drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81
  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(struct timer_list *t)
 448 {
 449         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 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         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1200         mod_timer(&tun->flow_gc_timer,
1201                   round_jiffies_up(jiffies + tun->ageing_time));
1202 }
1203
1204 static void tun_flow_uninit(struct tun_struct *tun)
1205 {
1206         del_timer_sync(&tun->flow_gc_timer);
1207         tun_flow_flush(tun);
1208 }
1209
1210 #define MIN_MTU 68
1211 #define MAX_MTU 65535
1212
1213 /* Initialize net device. */
1214 static void tun_net_init(struct net_device *dev)
1215 {
1216         struct tun_struct *tun = netdev_priv(dev);
1217
1218         switch (tun->flags & TUN_TYPE_MASK) {
1219         case IFF_TUN:
1220                 dev->netdev_ops = &tun_netdev_ops;
1221
1222                 /* Point-to-Point TUN Device */
1223                 dev->hard_header_len = 0;
1224                 dev->addr_len = 0;
1225                 dev->mtu = 1500;
1226
1227                 /* Zero header length */
1228                 dev->type = ARPHRD_NONE;
1229                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1230                 break;
1231
1232         case IFF_TAP:
1233                 dev->netdev_ops = &tap_netdev_ops;
1234                 /* Ethernet TAP Device */
1235                 ether_setup(dev);
1236                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1237                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1238
1239                 eth_hw_addr_random(dev);
1240
1241                 break;
1242         }
1243
1244         dev->min_mtu = MIN_MTU;
1245         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1246 }
1247
1248 /* Character device part */
1249
1250 /* Poll */
1251 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1252 {
1253         struct tun_file *tfile = file->private_data;
1254         struct tun_struct *tun = tun_get(tfile);
1255         struct sock *sk;
1256         unsigned int mask = 0;
1257
1258         if (!tun)
1259                 return POLLERR;
1260
1261         sk = tfile->socket.sk;
1262
1263         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1264
1265         poll_wait(file, sk_sleep(sk), wait);
1266
1267         if (!skb_array_empty(&tfile->tx_array))
1268                 mask |= POLLIN | POLLRDNORM;
1269
1270         if (tun->dev->flags & IFF_UP &&
1271             (sock_writeable(sk) ||
1272              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1273               sock_writeable(sk))))
1274                 mask |= POLLOUT | POLLWRNORM;
1275
1276         if (tun->dev->reg_state != NETREG_REGISTERED)
1277                 mask = POLLERR;
1278
1279         tun_put(tun);
1280         return mask;
1281 }
1282
1283 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1284                                             size_t len,
1285                                             const struct iov_iter *it)
1286 {
1287         struct sk_buff *skb;
1288         size_t linear;
1289         int err;
1290         int i;
1291
1292         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1293                 return ERR_PTR(-ENOMEM);
1294
1295         local_bh_disable();
1296         skb = napi_get_frags(&tfile->napi);
1297         local_bh_enable();
1298         if (!skb)
1299                 return ERR_PTR(-ENOMEM);
1300
1301         linear = iov_iter_single_seg_count(it);
1302         err = __skb_grow(skb, linear);
1303         if (err)
1304                 goto free;
1305
1306         skb->len = len;
1307         skb->data_len = len - linear;
1308         skb->truesize += skb->data_len;
1309
1310         for (i = 1; i < it->nr_segs; i++) {
1311                 size_t fragsz = it->iov[i].iov_len;
1312                 unsigned long offset;
1313                 struct page *page;
1314                 void *data;
1315
1316                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1317                         err = -EINVAL;
1318                         goto free;
1319                 }
1320
1321                 local_bh_disable();
1322                 data = napi_alloc_frag(fragsz);
1323                 local_bh_enable();
1324                 if (!data) {
1325                         err = -ENOMEM;
1326                         goto free;
1327                 }
1328
1329                 page = virt_to_head_page(data);
1330                 offset = data - page_address(page);
1331                 skb_fill_page_desc(skb, i - 1, page, offset, fragsz);
1332         }
1333
1334         return skb;
1335 free:
1336         /* frees skb and all frags allocated with napi_alloc_frag() */
1337         napi_free_frags(&tfile->napi);
1338         return ERR_PTR(err);
1339 }
1340
1341 /* prepad is the amount to reserve at front.  len is length after that.
1342  * linear is a hint as to how much to copy (usually headers). */
1343 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1344                                      size_t prepad, size_t len,
1345                                      size_t linear, int noblock)
1346 {
1347         struct sock *sk = tfile->socket.sk;
1348         struct sk_buff *skb;
1349         int err;
1350
1351         /* Under a page?  Don't bother with paged skb. */
1352         if (prepad + len < PAGE_SIZE || !linear)
1353                 linear = len;
1354
1355         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1356                                    &err, 0);
1357         if (!skb)
1358                 return ERR_PTR(err);
1359
1360         skb_reserve(skb, prepad);
1361         skb_put(skb, linear);
1362         skb->data_len = len - linear;
1363         skb->len += len - linear;
1364
1365         return skb;
1366 }
1367
1368 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1369                            struct sk_buff *skb, int more)
1370 {
1371         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1372         struct sk_buff_head process_queue;
1373         u32 rx_batched = tun->rx_batched;
1374         bool rcv = false;
1375
1376         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1377                 local_bh_disable();
1378                 netif_receive_skb(skb);
1379                 local_bh_enable();
1380                 return;
1381         }
1382
1383         spin_lock(&queue->lock);
1384         if (!more || skb_queue_len(queue) == rx_batched) {
1385                 __skb_queue_head_init(&process_queue);
1386                 skb_queue_splice_tail_init(queue, &process_queue);
1387                 rcv = true;
1388         } else {
1389                 __skb_queue_tail(queue, skb);
1390         }
1391         spin_unlock(&queue->lock);
1392
1393         if (rcv) {
1394                 struct sk_buff *nskb;
1395
1396                 local_bh_disable();
1397                 while ((nskb = __skb_dequeue(&process_queue)))
1398                         netif_receive_skb(nskb);
1399                 netif_receive_skb(skb);
1400                 local_bh_enable();
1401         }
1402 }
1403
1404 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1405                               int len, int noblock, bool zerocopy)
1406 {
1407         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1408                 return false;
1409
1410         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1411                 return false;
1412
1413         if (!noblock)
1414                 return false;
1415
1416         if (zerocopy)
1417                 return false;
1418
1419         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1420             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1421                 return false;
1422
1423         return true;
1424 }
1425
1426 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1427                                      struct tun_file *tfile,
1428                                      struct iov_iter *from,
1429                                      struct virtio_net_hdr *hdr,
1430                                      int len, int *skb_xdp)
1431 {
1432         struct page_frag *alloc_frag = &current->task_frag;
1433         struct sk_buff *skb;
1434         struct bpf_prog *xdp_prog;
1435         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1436         unsigned int delta = 0;
1437         char *buf;
1438         size_t copied;
1439         bool xdp_xmit = false;
1440         int err, pad = TUN_RX_PAD;
1441
1442         rcu_read_lock();
1443         xdp_prog = rcu_dereference(tun->xdp_prog);
1444         if (xdp_prog)
1445                 pad += TUN_HEADROOM;
1446         buflen += SKB_DATA_ALIGN(len + pad);
1447         rcu_read_unlock();
1448
1449         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1450         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1451                 return ERR_PTR(-ENOMEM);
1452
1453         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1454         copied = copy_page_from_iter(alloc_frag->page,
1455                                      alloc_frag->offset + pad,
1456                                      len, from);
1457         if (copied != len)
1458                 return ERR_PTR(-EFAULT);
1459
1460         /* There's a small window that XDP may be set after the check
1461          * of xdp_prog above, this should be rare and for simplicity
1462          * we do XDP on skb in case the headroom is not enough.
