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_cleanup_tx_array(struct tun_file *tfile)
 615 {
 616         if (tfile->tx_array.ring.queue) {
 617                 skb_array_cleanup(&tfile->tx_array);
 618                 memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
 619         }
 620 }
 621
 622 static void __tun_detach(struct tun_file *tfile, bool clean)
 623 {
 624         struct tun_file *ntfile;
 625         struct tun_struct *tun;
 626
 627         tun = rtnl_dereference(tfile->tun);
 628
 629         if (tun && clean) {
 630                 tun_napi_disable(tun, tfile);
 631                 tun_napi_del(tun, tfile);
 632         }
 633
 634         if (tun && !tfile->detached) {
 635                 u16 index = tfile->queue_index;
 636                 BUG_ON(index >= tun->numqueues);
 637
 638                 rcu_assign_pointer(tun->tfiles[index],
 639                                    tun->tfiles[tun->numqueues - 1]);
 640                 ntfile = rtnl_dereference(tun->tfiles[index]);
 641                 ntfile->queue_index = index;
 642
 643                 --tun->numqueues;
 644                 if (clean) {
 645                         RCU_INIT_POINTER(tfile->tun, NULL);
 646                         sock_put(&tfile->sk);
 647                 } else
 648                         tun_disable_queue(tun, tfile);
 649
 650                 synchronize_net();
 651                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 652                 /* Drop read queue */
 653                 tun_queue_purge(tfile);
 654                 tun_set_real_num_queues(tun);
 655         } else if (tfile->detached && clean) {
 656                 tun = tun_enable_queue(tfile);
 657                 sock_put(&tfile->sk);
 658         }
 659
 660         if (clean) {
 661                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 662                         netif_carrier_off(tun->dev);
 663
 664                         if (!(tun->flags & IFF_PERSIST) &&
 665                             tun->dev->reg_state == NETREG_REGISTERED)
 666                                 unregister_netdevice(tun->dev);
 667                 }
 668                 tun_cleanup_tx_array(tfile);
 669                 sock_put(&tfile->sk);
 670         }
 671 }
 672
 673 static void tun_detach(struct tun_file *tfile, bool clean)
 674 {
 675         rtnl_lock();
 676         __tun_detach(tfile, clean);
 677         rtnl_unlock();
 678 }
 679
 680 static void tun_detach_all(struct net_device *dev)
 681 {
 682         struct tun_struct *tun = netdev_priv(dev);
 683         struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
 684         struct tun_file *tfile, *tmp;
 685         int i, n = tun->numqueues;
 686
 687         for (i = 0; i < n; i++) {
 688                 tfile = rtnl_dereference(tun->tfiles[i]);
 689                 BUG_ON(!tfile);
 690                 tun_napi_disable(tun, tfile);
 691                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 692                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 693                 RCU_INIT_POINTER(tfile->tun, NULL);
 694                 --tun->numqueues;
 695         }
 696         list_for_each_entry(tfile, &tun->disabled, next) {
 697                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 698                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 699                 RCU_INIT_POINTER(tfile->tun, NULL);
 700         }
 701         BUG_ON(tun->numqueues != 0);
 702
 703         synchronize_net();
 704         for (i = 0; i < n; i++) {
 705                 tfile = rtnl_dereference(tun->tfiles[i]);
 706                 tun_napi_del(tun, tfile);
 707                 /* Drop read queue */
 708                 tun_queue_purge(tfile);
 709                 sock_put(&tfile->sk);
 710                 tun_cleanup_tx_array(tfile);
 711         }
 712         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 713                 tun_enable_queue(tfile);
 714                 tun_queue_purge(tfile);
 715                 sock_put(&tfile->sk);
 716                 tun_cleanup_tx_array(tfile);
 717         }
 718         BUG_ON(tun->numdisabled != 0);
 719
 720         if (xdp_prog)
 721                 bpf_prog_put(xdp_prog);
 722
 723         if (tun->flags & IFF_PERSIST)
 724                 module_put(THIS_MODULE);
 725 }
 726
 727 static int tun_attach(struct tun_struct *tun, struct file *file,
 728                       bool skip_filter, bool napi)
 729 {
 730         struct tun_file *tfile = file->private_data;
 731         struct net_device *dev = tun->dev;
 732         int err;
 733
 734         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 735         if (err < 0)
 736                 goto out;
 737
 738         err = -EINVAL;
 739         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 740                 goto out;
 741
 742         err = -EBUSY;
 743         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 744                 goto out;
 745
 746         err = -E2BIG;
 747         if (!tfile->detached &&
 748             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 749                 goto out;
 750
 751         err = 0;
 752
 753         /* Re-attach the filter to persist device */
 754         if (!skip_filter && (tun->filter_attached == true)) {
 755                 lock_sock(tfile->socket.sk);
 756                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 757                 release_sock(tfile->socket.sk);
 758                 if (!err)
 759                         goto out;
 760         }
 761
 762         if (!tfile->detached &&
 763             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 764                 err = -ENOMEM;
 765                 goto out;
 766         }
 767
 768         tfile->queue_index = tun->numqueues;
 769         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 770         rcu_assign_pointer(tfile->tun, tun);
 771         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 772         tun->numqueues++;
 773
 774         if (tfile->detached) {
 775                 tun_enable_queue(tfile);
 776         } else {
 777                 sock_hold(&tfile->sk);
 778                 tun_napi_init(tun, tfile, napi);
 779         }
 780
 781         tun_set_real_num_queues(tun);
 782
 783         /* device is allowed to go away first, so no need to hold extra
 784          * refcnt.
