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