1463          */
1464         if (hdr->gso_type || !xdp_prog)
1465                 *skb_xdp = 1;
1466         else
1467                 *skb_xdp = 0;
1468
1469         rcu_read_lock();
1470         xdp_prog = rcu_dereference(tun->xdp_prog);
1471         if (xdp_prog && !*skb_xdp) {
1472                 struct xdp_buff xdp;
1473                 void *orig_data;
1474                 u32 act;
1475
1476                 xdp.data_hard_start = buf;
1477                 xdp.data = buf + pad;
1478                 xdp_set_data_meta_invalid(&xdp);
1479                 xdp.data_end = xdp.data + len;
1480                 orig_data = xdp.data;
1481                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1482
1483                 switch (act) {
1484                 case XDP_REDIRECT:
1485                         get_page(alloc_frag->page);
1486                         alloc_frag->offset += buflen;
1487                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1488                         if (err)
1489                                 goto err_redirect;
1490                         rcu_read_unlock();
1491                         return NULL;
1492                 case XDP_TX:
1493                         xdp_xmit = true;
1494                         /* fall through */
1495                 case XDP_PASS:
1496                         delta = orig_data - xdp.data;
1497                         break;
1498                 default:
1499                         bpf_warn_invalid_xdp_action(act);
1500                         /* fall through */
1501                 case XDP_ABORTED:
1502                         trace_xdp_exception(tun->dev, xdp_prog, act);
1503                         /* fall through */
1504                 case XDP_DROP:
1505                         goto err_xdp;
1506                 }
1507         }
1508
1509         skb = build_skb(buf, buflen);
1510         if (!skb) {
1511                 rcu_read_unlock();
1512                 return ERR_PTR(-ENOMEM);
1513         }
1514
1515         skb_reserve(skb, pad - delta);
1516         skb_put(skb, len + delta);
1517         get_page(alloc_frag->page);
1518         alloc_frag->offset += buflen;
1519
1520         if (xdp_xmit) {
1521                 skb->dev = tun->dev;
1522                 generic_xdp_tx(skb, xdp_prog);
1523                 rcu_read_unlock();
1524                 return NULL;
1525         }
1526
1527         rcu_read_unlock();
1528
1529         return skb;
1530
1531 err_redirect:
1532         put_page(alloc_frag->page);
1533 err_xdp:
1534         rcu_read_unlock();
1535         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1536         return NULL;
1537 }
1538
1539 /* Get packet from user space buffer */
1540 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1541                             void *msg_control, struct iov_iter *from,
1542                             int noblock, bool more)
1543 {
1544         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1545         struct sk_buff *skb;
1546         size_t total_len = iov_iter_count(from);
1547         size_t len = total_len, align = tun->align, linear;
1548         struct virtio_net_hdr gso = { 0 };
1549         struct tun_pcpu_stats *stats;
1550         int good_linear;
1551         int copylen;
1552         bool zerocopy = false;
1553         int err;
1554         u32 rxhash;
1555         int skb_xdp = 1;
1556         bool frags = tun_napi_frags_enabled(tun);
1557
1558         if (!(tun->dev->flags & IFF_UP))
1559                 return -EIO;
1560
1561         if (!(tun->flags & IFF_NO_PI)) {
1562                 if (len < sizeof(pi))
1563                         return -EINVAL;
1564                 len -= sizeof(pi);
1565
1566                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1567                         return -EFAULT;
1568         }
1569
1570         if (tun->flags & IFF_VNET_HDR) {
1571                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1572
1573                 if (len < vnet_hdr_sz)
1574                         return -EINVAL;
1575                 len -= vnet_hdr_sz;
1576
1577                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1578                         return -EFAULT;
1579
1580                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1581                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1582                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1583
1584                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1585                         return -EINVAL;
1586                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1587         }
1588
1589         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1590                 align += NET_IP_ALIGN;
1591                 if (unlikely(len < ETH_HLEN ||
1592                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1593                         return -EINVAL;
1594         }
1595
1596         good_linear = SKB_MAX_HEAD(align);
1597
1598         if (msg_control) {
1599                 struct iov_iter i = *from;
1600
1601                 /* There are 256 bytes to be copied in skb, so there is
1602                  * enough room for skb expand head in case it is used.
1603                  * The rest of the buffer is mapped from userspace.
1604                  */
1605                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1606                 if (copylen > good_linear)
1607                         copylen = good_linear;
1608                 linear = copylen;
1609                 iov_iter_advance(&i, copylen);
1610                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1611                         zerocopy = true;
1612         }
1613
1614         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1615                 /* For the packet that is not easy to be processed
1616                  * (e.g gso or jumbo packet), we will do it at after
1617                  * skb was created with generic XDP routine.
1618                  */
1619                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1620                 if (IS_ERR(skb)) {
1621                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1622                         return PTR_ERR(skb);
1623                 }
1624                 if (!skb)
1625                         return total_len;
1626         } else {
1627                 if (!zerocopy) {
1628                         copylen = len;
1629                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1630                                 linear = good_linear;
1631                         else
1632                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1633                 }
1634
1635                 if (frags) {
1636                         mutex_lock(&tfile->napi_mutex);
1637                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1638                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1639                          * If zerocopy is enabled, then this layout will be
1640                          * overwritten by zerocopy_sg_from_iter().