 785          */
 786
 787 out:
 788         return err;
 789 }
 790
 791 static struct tun_struct *tun_get(struct tun_file *tfile)
 792 {
 793         struct tun_struct *tun;
 794
 795         rcu_read_lock();
 796         tun = rcu_dereference(tfile->tun);
 797         if (tun)
 798                 dev_hold(tun->dev);
 799         rcu_read_unlock();
 800
 801         return tun;
 802 }
 803
 804 static void tun_put(struct tun_struct *tun)
 805 {
 806         dev_put(tun->dev);
 807 }
 808
 809 /* TAP filtering */
 810 static void addr_hash_set(u32 *mask, const u8 *addr)
 811 {
 812         int n = ether_crc(ETH_ALEN, addr) >> 26;
 813         mask[n >> 5] |= (1 << (n & 31));
 814 }
 815
 816 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 817 {
 818         int n = ether_crc(ETH_ALEN, addr) >> 26;
 819         return mask[n >> 5] & (1 << (n & 31));
 820 }
 821
 822 static int update_filter(struct tap_filter *filter, void __user *arg)
 823 {
 824         struct { u8 u[ETH_ALEN]; } *addr;
 825         struct tun_filter uf;
 826         int err, alen, n, nexact;
 827
 828         if (copy_from_user(&uf, arg, sizeof(uf)))
 829                 return -EFAULT;
 830
 831         if (!uf.count) {
 832                 /* Disabled */
 833                 filter->count = 0;
 834                 return 0;
 835         }
 836
 837         alen = ETH_ALEN * uf.count;
 838         addr = memdup_user(arg + sizeof(uf), alen);
 839         if (IS_ERR(addr))
 840                 return PTR_ERR(addr);
 841
 842         /* The filter is updated without holding any locks. Which is
 843          * perfectly safe. We disable it first and in the worst
 844          * case we'll accept a few undesired packets. */
 845         filter->count = 0;
 846         wmb();
 847
 848         /* Use first set of addresses as an exact filter */
 849         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 850                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 851
 852         nexact = n;
 853
 854         /* Remaining multicast addresses are hashed,
 855          * unicast will leave the filter disabled. */
 856         memset(filter->mask, 0, sizeof(filter->mask));
 857         for (; n < uf.count; n++) {
 858                 if (!is_multicast_ether_addr(addr[n].u)) {
 859                         err = 0; /* no filter */
 860                         goto free_addr;
 861                 }
 862                 addr_hash_set(filter->mask, addr[n].u);
 863         }
 864
 865         /* For ALLMULTI just set the mask to all ones.
 866          * This overrides the mask populated above. */
 867         if ((uf.flags & TUN_FLT_ALLMULTI))
 868                 memset(filter->mask, ~0, sizeof(filter->mask));
 869
 870         /* Now enable the filter */
 871         wmb();
 872         filter->count = nexact;
 873
 874         /* Return the number of exact filters */
 875         err = nexact;
 876 free_addr:
 877         kfree(addr);
 878         return err;
 879 }
 880
 881 /* Returns: 0 - drop, !=0 - accept */
 882 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 883 {
 884         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 885          * at this point. */
 886         struct ethhdr *eh = (struct ethhdr *) skb->data;
 887         int i;
 888
 889         /* Exact match */
 890         for (i = 0; i < filter->count; i++)
 891                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 892                         return 1;
 893
 894         /* Inexact match (multicast only) */
 895         if (is_multicast_ether_addr(eh->h_dest))
 896                 return addr_hash_test(filter->mask, eh->h_dest);
 897
 898         return 0;
 899 }
 900
 901 /*
 902  * Checks whether the packet is accepted or not.
