         unsigned int            flags;
 213         kuid_t                  owner;
 214         kgid_t                  group;
 215
 216         struct net_device       *dev;
 217         netdev_features_t       set_features;
 218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 219                           NETIF_F_TSO6)
 220
 221         int                     align;
 222         int                     vnet_hdr_sz;
 223         int                     sndbuf;
 224         struct tap_filter       txflt;
 225         struct sock_fprog       fprog;
 226         /* protected by rtnl lock */
 227         bool                    filter_attached;
 228 #ifdef TUN_DEBUG
 229         int debug;
 230 #endif
 231         spinlock_t lock;
 232         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 233         struct timer_list flow_gc_timer;
 234         unsigned long ageing_time;
 235         unsigned int numdisabled;
 236         struct list_head disabled;
 237         void *security;
 238         u32 flow_count;
 239         u32 rx_batched;
 240         struct tun_pcpu_stats __percpu *pcpu_stats;
 241         struct bpf_prog __rcu *xdp_prog;
 242         struct tun_prog __rcu *steering_prog;
 243         struct tun_prog __rcu *filter_prog;
 244 };
 245
 246 struct veth {
 247         __be16 h_vlan_proto;
 248         __be16 h_vlan_TCI;
 249 };
 250
 251 bool tun_is_xdp_buff(void *ptr)
 252 {
 253         return (unsigned long)ptr & TUN_XDP_FLAG;
 254 }
 255 EXPORT_SYMBOL(tun_is_xdp_buff);
 256
 257 void *tun_xdp_to_ptr(void *ptr)
 258 {
 259         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 260 }
 261 EXPORT_SYMBOL(tun_xdp_to_ptr);
 262
 263 void *tun_ptr_to_xdp(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_ptr_to_xdp);
 268
 269 static int tun_napi_receive(struct napi_struct *napi, int budget)
 270 {
 271         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 272         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 273         struct sk_buff_head process_queue;
 274         struct sk_buff *skb;
 275         int received = 0;
 276
 277         __skb_queue_head_init(&process_queue);
 278
 279         spin_lock(&queue->lock);
 280         skb_queue_splice_tail_init(queue, &process_queue);
 281         spin_unlock(&queue->lock);
 282
 283         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 284                 napi_gro_receive(napi, skb);
 285                 ++received;
 286         }
 287
 288         if (!skb_queue_empty(&process_queue)) {
 289                 spin_lock(&queue->lock);
 290                 skb_queue_splice(&process_queue, queue);
 291                 spin_unlock(&queue->lock);
 292         }
 293
 294         return received;
 295 }
 296
 297 static int tun_napi_poll(struct napi_struct *napi, int budget)
 298 {
 299         unsigned int received;
 300
 301         received = tun_napi_receive(napi, budget);
 302
 303         if (received < budget)
 304                 napi_complete_done(napi, received);
 305
 306         return received;
 307 }
 308
 309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 310                           bool napi_en)
 311 {
 312         tfile->napi_enabled = napi_en;
 313         if (napi_en) {
 314                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 315                                NAPI_POLL_WEIGHT);
 316                 napi_enable(&tfile->napi);
 317                 mutex_init(&tfile->napi_mutex);
 318         }
 319 }
 320
 321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 322 {
 323         if (tfile->napi_enabled)
 324                 napi_disable(&tfile->napi);
 325 }
 326
 327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 netif_napi_del(&tfile->napi);
 331 }
 332
 333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 334 {
 335         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 336 }
 337
 338 #ifdef CONFIG_TUN_VNET_CROSS_LE
 339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 340 {
 341         return tun->flags & TUN_VNET_BE ? false :
 342                 virtio_legacy_is_little_endian();
 343 }
 344
 345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 346 {
 347         int be = !!(tun->flags & TUN_VNET_BE);
 348
 349         if (put_user(be, argp))
 350                 return -EFAULT;
 351
 352         return 0;
 353 }
 354
 355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 356 {
 357         int be;
 358
 359         if (get_user(be, argp))
 360                 return -EFAULT;
 361
 362         if (be)
 363                 tun->flags |= TUN_VNET_BE;
 364         else
 365                 tun->flags &= ~TUN_VNET_BE;
 366
 367         return 0;
 368 }
 369 #else
 370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 371 {
 372         return virtio_legacy_is_little_endian();
 373 }
 374
 375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 376 {
 377         return -EINVAL;
 378 }
 379
 380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 385
 386 static inline bool tun_is_little_endian(struct tun_struct *tun)
 387 {
 388         return tun->flags & TUN_VNET_LE ||
 389                 tun_legacy_is_little_endian(tun);
 390 }
 391
 392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 393 {
 394         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 395 }
 396
 397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 398 {
 399         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline u32 tun_hashfn(u32 rxhash)
 403 {
 404         return rxhash & 0x3ff;
 405 }
 406
 407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 408 {
 409         struct tun_flow_entry *e;
 410
 411         hlist_for_each_entry_rcu(e, head, hash_link) {
 412                 if (e->rxhash == rxhash)
 413                         return e;
 414         }
 415         return NULL;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 419                                               struct hlist_head *head,
 420                                               u32 rxhash, u16 queue_index)
 421 {
 422         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 423
 424         if (e) {
 425                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 426                           rxhash, queue_index);
 427                 e->updated = jiffies;
 428                 e->rxhash = rxhash;
 429                 e->rps_rxhash = 0;
 430                 e->queue_index = queue_index;
 431                 e->tun = tun;
 432                 hlist_add_head_rcu(&e->hash_link, head);
 433                 ++tun->flow_count;
 434         }
 435         return e;
 436 }
 437
 438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 439 {
 440         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 441                   e->rxhash, e->queue_index);
 442         hlist_del_rcu(&e->hash_link);
 443         kfree_rcu(e, rcu);
 444         --tun->flow_count;
 445 }
 446
 447 static void tun_flow_flush(struct tun_struct *tun)
 448 {
 449         int i;
 450
 451         spin_lock_bh(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 457                         tun_flow_delete(tun, e);
 458         }
 459         spin_unlock_bh(&tun->lock);
 460 }
 461
 462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 463 {
 464         int i;
 465
 466         spin_lock_bh(&tun->lock);
 467         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 468                 struct tun_flow_entry *e;
 469                 struct hlist_node *n;
 470
 471                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 472                         if (e->queue_index == queue_index)
 473                                 tun_flow_delete(tun, e);
 474                 }
 475         }
 476         spin_unlock_bh(&tun->lock);
 477 }
 478
 479 static void tun_flow_cleanup(struct timer_list *t)
 480 {
 481         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 482         unsigned long delay = tun->ageing_time;
 483         unsigned long next_timer = jiffies + delay;
 484         unsigned long count = 0;
 485         int i;
 486
 487         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 488
 489         spin_lock(&tun->lock);
 490         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 491                 struct tun_flow_entry *e;
 492                 struct hlist_node *n;
 493
 494                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 495                         unsigned long this_timer;
 496
 497                         this_timer = e->updated + delay;
 498                         if (time_before_eq(this_timer, jiffies)) {
 499                                 tun_flow_delete(tun, e);
 500                                 continue;
 501                         }
 502                         count++;
 503                         if (time_before(this_timer, next_timer))
 504                                 next_timer = this_timer;
 505                 }
 506         }
 507
 508         if (count)
 509                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 510         spin_unlock(&tun->lock);
 511 }
 512
 513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 514                             struct tun_file *tfile)
 515 {
 516         struct hlist_head *head;
 517         struct tun_flow_entry *e;
 518         unsigned long delay = tun->ageing_time;
 519         u16 queue_index = tfile->queue_index;
 520
 521         if (!rxhash)
 522                 return;
 523         else
 524                 head = &tun->flows[tun_hashfn(rxhash)];
 525
 526         rcu_read_lock();
 527
 528         /* We may get a very small possibility of OOO during switching, not
 529          * worth to optimize.*/
 530         if (tun->numqueues == 1 || tfile->detached)
 531                 goto unlock;
 532
 533         e = tun_flow_find(head, rxhash);
 534         if (likely(e)) {
 535                 /* TODO: keep queueing to old queue until it's empty? */
 536                 e->queue_index = queue_index;
 537                 e->updated = jiffies;
 538                 sock_rps_record_flow_hash(e->rps_rxhash);
 539         } else {
 540                 spin_lock_bh(&tun->lock);
 541                 if (!tun_flow_find(head, rxhash) &&
 542                     tun->flow_count < MAX_TAP_FLOWS)
 543                         tun_flow_create(tun, head, rxhash, queue_index);
 544
 545                 if (!timer_pending(&tun->flow_gc_timer))
 546                         mod_timer(&tun->flow_gc_timer,
 547                                   round_jiffies_up(jiffies + delay));
 548                 spin_unlock_bh(&tun->lock);
 549         }
 550
 551 unlock:
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash first. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here. If we could not get rxhash, then we would
 570  * hope the rxq no. may help here.