1641                          */
1642                         zerocopy = false;
1643                 } else {
1644                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1645                                             noblock);
1646                 }
1647
1648                 if (IS_ERR(skb)) {
1649                         if (PTR_ERR(skb) != -EAGAIN)
1650                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1651                         if (frags)
1652                                 mutex_unlock(&tfile->napi_mutex);
1653                         return PTR_ERR(skb);
1654                 }
1655
1656                 if (zerocopy)
1657                         err = zerocopy_sg_from_iter(skb, from);
1658                 else
1659                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1660
1661                 if (err) {
1662                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1663                         kfree_skb(skb);
1664                         if (frags) {
1665                                 tfile->napi.skb = NULL;
1666                                 mutex_unlock(&tfile->napi_mutex);
1667                         }
1668
1669                         return -EFAULT;
1670                 }
1671         }
1672
1673         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1674                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1675                 kfree_skb(skb);
1676                 if (frags) {
1677                         tfile->napi.skb = NULL;
1678                         mutex_unlock(&tfile->napi_mutex);
1679                 }
1680
1681                 return -EINVAL;
1682         }
1683
1684         switch (tun->flags & TUN_TYPE_MASK) {
1685         case IFF_TUN:
1686                 if (tun->flags & IFF_NO_PI) {
1687                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1688
1689                         switch (ip_version) {
1690                         case 4:
1691                                 pi.proto = htons(ETH_P_IP);
1692                                 break;
1693                         case 6:
1694                                 pi.proto = htons(ETH_P_IPV6);
1695                                 break;
1696                         default:
1697                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1698                                 kfree_skb(skb);
1699                                 return -EINVAL;
1700                         }
1701                 }
1702
1703                 skb_reset_mac_header(skb);
1704                 skb->protocol = pi.proto;
1705                 skb->dev = tun->dev;
1706                 break;
1707         case IFF_TAP:
1708                 if (!frags)
1709                         skb->protocol = eth_type_trans(skb, tun->dev);
1710                 break;
1711         }
1712
1713         /* copy skb_ubuf_info for callback when skb has no error */
1714         if (zerocopy) {
1715                 skb_shinfo(skb)->destructor_arg = msg_control;
1716                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1717                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1718         } else if (msg_control) {
1719                 struct ubuf_info *uarg = msg_control;
1720                 uarg->callback(uarg, false);
1721         }
1722
1723         skb_reset_network_header(skb);
1724         skb_probe_transport_header(skb, 0);
1725
1726         if (skb_xdp) {
1727                 struct bpf_prog *xdp_prog;
1728                 int ret;
1729
1730                 rcu_read_lock();
1731                 xdp_prog = rcu_dereference(tun->xdp_prog);
1732                 if (xdp_prog) {
1733                         ret = do_xdp_generic(xdp_prog, skb);
1734                         if (ret != XDP_PASS) {
1735                                 rcu_read_unlock();
1736                                 return total_len;
1737                         }
1738                 }
1739                 rcu_read_unlock();
1740         }
1741
1742         rxhash = __skb_get_hash_symmetric(skb);
1743
1744         if (frags) {
1745                 /* Exercise flow dissector code path. */
1746                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1747
1748                 if (unlikely(headlen > skb_headlen(skb))) {
1749                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1750                         napi_free_frags(&tfile->napi);
1751                         mutex_unlock(&tfile->napi_mutex);
1752                         WARN_ON(1);
1753                         return -ENOMEM;
1754                 }
1755
1756                 local_bh_disable();
1757                 napi_gro_frags(&tfile->napi);
1758                 local_bh_enable();
1759                 mutex_unlock(&tfile->napi_mutex);
1760         } else if (tfile->napi_enabled) {
1761                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1762                 int queue_len;
1763
1764                 spin_lock_bh(&queue->lock);
1765                 __skb_queue_tail(queue, skb);
1766                 queue_len = skb_queue_len(queue);
1767                 spin_unlock(&queue->lock);
1768
1769                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1770                         napi_schedule(&tfile->napi);
1771
1772                 local_bh_enable();
1773         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1774                 tun_rx_batched(tun, tfile, skb, more);
1775         } else {
1776                 netif_rx_ni(skb);
1777         }
1778
1779         stats = get_cpu_ptr(tun->pcpu_stats);
1780         u64_stats_update_begin(&stats->syncp);
1781         stats->rx_packets++;
1782         stats->rx_bytes += len;
1783         u64_stats_update_end(&stats->syncp);
1784         put_cpu_ptr(stats);
1785
1786         tun_flow_update(tun, rxhash, tfile);
1787         return total_len;
1788 }
1789
1790 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1791 {
1792         struct file *file = iocb->ki_filp;
1793         struct tun_file *tfile = file->private_data;
1794         struct tun_struct *tun = tun_get(tfile);
1795         ssize_t result;
1796
1797         if (!tun)
1798                 return -EBADFD;
1799
1800         result = tun_get_user(tun, tfile, NULL, from,
1801                               file->f_flags & O_NONBLOCK, false);
1802
1803         tun_put(tun);
1804         return result;
1805 }
1806
1807 /* Put packet to the user space buffer */
1808 static ssize_t tun_put_user(struct tun_struct *tun,
1809                             struct tun_file *tfile,
1810                             struct sk_buff *skb,
1811                             struct iov_iter *iter)
1812 {
1813         struct tun_pi pi = { 0, skb->protocol };
1814         struct tun_pcpu_stats *stats;
1815         ssize_t total;
1816         int vlan_offset = 0;
1817         int vlan_hlen = 0;
1818         int vnet_hdr_sz = 0;
1819
1820         if (skb_vlan_tag_present(skb))
1821                 vlan_hlen = VLAN_HLEN;
1822
1823         if (tun->flags & IFF_VNET_HDR)
1824                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1825
1826         total = skb->len + vlan_hlen + vnet_hdr_sz;
1827
1828         if (!(tun->flags & IFF_NO_PI)) {
1829                 if (iov_iter_count(iter) < sizeof(pi))
1830                         return -EINVAL;
1831
1832                 total += sizeof(pi);
1833                 if (iov_iter_count(iter) < total) {
1834                         /* Packet will be striped */
1835                         pi.flags |= TUN_PKT_STRIP;
1836                 }
1837
1838                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1839                         return -EFAULT;
1840         }
1841
1842         if (vnet_hdr_sz) {
1843                 struct virtio_net_hdr gso;
1844
1845                 if (iov_iter_count(iter) < vnet_hdr_sz)
1846                         return -EINVAL;
1847
1848                 if (virtio_net_hdr_from_skb(skb, &gso,
1849                                             tun_is_little_endian(tun), true)) {
1850                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1851                         pr_err("unexpected GSO type: "
1852                                "0x%x, gso_size %d, hdr_len %d\n",
1853                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1854                                tun16_to_cpu(tun, gso.hdr_len));
1855                         print_hex_dump(KERN_ERR, "tun: ",
1856                                        DUMP_PREFIX_NONE,
1857                                        16, 1, skb->head,
1858                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1859                         WARN_ON_ONCE(1);
1860                         return -EINVAL;
1861                 }
1862
1863                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1864                         return -EFAULT;
1865
1866                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1867         }
1868
1869         if (vlan_hlen) {
1870                 int ret;
1871                 struct {
1872                         __be16 h_vlan_proto;
1873                         __be16 h_vlan_TCI;