 903  * Returns: 0 - drop, !=0 - accept
 904  */
 905 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 906 {
 907         if (!filter->count)
 908                 return 1;
 909
 910         return run_filter(filter, skb);
 911 }
 912
 913 /* Network device part of the driver */
 914
 915 static const struct ethtool_ops tun_ethtool_ops;
 916
 917 /* Net device detach from fd. */
 918 static void tun_net_uninit(struct net_device *dev)
 919 {
 920         tun_detach_all(dev);
 921 }
 922
 923 /* Net device open. */
 924 static int tun_net_open(struct net_device *dev)
 925 {
 926         struct tun_struct *tun = netdev_priv(dev);
 927         int i;
 928
 929         netif_tx_start_all_queues(dev);
 930
 931         for (i = 0; i < tun->numqueues; i++) {
 932                 struct tun_file *tfile;
 933
 934                 tfile = rtnl_dereference(tun->tfiles[i]);
 935                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
 936         }
 937
 938         return 0;
 939 }
 940
 941 /* Net device close. */
 942 static int tun_net_close(struct net_device *dev)
 943 {
 944         netif_tx_stop_all_queues(dev);
 945         return 0;
 946 }
 947
 948 /* Net device start xmit */
 949 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 950 {
 951         struct tun_struct *tun = netdev_priv(dev);
 952         int txq = skb->queue_mapping;
 953         struct tun_file *tfile;
 954         u32 numqueues = 0;
 955
 956         rcu_read_lock();
 957         tfile = rcu_dereference(tun->tfiles[txq]);
 958         numqueues = READ_ONCE(tun->numqueues);
 959
 960         /* Drop packet if interface is not attached */
 961         if (txq >= numqueues)
 962                 goto drop;
 963
 964 #ifdef CONFIG_RPS
 965         if (numqueues == 1 && static_key_false(&rps_needed)) {
 966                 /* Select queue was not called for the skbuff, so we extract the
 967                  * RPS hash and save it into the flow_table here.
 968                  */
 969                 __u32 rxhash;
 970
 971                 rxhash = __skb_get_hash_symmetric(skb);
 972                 if (rxhash) {
 973                         struct tun_flow_entry *e;
 974                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 975                                         rxhash);
 976                         if (e)
 977                                 tun_flow_save_rps_rxhash(e, rxhash);
 978                 }
 979         }
 980 #endif
 981
 982         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 983
 984         BUG_ON(!tfile);
 985
 986         /* Drop if the filter does not like it.
 987          * This is a noop if the filter is disabled.
 988          * Filter can be enabled only for the TAP devices. */
 989         if (!check_filter(&tun->txflt, skb))
 990                 goto drop;
 991
 992         if (tfile->socket.sk->sk_filter &&
 993             sk_filter(tfile->socket.sk, skb))
 994                 goto drop;
 995
 996         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
 997                 goto drop;
 998
 999         skb_tx_timestamp(skb);
1000
1001         /* Orphan the skb - required as we might hang on to it
1002          * for indefinite time.
1003          */
1004         skb_orphan(skb);
1005
1006         nf_reset(skb);
1007
1008         if (skb_array_produce(&tfile->tx_array, skb))
1009                 goto drop;
1010
1011         /* Notify and wake up reader process */
1012         if (tfile->flags & TUN_FASYNC)
1013                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1014         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1015
1016         rcu_read_unlock();
1017         return NETDEV_TX_OK;
1018
1019 drop:
1020         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1021         skb_tx_error(skb);
1022         kfree_skb(skb);
1023         rcu_read_unlock();
1024         return NET_XMIT_DROP;
1025 }
1026
1027 static void tun_net_mclist(struct net_device *dev)
1028 {
1029         /*
1030          * This callback is supposed to deal with mc filter in
1031          * _rx_ path and has nothing to do with the _tx_ path.
1032          * In rx path we always accept everything userspace gives us.
1033          */
1034 }
1035
1036 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1037         netdev_features_t features)
1038 {
1039         struct tun_struct *tun = netdev_priv(dev);
1040
1041         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1042 }
1043 #ifdef CONFIG_NET_POLL_CONTROLLER
1044 static void tun_poll_controller(struct net_device *dev)
1045 {
1046         /*
1047          * Tun only receives frames when:
1048          * 1) the char device endpoint gets data from user space
1049          * 2) the tun socket gets a sendmsg call from user space
1050          * If NAPI is not enabled, since both of those are synchronous
1051          * operations, we are guaranteed never to have pending data when we poll
1052          * for it so there is nothing to do here but return.
1053          * We need this though so netpoll recognizes us as an interface that
1054          * supports polling, which enables bridge devices in virt setups to
1055          * still use netconsole
1056          * If NAPI is enabled, however, we need to schedule polling for all
1057          * queues unless we are using napi_gro_frags(), which we call in
1058          * process context and not in NAPI context.