 571  */
 572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 573 {
 574         struct tun_flow_entry *e;
 575         u32 txq = 0;
 576         u32 numqueues = 0;
 577
 578         numqueues = READ_ONCE(tun->numqueues);
 579
 580         txq = __skb_get_hash_symmetric(skb);
 581         if (txq) {
 582                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 583                 if (e) {
 584                         tun_flow_save_rps_rxhash(e, txq);
 585                         txq = e->queue_index;
 586                 } else
 587                         /* use multiply and shift instead of expensive divide */
 588                         txq = ((u64)txq * numqueues) >> 32;
 589         } else if (likely(skb_rx_queue_recorded(skb))) {
 590                 txq = skb_get_rx_queue(skb);
 591                 while (unlikely(txq >= numqueues))
 592                         txq -= numqueues;
 593         }
 594
 595         return txq;
 596 }
 597
 598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 599 {
 600         struct tun_prog *prog;
 601         u16 ret = 0;
 602
 603         prog = rcu_dereference(tun->steering_prog);
 604         if (prog)
 605                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 606
 607         return ret % tun->numqueues;
 608 }
 609
 610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 611                             void *accel_priv, select_queue_fallback_t fallback)
 612 {
 613         struct tun_struct *tun = netdev_priv(dev);
 614         u16 ret;
 615
 616         rcu_read_lock();
 617         if (rcu_dereference(tun->steering_prog))
 618                 ret = tun_ebpf_select_queue(tun, skb);
 619         else
 620                 ret = tun_automq_select_queue(tun, skb);
 621         rcu_read_unlock();
 622
 623         return ret;
 624 }
 625
 626 static inline bool tun_not_capable(struct tun_struct *tun)
 627 {
 628         const struct cred *cred = current_cred();
 629         struct net *net = dev_net(tun->dev);
 630
 631         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 632                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 633                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 634 }
 635
 636 static void tun_set_real_num_queues(struct tun_struct *tun)
 637 {
 638         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 639         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 640 }
 641
 642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 643 {
 644         tfile->detached = tun;
 645         list_add_tail(&tfile->next, &tun->disabled);
 646         ++tun->numdisabled;
 647 }
 648
 649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 650 {
 651         struct tun_struct *tun = tfile->detached;
 652
 653         tfile->detached = NULL;
 654         list_del_init(&tfile->next);
 655         --tun->numdisabled;
 656         return tun;
 657 }
 658
 659 void tun_ptr_free(void *ptr)
 660 {
 661         if (!ptr)
 662                 return;
 663         if (tun_is_xdp_buff(ptr)) {
 664                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
 665
 666                 put_page(virt_to_head_page(xdp->data));
 667         } else {
 668                 __skb_array_destroy_skb(ptr);
 669         }
 670 }
 671 EXPORT_SYMBOL_GPL(tun_ptr_free);
 672
 673 static void tun_queue_purge(struct tun_file *tfile)
 674 {
 675         void *ptr;
 676
 677         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 678                 tun_ptr_free(ptr);
 679
 680         skb_queue_purge(&tfile->sk.sk_write_queue);
 681         skb_queue_purge(&tfile->sk.sk_error_queue);
 682 }
 683
 684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 685 {
 686         if (tfile->tx_ring.queue) {
 687                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 688                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 690         }
 691 }
 692
 693 static void __tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_file *ntfile;
 696         struct tun_struct *tun;
 697
 698         tun = rtnl_dereference(tfile->tun);
 699
 700         if (tun && clean) {
 701                 tun_napi_disable(tun, tfile);
 702                 tun_napi_del(tun, tfile);
 703         }
 704
 705         if (tun && !tfile->detached) {
 706                 u16 index = tfile->queue_index;
 707                 BUG_ON(index >= tun->numqueues);
 708
 709                 rcu_assign_pointer(tun->tfiles[index],
 710                                    tun->tfiles[tun->numqueues - 1]);
 711                 ntfile = rtnl_dereference(tun->tfiles[index]);
 712                 ntfile->queue_index = index;
 713
 714                 --tun->numqueues;
 715                 if (clean) {
 716                         RCU_INIT_POINTER(tfile->tun, NULL);
 717                         sock_put(&tfile->sk);
 718                 } else
 719                         tun_disable_queue(tun, tfile);
 720
 721                 synchronize_net();
 722                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 723                 /* Drop read queue */
 724                 tun_queue_purge(tfile);
 725                 tun_set_real_num_queues(tun);
 726         } else if (tfile->detached && clean) {
 727                 tun = tun_enable_queue(tfile);
 728                 sock_put(&tfile->sk);
 729         }
 730
 731         if (clean) {
 732                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 733                         netif_carrier_off(tun->dev);
 734
 735                         if (!(tun->flags & IFF_PERSIST) &&
 736                             tun->dev->reg_state == NETREG_REGISTERED)
 737                                 unregister_netdevice(tun->dev);
 738                 }
 739                 tun_cleanup_tx_ring(tfile);
 740                 sock_put(&tfile->sk);
 741         }
 742 }
 743
 744 static void tun_detach(struct tun_file *tfile, bool clean)
 745 {
 746         struct tun_struct *tun;
 747         struct net_device *dev;
 748
 749         rtnl_lock();
 750         tun = rtnl_dereference(tfile->tun);
 751         dev = tun ? tun->dev : NULL;
 752         __tun_detach(tfile, clean);
 753         if (dev)
 754                 netdev_state_change(dev);
 755         rtnl_unlock();
 756 }
 757
 758 static void tun_detach_all(struct net_device *dev)
 759 {
 760         struct tun_struct *tun = netdev_priv(dev);
 761         struct tun_file *tfile, *tmp;
 762         int i, n = tun->numqueues;
 763
 764         for (i = 0; i < n; i++) {
 765                 tfile = rtnl_dereference(tun->tfiles[i]);
 766                 BUG_ON(!tfile);
 767                 tun_napi_disable(tun, tfile);
 768                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 769                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 770                 RCU_INIT_POINTER(tfile->tun, NULL);
 771                 --tun->numqueues;
 772         }
 773         list_for_each_entry(tfile, &tun->disabled, next) {
 774                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 775                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 776                 RCU_INIT_POINTER(tfile->tun, NULL);
 777         }
 778         BUG_ON(tun->numqueues != 0);
 779
 780         synchronize_net();
 781         for (i = 0; i < n; i++) {
 782                 tfile = rtnl_dereference(tun->tfiles[i]);
 783                 tun_napi_del(tun, tfile);
 784                 /* Drop read queue */
 785                 tun_queue_purge(tfile);
 786                 sock_put(&tfile->sk);
 787                 tun_cleanup_tx_ring(tfile);
 788         }
 789         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 790                 tun_enable_queue(tfile);
 791                 tun_queue_purge(tfile);
 792                 sock_put(&tfile->sk);
 793                 tun_cleanup_tx_ring(tfile);
 794         }
 795         BUG_ON(tun->numdisabled != 0);
 796
 797         if (tun->flags & IFF_PERSIST)
 798                 module_put(THIS_MODULE);
 799 }
 800
 801 static int tun_attach(struct tun_struct *tun, struct file *file,
 802                       bool skip_filter, bool napi)
 803 {
 804         struct tun_file *tfile = file->private_data;
 805         struct net_device *dev = tun->dev;
 806         int err;
 807
 808         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 809         if (err < 0)
 810                 goto out;
 811
 812         err = -EINVAL;
 813         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 814                 goto out;
 815
 816         err = -EBUSY;
 817         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 818                 goto out;
 819
 820         err = -E2BIG;
 821         if (!tfile->detached &&
 822             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 823                 goto out;
 824
 825         err = 0;
 826
 827         /* Re-attach the filter to persist device */
 828         if (!skip_filter && (tun->filter_attached == true)) {
 829                 lock_sock(tfile->socket.sk);
 830                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 831                 release_sock(tfile->socket.sk);
 832                 if (!err)
 833                         goto out;
 834         }
 835
 836         if (!tfile->detached &&
 837             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 838                 err = -ENOMEM;
 839                 goto out;
 840         }
 841
 842         tfile->queue_index = tun->numqueues;
 843         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 844
 845         if (tfile->detached) {
 846                 /* Re-attach detached tfile, updating XDP queue_index */
 847                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 848
 849                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 850                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 851         } else {
 852                 /* Setup XDP RX-queue info, for new tfile getting attached */
 853                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 854                                        tun->dev, tfile->queue_index);
 855                 if (err < 0)
 856                         goto out;
 857                 err = 0;
 858         }
 859
 860         rcu_assign_pointer(tfile->tun, tun);
 861         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 862         tun->numqueues++;
 863
 864         if (tfile->detached) {
 865                 tun_enable_queue(tfile);
 866         } else {
 867                 sock_hold(&tfile->sk);
 868                 tun_napi_init(tun, tfile, napi);
 869         }
 870
 871         tun_set_real_num_queues(tun);
 872
 873         /* device is allowed to go away first, so no need to hold extra
 874          * refcnt.
 875          */
 876
 877 out:
 878         return err;
 879 }
 880
 881 static struct tun_struct *tun_get(struct tun_file *tfile)
 882 {
 883         struct tun_struct *tun;
 884
 885         rcu_read_lock();
 886         tun = rcu_dereference(tfile->tun);
 887         if (tun)
 888                 dev_hold(tun->dev);
 889         rcu_read_unlock();
 890
 891         return tun;
 892 }
 893
 894 static void tun_put(struct tun_struct *tun)
 895 {
 896         dev_put(tun->dev);
 897 }
 898
 899 /* TAP filtering */
 900 static void addr_hash_set(u32 *mask, const u8 *addr)
 901 {
 902         int n = ether_crc(ETH_ALEN, addr) >> 26;
 903         mask[n >> 5] |= (1 << (n & 31));
 904 }
 905
 906 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 907 {
 908         int n = ether_crc(ETH_ALEN, addr) >> 26;
 909         return mask[n >> 5] & (1 << (n & 31));
 910 }
 911
 912 static int update_filter(struct tap_filter *filter, void __user *arg)
 913 {
 914         struct { u8 u[ETH_ALEN]; } *addr;
 915         struct tun_filter uf;
 916         int err, alen, n, nexact;
 917
 918         if (copy_from_user(&uf, arg, sizeof(uf)))
 919                 return -EFAULT;
 920
 921         if (!uf.count) {
 922                 /* Disabled */
 923                 filter->count = 0;
 924                 return 0;
 925         }
 926
 927         alen = ETH_ALEN * uf.count;
 928         addr = memdup_user(arg + sizeof(uf), alen);
 929         if (IS_ERR(addr))
 930                 return PTR_ERR(addr);
 931
 932         /* The filter is updated without holding any locks. Which is
 933          * perfectly safe. We disable it first and in the worst
 934          * case we'll accept a few undesired packets. */
 935         filter->count = 0;
 936         wmb();
 937
 938         /* Use first set of addresses as an exact filter */
 939         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 940                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 941
 942         nexact = n;
 943
 944         /* Remaining multicast addresses are hashed,
 945          * unicast will leave the filter disabled. */
 946         memset(filter->mask, 0, sizeof(filter->mask));
 947         for (; n < uf.count; n++) {
 948                 if (!is_multicast_ether_addr(addr[n].u)) {
 949                         err = 0; /* no filter */
 950                         goto free_addr;
 951                 }
 952                 addr_hash_set(filter->mask, addr[n].u);
 953         }
 954
 955         /* For ALLMULTI just set the mask to all ones.
 956          * This overrides the mask populated above. */
 957         if ((uf.flags & TUN_FLT_ALLMULTI))
 958                 memset(filter->mask, ~0, sizeof(filter->mask));
 959
 960         /* Now enable the filter */
 961         wmb();
 962         filter->count = nexact;
 963
 964         /* Return the number of exact filters */
 965         err = nexact;
 966 free_addr:
 967         kfree(addr);
 968         return err;
 969 }
 970
 971 /* Returns: 0 - drop, !=0 - accept */
 972 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 973 {
 974         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 975          * at this point. */
 976         struct ethhdr *eh = (struct ethhdr *) skb->data;
 977         int i;
 978
 979         /* Exact match */
 980         for (i = 0; i < filter->count; i++)
 981                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 982                         return 1;
 983
 984         /* Inexact match (multicast only) */
 985         if (is_multicast_ether_addr(eh->h_dest))
 986                 return addr_hash_test(filter->mask, eh->h_dest);
 987
 988         return 0;
 989 }
 990
 991 /*
 992  * Checks whether the packet is accepted or not.
 993  * Returns: 0 - drop, !=0 - accept
 994  */
 995 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 996 {
 997         if (!filter->count)
 998                 return 1;
 999
1000         return run_filter(filter, skb);
1001 }
1002
1003 /* Network device part of the driver */
1004
1005 static const struct ethtool_ops tun_ethtool_ops;
1006
1007 /* Net device detach from fd. */
1008 static void tun_net_uninit(struct net_device *dev)
1009 {
1010         tun_detach_all(dev);
1011 }
1012
1013 /* Net device open. */
1014 static int tun_net_open(struct net_device *dev)
1015 {
1016         struct tun_struct *tun = netdev_priv(dev);
1017         int i;
1018
1019         netif_tx_start_all_queues(dev);
1020
1021         for (i = 0; i < tun->numqueues; i++) {
1022                 struct tun_file *tfile;
1023
1024                 tfile = rtnl_dereference(tun->tfiles[i]);
1025                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1026         }
1027
1028         return 0;
1029 }
1030
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1033 {
1034         netif_tx_stop_all_queues(dev);
1035         return 0;
1036 }
1037
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 {
1041 #ifdef CONFIG_RPS
1042         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1043                 /* Select queue was not called for the skbuff, so we extract the
1044                  * RPS hash and save it into the flow_table here.
1045                  */
1046                 __u32 rxhash;
1047
1048                 rxhash = __skb_get_hash_symmetric(skb);
1049                 if (rxhash) {
1050                         struct tun_flow_entry *e;
1051                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1052                                         rxhash);
1053                         if (e)
1054                                 tun_flow_save_rps_rxhash(e, rxhash);
1055                 }
1056         }
1057 #endif
1058 }
1059
1060 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1061                                     struct sk_buff *skb,
1062                                     int len)
1063 {
1064         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1065
1066         if (prog)
1067                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1068
1069         return len;
1070 }
1071
1072 /* Net device start xmit */
1073 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1074 {
1075         struct tun_struct *tun = netdev_priv(dev);
1076         int txq = skb->queue_mapping;
1077         struct tun_file *tfile;
1078         int len = skb->len;
1079
1080         rcu_read_lock();
1081         tfile = rcu_dereference(tun->tfiles[txq]);
1082
1083         /* Drop packet if interface is not attached */
1084         if (txq >= tun->numqueues)
1085                 goto drop;
1086
1087         if (!rcu_dereference(tun->steering_prog))
1088                 tun_automq_xmit(tun, skb);
1089
1090         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1091
1092         BUG_ON(!tfile);
1093
1094         /* Drop if the filter does not like it.
1095          * This is a noop if the filter is disabled.