1874                 } veth;
1875
1876                 veth.h_vlan_proto = skb->vlan_proto;
1877                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1878
1879                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1880
1881                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1882                 if (ret || !iov_iter_count(iter))
1883                         goto done;
1884
1885                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1886                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1887                         goto done;
1888         }
1889
1890         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1891
1892 done:
1893         /* caller is in process context, */
1894         stats = get_cpu_ptr(tun->pcpu_stats);
1895         u64_stats_update_begin(&stats->syncp);
1896         stats->tx_packets++;
1897         stats->tx_bytes += skb->len + vlan_hlen;
1898         u64_stats_update_end(&stats->syncp);
1899         put_cpu_ptr(tun->pcpu_stats);
1900
1901         return total;
1902 }
1903
1904 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1905                                      int *err)
1906 {
1907         DECLARE_WAITQUEUE(wait, current);
1908         struct sk_buff *skb = NULL;
1909         int error = 0;
1910
1911         skb = skb_array_consume(&tfile->tx_array);
1912         if (skb)
1913                 goto out;
1914         if (noblock) {
1915                 error = -EAGAIN;
1916                 goto out;
1917         }
1918
1919         add_wait_queue(&tfile->wq.wait, &wait);
1920         current->state = TASK_INTERRUPTIBLE;
1921
1922         while (1) {
1923                 skb = skb_array_consume(&tfile->tx_array);
1924                 if (skb)
1925                         break;
1926                 if (signal_pending(current)) {
1927                         error = -ERESTARTSYS;
1928                         break;
1929                 }
1930                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1931                         error = -EFAULT;
1932                         break;
1933                 }
1934
1935                 schedule();
1936         }
1937
1938         current->state = TASK_RUNNING;
1939         remove_wait_queue(&tfile->wq.wait, &wait);
1940
1941 out:
1942         *err = error;
1943         return skb;
1944 }
1945
1946 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1947                            struct iov_iter *to,
1948                            int noblock, struct sk_buff *skb)
1949 {
1950         ssize_t ret;
1951         int err;
1952
1953         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1954
1955         if (!iov_iter_count(to))
1956                 return 0;
1957
1958         if (!skb) {
1959                 /* Read frames from ring */
1960                 skb = tun_ring_recv(tfile, noblock, &err);
1961                 if (!skb)
1962                         return err;
1963         }
1964
1965         ret = tun_put_user(tun, tfile, skb, to);
1966         if (unlikely(ret < 0))
1967                 kfree_skb(skb);
1968         else
1969                 consume_skb(skb);
1970
1971         return ret;
1972 }
1973
1974 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1975 {
1976         struct file *file = iocb->ki_filp;
1977         struct tun_file *tfile = file->private_data;
1978         struct tun_struct *tun = tun_get(tfile);
1979         ssize_t len = iov_iter_count(to), ret;
1980
1981         if (!tun)
1982                 return -EBADFD;
1983         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1984         ret = min_t(ssize_t, ret, len);
1985         if (ret > 0)
1986                 iocb->ki_pos = ret;
1987         tun_put(tun);
1988         return ret;
1989 }
1990
1991 static void tun_free_netdev(struct net_device *dev)
1992 {
1993         struct tun_struct *tun = netdev_priv(dev);
1994
1995         BUG_ON(!(list_empty(&tun->disabled)));
1996         free_percpu(tun->pcpu_stats);
1997         tun_flow_uninit(tun);
1998         security_tun_dev_free_security(tun->security);
1999 }
2000
2001 static void tun_setup(struct net_device *dev)
2002 {
2003         struct tun_struct *tun = netdev_priv(dev);
2004
2005         tun->owner = INVALID_UID;
2006         tun->group = INVALID_GID;
2007
2008         dev->ethtool_ops = &tun_ethtool_ops;
2009         dev->needs_free_netdev = true;
2010         dev->priv_destructor = tun_free_netdev;
2011         /* We prefer our own queue length */
2012         dev->tx_queue_len = TUN_READQ_SIZE;
2013 }
2014
2015 /* Trivial set of netlink ops to allow deleting tun or tap
2016  * device with netlink.
2017  */
2018 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2019                         struct netlink_ext_ack *extack)
2020 {
2021         return -EINVAL;
2022 }
2023
2024 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2025         .kind           = DRV_NAME,
2026         .priv_size      = sizeof(struct tun_struct),
2027         .setup          = tun_setup,
2028         .validate       = tun_validate,
2029 };
2030
2031 static void tun_sock_write_space(struct sock *sk)
2032 {
2033         struct tun_file *tfile;
2034         wait_queue_head_t *wqueue;
2035
2036         if (!sock_writeable(sk))
2037                 return;
2038
2039         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2040                 return;
2041
2042         wqueue = sk_sleep(sk);
2043         if (wqueue && waitqueue_active(wqueue))
2044                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
2045                                                 POLLWRNORM | POLLWRBAND);
2046
2047         tfile = container_of(sk, struct tun_file, sk);
2048         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2049 }
2050
2051 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2052 {
2053         int ret;
2054         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2055         struct tun_struct *tun = tun_get(tfile);
2056
2057         if (!tun)
2058                 return -EBADFD;
2059
2060         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2061                            m->msg_flags & MSG_DONTWAIT,
2062                            m->msg_flags & MSG_MORE);
2063         tun_put(tun);
2064         return ret;
2065 }
2066
2067 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2068                        int flags)
2069 {
2070         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2071         struct tun_struct *tun = tun_get(tfile);
2072         int ret;
2073
2074         if (!tun)
2075                 return -EBADFD;
2076
2077         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2078                 ret = -EINVAL;
2079                 goto out;
2080         }
2081         if (flags & MSG_ERRQUEUE) {
2082                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2083                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2084                 goto out;
2085         }
2086         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT,
2087                           m->msg_control);
2088         if (ret > (ssize_t)total_len) {
2089                 m->msg_flags |= MSG_TRUNC;
2090                 ret = flags & MSG_TRUNC ? ret : total_len;
2091         }
2092 out:
2093         tun_put(tun);
2094         return ret;
2095 }
2096
2097 static int tun_peek_len(struct socket *sock)
2098 {
2099         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2100         struct tun_struct *tun;
2101         int ret = 0;
2102
2103         tun = tun_get(tfile);
2104         if (!tun)
2105                 return 0;
2106
2107         ret = skb_array_peek_len(&tfile->tx_array);
2108         tun_put(tun);
2109
2110         return ret;
2111 }
2112
2113 /* Ops structure to mimic raw sockets with tun */
2114 static const struct proto_ops tun_socket_ops = {
2115         .peek_len = tun_peek_len,
2116         .sendmsg = tun_sendmsg,
2117         .recvmsg = tun_recvmsg,
2118 };
2119
2120 static struct proto tun_proto = {
2121         .name           = "tun",
2122         .owner          = THIS_MODULE,
2123         .obj_size       = sizeof(struct tun_file),
2124 };
2125
2126 static int tun_flags(struct tun_struct *tun)
2127 {
2128         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2129 }
2130
2131 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2132                               char *buf)
2133 {
2134         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2135         return sprintf(buf, "0x%x\n", tun_flags(tun));
2136 }
2137
2138 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2139                               char *buf)
2140 {
2141         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2142         return uid_valid(tun->owner)?
2143                 sprintf(buf, "%u\n",
2144                         from_kuid_munged(current_user_ns(), tun->owner)):
2145                 sprintf(buf, "-1\n");
2146 }
2147
2148 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2149                               char *buf)
2150 {
2151         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2152         return gid_valid(tun->group) ?