1059          */
1060         struct tun_struct *tun = netdev_priv(dev);
1061
1062         if (tun->flags & IFF_NAPI) {
1063                 struct tun_file *tfile;
1064                 int i;
1065
1066                 if (tun_napi_frags_enabled(tun))
1067                         return;
1068
1069                 rcu_read_lock();
1070                 for (i = 0; i < tun->numqueues; i++) {
1071                         tfile = rcu_dereference(tun->tfiles[i]);
1072                         if (tfile->napi_enabled)
1073                                 napi_schedule(&tfile->napi);
1074                 }
1075                 rcu_read_unlock();
1076         }
1077         return;
1078 }
1079 #endif
1080
1081 static void tun_set_headroom(struct net_device *dev, int new_hr)
1082 {
1083         struct tun_struct *tun = netdev_priv(dev);
1084
1085         if (new_hr < NET_SKB_PAD)
1086                 new_hr = NET_SKB_PAD;
1087
1088         tun->align = new_hr;
1089 }
1090
1091 static void
1092 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1093 {
1094         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1095         struct tun_struct *tun = netdev_priv(dev);
1096         struct tun_pcpu_stats *p;
1097         int i;
1098
1099         for_each_possible_cpu(i) {
1100                 u64 rxpackets, rxbytes, txpackets, txbytes;
1101                 unsigned int start;
1102
1103                 p = per_cpu_ptr(tun->pcpu_stats, i);
1104                 do {
1105                         start = u64_stats_fetch_begin(&p->syncp);
1106                         rxpackets       = p->rx_packets;
1107                         rxbytes         = p->rx_bytes;
1108                         txpackets       = p->tx_packets;
1109                         txbytes         = p->tx_bytes;
1110                 } while (u64_stats_fetch_retry(&p->syncp, start));
1111
1112                 stats->rx_packets       += rxpackets;
1113                 stats->rx_bytes         += rxbytes;
1114                 stats->tx_packets       += txpackets;
1115                 stats->tx_bytes         += txbytes;
1116
1117                 /* u32 counters */
1118                 rx_dropped      += p->rx_dropped;
1119                 rx_frame_errors += p->rx_frame_errors;
1120                 tx_dropped      += p->tx_dropped;
1121         }
1122         stats->rx_dropped  = rx_dropped;
1123         stats->rx_frame_errors = rx_frame_errors;
1124         stats->tx_dropped = tx_dropped;
1125 }
1126
1127 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1128                        struct netlink_ext_ack *extack)
1129 {
1130         struct tun_struct *tun = netdev_priv(dev);
1131         struct bpf_prog *old_prog;
1132
1133         old_prog = rtnl_dereference(tun->xdp_prog);
1134         rcu_assign_pointer(tun->xdp_prog, prog);
1135         if (old_prog)
1136                 bpf_prog_put(old_prog);
1137
1138         return 0;
1139 }
1140
1141 static u32 tun_xdp_query(struct net_device *dev)
1142 {
1143         struct tun_struct *tun = netdev_priv(dev);
1144         const struct bpf_prog *xdp_prog;
1145
1146         xdp_prog = rtnl_dereference(tun->xdp_prog);
1147         if (xdp_prog)
1148                 return xdp_prog->aux->id;
1149
1150         return 0;
1151 }
1152
1153 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1154 {
1155         switch (xdp->command) {
1156         case XDP_SETUP_PROG:
1157                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1158         case XDP_QUERY_PROG:
1159                 xdp->prog_id = tun_xdp_query(dev);
1160                 xdp->prog_attached = !!xdp->prog_id;
1161                 return 0;
1162         default:
1163                 return -EINVAL;
1164         }
1165 }
1166
1167 static const struct net_device_ops tun_netdev_ops = {
1168         .ndo_uninit             = tun_net_uninit,
1169         .ndo_open               = tun_net_open,
1170         .ndo_stop               = tun_net_close,
1171         .ndo_start_xmit         = tun_net_xmit,
1172         .ndo_fix_features       = tun_net_fix_features,
1173         .ndo_select_queue       = tun_select_queue,
1174 #ifdef CONFIG_NET_POLL_CONTROLLER
1175         .ndo_poll_controller    = tun_poll_controller,
1176 #endif
1177         .ndo_set_rx_headroom    = tun_set_headroom,
1178         .ndo_get_stats64        = tun_net_get_stats64,
1179 };
1180
1181 static const struct net_device_ops tap_netdev_ops = {
1182         .ndo_uninit             = tun_net_uninit,
1183         .ndo_open               = tun_net_open,
1184         .ndo_stop               = tun_net_close,
1185         .ndo_start_xmit         = tun_net_xmit,
1186         .ndo_fix_features       = tun_net_fix_features,
1187         .ndo_set_rx_mode        = tun_net_mclist,
1188         .ndo_set_mac_address    = eth_mac_addr,
1189         .ndo_validate_addr      = eth_validate_addr,
1190         .ndo_select_queue       = tun_select_queue,
1191 #ifdef CONFIG_NET_POLL_CONTROLLER
1192         .ndo_poll_controller    = tun_poll_controller,
1193 #endif
1194         .ndo_features_check     = passthru_features_check,