1096          * Filter can be enabled only for the TAP devices. */
1097         if (!check_filter(&tun->txflt, skb))
1098                 goto drop;
1099
1100         if (tfile->socket.sk->sk_filter &&
1101             sk_filter(tfile->socket.sk, skb))
1102                 goto drop;
1103
1104         len = run_ebpf_filter(tun, skb, len);
1105
1106         /* Trim extra bytes since we may insert vlan proto & TCI
1107          * in tun_put_user().
1108          */
1109         len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1110         if (len <= 0 || pskb_trim(skb, len))
1111                 goto drop;
1112
1113         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1114                 goto drop;
1115
1116         skb_tx_timestamp(skb);
1117
1118         /* Orphan the skb - required as we might hang on to it
1119          * for indefinite time.
1120          */
1121         skb_orphan(skb);
1122
1123         nf_reset(skb);
1124
1125         if (ptr_ring_produce(&tfile->tx_ring, skb))
1126                 goto drop;
1127
1128         /* Notify and wake up reader process */
1129         if (tfile->flags & TUN_FASYNC)
1130                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1131         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1132
1133         rcu_read_unlock();
1134         return NETDEV_TX_OK;
1135
1136 drop:
1137         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1138         skb_tx_error(skb);
1139         kfree_skb(skb);
1140         rcu_read_unlock();
1141         return NET_XMIT_DROP;
1142 }
1143
1144 static void tun_net_mclist(struct net_device *dev)
1145 {
1146         /*
1147          * This callback is supposed to deal with mc filter in
1148          * _rx_ path and has nothing to do with the _tx_ path.
1149          * In rx path we always accept everything userspace gives us.
1150          */
1151 }
1152
1153 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1154         netdev_features_t features)
1155 {
1156         struct tun_struct *tun = netdev_priv(dev);
1157
1158         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1159 }
1160 #ifdef CONFIG_NET_POLL_CONTROLLER
1161 static void tun_poll_controller(struct net_device *dev)
1162 {
1163         /*
1164          * Tun only receives frames when:
1165          * 1) the char device endpoint gets data from user space
1166          * 2) the tun socket gets a sendmsg call from user space
1167          * If NAPI is not enabled, since both of those are synchronous
1168          * operations, we are guaranteed never to have pending data when we poll
1169          * for it so there is nothing to do here but return.
1170          * We need this though so netpoll recognizes us as an interface that
1171          * supports polling, which enables bridge devices in virt setups to
1172          * still use netconsole
1173          * If NAPI is enabled, however, we need to schedule polling for all
1174          * queues unless we are using napi_gro_frags(), which we call in
1175          * process context and not in NAPI context.
1176          */
1177         struct tun_struct *tun = netdev_priv(dev);
1178
1179         if (tun->flags & IFF_NAPI) {
1180                 struct tun_file *tfile;
1181                 int i;
1182
1183                 if (tun_napi_frags_enabled(tun))
1184                         return;
1185
1186                 rcu_read_lock();
1187                 for (i = 0; i < tun->numqueues; i++) {
1188                         tfile = rcu_dereference(tun->tfiles[i]);
1189                         if (tfile->napi_enabled)
1190                                 napi_schedule(&tfile->napi);
1191                 }
1192                 rcu_read_unlock();
1193         }
1194         return;
1195 }
1196 #endif
1197
1198 static void tun_set_headroom(struct net_device *dev, int new_hr)
1199 {
1200         struct tun_struct *tun = netdev_priv(dev);
1201
1202         if (new_hr < NET_SKB_PAD)
1203                 new_hr = NET_SKB_PAD;
1204
1205         tun->align = new_hr;
1206 }
1207
1208 static void
1209 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1210 {
1211         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1212         struct tun_struct *tun = netdev_priv(dev);
1213         struct tun_pcpu_stats *p;
1214         int i;
1215
1216         for_each_possible_cpu(i) {
1217                 u64 rxpackets, rxbytes, txpackets, txbytes;
1218                 unsigned int start;
1219
1220                 p = per_cpu_ptr(tun->pcpu_stats, i);
1221                 do {
1222                         start = u64_stats_fetch_begin(&p->syncp);
1223                         rxpackets       = p->rx_packets;
1224                         rxbytes         = p->rx_bytes;
1225                         txpackets       = p->tx_packets;
1226                         txbytes         = p->tx_bytes;
1227                 } while (u64_stats_fetch_retry(&p->syncp, start));
1228
1229                 stats->rx_packets       += rxpackets;
1230                 stats->rx_bytes         += rxbytes;
1231                 stats->tx_packets       += txpackets;
1232                 stats->tx_bytes         += txbytes;
1233
1234                 /* u32 counters */
1235                 rx_dropped      += p->rx_dropped;
1236                 rx_frame_errors += p->rx_frame_errors;
1237                 tx_dropped      += p->tx_dropped;
1238         }
1239         stats->rx_dropped  = rx_dropped;
1240         stats->rx_frame_errors = rx_frame_errors;
1241         stats->tx_dropped = tx_dropped;
1242 }
1243
1244 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1245                        struct netlink_ext_ack *extack)
1246 {
1247         struct tun_struct *tun = netdev_priv(dev);
1248         struct bpf_prog *old_prog;
1249
1250         old_prog = rtnl_dereference(tun->xdp_prog);
1251         rcu_assign_pointer(tun->xdp_prog, prog);
1252         if (old_prog)
1253                 bpf_prog_put(old_prog);
1254
1255         return 0;
1256 }
1257
1258 static u32 tun_xdp_query(struct net_device *dev)
1259 {
1260         struct tun_struct *tun = netdev_priv(dev);
1261         const struct bpf_prog *xdp_prog;
1262
1263         xdp_prog = rtnl_dereference(tun->xdp_prog);
1264         if (xdp_prog)
1265                 return xdp_prog->aux->id;
1266
1267         return 0;
1268 }
1269
1270 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1271 {
1272         switch (xdp->command) {
1273         case XDP_SETUP_PROG:
1274                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1275         case XDP_QUERY_PROG:
1276                 xdp->prog_id = tun_xdp_query(dev);
1277                 xdp->prog_attached = !!xdp->prog_id;
1278                 return 0;
1279         default:
1280                 return -EINVAL;
1281         }
1282 }
1283
1284 static const struct net_device_ops tun_netdev_ops = {
1285         .ndo_uninit             = tun_net_uninit,
1286         .ndo_open               = tun_net_open,
1287         .ndo_stop               = tun_net_close,
1288         .ndo_start_xmit         = tun_net_xmit,
1289         .ndo_fix_features       = tun_net_fix_features,
1290         .ndo_select_queue       = tun_select_queue,
1291 #ifdef CONFIG_NET_POLL_CONTROLLER
1292         .ndo_poll_controller    = tun_poll_controller,
1293 #endif
1294         .ndo_set_rx_headroom    = tun_set_headroom,
1295         .ndo_get_stats64        = tun_net_get_stats64,
1296 };
1297
1298 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1299 {
1300         struct tun_struct *tun = netdev_priv(dev);
1301         struct xdp_buff *buff = xdp->data_hard_start;
1302         int headroom = xdp->data - xdp->data_hard_start;
1303         struct tun_file *tfile;
1304         u32 numqueues;
1305         int ret = 0;
1306
1307         /* Assure headroom is available and buff is properly aligned */
1308         if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1309                 return -ENOSPC;
1310
1311         *buff = *xdp;
1312
1313         rcu_read_lock();
1314
1315         numqueues = READ_ONCE(tun->numqueues);
1316         if (!numqueues) {
1317                 ret = -ENOSPC;
1318                 goto out;
1319         }
1320
1321         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1322                                             numqueues]);
1323         /* Encode the XDP flag into lowest bit for consumer to differ
1324          * XDP buffer from sk_buff.
1325          */
1326         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1327                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1328                 ret = -ENOSPC;
1329         }
1330
1331 out:
1332         rcu_read_unlock();
1333         return ret;
1334 }
1335
1336 static void tun_xdp_flush(struct net_device *dev)
1337 {
1338         struct tun_struct *tun = netdev_priv(dev);
1339         struct tun_file *tfile;
1340         u32 numqueues;
1341
1342         rcu_read_lock();
1343
1344         numqueues = READ_ONCE(tun->numqueues);
1345         if (!numqueues)
1346                 goto out;
1347
1348         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1349                                             numqueues]);
1350         /* Notify and wake up reader process */
1351         if (tfile->flags & TUN_FASYNC)
1352                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1353         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1354
1355 out:
1356         rcu_read_unlock();
1357 }
1358
1359 static const struct net_device_ops tap_netdev_ops = {
1360         .ndo_uninit             = tun_net_uninit,
1361         .ndo_open               = tun_net_open,
1362         .ndo_stop               = tun_net_close,
1363         .ndo_start_xmit         = tun_net_xmit,
1364         .ndo_fix_features       = tun_net_fix_features,
1365         .ndo_set_rx_mode        = tun_net_mclist,
1366         .ndo_set_mac_address    = eth_mac_addr,
1367         .ndo_validate_addr      = eth_validate_addr,
1368         .ndo_select_queue       = tun_select_queue,
1369 #ifdef CONFIG_NET_POLL_CONTROLLER
1370         .ndo_poll_controller    = tun_poll_controller,
1371 #endif
1372         .ndo_features_check     = passthru_features_check,
1373         .ndo_set_rx_headroom    = tun_set_headroom,
1374         .ndo_get_stats64        = tun_net_get_stats64,
1375         .ndo_bpf                = tun_xdp,
1376         .ndo_xdp_xmit           = tun_xdp_xmit,
1377         .ndo_xdp_flush          = tun_xdp_flush,
1378 };
1379
1380 static void tun_flow_init(struct tun_struct *tun)
1381 {
1382         int i;
1383
1384         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1385                 INIT_HLIST_HEAD(&tun->flows[i]);
1386
1387         tun->ageing_time = TUN_FLOW_EXPIRE;
1388         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1389         mod_timer(&tun->flow_gc_timer,
1390                   round_jiffies_up(jiffies + tun->ageing_time));
1391 }
1392
1393 static void tun_flow_uninit(struct tun_struct *tun)
1394 {
1395         del_timer_sync(&tun->flow_gc_timer);
1396         tun_flow_flush(tun);
1397 }
1398
1399 #define MIN_MTU 68
1400 #define MAX_MTU 65535
1401
1402 /* Initialize net device. */
1403 static void tun_net_init(struct net_device *dev)
1404 {
1405         struct tun_struct *tun = netdev_priv(dev);
1406
1407         switch (tun->flags & TUN_TYPE_MASK) {
1408         case IFF_TUN:
1409                 dev->netdev_ops = &tun_netdev_ops;
1410
1411                 /* Point-to-Point TUN Device */
1412                 dev->hard_header_len = 0;
1413                 dev->addr_len = 0;
1414                 dev->mtu = 1500;
1415
1416                 /* Zero header length */
1417                 dev->type = ARPHRD_NONE;
1418                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1419                 break;
1420
1421         case IFF_TAP:
1422                 dev->netdev_ops = &tap_netdev_ops;
1423                 /* Ethernet TAP Device */
1424                 ether_setup(dev);
1425                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1426                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1427
1428                 eth_hw_addr_random(dev);
1429
1430                 break;
1431         }
1432
1433         dev->min_mtu = MIN_MTU;
1434         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1435 }
1436
1437 /* Character device part */
1438
1439 /* Poll */
1440 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1441 {
1442         struct tun_file *tfile = file->private_data;
1443         struct tun_struct *tun = tun_get(tfile);
1444         struct sock *sk;
1445         __poll_t mask = 0;
1446
1447         if (!tun)
1448                 return EPOLLERR;
1449
1450         sk = tfile->socket.sk;
1451
1452         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1453
1454         poll_wait(file, sk_sleep(sk), wait);
1455
1456         if (!ptr_ring_empty(&tfile->tx_ring))
1457                 mask |= EPOLLIN | EPOLLRDNORM;
1458
1459         if (tun->dev->flags & IFF_UP &&
1460             (sock_writeable(sk) ||
1461              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1462               sock_writeable(sk))))
1463                 mask |= EPOLLOUT | EPOLLWRNORM;
1464
1465         if (tun->dev->reg_state != NETREG_REGISTERED)
1466                 mask = EPOLLERR;
1467
1468         tun_put(tun);
1469         return mask;
1470 }
1471
1472 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1473                                             size_t len,
1474                                             const struct iov_iter *it)
1475 {
1476         struct sk_buff *skb;
1477         size_t linear;
1478         int err;
1479         int i;
1480
1481         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1482                 return ERR_PTR(-ENOMEM);
1483
1484         local_bh_disable();
1485         skb = napi_get_frags(&tfile->napi);
1486         local_bh_enable();
1487         if (!skb)
1488                 return ERR_PTR(-ENOMEM);
1489
1490         linear = iov_iter_single_seg_count(it);
1491         err = __skb_grow(skb, linear);
1492         if (err)
1493                 goto free;
1494
1495         skb->len = len;
1496         skb->data_len = len - linear;
1497         skb->truesize += skb->data_len;
1498
1499         for (i = 1; i < it->nr_segs; i++) {
1500                 struct page_frag *pfrag = &current->task_frag;
1501                 size_t fragsz = it->iov[i].iov_len;
1502
1503                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1504                         err = -EINVAL;
1505                         goto free;
1506                 }
1507
1508                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1509                         err = -ENOMEM;
1510                         goto free;
1511                 }
1512
1513                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1514                                    pfrag->offset, fragsz);
1515                 page_ref_inc(pfrag->page);
1516                 pfrag->offset += fragsz;
1517         }
1518
1519         return skb;
1520 free:
1521         /* frees skb and all frags allocated with napi_alloc_frag() */
1522         napi_free_frags(&tfile->napi);
1523         return ERR_PTR(err);
1524 }
1525
1526 /* prepad is the amount to reserve at front.  len is length after that.