2153                 sprintf(buf, "%u\n",
2154                         from_kgid_munged(current_user_ns(), tun->group)):
2155                 sprintf(buf, "-1\n");
2156 }
2157
2158 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2159 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2160 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2161
2162 static struct attribute *tun_dev_attrs[] = {
2163         &dev_attr_tun_flags.attr,
2164         &dev_attr_owner.attr,
2165         &dev_attr_group.attr,
2166         NULL
2167 };
2168
2169 static const struct attribute_group tun_attr_group = {
2170         .attrs = tun_dev_attrs
2171 };
2172
2173 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2174 {
2175         struct tun_struct *tun;
2176         struct tun_file *tfile = file->private_data;
2177         struct net_device *dev;
2178         int err;
2179
2180         if (tfile->detached)
2181                 return -EINVAL;
2182
2183         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2184                 if (!capable(CAP_NET_ADMIN))
2185                         return -EPERM;
2186
2187                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2188                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2189                         return -EINVAL;
2190         }
2191
2192         dev = __dev_get_by_name(net, ifr->ifr_name);
2193         if (dev) {
2194                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2195                         return -EBUSY;
2196                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2197                         tun = netdev_priv(dev);
2198                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2199                         tun = netdev_priv(dev);
2200                 else
2201                         return -EINVAL;
2202
2203                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2204                     !!(tun->flags & IFF_MULTI_QUEUE))
2205                         return -EINVAL;
2206
2207                 if (tun_not_capable(tun))
2208                         return -EPERM;
2209                 err = security_tun_dev_open(tun->security);
2210                 if (err < 0)
2211                         return err;
2212
2213                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2214                                  ifr->ifr_flags & IFF_NAPI);
2215                 if (err < 0)
2216                         return err;
2217
2218                 if (tun->flags & IFF_MULTI_QUEUE &&
2219                     (tun->numqueues + tun->numdisabled > 1)) {
2220                         /* One or more queue has already been attached, no need
2221                          * to initialize the device again.
2222                          */
2223                         return 0;
2224                 }
2225         }
2226         else {
2227                 char *name;
2228                 unsigned long flags = 0;
2229                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2230                              MAX_TAP_QUEUES : 1;
2231
2232                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2233                         return -EPERM;
2234                 err = security_tun_dev_create();
2235                 if (err < 0)
2236                         return err;
2237
2238                 /* Set dev type */
2239                 if (ifr->ifr_flags & IFF_TUN) {
2240                         /* TUN device */
2241                         flags |= IFF_TUN;
2242                         name = "tun%d";
2243                 } else if (ifr->ifr_flags & IFF_TAP) {
2244                         /* TAP device */
2245                         flags |= IFF_TAP;
2246                         name = "tap%d";
2247                 } else
2248                         return -EINVAL;
2249
2250                 if (*ifr->ifr_name)
2251                         name = ifr->ifr_name;
2252
2253                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2254                                        NET_NAME_UNKNOWN, tun_setup, queues,
2255                                        queues);
2256
2257                 if (!dev)
2258                         return -ENOMEM;
2259                 err = dev_get_valid_name(net, dev, name);
2260                 if (err < 0)
2261                         goto err_free_dev;
2262
2263                 dev_net_set(dev, net);
2264                 dev->rtnl_link_ops = &tun_link_ops;
2265                 dev->ifindex = tfile->ifindex;
2266                 dev->sysfs_groups[0] = &tun_attr_group;
2267
2268                 tun = netdev_priv(dev);
2269                 tun->dev = dev;
2270                 tun->flags = flags;
2271                 tun->txflt.count = 0;
2272                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2273
2274                 tun->align = NET_SKB_PAD;
2275                 tun->filter_attached = false;
2276                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2277                 tun->rx_batched = 0;
2278
2279                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2280                 if (!tun->pcpu_stats) {
2281                         err = -ENOMEM;
2282                         goto err_free_dev;
2283                 }
2284
2285                 spin_lock_init(&tun->lock);
2286
2287                 err = security_tun_dev_alloc_security(&tun->security);
2288                 if (err < 0)
2289                         goto err_free_stat;
2290
2291                 tun_net_init(dev);
2292                 tun_flow_init(tun);
2293
2294                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2295                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2296                                    NETIF_F_HW_VLAN_STAG_TX;
2297                 dev->features = dev->hw_features | NETIF_F_LLTX;
2298                 dev->vlan_features = dev->features &
2299                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2300                                        NETIF_F_HW_VLAN_STAG_TX);
2301
2302                 INIT_LIST_HEAD(&tun->disabled);
2303                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2304                 if (err < 0)
2305                         goto err_free_flow;
2306
2307                 err = register_netdevice(tun->dev);
2308                 if (err < 0)
2309                         goto err_detach;
2310         }
2311
2312         netif_carrier_on(tun->dev);
2313
2314         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2315
2316         tun->flags = (tun->flags & ~TUN_FEATURES) |
2317                 (ifr->ifr_flags & TUN_FEATURES);
2318
2319         /* Make sure persistent devices do not get stuck in
2320          * xoff state.
2321          */
2322         if (netif_running(tun->dev))
2323                 netif_tx_wake_all_queues(tun->dev);
2324
2325         strcpy(ifr->ifr_name, tun->dev->name);
2326         return 0;
2327
2328 err_detach:
2329         tun_detach_all(dev);
2330         /* register_netdevice() already called tun_free_netdev() */
2331         goto err_free_dev;
2332
2333 err_free_flow:
2334         tun_flow_uninit(tun);
2335         security_tun_dev_free_security(tun->security);
2336 err_free_stat:
2337         free_percpu(tun->pcpu_stats);
2338 err_free_dev:
2339         free_netdev(dev);
2340         return err;
2341 }
2342
2343 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2344                        struct ifreq *ifr)
2345 {
2346         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2347
2348         strcpy(ifr->ifr_name, tun->dev->name);
2349
2350         ifr->ifr_flags = tun_flags(tun);
2351
2352 }
2353
2354 /* This is like a cut-down ethtool ops, except done via tun fd so no
2355  * privs required. */
2356 static int set_offload(struct tun_struct *tun, unsigned long arg)
2357 {
2358         netdev_features_t features = 0;
2359
2360         if (arg & TUN_F_CSUM) {
2361                 features |= NETIF_F_HW_CSUM;
2362                 arg &= ~TUN_F_CSUM;
2363
2364                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2365                         if (arg & TUN_F_TSO_ECN) {
2366                                 features |= NETIF_F_TSO_ECN;
2367                                 arg &= ~TUN_F_TSO_ECN;
2368                         }
2369                         if (arg & TUN_F_TSO4)
2370                                 features |= NETIF_F_TSO;
2371                         if (arg & TUN_F_TSO6)