1195         .ndo_set_rx_headroom    = tun_set_headroom,
1196         .ndo_get_stats64        = tun_net_get_stats64,
1197         .ndo_bpf                = tun_xdp,
1198 };
1199
1200 static void tun_flow_init(struct tun_struct *tun)
1201 {
1202         int i;
1203
1204         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1205                 INIT_HLIST_HEAD(&tun->flows[i]);
1206
1207         tun->ageing_time = TUN_FLOW_EXPIRE;
1208         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1209         mod_timer(&tun->flow_gc_timer,
1210                   round_jiffies_up(jiffies + tun->ageing_time));
1211 }
1212
1213 static void tun_flow_uninit(struct tun_struct *tun)
1214 {
1215         del_timer_sync(&tun->flow_gc_timer);
1216         tun_flow_flush(tun);
1217 }
1218
1219 #define MIN_MTU 68
1220 #define MAX_MTU 65535
1221
1222 /* Initialize net device. */
1223 static void tun_net_init(struct net_device *dev)
1224 {
1225         struct tun_struct *tun = netdev_priv(dev);
1226
1227         switch (tun->flags & TUN_TYPE_MASK) {
1228         case IFF_TUN:
1229                 dev->netdev_ops = &tun_netdev_ops;
1230
1231                 /* Point-to-Point TUN Device */
1232                 dev->hard_header_len = 0;
1233                 dev->addr_len = 0;
1234                 dev->mtu = 1500;
1235
1236                 /* Zero header length */
1237                 dev->type = ARPHRD_NONE;
1238                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1239                 break;
1240
1241         case IFF_TAP:
1242                 dev->netdev_ops = &tap_netdev_ops;
1243                 /* Ethernet TAP Device */
1244                 ether_setup(dev);
1245                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1246                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1247
1248                 eth_hw_addr_random(dev);
1249
1250                 break;
1251         }
1252
1253         dev->min_mtu = MIN_MTU;
1254         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1255 }
1256
1257 /* Character device part */
1258
1259 /* Poll */
1260 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1261 {
1262         struct tun_file *tfile = file->private_data;
1263         struct tun_struct *tun = tun_get(tfile);
1264         struct sock *sk;
1265         unsigned int mask = 0;
1266
1267         if (!tun)
1268                 return POLLERR;
1269
1270         sk = tfile->socket.sk;
1271
1272         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1273
1274         poll_wait(file, sk_sleep(sk), wait);
1275
1276         if (!skb_array_empty(&tfile->tx_array))
1277                 mask |= POLLIN | POLLRDNORM;
1278
1279         if (tun->dev->flags & IFF_UP &&
1280             (sock_writeable(sk) ||
1281              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1282               sock_writeable(sk))))
1283                 mask |= POLLOUT | POLLWRNORM;
1284
1285         if (tun->dev->reg_state != NETREG_REGISTERED)
1286                 mask = POLLERR;
1287
1288         tun_put(tun);
1289         return mask;
1290 }
1291
1292 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1293                                             size_t len,
1294                                             const struct iov_iter *it)
1295 {
1296         struct sk_buff *skb;
1297         size_t linear;
1298         int err;
1299         int i;
1300
1301         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1302                 return ERR_PTR(-ENOMEM);
1303
1304         local_bh_disable();
1305         skb = napi_get_frags(&tfile->napi);
1306         local_bh_enable();
1307         if (!skb)
1308                 return ERR_PTR(-ENOMEM);
1309
1310         linear = iov_iter_single_seg_count(it);
1311         err = __skb_grow(skb, linear);
1312         if (err)
1313                 goto free;
1314
1315         skb->len = len;
1316         skb->data_len = len - linear;
1317         skb->truesize += skb->data_len;
1318
1319         for (i = 1; i < it->nr_segs; i++) {
1320                 struct page_frag *pfrag = &current->task_frag;
1321                 size_t fragsz = it->iov[i].iov_len;
1322
1323                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1324                         err = -EINVAL;
1325                         goto free;
1326                 }
1327
1328                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1329                         err = -ENOMEM;
1330                         goto free;
1331                 }
1332
1333                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1334                                    pfrag->offset, fragsz);
1335                 page_ref_inc(pfrag->page);
1336                 pfrag->offset += fragsz;
1337         }
1338
1339         return skb;
1340 free:
1341         /* frees skb and all frags allocated with napi_alloc_frag() */
1342         napi_free_frags(&tfile->napi);
1343         return ERR_PTR(err);
1344 }
1345
1346 /* prepad is the amount to reserve at front.  len is length after that.