1527  * linear is a hint as to how much to copy (usually headers). */
1528 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1529                                      size_t prepad, size_t len,
1530                                      size_t linear, int noblock)
1531 {
1532         struct sock *sk = tfile->socket.sk;
1533         struct sk_buff *skb;
1534         int err;
1535
1536         /* Under a page?  Don't bother with paged skb. */
1537         if (prepad + len < PAGE_SIZE || !linear)
1538                 linear = len;
1539
1540         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1541                                    &err, 0);
1542         if (!skb)
1543                 return ERR_PTR(err);
1544
1545         skb_reserve(skb, prepad);
1546         skb_put(skb, linear);
1547         skb->data_len = len - linear;
1548         skb->len += len - linear;
1549
1550         return skb;
1551 }
1552
1553 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1554                            struct sk_buff *skb, int more)
1555 {
1556         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1557         struct sk_buff_head process_queue;
1558         u32 rx_batched = tun->rx_batched;
1559         bool rcv = false;
1560
1561         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1562                 local_bh_disable();
1563                 netif_receive_skb(skb);
1564                 local_bh_enable();
1565                 return;
1566         }
1567
1568         spin_lock(&queue->lock);
1569         if (!more || skb_queue_len(queue) == rx_batched) {
1570                 __skb_queue_head_init(&process_queue);
1571                 skb_queue_splice_tail_init(queue, &process_queue);
1572                 rcv = true;
1573         } else {
1574                 __skb_queue_tail(queue, skb);
1575         }
1576         spin_unlock(&queue->lock);
1577
1578         if (rcv) {
1579                 struct sk_buff *nskb;
1580
1581                 local_bh_disable();
1582                 while ((nskb = __skb_dequeue(&process_queue)))
1583                         netif_receive_skb(nskb);
1584                 netif_receive_skb(skb);
1585                 local_bh_enable();
1586         }
1587 }
1588
1589 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1590                               int len, int noblock, bool zerocopy)
1591 {
1592         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1593                 return false;
1594
1595         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1596                 return false;
1597
1598         if (!noblock)
1599                 return false;
1600
1601         if (zerocopy)
1602                 return false;
1603
1604         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1605             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1606                 return false;
1607
1608         return true;
1609 }
1610
1611 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1612                                      struct tun_file *tfile,
1613                                      struct iov_iter *from,
1614                                      struct virtio_net_hdr *hdr,
1615                                      int len, int *skb_xdp)
1616 {
1617         struct page_frag *alloc_frag = &current->task_frag;
1618         struct sk_buff *skb;
1619         struct bpf_prog *xdp_prog;
1620         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1621         unsigned int delta = 0;
1622         char *buf;
1623         size_t copied;
1624         int err, pad = TUN_RX_PAD;
1625
1626         rcu_read_lock();
1627         xdp_prog = rcu_dereference(tun->xdp_prog);
1628         if (xdp_prog)
1629                 pad += TUN_HEADROOM;
1630         buflen += SKB_DATA_ALIGN(len + pad);
1631         rcu_read_unlock();
1632
1633         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1634         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1635                 return ERR_PTR(-ENOMEM);
1636
1637         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1638         copied = copy_page_from_iter(alloc_frag->page,
1639                                      alloc_frag->offset + pad,
1640                                      len, from);
1641         if (copied != len)
1642                 return ERR_PTR(-EFAULT);
1643
1644         /* There's a small window that XDP may be set after the check
1645          * of xdp_prog above, this should be rare and for simplicity
1646          * we do XDP on skb in case the headroom is not enough.
1647          */
1648         if (hdr->gso_type || !xdp_prog)
1649                 *skb_xdp = 1;
1650         else
1651                 *skb_xdp = 0;
1652
1653         preempt_disable();
1654         rcu_read_lock();
1655         xdp_prog = rcu_dereference(tun->xdp_prog);
1656         if (xdp_prog && !*skb_xdp) {
1657                 struct xdp_buff xdp;
1658                 void *orig_data;
1659                 u32 act;
1660
1661                 xdp.data_hard_start = buf;
1662                 xdp.data = buf + pad;
1663                 xdp_set_data_meta_invalid(&xdp);
1664                 xdp.data_end = xdp.data + len;
1665                 xdp.rxq = &tfile->xdp_rxq;
1666                 orig_data = xdp.data;
1667                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1668
1669                 switch (act) {
1670                 case XDP_REDIRECT:
1671                         get_page(alloc_frag->page);
1672                         alloc_frag->offset += buflen;
1673                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1674                         xdp_do_flush_map();
1675                         if (err)
1676                                 goto err_redirect;
1677                         rcu_read_unlock();
1678                         preempt_enable();
1679                         return NULL;
1680                 case XDP_TX:
1681                         get_page(alloc_frag->page);
1682                         alloc_frag->offset += buflen;
1683                         if (tun_xdp_xmit(tun->dev, &xdp))
1684                                 goto err_redirect;
1685                         tun_xdp_flush(tun->dev);
1686                         rcu_read_unlock();
1687                         preempt_enable();
1688                         return NULL;
1689                 case XDP_PASS:
1690                         delta = orig_data - xdp.data;
1691                         break;
1692                 default:
1693                         bpf_warn_invalid_xdp_action(act);
1694                         /* fall through */
1695                 case XDP_ABORTED:
1696                         trace_xdp_exception(tun->dev, xdp_prog, act);
1697                         /* fall through */
1698                 case XDP_DROP:
1699                         goto err_xdp;
1700                 }
1701         }
1702
1703         skb = build_skb(buf, buflen);
1704         if (!skb) {
1705                 rcu_read_unlock();
1706                 preempt_enable();
1707                 return ERR_PTR(-ENOMEM);
1708         }
1709
1710         skb_reserve(skb, pad - delta);
1711         skb_put(skb, len + delta);
1712         get_page(alloc_frag->page);
1713         alloc_frag->offset += buflen;
1714
1715         rcu_read_unlock();
1716         preempt_enable();
1717
1718         return skb;
1719
1720 err_redirect:
1721         put_page(alloc_frag->page);
1722 err_xdp:
1723         rcu_read_unlock();
1724         preempt_enable();
1725         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1726         return NULL;
1727 }
1728
1729 /* Get packet from user space buffer */
1730 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1731                             void *msg_control, struct iov_iter *from,
1732                             int noblock, bool more)
1733 {
1734         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1735         struct sk_buff *skb;
1736         size_t total_len = iov_iter_count(from);
1737         size_t len = total_len, align = tun->align, linear;
1738         struct virtio_net_hdr gso = { 0 };
1739         struct tun_pcpu_stats *stats;
1740         int good_linear;
1741         int copylen;
1742         bool zerocopy = false;
1743         int err;
1744         u32 rxhash = 0;
1745         int skb_xdp = 1;
1746         bool frags = tun_napi_frags_enabled(tun);
1747
1748         if (!(tun->dev->flags & IFF_UP))
1749                 return -EIO;
1750
1751         if (!(tun->flags & IFF_NO_PI)) {
1752                 if (len < sizeof(pi))
1753                         return -EINVAL;
1754                 len -= sizeof(pi);
1755
1756                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1757                         return -EFAULT;
1758         }
1759
1760         if (tun->flags & IFF_VNET_HDR) {
1761                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1762
1763                 if (len < vnet_hdr_sz)
1764                         return -EINVAL;
1765                 len -= vnet_hdr_sz;
1766
1767                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1768                         return -EFAULT;
1769
1770                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1771                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1772                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1773
1774                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1775                         return -EINVAL;
1776                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1777         }
1778
1779         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1780                 align += NET_IP_ALIGN;
1781                 if (unlikely(len < ETH_HLEN ||
1782                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1783                         return -EINVAL;
1784         }
1785
1786         good_linear = SKB_MAX_HEAD(align);
1787
1788         if (msg_control) {
1789                 struct iov_iter i = *from;
1790
1791                 /* There are 256 bytes to be copied in skb, so there is
1792                  * enough room for skb expand head in case it is used.
1793                  * The rest of the buffer is mapped from userspace.
1794                  */
1795                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1796                 if (copylen > good_linear)
1797                         copylen = good_linear;
1798                 linear = copylen;
1799                 iov_iter_advance(&i, copylen);
1800                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1801                         zerocopy = true;
1802         }
1803
1804         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1805                 /* For the packet that is not easy to be processed
1806                  * (e.g gso or jumbo packet), we will do it at after
1807                  * skb was created with generic XDP routine.
1808                  */
1809                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1810                 if (IS_ERR(skb)) {
1811                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1812                         return PTR_ERR(skb);
1813                 }
1814                 if (!skb)
1815                         return total_len;
1816         } else {
1817                 if (!zerocopy) {
1818                         copylen = len;
1819                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1820                                 linear = good_linear;
1821                         else
1822                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1823                 }
1824
1825                 if (frags) {
1826                         mutex_lock(&tfile->napi_mutex);
1827                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1828                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1829                          * If zerocopy is enabled, then this layout will be
1830                          * overwritten by zerocopy_sg_from_iter().