2372                                 features |= NETIF_F_TSO6;
2373                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2374                 }
2375         }
2376
2377         /* This gives the user a way to test for new features in future by
2378          * trying to set them. */
2379         if (arg)
2380                 return -EINVAL;
2381
2382         tun->set_features = features;
2383         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2384         tun->dev->wanted_features |= features;
2385         netdev_update_features(tun->dev);
2386
2387         return 0;
2388 }
2389
2390 static void tun_detach_filter(struct tun_struct *tun, int n)
2391 {
2392         int i;
2393         struct tun_file *tfile;
2394
2395         for (i = 0; i < n; i++) {
2396                 tfile = rtnl_dereference(tun->tfiles[i]);
2397                 lock_sock(tfile->socket.sk);
2398                 sk_detach_filter(tfile->socket.sk);
2399                 release_sock(tfile->socket.sk);
2400         }
2401
2402         tun->filter_attached = false;
2403 }
2404
2405 static int tun_attach_filter(struct tun_struct *tun)
2406 {
2407         int i, ret = 0;
2408         struct tun_file *tfile;
2409
2410         for (i = 0; i < tun->numqueues; i++) {
2411                 tfile = rtnl_dereference(tun->tfiles[i]);
2412                 lock_sock(tfile->socket.sk);
2413                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2414                 release_sock(tfile->socket.sk);
2415                 if (ret) {
2416                         tun_detach_filter(tun, i);
2417                         return ret;
2418                 }
2419         }
2420
2421         tun->filter_attached = true;
2422         return ret;
2423 }
2424
2425 static void tun_set_sndbuf(struct tun_struct *tun)
2426 {
2427         struct tun_file *tfile;
2428         int i;
2429
2430         for (i = 0; i < tun->numqueues; i++) {
2431                 tfile = rtnl_dereference(tun->tfiles[i]);
2432                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2433         }
2434 }
2435
2436 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2437 {
2438         struct tun_file *tfile = file->private_data;
2439         struct tun_struct *tun;
2440         int ret = 0;
2441
2442         rtnl_lock();
2443
2444         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2445                 tun = tfile->detached;
2446                 if (!tun) {
2447                         ret = -EINVAL;
2448                         goto unlock;
2449                 }
2450                 ret = security_tun_dev_attach_queue(tun->security);
2451                 if (ret < 0)
2452                         goto unlock;
2453                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2454         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2455                 tun = rtnl_dereference(tfile->tun);
2456                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2457                         ret = -EINVAL;
2458                 else
2459                         __tun_detach(tfile, false);
2460         } else
2461                 ret = -EINVAL;
2462
2463 unlock:
2464         rtnl_unlock();
2465         return ret;
2466 }
2467
2468 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2469                             unsigned long arg, int ifreq_len)
2470 {
2471         struct tun_file *tfile = file->private_data;
2472         struct tun_struct *tun;
2473         void __user* argp = (void __user*)arg;
2474         struct ifreq ifr;
2475         kuid_t owner;
2476         kgid_t group;
2477         int sndbuf;
2478         int vnet_hdr_sz;
2479         unsigned int ifindex;
2480         int le;
2481         int ret;
2482
2483         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2484                 if (copy_from_user(&ifr, argp, ifreq_len))
2485                         return -EFAULT;
2486         } else {
2487                 memset(&ifr, 0, sizeof(ifr));
2488         }
2489         if (cmd == TUNGETFEATURES) {
2490                 /* Currently this just means: "what IFF flags are valid?".
2491                  * This is needed because we never checked for invalid flags on
2492                  * TUNSETIFF.
2493                  */
2494                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2495                                 (unsigned int __user*)argp);
2496         } else if (cmd == TUNSETQUEUE)
2497                 return tun_set_queue(file, &ifr);
2498
2499         ret = 0;
2500         rtnl_lock();
2501
2502         tun = tun_get(tfile);
2503         if (cmd == TUNSETIFF) {
2504                 ret = -EEXIST;
2505                 if (tun)
2506                         goto unlock;
2507
2508                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2509
2510                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2511
2512                 if (ret)
2513                         goto unlock;
2514
2515                 if (copy_to_user(argp, &ifr, ifreq_len))
2516                         ret = -EFAULT;
2517                 goto unlock;
2518         }
2519         if (cmd == TUNSETIFINDEX) {
2520                 ret = -EPERM;
2521                 if (tun)
2522                         goto unlock;
2523
2524                 ret = -EFAULT;
2525                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2526                         goto unlock;
2527
2528                 ret = 0;
2529                 tfile->ifindex = ifindex;
2530                 goto unlock;
2531         }
2532
2533         ret = -EBADFD;
2534         if (!tun)
2535                 goto unlock;
2536
2537         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2538
2539         ret = 0;
2540         switch (cmd) {
2541         case TUNGETIFF:
2542                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2543
2544                 if (tfile->detached)
2545                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2546                 if (!tfile->socket.sk->sk_filter)
2547                         ifr.ifr_flags |= IFF_NOFILTER;
2548
2549                 if (copy_to_user(argp, &ifr, ifreq_len))
2550                         ret = -EFAULT;
2551                 break;
2552
2553         case TUNSETNOCSUM:
2554                 /* Disable/Enable checksum */
2555
2556                 /* [unimplemented] */
2557                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2558                           arg ? "disabled" : "enabled");
2559                 break;
2560
2561         case TUNSETPERSIST:
2562                 /* Disable/Enable persist mode. Keep an extra reference to the
2563                  * module to prevent the module being unprobed.
2564                  */
2565                 if (arg && !(tun->flags & IFF_PERSIST)) {
2566                         tun->flags |= IFF_PERSIST;
2567                         __module_get(THIS_MODULE);
2568                 }
2569                 if (!arg && (tun->flags & IFF_PERSIST)) {
2570                         tun->flags &= ~IFF_PERSIST;
2571                         module_put(THIS_MODULE);
2572                 }
2573
2574                 tun_debug(KERN_INFO, tun, "persist %s\n",
2575                           arg ? "enabled" : "disabled");
2576                 break;
2577
2578         case TUNSETOWNER:
2579                 /* Set owner of the device */
2580                 owner = make_kuid(current_user_ns(), arg);
2581                 if (!uid_valid(owner)) {
2582                         ret = -EINVAL;
2583                         break;
2584                 }
2585                 tun->owner = owner;
2586                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2587                           from_kuid(&init_user_ns, tun->owner));
2588                 break;
2589
2590         case TUNSETGROUP:
2591                 /* Set group of the device */
2592                 group = make_kgid(current_user_ns(), arg);
2593                 if (!gid_valid(group)) {
2594                         ret = -EINVAL;
2595                         break;
2596                 }
2597                 tun->group = group;
2598                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2599                           from_kgid(&init_user_ns, tun->group));
2600                 break;
2601
2602         case TUNSETLINK:
2603                 /* Only allow setting the type when the interface is down */
2604                 if (tun->dev->flags & IFF_UP) {
2605                         tun_debug(KERN_INFO, tun,