1347  * linear is a hint as to how much to copy (usually headers). */
1348 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1349                                      size_t prepad, size_t len,
1350                                      size_t linear, int noblock)
1351 {
1352         struct sock *sk = tfile->socket.sk;
1353         struct sk_buff *skb;
1354         int err;
1355
1356         /* Under a page?  Don't bother with paged skb. */
1357         if (prepad + len < PAGE_SIZE || !linear)
1358                 linear = len;
1359
1360         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1361                                    &err, 0);
1362         if (!skb)
1363                 return ERR_PTR(err);
1364
1365         skb_reserve(skb, prepad);
1366         skb_put(skb, linear);
1367         skb->data_len = len - linear;
1368         skb->len += len - linear;
1369
1370         return skb;
1371 }
1372
1373 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1374                            struct sk_buff *skb, int more)
1375 {
1376         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1377         struct sk_buff_head process_queue;
1378         u32 rx_batched = tun->rx_batched;
1379         bool rcv = false;
1380
1381         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1382                 local_bh_disable();
1383                 netif_receive_skb(skb);
1384                 local_bh_enable();
1385                 return;
1386         }
1387
1388         spin_lock(&queue->lock);
1389         if (!more || skb_queue_len(queue) == rx_batched) {
1390                 __skb_queue_head_init(&process_queue);
1391                 skb_queue_splice_tail_init(queue, &process_queue);
1392                 rcv = true;
1393         } else {
1394                 __skb_queue_tail(queue, skb);
1395         }
1396         spin_unlock(&queue->lock);
1397
1398         if (rcv) {
1399                 struct sk_buff *nskb;
1400
1401                 local_bh_disable();
1402                 while ((nskb = __skb_dequeue(&process_queue)))
1403                         netif_receive_skb(nskb);
1404                 netif_receive_skb(skb);
1405                 local_bh_enable();
1406         }
1407 }
1408
1409 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1410                               int len, int noblock, bool zerocopy)
1411 {
1412         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1413                 return false;
1414
1415         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1416                 return false;
1417
1418         if (!noblock)
1419                 return false;
1420
1421         if (zerocopy)
1422                 return false;
1423
1424         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1425             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1426                 return false;
1427
1428         return true;
1429 }
1430
1431 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1432                                      struct tun_file *tfile,
1433                                      struct iov_iter *from,
1434                                      struct virtio_net_hdr *hdr,
1435                                      int len, int *skb_xdp)
1436 {
1437         struct page_frag *alloc_frag = &current->task_frag;
1438         struct sk_buff *skb;
1439         struct bpf_prog *xdp_prog;
1440         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1441         unsigned int delta = 0;
1442         char *buf;
1443         size_t copied;
1444         bool xdp_xmit = false;
1445         int err, pad = TUN_RX_PAD;
1446
1447         rcu_read_lock();
1448         xdp_prog = rcu_dereference(tun->xdp_prog);
1449         if (xdp_prog)
1450                 pad += TUN_HEADROOM;
1451         buflen += SKB_DATA_ALIGN(len + pad);
1452         rcu_read_unlock();
1453
1454         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1455         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1456                 return ERR_PTR(-ENOMEM);
1457
1458         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1459         copied = copy_page_from_iter(alloc_frag->page,
1460                                      alloc_frag->offset + pad,
1461                                      len, from);
1462         if (copied != len)
1463                 return ERR_PTR(-EFAULT);
1464
1465         /* There's a small window that XDP may be set after the check
1466          * of xdp_prog above, this should be rare and for simplicity
1467          * we do XDP on skb in case the headroom is not enough.
1468          */
1469         if (hdr->gso_type || !xdp_prog)
1470                 *skb_xdp = 1;
1471         else
1472                 *skb_xdp = 0;
1473
1474         rcu_read_lock();
1475         xdp_prog = rcu_dereference(tun->xdp_prog);
1476         if (xdp_prog && !*skb_xdp) {
1477                 struct xdp_buff xdp;
1478                 void *orig_data;
1479                 u32 act;
1480
1481                 xdp.data_hard_start = buf;
1482                 xdp.data = buf + pad;
1483                 xdp_set_data_meta_invalid(&xdp);
1484                 xdp.data_end = xdp.data + len;
1485                 orig_data = xdp.data;
1486                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1487
1488                 switch (act) {
1489                 case XDP_REDIRECT:
1490                         get_page(alloc_frag->page);
1491                         alloc_frag->offset += buflen;
1492                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1493                         if (err)
1494                                 goto err_redirect;
1495                         rcu_read_unlock();
1496                         return NULL;
1497                 case XDP_TX:
1498                         xdp_xmit = true;
1499                         /* fall through */