1831                          */
1832                         zerocopy = false;
1833                 } else {
1834                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1835                                             noblock);
1836                 }
1837
1838                 if (IS_ERR(skb)) {
1839                         if (PTR_ERR(skb) != -EAGAIN)
1840                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1841                         if (frags)
1842                                 mutex_unlock(&tfile->napi_mutex);
1843                         return PTR_ERR(skb);
1844                 }
1845
1846                 if (zerocopy)
1847                         err = zerocopy_sg_from_iter(skb, from);
1848                 else
1849                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1850
1851                 if (err) {
1852                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1853                         kfree_skb(skb);
1854                         if (frags) {
1855                                 tfile->napi.skb = NULL;
1856                                 mutex_unlock(&tfile->napi_mutex);
1857                         }
1858
1859                         return -EFAULT;
1860                 }
1861         }
1862
1863         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1864                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1865                 kfree_skb(skb);
1866                 if (frags) {
1867                         tfile->napi.skb = NULL;
1868                         mutex_unlock(&tfile->napi_mutex);
1869                 }
1870
1871                 return -EINVAL;
1872         }
1873
1874         switch (tun->flags & TUN_TYPE_MASK) {
1875         case IFF_TUN:
1876                 if (tun->flags & IFF_NO_PI) {
1877                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1878
1879                         switch (ip_version) {
1880                         case 4:
1881                                 pi.proto = htons(ETH_P_IP);
1882                                 break;
1883                         case 6:
1884                                 pi.proto = htons(ETH_P_IPV6);
1885                                 break;
1886                         default:
1887                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1888                                 kfree_skb(skb);
1889                                 return -EINVAL;
1890                         }
1891                 }
1892
1893                 skb_reset_mac_header(skb);
1894                 skb->protocol = pi.proto;
1895                 skb->dev = tun->dev;
1896                 break;
1897         case IFF_TAP:
1898                 if (!frags)
1899                         skb->protocol = eth_type_trans(skb, tun->dev);
1900                 break;
1901         }
1902
1903         /* copy skb_ubuf_info for callback when skb has no error */
1904         if (zerocopy) {
1905                 skb_shinfo(skb)->destructor_arg = msg_control;
1906                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1907                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1908         } else if (msg_control) {
1909                 struct ubuf_info *uarg = msg_control;
1910                 uarg->callback(uarg, false);
1911         }
1912
1913         skb_reset_network_header(skb);
1914         skb_probe_transport_header(skb, 0);
1915
1916         if (skb_xdp) {
1917                 struct bpf_prog *xdp_prog;
1918                 int ret;
1919
1920                 rcu_read_lock();
1921                 xdp_prog = rcu_dereference(tun->xdp_prog);
1922                 if (xdp_prog) {
1923                         ret = do_xdp_generic(xdp_prog, skb);
1924                         if (ret != XDP_PASS) {
1925                                 rcu_read_unlock();
1926                                 return total_len;
1927                         }
1928                 }
1929                 rcu_read_unlock();
1930         }
1931
1932         rcu_read_lock();
1933         if (!rcu_dereference(tun->steering_prog))
1934                 rxhash = __skb_get_hash_symmetric(skb);
1935         rcu_read_unlock();
1936
1937         if (frags) {
1938                 /* Exercise flow dissector code path. */
1939                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1940
1941                 if (unlikely(headlen > skb_headlen(skb))) {
1942                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1943                         napi_free_frags(&tfile->napi);
1944                         mutex_unlock(&tfile->napi_mutex);
1945                         WARN_ON(1);
1946                         return -ENOMEM;
1947                 }
1948
1949                 local_bh_disable();
1950                 napi_gro_frags(&tfile->napi);
1951                 local_bh_enable();
1952                 mutex_unlock(&tfile->napi_mutex);
1953         } else if (tfile->napi_enabled) {
1954                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1955                 int queue_len;
1956
1957                 spin_lock_bh(&queue->lock);
1958                 __skb_queue_tail(queue, skb);
1959                 queue_len = skb_queue_len(queue);
1960                 spin_unlock(&queue->lock);
1961
1962                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1963                         napi_schedule(&tfile->napi);
1964
1965                 local_bh_enable();
1966         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1967                 tun_rx_batched(tun, tfile, skb, more);
1968         } else {
1969                 netif_rx_ni(skb);
1970         }
1971
1972         stats = get_cpu_ptr(tun->pcpu_stats);
1973         u64_stats_update_begin(&stats->syncp);
1974         stats->rx_packets++;
1975         stats->rx_bytes += len;
1976         u64_stats_update_end(&stats->syncp);
1977         put_cpu_ptr(stats);
1978
1979         if (rxhash)
1980                 tun_flow_update(tun, rxhash, tfile);
1981
1982         return total_len;
1983 }
1984
1985 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1986 {
1987         struct file *file = iocb->ki_filp;
1988         struct tun_file *tfile = file->private_data;
1989         struct tun_struct *tun = tun_get(tfile);
1990         ssize_t result;
1991
1992         if (!tun)
1993                 return -EBADFD;
1994
1995         result = tun_get_user(tun, tfile, NULL, from,
1996                               file->f_flags & O_NONBLOCK, false);
1997
1998         tun_put(tun);
1999         return result;
2000 }
2001
2002 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2003                                 struct tun_file *tfile,
2004                                 struct xdp_buff *xdp,
2005                                 struct iov_iter *iter)
2006 {
2007         int vnet_hdr_sz = 0;
2008         size_t size = xdp->data_end - xdp->data;
2009         struct tun_pcpu_stats *stats;
2010         size_t ret;
2011
2012         if (tun->flags & IFF_VNET_HDR) {
2013                 struct virtio_net_hdr gso = { 0 };
2014
2015                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2016                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2017                         return -EINVAL;
2018                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2019                              sizeof(gso)))
2020                         return -EFAULT;
2021                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2022         }
2023
2024         ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2025
2026         stats = get_cpu_ptr(tun->pcpu_stats);
2027         u64_stats_update_begin(&stats->syncp);
2028         stats->tx_packets++;
2029         stats->tx_bytes += ret;
2030         u64_stats_update_end(&stats->syncp);
2031         put_cpu_ptr(tun->pcpu_stats);
2032
2033         return ret;
2034 }
2035
2036 /* Put packet to the user space buffer */
2037 static ssize_t tun_put_user(struct tun_struct *tun,
2038                             struct tun_file *tfile,
2039                             struct sk_buff *skb,
2040                             struct iov_iter *iter)
2041 {
2042         struct tun_pi pi = { 0, skb->protocol };
2043         struct tun_pcpu_stats *stats;
2044         ssize_t total;
2045         int vlan_offset = 0;
2046         int vlan_hlen = 0;
2047         int vnet_hdr_sz = 0;
2048
2049         if (skb_vlan_tag_present(skb))
2050                 vlan_hlen = VLAN_HLEN;
2051
2052         if (tun->flags & IFF_VNET_HDR)
2053                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2054
2055         total = skb->len + vlan_hlen + vnet_hdr_sz;
2056
2057         if (!(tun->flags & IFF_NO_PI)) {
2058                 if (iov_iter_count(iter) < sizeof(pi))
2059                         return -EINVAL;
2060
2061                 total += sizeof(pi);
2062                 if (iov_iter_count(iter) < total) {
2063                         /* Packet will be striped */
2064                         pi.flags |= TUN_PKT_STRIP;
2065                 }
2066
2067                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2068                         return -EFAULT;
2069         }
2070
2071         if (vnet_hdr_sz) {
2072                 struct virtio_net_hdr gso;
2073
2074                 if (iov_iter_count(iter) < vnet_hdr_sz)
2075                         return -EINVAL;
2076
2077                 if (virtio_net_hdr_from_skb(skb, &gso,
2078                                             tun_is_little_endian(tun), true)) {
2079                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2080                         pr_err("unexpected GSO type: "
2081                                "0x%x, gso_size %d, hdr_len %d\n",
2082                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2083                                tun16_to_cpu(tun, gso.hdr_len));
2084                         print_hex_dump(KERN_ERR, "tun: ",
2085                                        DUMP_PREFIX_NONE,
2086                                        16, 1, skb->head,
2087                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2088                         WARN_ON_ONCE(1);
2089                         return -EINVAL;
2090                 }
2091
2092                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2093                         return -EFAULT;
2094
2095                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2096         }
2097
2098         if (vlan_hlen) {
2099                 int ret;
2100                 struct veth veth;
2101
2102                 veth.h_vlan_proto = skb->vlan_proto;
2103                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2104
2105                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2106
2107                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2108                 if (ret || !iov_iter_count(iter))
2109                         goto done;
2110
2111                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2112                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2113                         goto done;
2114         }
2115
2116         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2117
2118 done:
2119         /* caller is in process context, */
2120         stats = get_cpu_ptr(tun->pcpu_stats);
2121         u64_stats_update_begin(&stats->syncp);
2122         stats->tx_packets++;
2123         stats->tx_bytes += skb->len + vlan_hlen;
2124         u64_stats_update_end(&stats->syncp);
2125         put_cpu_ptr(tun->pcpu_stats);
2126
2127         return total;
2128 }
2129
2130 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2131 {
2132         DECLARE_WAITQUEUE(wait, current);
2133         void *ptr = NULL;
2134         int error = 0;
2135
2136         ptr = ptr_ring_consume(&tfile->tx_ring);
2137         if (ptr)
2138                 goto out;
2139         if (noblock) {
2140                 error = -EAGAIN;
2141                 goto out;
2142         }
2143
2144         add_wait_queue(&tfile->wq.wait, &wait);
2145         current->state = TASK_INTERRUPTIBLE;
2146
2147         while (1) {
2148                 ptr = ptr_ring_consume(&tfile->tx_ring);
2149                 if (ptr)
2150                         break;
2151                 if (signal_pending(current)) {
2152                         error = -ERESTARTSYS;
2153                         break;
2154                 }
2155                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2156                         error = -EFAULT;
2157                         break;
2158                 }
2159
2160                 schedule();
2161         }
2162
2163         current->state = TASK_RUNNING;
2164         remove_wait_queue(&tfile->wq.wait, &wait);
2165
2166 out:
2167         *err = error;
2168         return ptr;
2169 }
2170
2171 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2172                            struct iov_iter *to,
2173                            int noblock, void *ptr)
2174 {
2175         ssize_t ret;
2176         int err;
2177
2178         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2179
2180         if (!iov_iter_count(to)) {
2181                 tun_ptr_free(ptr);
2182                 return 0;
2183         }
2184
2185         if (!ptr) {
2186                 /* Read frames from ring */
2187                 ptr = tun_ring_recv(tfile, noblock, &err);
2188                 if (!ptr)
2189                         return err;
2190         }
2191
2192         if (tun_is_xdp_buff(ptr)) {
2193                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2194
2195                 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2196                 put_page(virt_to_head_page(xdp->data));
2197         } else {
2198                 struct sk_buff *skb = ptr;
2199
2200                 ret = tun_put_user(tun, tfile, skb, to);
2201                 if (unlikely(ret < 0))
2202                         kfree_skb(skb);
2203                 else
2204                         consume_skb(skb);
2205         }
2206
2207         return ret;
2208 }
2209
2210 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2211 {
2212         struct file *file = iocb->ki_filp;
2213         struct tun_file *tfile = file->private_data;
2214         struct tun_struct *tun = tun_get(tfile);
2215         ssize_t len = iov_iter_count(to), ret;
2216
2217         if (!tun)
2218                 return -EBADFD;
2219         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2220         ret = min_t(ssize_t, ret, len);
2221         if (ret > 0)
2222                 iocb->ki_pos = ret;
2223         tun_put(tun);
2224         return ret;
2225 }
2226
2227 static void tun_prog_free(struct rcu_head *rcu)
2228 {
2229         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2230
2231         bpf_prog_destroy(prog->prog);
2232         kfree(prog);
2233 }
2234
2235 static int __tun_set_ebpf(struct tun_struct *tun,
2236                           struct tun_prog __rcu **prog_p,
2237                           struct bpf_prog *prog)
2238 {
2239         struct tun_prog *old, *new = NULL;
2240
2241         if (prog) {
2242                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2243                 if (!new)
2244                         return -ENOMEM;
2245                 new->prog = prog;
2246         }
2247
2248         spin_lock_bh(&tun->lock);
2249         old = rcu_dereference_protected(*prog_p,
2250                                         lockdep_is_held(&tun->lock));
2251         rcu_assign_pointer(*prog_p, new);
2252         spin_unlock_bh(&tun->lock);
2253
2254         if (old)
2255                 call_rcu(&old->rcu, tun_prog_free);
2256
2257         return 0;
2258 }
2259
2260 static void tun_free_netdev(struct net_device *dev)
2261 {
2262         struct tun_struct *tun = netdev_priv(dev);
2263
2264         BUG_ON(!(list_empty(&tun->disabled)));
2265         free_percpu(tun->pcpu_stats);
2266         tun_flow_uninit(tun);
2267         security_tun_dev_free_security(tun->security);
2268         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2269         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2270 }
2271
2272 static void tun_setup(struct net_device *dev)
2273 {
2274         struct tun_struct *tun = netdev_priv(dev);
2275
2276         tun->owner = INVALID_UID;
2277         tun->group = INVALID_GID;
2278
2279         dev->ethtool_ops = &tun_ethtool_ops;
2280         dev->needs_free_netdev = true;
2281         dev->priv_destructor = tun_free_netdev;
2282         /* We prefer our own queue length */
2283         dev->tx_queue_len = TUN_READQ_SIZE;
2284 }
2285
2286 /* Trivial set of netlink ops to allow deleting tun or tap
2287  * device with netlink.