2606                                   "Linktype set failed because interface is up\n");
2607                         ret = -EBUSY;
2608                 } else {
2609                         tun->dev->type = (int) arg;
2610                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2611                                   tun->dev->type);
2612                         ret = 0;
2613                 }
2614                 break;
2615
2616 #ifdef TUN_DEBUG
2617         case TUNSETDEBUG:
2618                 tun->debug = arg;
2619                 break;
2620 #endif
2621         case TUNSETOFFLOAD:
2622                 ret = set_offload(tun, arg);
2623                 break;
2624
2625         case TUNSETTXFILTER:
2626                 /* Can be set only for TAPs */
2627                 ret = -EINVAL;
2628                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2629                         break;
2630                 ret = update_filter(&tun->txflt, (void __user *)arg);
2631                 break;
2632
2633         case SIOCGIFHWADDR:
2634                 /* Get hw address */
2635                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2636                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2637                 if (copy_to_user(argp, &ifr, ifreq_len))
2638                         ret = -EFAULT;
2639                 break;
2640
2641         case SIOCSIFHWADDR:
2642                 /* Set hw address */
2643                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2644                           ifr.ifr_hwaddr.sa_data);
2645
2646                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2647                 break;
2648
2649         case TUNGETSNDBUF:
2650                 sndbuf = tfile->socket.sk->sk_sndbuf;
2651                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2652                         ret = -EFAULT;
2653                 break;
2654
2655         case TUNSETSNDBUF:
2656                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2657                         ret = -EFAULT;
2658                         break;
2659                 }
2660                 if (sndbuf <= 0) {
2661                         ret = -EINVAL;
2662                         break;
2663                 }
2664
2665                 tun->sndbuf = sndbuf;
2666                 tun_set_sndbuf(tun);
2667                 break;
2668
2669         case TUNGETVNETHDRSZ:
2670                 vnet_hdr_sz = tun->vnet_hdr_sz;
2671                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2672                         ret = -EFAULT;
2673                 break;
2674
2675         case TUNSETVNETHDRSZ:
2676                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2677                         ret = -EFAULT;
2678                         break;
2679                 }
2680                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2681                         ret = -EINVAL;
2682                         break;
2683                 }
2684
2685                 tun->vnet_hdr_sz = vnet_hdr_sz;
2686                 break;
2687
2688         case TUNGETVNETLE:
2689                 le = !!(tun->flags & TUN_VNET_LE);
2690                 if (put_user(le, (int __user *)argp))
2691                         ret = -EFAULT;
2692                 break;
2693
2694         case TUNSETVNETLE:
2695                 if (get_user(le, (int __user *)argp)) {
2696                         ret = -EFAULT;
2697                         break;
2698                 }
2699                 if (le)
2700                         tun->flags |= TUN_VNET_LE;
2701                 else
2702                         tun->flags &= ~TUN_VNET_LE;
2703                 break;
2704
2705         case TUNGETVNETBE:
2706                 ret = tun_get_vnet_be(tun, argp);
2707                 break;
2708
2709         case TUNSETVNETBE:
2710                 ret = tun_set_vnet_be(tun, argp);
2711                 break;
2712
2713         case TUNATTACHFILTER:
2714                 /* Can be set only for TAPs */
2715                 ret = -EINVAL;
2716                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2717                         break;
2718                 ret = -EFAULT;
2719                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2720                         break;
2721
2722                 ret = tun_attach_filter(tun);
2723                 break;
2724
2725         case TUNDETACHFILTER:
2726                 /* Can be set only for TAPs */
2727                 ret = -EINVAL;
2728                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2729                         break;
2730                 ret = 0;
2731                 tun_detach_filter(tun, tun->numqueues);
2732                 break;
2733
2734         case TUNGETFILTER:
2735                 ret = -EINVAL;
2736                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2737                         break;
2738                 ret = -EFAULT;
2739                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2740                         break;
2741                 ret = 0;
2742                 break;
2743
2744         default:
2745                 ret = -EINVAL;
2746                 break;
2747         }
2748
2749 unlock:
2750         rtnl_unlock();
2751         if (tun)
2752                 tun_put(tun);
2753         return ret;
2754 }
2755
2756 static long tun_chr_ioctl(struct file *file,
2757                           unsigned int cmd, unsigned long arg)
2758 {
2759         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2760 }
2761
2762 #ifdef CONFIG_COMPAT
2763 static long tun_chr_compat_ioctl(struct file *file,
2764                          unsigned int cmd, unsigned long arg)
2765 {
2766         switch (cmd) {
2767         case TUNSETIFF:
2768         case TUNGETIFF:
2769         case TUNSETTXFILTER:
2770         case TUNGETSNDBUF:
2771         case TUNSETSNDBUF:
2772         case SIOCGIFHWADDR:
2773         case SIOCSIFHWADDR:
2774                 arg = (unsigned long)compat_ptr(arg);
2775                 break;
2776         default:
2777                 arg = (compat_ulong_t)arg;
2778                 break;
2779         }
2780
2781         /*
2782          * compat_ifreq is shorter than ifreq, so we must not access beyond
2783          * the end of that structure. All fields that are used in this
2784          * driver are compatible though, we don't need to convert the
2785          * contents.
2786          */
2787         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2788 }
2789 #endif /* CONFIG_COMPAT */
2790
2791 static int tun_chr_fasync(int fd, struct file *file, int on)
2792 {
2793         struct tun_file *tfile = file->private_data;
2794         int ret;
2795
2796         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2797                 goto out;
2798
2799         if (on) {
2800                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2801                 tfile->flags |= TUN_FASYNC;
2802         } else
2803                 tfile->flags &= ~TUN_FASYNC;
2804         ret = 0;
2805 out:
2806         return ret;
2807 }
2808
2809 static int tun_chr_open(struct inode *inode, struct file * file)
2810 {
2811         struct net *net = current->nsproxy->net_ns;
2812         struct tun_file *tfile;
2813
2814         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2815
2816         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2817                                             &tun_proto, 0);
2818         if (!tfile)
2819                 return -ENOMEM;
2820         RCU_INIT_POINTER(tfile->tun, NULL);
2821         tfile->flags = 0;
2822         tfile->ifindex = 0;
2823
2824         init_waitqueue_head(&tfile->wq.wait);
2825         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2826
2827         tfile->socket.file = file;
2828         tfile->socket.ops = &tun_socket_ops;
2829
2830         sock_init_data(&tfile->socket, &tfile->sk);
2831
2832         tfile->sk.sk_write_space = tun_sock_write_space;
2833         tfile->sk.sk_sndbuf = INT_MAX;
2834
2835         file->private_data = tfile;
2836         INIT_LIST_HEAD(&tfile->next);
2837
2838         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2839
2840         return 0;
2841 }
2842
2843 static int tun_chr_close(struct inode *inode, struct file *file)
2844 {
2845         struct tun_file *tfile = file->private_data;
2846
2847         tun_detach(tfile, true);
2848