1500                 case XDP_PASS:
1501                         delta = orig_data - xdp.data;
1502                         break;
1503                 default:
1504                         bpf_warn_invalid_xdp_action(act);
1505                         /* fall through */
1506                 case XDP_ABORTED:
1507                         trace_xdp_exception(tun->dev, xdp_prog, act);
1508                         /* fall through */
1509                 case XDP_DROP:
1510                         goto err_xdp;
1511                 }
1512         }
1513
1514         skb = build_skb(buf, buflen);
1515         if (!skb) {
1516                 rcu_read_unlock();
1517                 return ERR_PTR(-ENOMEM);
1518         }
1519
1520         skb_reserve(skb, pad - delta);
1521         skb_put(skb, len + delta);
1522         get_page(alloc_frag->page);
1523         alloc_frag->offset += buflen;
1524
1525         if (xdp_xmit) {
1526                 skb->dev = tun->dev;
1527                 generic_xdp_tx(skb, xdp_prog);
1528                 rcu_read_unlock();
1529                 return NULL;
1530         }
1531
1532         rcu_read_unlock();
1533
1534         return skb;
1535
1536 err_redirect:
1537         put_page(alloc_frag->page);
1538 err_xdp:
1539         rcu_read_unlock();
1540         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1541         return NULL;
1542 }
1543
1544 /* Get packet from user space buffer */
1545 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1546                             void *msg_control, struct iov_iter *from,
1547                             int noblock, bool more)
1548 {
1549         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1550         struct sk_buff *skb;
1551         size_t total_len = iov_iter_count(from);
1552         size_t len = total_len, align = tun->align, linear;
1553         struct virtio_net_hdr gso = { 0 };
1554         struct tun_pcpu_stats *stats;
1555         int good_linear;
1556         int copylen;
1557         bool zerocopy = false;
1558         int err;
1559         u32 rxhash;
1560         int skb_xdp = 1;
1561         bool frags = tun_napi_frags_enabled(tun);
1562
1563         if (!(tun->dev->flags & IFF_UP))
1564                 return -EIO;
1565
1566         if (!(tun->flags & IFF_NO_PI)) {
1567                 if (len < sizeof(pi))
1568                         return -EINVAL;
1569                 len -= sizeof(pi);
1570
1571                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1572                         return -EFAULT;
1573         }
1574
1575         if (tun->flags & IFF_VNET_HDR) {
1576                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1577
1578                 if (len < vnet_hdr_sz)
1579                         return -EINVAL;
1580                 len -= vnet_hdr_sz;
1581
1582                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1583                         return -EFAULT;
1584
1585                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1586                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1587                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1588
1589                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1590                         return -EINVAL;
1591                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1592         }
1593
1594         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1595                 align += NET_IP_ALIGN;
1596                 if (unlikely(len < ETH_HLEN ||
1597                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1598                         return -EINVAL;
1599         }
1600
1601         good_linear = SKB_MAX_HEAD(align);
1602
1603         if (msg_control) {
1604                 struct iov_iter i = *from;
1605
1606                 /* There are 256 bytes to be copied in skb, so there is
1607                  * enough room for skb expand head in case it is used.
1608                  * The rest of the buffer is mapped from userspace.
1609                  */
1610                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1611                 if (copylen > good_linear)
1612                         copylen = good_linear;
1613                 linear = copylen;
1614                 iov_iter_advance(&i, copylen);
1615                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1616                         zerocopy = true;
1617         }
1618
1619         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1620                 /* For the packet that is not easy to be processed
1621                  * (e.g gso or jumbo packet), we will do it at after
1622                  * skb was created with generic XDP routine.
1623                  */
1624                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1625                 if (IS_ERR(skb)) {
1626                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1627                         return PTR_ERR(skb);
1628                 }
1629                 if (!skb)
1630                         return total_len;
1631         } else {
1632                 if (!zerocopy) {
1633                         copylen = len;
1634                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1635                                 linear = good_linear;
1636                         else
1637                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1638                 }
1639
1640                 if (frags) {
1641                         mutex_lock(&tfile->napi_mutex);
1642                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1643                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1644                          * If zerocopy is enabled, then this layout will be
1645                          * overwritten by zerocopy_sg_from_iter().