2288  */
2289 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2290                         struct netlink_ext_ack *extack)
2291 {
2292         return -EINVAL;
2293 }
2294
2295 static size_t tun_get_size(const struct net_device *dev)
2296 {
2297         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2298         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2299
2300         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2301                nla_total_size(sizeof(gid_t)) + /* GROUP */
2302                nla_total_size(sizeof(u8)) + /* TYPE */
2303                nla_total_size(sizeof(u8)) + /* PI */
2304                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2305                nla_total_size(sizeof(u8)) + /* PERSIST */
2306                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2307                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2308                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2309                0;
2310 }
2311
2312 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2313 {
2314         struct tun_struct *tun = netdev_priv(dev);
2315
2316         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2317                 goto nla_put_failure;
2318         if (uid_valid(tun->owner) &&
2319             nla_put_u32(skb, IFLA_TUN_OWNER,
2320                         from_kuid_munged(current_user_ns(), tun->owner)))
2321                 goto nla_put_failure;
2322         if (gid_valid(tun->group) &&
2323             nla_put_u32(skb, IFLA_TUN_GROUP,
2324                         from_kgid_munged(current_user_ns(), tun->group)))
2325                 goto nla_put_failure;
2326         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2327                 goto nla_put_failure;
2328         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2329                 goto nla_put_failure;
2330         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2331                 goto nla_put_failure;
2332         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2333                        !!(tun->flags & IFF_MULTI_QUEUE)))
2334                 goto nla_put_failure;
2335         if (tun->flags & IFF_MULTI_QUEUE) {
2336                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2337                         goto nla_put_failure;
2338                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2339                                 tun->numdisabled))
2340                         goto nla_put_failure;
2341         }
2342
2343         return 0;
2344
2345 nla_put_failure:
2346         return -EMSGSIZE;
2347 }
2348
2349 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2350         .kind           = DRV_NAME,
2351         .priv_size      = sizeof(struct tun_struct),
2352         .setup          = tun_setup,
2353         .validate       = tun_validate,
2354         .get_size       = tun_get_size,
2355         .fill_info      = tun_fill_info,
2356 };
2357
2358 static void tun_sock_write_space(struct sock *sk)
2359 {
2360         struct tun_file *tfile;
2361         wait_queue_head_t *wqueue;
2362
2363         if (!sock_writeable(sk))
2364                 return;
2365
2366         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2367                 return;
2368
2369         wqueue = sk_sleep(sk);
2370         if (wqueue && waitqueue_active(wqueue))
2371                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2372                                                 EPOLLWRNORM | EPOLLWRBAND);
2373
2374         tfile = container_of(sk, struct tun_file, sk);
2375         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2376 }
2377
2378 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2379 {
2380         int ret;
2381         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2382         struct tun_struct *tun = tun_get(tfile);
2383
2384         if (!tun)
2385                 return -EBADFD;
2386
2387         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2388                            m->msg_flags & MSG_DONTWAIT,
2389                            m->msg_flags & MSG_MORE);
2390         tun_put(tun);
2391         return ret;
2392 }
2393
2394 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2395                        int flags)
2396 {
2397         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2398         struct tun_struct *tun = tun_get(tfile);
2399         void *ptr = m->msg_control;
2400         int ret;
2401
2402         if (!tun) {
2403                 ret = -EBADFD;
2404                 goto out_free;
2405         }
2406
2407         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2408                 ret = -EINVAL;
2409                 goto out_put_tun;
2410         }
2411         if (flags & MSG_ERRQUEUE) {
2412                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2413                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2414                 goto out;
2415         }
2416         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2417         if (ret > (ssize_t)total_len) {
2418                 m->msg_flags |= MSG_TRUNC;
2419                 ret = flags & MSG_TRUNC ? ret : total_len;
2420         }
2421 out:
2422         tun_put(tun);
2423         return ret;
2424
2425 out_put_tun:
2426         tun_put(tun);
2427 out_free:
2428         tun_ptr_free(ptr);
2429         return ret;
2430 }
2431
2432 static int tun_ptr_peek_len(void *ptr)
2433 {
2434         if (likely(ptr)) {
2435                 if (tun_is_xdp_buff(ptr)) {
2436                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2437
2438                         return xdp->data_end - xdp->data;
2439                 }
2440                 return __skb_array_len_with_tag(ptr);
2441         } else {
2442                 return 0;
2443         }
2444 }
2445
2446 static int tun_peek_len(struct socket *sock)
2447 {
2448         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2449         struct tun_struct *tun;
2450         int ret = 0;
2451
2452         tun = tun_get(tfile);
2453         if (!tun)
2454                 return 0;
2455
2456         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2457         tun_put(tun);
2458
2459         return ret;
2460 }
2461
2462 /* Ops structure to mimic raw sockets with tun */
2463 static const struct proto_ops tun_socket_ops = {
2464         .peek_len = tun_peek_len,
2465         .sendmsg = tun_sendmsg,
2466         .recvmsg = tun_recvmsg,
2467 };
2468
2469 static struct proto tun_proto = {
2470         .name           = "tun",
2471         .owner          = THIS_MODULE,
2472         .obj_size       = sizeof(struct tun_file),
2473 };
2474
2475 static int tun_flags(struct tun_struct *tun)
2476 {
2477         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2478 }
2479
2480 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2481                               char *buf)
2482 {
2483         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2484         return sprintf(buf, "0x%x\n", tun_flags(tun));
2485 }
2486
2487 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2488                               char *buf)
2489 {
2490         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2491         return uid_valid(tun->owner)?
2492                 sprintf(buf, "%u\n",
2493                         from_kuid_munged(current_user_ns(), tun->owner)):
2494                 sprintf(buf, "-1\n");
2495 }
2496
2497 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2498                               char *buf)
2499 {
2500         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2501         return gid_valid(tun->group) ?
2502                 sprintf(buf, "%u\n",
2503                         from_kgid_munged(current_user_ns(), tun->group)):
2504                 sprintf(buf, "-1\n");
2505 }
2506
2507 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2508 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2509 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2510
2511 static struct attribute *tun_dev_attrs[] = {
2512         &dev_attr_tun_flags.attr,
2513         &dev_attr_owner.attr,
2514         &dev_attr_group.attr,
2515         NULL
2516 };
2517
2518 static const struct attribute_group tun_attr_group = {
2519         .attrs = tun_dev_attrs
2520 };
2521
2522 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2523 {
2524         struct tun_struct *tun;
2525         struct tun_file *tfile = file->private_data;
2526         struct net_device *dev;
2527         int err;
2528
2529         if (tfile->detached)
2530                 return -EINVAL;
2531
2532         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2533                 if (!capable(CAP_NET_ADMIN))
2534                         return -EPERM;
2535
2536                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2537                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2538                         return -EINVAL;
2539         }
2540
2541         dev = __dev_get_by_name(net, ifr->ifr_name);
2542         if (dev) {
2543                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2544                         return -EBUSY;
2545                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2546                         tun = netdev_priv(dev);
2547                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2548                         tun = netdev_priv(dev);
2549                 else
2550                         return -EINVAL;
2551
2552                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2553                     !!(tun->flags & IFF_MULTI_QUEUE))
2554                         return -EINVAL;
2555
2556                 if (tun_not_capable(tun))
2557                         return -EPERM;
2558                 err = security_tun_dev_open(tun->security);
2559                 if (err < 0)
2560                         return err;
2561
2562                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2563                                  ifr->ifr_flags & IFF_NAPI);
2564                 if (err < 0)
2565                         return err;
2566
2567                 if (tun->flags & IFF_MULTI_QUEUE &&
2568                     (tun->numqueues + tun->numdisabled > 1)) {
2569                         /* One or more queue has already been attached, no need
2570                          * to initialize the device again.
2571                          */
2572                         netdev_state_change(dev);
2573                         return 0;
2574                 }
2575
2576                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2577                               (ifr->ifr_flags & TUN_FEATURES);
2578
2579                 netdev_state_change(dev);
2580         } else {
2581                 char *name;
2582                 unsigned long flags = 0;
2583                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2584                              MAX_TAP_QUEUES : 1;
2585
2586                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2587                         return -EPERM;
2588                 err = security_tun_dev_create();
2589                 if (err < 0)
2590                         return err;
2591
2592                 /* Set dev type */
2593                 if (ifr->ifr_flags & IFF_TUN) {
2594                         /* TUN device */
2595                         flags |= IFF_TUN;
2596                         name = "tun%d";
2597                 } else if (ifr->ifr_flags & IFF_TAP) {
2598                         /* TAP device */
2599                         flags |= IFF_TAP;
2600                         name = "tap%d";
2601                 } else
2602                         return -EINVAL;
2603
2604                 if (*ifr->ifr_name)
2605                         name = ifr->ifr_name;
2606
2607                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2608                                        NET_NAME_UNKNOWN, tun_setup, queues,
2609                                        queues);
2610
2611                 if (!dev)
2612                         return -ENOMEM;
2613                 err = dev_get_valid_name(net, dev, name);
2614                 if (err < 0)
2615                         goto err_free_dev;
2616
2617                 dev_net_set(dev, net);
2618                 dev->rtnl_link_ops = &tun_link_ops;
2619                 dev->ifindex = tfile->ifindex;
2620                 dev->sysfs_groups[0] = &tun_attr_group;
2621
2622                 tun = netdev_priv(dev);
2623                 tun->dev = dev;
2624                 tun->flags = flags;
2625                 tun->txflt.count = 0;
2626                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2627
2628                 tun->align = NET_SKB_PAD;
2629                 tun->filter_attached = false;
2630                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2631                 tun->rx_batched = 0;
2632                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2633
2634                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2635                 if (!tun->pcpu_stats) {
2636                         err = -ENOMEM;
2637                         goto err_free_dev;
2638                 }
2639
2640                 spin_lock_init(&tun->lock);
2641
2642                 err = security_tun_dev_alloc_security(&tun->security);
2643                 if (err < 0)
2644                         goto err_free_stat;
2645
2646                 tun_net_init(dev);
2647                 tun_flow_init(tun);
2648
2649                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2650                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2651                                    NETIF_F_HW_VLAN_STAG_TX;
2652                 dev->features = dev->hw_features | NETIF_F_LLTX;
2653                 dev->vlan_features = dev->features &
2654                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2655                                        NETIF_F_HW_VLAN_STAG_TX);
2656
2657                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2658                               (ifr->ifr_flags & TUN_FEATURES);
2659
2660                 INIT_LIST_HEAD(&tun->disabled);
2661                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2662                 if (err < 0)
2663                         goto err_free_flow;
2664
2665                 err = register_netdevice(tun->dev);
2666                 if (err < 0)
2667                         goto err_detach;
2668         }
2669
2670         netif_carrier_on(tun->dev);
2671
2672         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2673
2674         /* Make sure persistent devices do not get stuck in
2675          * xoff state.