2849         return 0;
2850 }
2851
2852 #ifdef CONFIG_PROC_FS
2853 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
2854 {
2855         struct tun_file *tfile = file->private_data;
2856         struct tun_struct *tun;
2857         struct ifreq ifr;
2858
2859         memset(&ifr, 0, sizeof(ifr));
2860
2861         rtnl_lock();
2862         tun = tun_get(tfile);
2863         if (tun)
2864                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2865         rtnl_unlock();
2866
2867         if (tun)
2868                 tun_put(tun);
2869
2870         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2871 }
2872 #endif
2873
2874 static const struct file_operations tun_fops = {
2875         .owner  = THIS_MODULE,
2876         .llseek = no_llseek,
2877         .read_iter  = tun_chr_read_iter,
2878         .write_iter = tun_chr_write_iter,
2879         .poll   = tun_chr_poll,
2880         .unlocked_ioctl = tun_chr_ioctl,
2881 #ifdef CONFIG_COMPAT
2882         .compat_ioctl = tun_chr_compat_ioctl,
2883 #endif
2884         .open   = tun_chr_open,
2885         .release = tun_chr_close,
2886         .fasync = tun_chr_fasync,
2887 #ifdef CONFIG_PROC_FS
2888         .show_fdinfo = tun_chr_show_fdinfo,
2889 #endif
2890 };
2891
2892 static struct miscdevice tun_miscdev = {
2893         .minor = TUN_MINOR,
2894         .name = "tun",
2895         .nodename = "net/tun",
2896         .fops = &tun_fops,
2897 };
2898
2899 /* ethtool interface */
2900
2901 static int tun_get_link_ksettings(struct net_device *dev,
2902                                   struct ethtool_link_ksettings *cmd)
2903 {
2904         ethtool_link_ksettings_zero_link_mode(cmd, supported);
2905         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2906         cmd->base.speed         = SPEED_10;
2907         cmd->base.duplex        = DUPLEX_FULL;
2908         cmd->base.port          = PORT_TP;
2909         cmd->base.phy_address   = 0;
2910         cmd->base.autoneg       = AUTONEG_DISABLE;
2911         return 0;
2912 }
2913
2914 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2915 {
2916         struct tun_struct *tun = netdev_priv(dev);
2917
2918         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2919         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2920
2921         switch (tun->flags & TUN_TYPE_MASK) {
2922         case IFF_TUN:
2923                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2924                 break;
2925         case IFF_TAP:
2926                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2927                 break;
2928         }
2929 }
2930
2931 static u32 tun_get_msglevel(struct net_device *dev)
2932 {
2933 #ifdef TUN_DEBUG
2934         struct tun_struct *tun = netdev_priv(dev);
2935         return tun->debug;
2936 #else
2937         return -EOPNOTSUPP;
2938 #endif
2939 }
2940
2941 static void tun_set_msglevel(struct net_device *dev, u32 value)
2942 {
2943 #ifdef TUN_DEBUG
2944         struct tun_struct *tun = netdev_priv(dev);
2945         tun->debug = value;
2946 #endif
2947 }
2948
2949 static int tun_get_coalesce(struct net_device *dev,
2950                             struct ethtool_coalesce *ec)
2951 {
2952         struct tun_struct *tun = netdev_priv(dev);
2953
2954         ec->rx_max_coalesced_frames = tun->rx_batched;
2955
2956         return 0;
2957 }
2958
2959 static int tun_set_coalesce(struct net_device *dev,
2960                             struct ethtool_coalesce *ec)
2961 {
2962         struct tun_struct *tun = netdev_priv(dev);
2963
2964         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2965                 tun->rx_batched = NAPI_POLL_WEIGHT;
2966         else
2967                 tun->rx_batched = ec->rx_max_coalesced_frames;
2968
2969         return 0;
2970 }
2971
2972 static const struct ethtool_ops tun_ethtool_ops = {
2973         .get_drvinfo    = tun_get_drvinfo,
2974         .get_msglevel   = tun_get_msglevel,
2975         .set_msglevel   = tun_set_msglevel,
2976         .get_link       = ethtool_op_get_link,
2977         .get_ts_info    = ethtool_op_get_ts_info,
2978         .get_coalesce   = tun_get_coalesce,
2979         .set_coalesce   = tun_set_coalesce,
2980         .get_link_ksettings = tun_get_link_ksettings,
2981 };
2982
2983 static int tun_queue_resize(struct tun_struct *tun)
2984 {
2985         struct net_device *dev = tun->dev;
2986         struct tun_file *tfile;
2987         struct skb_array **arrays;
2988         int n = tun->numqueues + tun->numdisabled;
2989         int ret, i;
2990
2991         arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2992         if (!arrays)
2993                 return -ENOMEM;
2994
2995         for (i = 0; i < tun->numqueues; i++) {
2996                 tfile = rtnl_dereference(tun->tfiles[i]);
2997                 arrays[i] = &tfile->tx_array;
2998         }
2999         list_for_each_entry(tfile, &tun->disabled, next)
3000                 arrays[i++] = &tfile->tx_array;
3001
3002         ret = skb_array_resize_multiple(arrays, n,
3003                                         dev->tx_queue_len, GFP_KERNEL);
3004
3005         kfree(arrays);
3006         return ret;
3007 }
3008
3009 static int tun_device_event(struct notifier_block *unused,
3010                             unsigned long event, void *ptr)
3011 {
3012         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3013         struct tun_struct *tun = netdev_priv(dev);
3014
3015         if (dev->rtnl_link_ops != &tun_link_ops)
3016                 return NOTIFY_DONE;
3017
3018         switch (event) {
3019         case NETDEV_CHANGE_TX_QUEUE_LEN:
3020                 if (tun_queue_resize(tun))
3021                         return NOTIFY_BAD;
3022                 break;
3023         default:
3024                 break;
3025         }
3026
3027         return NOTIFY_DONE;
3028 }
3029
3030 static struct notifier_block tun_notifier_block __read_mostly = {
3031         .notifier_call  = tun_device_event,
3032 };
3033
3034 static int __init tun_init(void)
3035 {
3036         int ret = 0;
3037
3038         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3039
3040         ret = rtnl_link_register(&tun_link_ops);
3041         if (ret) {
3042                 pr_err("Can't register link_ops\n");
3043                 goto err_linkops;
3044         }
3045
3046         ret = misc_register(&tun_miscdev);
3047         if (ret) {
3048                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3049                 goto err_misc;
3050         }
3051
3052         ret = register_netdevice_notifier(&tun_notifier_block);
3053         if (ret) {
3054                 pr_err("Can't register netdevice notifier\n");
3055                 goto err_notifier;
3056         }
3057
3058         return  0;
3059
3060 err_notifier:
3061         misc_deregister(&tun_miscdev);
3062 err_misc:
3063         rtnl_link_unregister(&tun_link_ops);
3064 err_linkops:
3065         return ret;
3066 }
3067
3068 static void tun_cleanup(void)
3069 {
3070         misc_deregister(&tun_miscdev);
3071         rtnl_link_unregister(&tun_link_ops);
3072         unregister_netdevice_notifier(&tun_notifier_block);
3073 }
3074
3075 /* Get an underlying socket object from tun file.  Returns error unless file is
3076  * attached to a device.  The returned object works like a packet socket, it
3077  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3078  * holding a reference to the file for as long as the socket is in use. */
3079 struct socket *tun_get_socket(struct file *file)
3080 {
3081         struct tun_file *tfile;
3082         if (file->f_op != &tun_fops)
3083                 return ERR_PTR(-EINVAL);
3084         tfile = file->private_data;
3085         if (!tfile)
3086                 return ERR_PTR(-EBADFD);
3087         return &tfile->socket;
3088 }
3089 EXPORT_SYMBOL_GPL(tun_get_socket);
3090
3091 struct skb_array *tun_get_skb_array(struct file *file)
3092 {
3093         struct tun_file *tfile;
3094
3095         if (file->f_op != &tun_fops)
3096                 return ERR_PTR(-EINVAL);
3097         tfile = file->private_data;
3098         if (!tfile)
3099                 return ERR_PTR(-EBADFD);
3100         return &tfile->tx_array;
3101 }
3102 EXPORT_SYMBOL_GPL(tun_get_skb_array);
3103
3104 module_init(tun_init);
3105 module_exit(tun_cleanup);
3106 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3107 MODULE_AUTHOR(DRV_COPYRIGHT);
3108 MODULE_LICENSE("GPL");
3109 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3110 MODULE_ALIAS("devname:net/tun");