2676          */
2677         if (netif_running(tun->dev))
2678                 netif_tx_wake_all_queues(tun->dev);
2679
2680         strcpy(ifr->ifr_name, tun->dev->name);
2681         return 0;
2682
2683 err_detach:
2684         tun_detach_all(dev);
2685         /* register_netdevice() already called tun_free_netdev() */
2686         goto err_free_dev;
2687
2688 err_free_flow:
2689         tun_flow_uninit(tun);
2690         security_tun_dev_free_security(tun->security);
2691 err_free_stat:
2692         free_percpu(tun->pcpu_stats);
2693 err_free_dev:
2694         free_netdev(dev);
2695         return err;
2696 }
2697
2698 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2699                        struct ifreq *ifr)
2700 {
2701         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2702
2703         strcpy(ifr->ifr_name, tun->dev->name);
2704
2705         ifr->ifr_flags = tun_flags(tun);
2706
2707 }
2708
2709 /* This is like a cut-down ethtool ops, except done via tun fd so no
2710  * privs required. */
2711 static int set_offload(struct tun_struct *tun, unsigned long arg)
2712 {
2713         netdev_features_t features = 0;
2714
2715         if (arg & TUN_F_CSUM) {
2716                 features |= NETIF_F_HW_CSUM;
2717                 arg &= ~TUN_F_CSUM;
2718
2719                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2720                         if (arg & TUN_F_TSO_ECN) {
2721                                 features |= NETIF_F_TSO_ECN;
2722                                 arg &= ~TUN_F_TSO_ECN;
2723                         }
2724                         if (arg & TUN_F_TSO4)
2725                                 features |= NETIF_F_TSO;
2726                         if (arg & TUN_F_TSO6)
2727                                 features |= NETIF_F_TSO6;
2728                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2729                 }
2730
2731                 arg &= ~TUN_F_UFO;
2732         }
2733
2734         /* This gives the user a way to test for new features in future by
2735          * trying to set them. */
2736         if (arg)
2737                 return -EINVAL;
2738
2739         tun->set_features = features;
2740         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2741         tun->dev->wanted_features |= features;
2742         netdev_update_features(tun->dev);
2743
2744         return 0;
2745 }
2746
2747 static void tun_detach_filter(struct tun_struct *tun, int n)
2748 {
2749         int i;
2750         struct tun_file *tfile;
2751
2752         for (i = 0; i < n; i++) {
2753                 tfile = rtnl_dereference(tun->tfiles[i]);
2754                 lock_sock(tfile->socket.sk);
2755                 sk_detach_filter(tfile->socket.sk);
2756                 release_sock(tfile->socket.sk);
2757         }
2758
2759         tun->filter_attached = false;
2760 }
2761
2762 static int tun_attach_filter(struct tun_struct *tun)
2763 {
2764         int i, ret = 0;
2765         struct tun_file *tfile;
2766
2767         for (i = 0; i < tun->numqueues; i++) {
2768                 tfile = rtnl_dereference(tun->tfiles[i]);
2769                 lock_sock(tfile->socket.sk);
2770                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2771                 release_sock(tfile->socket.sk);
2772                 if (ret) {
2773                         tun_detach_filter(tun, i);
2774                         return ret;
2775                 }
2776         }
2777
2778         tun->filter_attached = true;
2779         return ret;
2780 }
2781
2782 static void tun_set_sndbuf(struct tun_struct *tun)
2783 {
2784         struct tun_file *tfile;
2785         int i;
2786
2787         for (i = 0; i < tun->numqueues; i++) {
2788                 tfile = rtnl_dereference(tun->tfiles[i]);
2789                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2790         }
2791 }
2792
2793 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2794 {
2795         struct tun_file *tfile = file->private_data;
2796         struct tun_struct *tun;
2797         int ret = 0;
2798
2799         rtnl_lock();
2800
2801         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2802                 tun = tfile->detached;
2803                 if (!tun) {
2804                         ret = -EINVAL;
2805                         goto unlock;
2806                 }
2807                 ret = security_tun_dev_attach_queue(tun->security);
2808                 if (ret < 0)
2809                         goto unlock;
2810                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2811         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2812                 tun = rtnl_dereference(tfile->tun);
2813                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2814                         ret = -EINVAL;
2815                 else
2816                         __tun_detach(tfile, false);
2817         } else
2818                 ret = -EINVAL;
2819
2820         if (ret >= 0)
2821                 netdev_state_change(tun->dev);
2822
2823 unlock:
2824         rtnl_unlock();
2825         return ret;
2826 }
2827
2828 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2829                         void __user *data)
2830 {
2831         struct bpf_prog *prog;
2832         int fd;
2833
2834         if (copy_from_user(&fd, data, sizeof(fd)))
2835                 return -EFAULT;
2836
2837         if (fd == -1) {
2838                 prog = NULL;
2839         } else {
2840                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2841                 if (IS_ERR(prog))
2842                         return PTR_ERR(prog);
2843         }
2844
2845         return __tun_set_ebpf(tun, prog_p, prog);
2846 }
2847
2848 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2849                             unsigned long arg, int ifreq_len)
2850 {
2851         struct tun_file *tfile = file->private_data;
2852         struct tun_struct *tun;
2853         void __user* argp = (void __user*)arg;
2854         struct ifreq ifr;
2855         struct net *net;
2856         kuid_t owner;
2857         kgid_t group;
2858         int sndbuf;
2859         int vnet_hdr_sz;
2860         unsigned int ifindex;
2861         int le;
2862         int ret;
2863         bool do_notify = false;
2864
2865         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2866             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2867                 if (copy_from_user(&ifr, argp, ifreq_len))
2868                         return -EFAULT;
2869         } else {
2870                 memset(&ifr, 0, sizeof(ifr));
2871         }
2872         if (cmd == TUNGETFEATURES) {
2873                 /* Currently this just means: "what IFF flags are valid?".
2874                  * This is needed because we never checked for invalid flags on
2875                  * TUNSETIFF.
2876                  */
2877                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2878                                 (unsigned int __user*)argp);
2879         } else if (cmd == TUNSETQUEUE)
2880                 return tun_set_queue(file, &ifr);
2881
2882         ret = 0;
2883         rtnl_lock();
2884
2885         tun = tun_get(tfile);
2886         net = sock_net(&tfile->sk);
2887         if (cmd == TUNSETIFF) {
2888                 ret = -EEXIST;
2889                 if (tun)
2890                         goto unlock;
2891
2892                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2893
2894                 ret = tun_set_iff(net, file, &ifr);
2895
2896                 if (ret)
2897                         goto unlock;
2898
2899                 if (copy_to_user(argp, &ifr, ifreq_len))
2900                         ret = -EFAULT;
2901                 goto unlock;
2902         }
2903         if (cmd == TUNSETIFINDEX) {
2904                 ret = -EPERM;
2905                 if (tun)
2906                         goto unlock;
2907
2908                 ret = -EFAULT;
2909                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2910                         goto unlock;
2911
2912                 ret = 0;
2913                 tfile->ifindex = ifindex;
2914                 goto unlock;
2915         }
2916         if (cmd == SIOCGSKNS) {
2917                 ret = -EPERM;
2918                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2919                         goto unlock;
2920
2921                 ret = open_related_ns(&net->ns, get_net_ns);
2922                 goto unlock;
2923         }
2924
2925         ret = -EBADFD;
2926         if (!tun)
2927                 goto unlock;
2928
2929         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2930
2931         ret = 0;
2932         switch (cmd) {
2933         case TUNGETIFF:
2934                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2935
2936                 if (tfile->detached)
2937                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2938                 if (!tfile->socket.sk->sk_filter)
2939                         ifr.ifr_flags |= IFF_NOFILTER;
2940
2941                 if (copy_to_user(argp, &ifr, ifreq_len))
2942                         ret = -EFAULT;
2943                 break;
2944
2945         case TUNSETNOCSUM:
2946                 /* Disable/Enable checksum */
2947
2948                 /* [unimplemented] */
2949                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2950                           arg ? "disabled" : "enabled");
2951                 break;
2952
2953         case TUNSETPERSIST:
2954                 /* Disable/Enable persist mode. Keep an extra reference to the
2955                  * module to prevent the module being unprobed.
2956                  */
2957                 if (arg && !(tun->flags & IFF_PERSIST)) {
2958                         tun->flags |= IFF_PERSIST;
2959                         __module_get(THIS_MODULE);
2960                         do_notify = true;
2961                 }
2962                 if (!arg && (tun->flags & IFF_PERSIST)) {
2963                         tun->flags &= ~IFF_PERSIST;
2964                         module_put(THIS_MODULE);
2965                         do_notify = true;
2966                 }
2967
2968                 tun_debug(KERN_INFO, tun, "persist %s\n",
2969                           arg ? "enabled" : "disabled");
2970                 break;
2971
2972         case TUNSETOWNER:
2973                 /* Set owner of the device */
2974                 owner = make_kuid(current_user_ns(), arg);
2975                 if (!uid_valid(owner)) {
2976                         ret = -EINVAL;
2977                         break;
2978                 }
2979                 tun->owner = owner;
2980                 do_notify = true;
2981                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2982                           from_kuid(&init_user_ns, tun->owner));
2983                 break;
2984
2985         case TUNSETGROUP:
2986                 /* Set group of the device */
2987                 group = make_kgid(current_user_ns(), arg);
2988                 if (!gid_valid(group)) {
2989                         ret = -EINVAL;
2990                         break;
2991                 }
2992                 tun->group = group;
2993                 do_notify = true;
2994                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2995                           from_kgid(&init_user_ns, tun->group));
2996                 break;
2997
2998         case TUNSETLINK:
2999                 /* Only allow setting the type when the interface is down */
3000                 if (tun->dev->flags & IFF_UP) {
3001                         tun_debug(KERN_INFO, tun,
3002                                   "Linktype set failed because interface is up\n");
3003                         ret = -EBUSY;
3004                 } else {
3005                         tun->dev->type = (int) arg;
3006                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3007                                   tun->dev->type);
3008                         ret = 0;
3009                 }
3010                 break;
3011
3012 #ifdef TUN_DEBUG
3013         case TUNSETDEBUG:
3014                 tun->debug = arg;
3015                 break;
3016 #endif
3017         case TUNSETOFFLOAD:
3018                 ret = set_offload(tun, arg);
3019                 break;
3020
3021         case TUNSETTXFILTER:
3022                 /* Can be set only for TAPs */
3023                 ret = -EINVAL;
3024                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3025                         break;
3026                 ret = update_filter(&tun->txflt, (void __user *)arg);
3027                 break;
3028
3029         case SIOCGIFHWADDR:
3030                 /* Get hw address */
3031                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3032                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3033                 if (copy_to_user(argp, &ifr, ifreq_len))
3034                         ret = -EFAULT;
3035                 break;
3036
3037         case SIOCSIFHWADDR:
3038                 /* Set hw address */
3039                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3040                           ifr.ifr_hwaddr.sa_data);
3041
3042                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3043                 break;
3044
3045         case TUNGETSNDBUF:
3046                 sndbuf = tfile->socket.sk->sk_sndbuf;
3047                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3048                         ret = -EFAULT;
3049                 break;
3050
3051         case TUNSETSNDBUF:
3052                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3053                         ret = -EFAULT;
3054                         break;
3055                 }
3056                 if (sndbuf <= 0) {
3057                         ret = -EINVAL;
3058                         break;
3059                 }
3060
3061                 tun->sndbuf = sndbuf;
3062                 tun_set_sndbuf(tun);
3063                 break;
3064
3065         case TUNGETVNETHDRSZ:
3066                 vnet_hdr_sz = tun->vnet_hdr_sz;
3067                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3068                         ret = -EFAULT;
3069                 break;
3070
3071         case TUNSETVNETHDRSZ:
3072                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3073                         ret = -EFAULT;
3074                         break;
3075                 }
3076                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3077                         ret = -EINVAL;
3078                         break;
3079                 }
3080
3081                 tun->vnet_hdr_sz = vnet_hdr_sz;
3082                 break;
3083
3084         case TUNGETVNETLE:
3085                 le = !!(tun->flags & TUN_VNET_LE);
3086                 if (put_user(le, (int __user *)argp))
3087                         ret = -EFAULT;
3088                 break;
3089
3090         case TUNSETVNETLE:
3091                 if (get_user(le, (int __user *)argp)) {
3092                         ret = -EFAULT;
3093                         break;
3094                 }
3095                 if (le)
3096                         tun->flags |= TUN_VNET_LE;
3097                 else
3098                         tun->flags &= ~TUN_VNET_LE;
3099                 break;
3100
3101         case TUNGETVNETBE:
3102                 ret = tun_get_vnet_be(tun, argp);
3103                 break;
3104
3105         case TUNSETVNETBE:
3106                 ret = tun_set_vnet_be(tun, argp);
3107                 break;
3108
3109         case TUNATTACHFILTER:
3110                 /* Can be set only for TAPs */
3111                 ret = -EINVAL;
3112                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3113                         break;
3114                 ret = -EFAULT;
3115                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3116                         break;
3117
3118                 ret = tun_attach_filter(tun);
3119                 break;
3120
3121         case TUNDETACHFILTER:
3122                 /* Can be set only for TAPs */
3123                 ret = -EINVAL;
3124                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3125                         break;
3126                 ret = 0;
3127                 tun_detach_filter(tun, tun->numqueues);
3128                 break;
3129
3130         case TUNGETFILTER:
3131                 ret = -EINVAL;
3132                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3133                         break;
3134                 ret = -EFAULT;
3135                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3136                         break;
3137                 ret = 0;
3138                 break;
3139
3140         case TUNSETSTEERINGEBPF:
3141                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3142                 break;
3143
3144         case TUNSETFILTEREBPF:
3145                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3146                 break;
3147
3148         default:
3149                 ret = -EINVAL;
3150                 break;
3151         }
3152
3153         if (do_notify)
3154                 netdev_state_change(tun->dev);
3155
3156 unlock:
3157         rtnl_unlock();
3158         if (tun)
3159                 tun_put(tun);
3160         return ret;
3161 }
3162
3163 static long tun_chr_ioctl(struct file *file,
3164                           unsigned int cmd, unsigned long arg)
3165 {
3166         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3167 }
3168
3169 #ifdef CONFIG_COMPAT
3170 static long tun_chr_compat_ioctl(struct file *file,
3171                          unsigned int cmd, unsigned long arg)
3172 {
3173         switch (cmd) {
3174         case TUNSETIFF:
3175         case TUNGETIFF:
3176         case TUNSETTXFILTER:
3177         case TUNGETSNDBUF:
3178         case TUNSETSNDBUF:
3179         case SIOCGIFHWADDR:
3180         case SIOCSIFHWADDR:
3181                 arg = (unsigned long)compat_ptr(arg);
3182                 break;
3183         default:
3184                 arg = (compat_ulong_t)arg;
3185                 break;
3186         }
3187
3188         /*
3189          * compat_ifreq is shorter than ifreq, so we must not access beyond
3190          * the end of that structure. All fields that are used in this
3191          * driver are compatible though, we don't need to convert the
3192          * contents.
3193          */
3194         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3195 }
3196 #endif /* CONFIG_COMPAT */
3197
3198 static int tun_chr_fasync(int fd, struct file *file, int on)
3199 {
3200         struct tun_file *tfile = file->private_data;
3201         int ret;
3202
3203         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3204                 goto out;
3205
3206         if (on) {
3207                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3208                 tfile->flags |= TUN_FASYNC;
3209         } else
3210                 tfile->flags &= ~TUN_FASYNC;
3211         ret = 0;
3212 out:
3213         return ret;
3214 }
3215
3216 static int tun_chr_open(struct inode *inode, struct file * file)
3217 {
3218         struct net *net = current->nsproxy->net_ns;
3219         struct tun_file *tfile;
3220
3221         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3222
3223         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3224                                             &tun_proto, 0);
3225         if (!tfile)
3226                 return -ENOMEM;
3227         RCU_INIT_POINTER(tfile->tun, NULL);
3228         tfile->flags = 0;
3229         tfile->ifindex = 0;
3230
3231         init_waitqueue_head(&tfile->wq.wait);
3232         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3233
3234         tfile->socket.file = file;
3235         tfile->socket.ops = &tun_socket_ops;
3236
3237         sock_init_data(&tfile->socket, &tfile->sk);
3238
3239         tfile->sk.sk_write_space = tun_sock_write_space;
3240         tfile->sk.sk_sndbuf = INT_MAX;
3241
3242         file->private_data = tfile;
3243         INIT_LIST_HEAD(&tfile->next);
3244
3245         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3246
3247         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3248
3249         return 0;
3250 }
3251
3252 static int tun_chr_close(struct inode *inode, struct file *file)
3253 {
3254         struct tun_file *tfile = file->private_data;
3255
3256         tun_detach(tfile, true);
3257
3258         return 0;
3259 }
3260
3261 #ifdef CONFIG_PROC_FS
3262 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3263 {
3264         struct tun_file *tfile = file->private_data;
3265         struct tun_struct *tun;
3266         struct ifreq ifr;
3267
3268         memset(&ifr, 0, sizeof(ifr));
3269
3270         rtnl_lock();
3271         tun = tun_get(tfile);
3272         if (tun)
3273                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3274         rtnl_unlock();
3275
3276         if (tun)
3277                 tun_put(tun);
3278
3279         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3280 }
3281 #endif
3282
3283 static const struct file_operations tun_fops = {
3284         .owner  = THIS_MODULE,
3285         .llseek = no_llseek,
3286         .read_iter  = tun_chr_read_iter,
3287         .write_iter = tun_chr_write_iter,
3288         .poll   = tun_chr_poll,
3289         .unlocked_ioctl = tun_chr_ioctl,
3290 #ifdef CONFIG_COMPAT
3291         .compat_ioctl = tun_chr_compat_ioctl,
3292 #endif
3293         .open   = tun_chr_open,
3294         .release = tun_chr_close,
3295         .fasync = tun_chr_fasync,
3296 #ifdef CONFIG_PROC_FS
3297         .show_fdinfo = tun_chr_show_fdinfo,
3298 #endif
3299 };
3300
3301 static struct miscdevice tun_miscdev = {
3302         .minor = TUN_MINOR,
3303         .name = "tun",
3304         .nodename = "net/tun",
3305         .fops = &tun_fops,
3306 };
3307
3308 /* ethtool interface */
3309
3310 static int tun_get_link_ksettings(struct net_device *dev,
3311                                   struct ethtool_link_ksettings *cmd)
3312 {
3313         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3314         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3315         cmd->base.speed         = SPEED_10;
3316         cmd->base.duplex        = DUPLEX_FULL;
3317         cmd->base.port          = PORT_TP;
3318         cmd->base.phy_address   = 0;
3319         cmd->base.autoneg       = AUTONEG_DISABLE;
3320         return 0;
3321 }
3322
3323 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3324 {
3325         struct tun_struct *tun = netdev_priv(dev);
3326
3327         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3328         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3329
3330         switch (tun->flags & TUN_TYPE_MASK) {
3331         case IFF_TUN:
3332                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3333                 break;
3334         case IFF_TAP:
3335                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3336                 break;
3337         }
3338 }
3339
3340 static u32 tun_get_msglevel(struct net_device *dev)
3341 {
3342 #ifdef TUN_DEBUG
3343         struct tun_struct *tun = netdev_priv(dev);
3344         return tun->debug;
3345 #else
3346         return -EOPNOTSUPP;
3347 #endif
3348 }
3349
3350 static void tun_set_msglevel(struct net_device *dev, u32 value)
3351 {
3352 #ifdef TUN_DEBUG
3353         struct tun_struct *tun = netdev_priv(dev);
3354         tun->debug = value;
3355 #endif
3356 }
3357
3358 static int tun_get_coalesce(struct net_device *dev,
3359                             struct ethtool_coalesce *ec)
3360 {
3361         struct tun_struct *tun = netdev_priv(dev);
3362
3363         ec->rx_max_coalesced_frames = tun->rx_batched;
3364
3365         return 0;
3366 }
3367
3368 static int tun_set_coalesce(struct net_device *dev,
3369                             struct ethtool_coalesce *ec)
3370 {
3371         struct tun_struct *tun = netdev_priv(dev);
3372
3373         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3374                 tun->rx_batched = NAPI_POLL_WEIGHT;
3375         else
3376                 tun->rx_batched = ec->rx_max_coalesced_frames;
3377
3378         return 0;
3379 }
3380
3381 static const struct ethtool_ops tun_ethtool_ops = {
3382         .get_drvinfo    = tun_get_drvinfo,
3383         .get_msglevel   = tun_get_msglevel,
3384         .set_msglevel   = tun_set_msglevel,
3385         .get_link       = ethtool_op_get_link,
3386         .get_ts_info    = ethtool_op_get_ts_info,
3387         .get_coalesce   = tun_get_coalesce,
3388         .set_coalesce   = tun_set_coalesce,
3389         .get_link_ksettings = tun_get_link_ksettings,
3390 };
3391
3392 static int tun_queue_resize(struct tun_struct *tun)
3393 {
3394         struct net_device *dev = tun->dev;
3395         struct tun_file *tfile;
3396         struct ptr_ring **rings;
3397         int n = tun->numqueues + tun->numdisabled;
3398         int ret, i;
3399
3400         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3401         if (!rings)
3402                 return -ENOMEM;
3403
3404         for (i = 0; i < tun->numqueues; i++) {
3405                 tfile = rtnl_dereference(tun->tfiles[i]);
3406                 rings[i] = &tfile->tx_ring;
3407         }
3408         list_for_each_entry(tfile, &tun->disabled, next)
3409                 rings[i++] = &tfile->tx_ring;
3410
3411         ret = ptr_ring_resize_multiple(rings, n,
3412                                        dev->tx_queue_len, GFP_KERNEL,
3413                                        tun_ptr_free);
3414
3415         kfree(rings);
3416         return ret;
3417 }
3418
3419 static int tun_device_event(struct notifier_block *unused,
3420                             unsigned long event, void *ptr)
3421 {
3422         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3423         struct tun_struct *tun = netdev_priv(dev);
3424
3425         if (dev->rtnl_link_ops != &tun_link_ops)
3426                 return NOTIFY_DONE;
3427
3428         switch (event) {
3429         case NETDEV_CHANGE_TX_QUEUE_LEN:
3430                 if (tun_queue_resize(tun))
3431                         return NOTIFY_BAD;
3432                 break;
3433         default:
3434                 break;
3435         }
3436
3437         return NOTIFY_DONE;
3438 }
3439
3440 static struct notifier_block tun_notifier_block __read_mostly = {
3441         .notifier_call  = tun_device_event,
3442 };
3443
3444 static int __init tun_init(void)
3445 {
3446         int ret = 0;
3447
3448         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3449
3450         ret = rtnl_link_register(&tun_link_ops);
3451         if (ret) {
3452                 pr_err("Can't register link_ops\n");
3453                 goto err_linkops;
3454         }
3455
3456         ret = misc_register(&tun_miscdev);
3457         if (ret) {
3458                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3459                 goto err_misc;
3460         }
3461
3462         ret = register_netdevice_notifier(&tun_notifier_block);
3463         if (ret) {
3464                 pr_err("Can't register netdevice notifier\n");
3465                 goto err_notifier;
3466         }
3467
3468         return  0;
3469
3470 err_notifier:
3471         misc_deregister(&tun_miscdev);
3472 err_misc:
3473         rtnl_link_unregister(&tun_link_ops);
3474 err_linkops:
3475         return ret;
3476 }
3477
3478 static void tun_cleanup(void)
3479 {
3480         misc_deregister(&tun_miscdev);
3481         rtnl_link_unregister(&tun_link_ops);
3482         unregister_netdevice_notifier(&tun_notifier_block);
3483 }
3484
3485 /* Get an underlying socket object from tun file.  Returns error unless file is
3486  * attached to a device.  The returned object works like a packet socket, it
3487  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3488  * holding a reference to the file for as long as the socket is in use. */
3489 struct socket *tun_get_socket(struct file *file)
3490 {
3491         struct tun_file *tfile;
3492         if (file->f_op != &tun_fops)
3493                 return ERR_PTR(-EINVAL);
3494         tfile = file->private_data;
3495         if (!tfile)
3496                 return ERR_PTR(-EBADFD);
3497         return &tfile->socket;
3498 }
3499 EXPORT_SYMBOL_GPL(tun_get_socket);
3500
3501 struct ptr_ring *tun_get_tx_ring(struct file *file)
3502 {
3503         struct tun_file *tfile;
3504
3505         if (file->f_op != &tun_fops)
3506                 return ERR_PTR(-EINVAL);
3507         tfile = file->private_data;
3508         if (!tfile)
3509                 return ERR_PTR(-EBADFD);
3510         return &tfile->tx_ring;
3511 }
3512 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3513
3514 module_init(tun_init);
3515 module_exit(tun_cleanup);
3516 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3517 MODULE_AUTHOR(DRV_COPYRIGHT);
3518 MODULE_LICENSE("GPL");
3519 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3520 MODULE_ALIAS("devname:net/tun");