1646                          */
1647                         zerocopy = false;
1648                 } else {
1649                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1650                                             noblock);
1651                 }
1652
1653                 if (IS_ERR(skb)) {
1654                         if (PTR_ERR(skb) != -EAGAIN)
1655                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1656                         if (frags)
1657                                 mutex_unlock(&tfile->napi_mutex);
1658                         return PTR_ERR(skb);
1659                 }
1660
1661                 if (zerocopy)
1662                         err = zerocopy_sg_from_iter(skb, from);
1663                 else
1664                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1665
1666                 if (err) {
1667                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1668                         kfree_skb(skb);
1669                         if (frags) {
1670                                 tfile->napi.skb = NULL;
1671                                 mutex_unlock(&tfile->napi_mutex);
1672                         }
1673
1674                         return -EFAULT;
1675                 }
1676         }
1677
1678         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1679                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1680                 kfree_skb(skb);
1681                 if (frags) {
1682                         tfile->napi.skb = NULL;
1683                         mutex_unlock(&tfile->napi_mutex);
1684                 }
1685
1686                 return -EINVAL;
1687         }
1688
1689         switch (tun->flags & TUN_TYPE_MASK) {
1690         case IFF_TUN:
1691                 if (tun->flags & IFF_NO_PI) {
1692                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1693
1694                         switch (ip_version) {
1695                         case 4:
1696                                 pi.proto = htons(ETH_P_IP);
1697                                 break;
1698                         case 6:
1699                                 pi.proto = htons(ETH_P_IPV6);
1700                                 break;
1701                         default:
1702                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1703                                 kfree_skb(skb);
1704                                 return -EINVAL;
1705                         }
1706                 }
1707
1708                 skb_reset_mac_header(skb);
1709                 skb->protocol = pi.proto;
1710                 skb->dev = tun->dev;
1711                 break;
1712         case IFF_TAP:
1713                 if (!frags)
1714                         skb->protocol = eth_type_trans(skb, tun->dev);
1715                 break;
1716         }
1717
1718         /* copy skb_ubuf_info for callback when skb has no error */
1719         if (zerocopy) {
1720                 skb_shinfo(skb)->destructor_arg = msg_control;
1721                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1722                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1723         } else if (msg_control) {
1724                 struct ubuf_info *uarg = msg_control;
1725                 uarg->callback(uarg, false);
1726         }
1727
1728         skb_reset_network_header(skb);
1729         skb_probe_transport_header(skb, 0);
1730
1731         if (skb_xdp) {
1732                 struct bpf_prog *xdp_prog;
1733                 int ret;
1734
1735                 rcu_read_lock();
1736                 xdp_prog = rcu_dereference(tun->xdp_prog);
1737                 if (xdp_prog) {
1738                         ret = do_xdp_generic(xdp_prog, skb);
1739                         if (ret != XDP_PASS) {
1740                                 rcu_read_unlock();
1741                                 return total_len;
1742                         }
1743                 }
1744                 rcu_read_unlock();
1745         }
1746
1747         rxhash = __skb_get_hash_symmetric(skb);
1748
1749         if (frags) {
1750                 /* Exercise flow dissector code path. */
1751                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1752
1753                 if (unlikely(headlen > skb_headlen(skb))) {
1754                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1755                         napi_free_frags(&tfile->napi);
1756                         mutex_unlock(&tfile->napi_mutex);
1757                         WARN_ON(1);
1758                         return -ENOMEM;
1759                 }
1760
1761                 local_bh_disable();
1762                 napi_gro_frags(&tfile->napi);
1763                 local_bh_enable();
1764                 mutex_unlock(&tfile->napi_mutex);
1765         } else if (tfile->napi_enabled) {
1766                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1767                 int queue_len;
1768
1769                 spin_lock_bh(&queue->lock);
1770                 __skb_queue_tail(queue, skb);
1771                 queue_len = skb_queue_len(queue);
1772                 spin_unlock(&queue->lock);
1773
1774                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1775                         napi_schedule(&tfile->napi);
1776
1777                 local_bh_enable();
1778         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1779                 tun_rx_batched(tun, tfile, skb, more);
1780         } else {
1781                 netif_rx_ni(skb);
1782         }
1783
1784         stats = get_cpu_ptr(tun->pcpu_stats);
1785         u64_stats_update_begin(&stats->syncp);
1786         stats->rx_packets++;
1787         stats->rx_bytes += len;
1788         u64_stats_update_end(&stats->syncp);
1789         put_cpu_ptr(stats);
1790
1791         tun_flow_update(tun, rxhash, tfile);
1792         return total_len;
1793 }
1794
1795 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1796 {
1797         struct file *file = iocb->ki_filp;
1798         struct tun_file *tfile = file->private_data;
1799         struct tun_struct *tun = tun_get(tfile);
1800         ssize_t result;
1801
1802         if (!tun)
1803                 return -EBADFD;
1804
1805         result = tun_get_user(tun, tfile, NULL, from,
1806                               file->f_flags & O_NONBLOCK, false);
1807
1808         tun_put(tun);
1809         return result;
1810 }
1811
1812 /* Put packet to the user space buffer */
1813 static ssize_t tun_put_user(struct tun_struct *tun,
1814                             struct tun_file *tfile,
1815                             struct sk_buff *skb,
1816                             struct iov_iter *iter)
1817 {
1818         struct tun_pi pi = { 0, skb->protocol };
1819         struct tun_pcpu_stats *stats;
1820         ssize_t total;
1821         int vlan_offset = 0;
1822         int vlan_hlen = 0;
1823         int vnet_hdr_sz = 0;
1824
1825         if (skb_vlan_tag_present(skb))
1826                 vlan_hlen = VLAN_HLEN;
1827
1828         if (tun->flags & IFF_VNET_HDR)
1829                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1830
1831         total = skb->len + vlan_hlen + vnet_hdr_sz;
1832
1833         if (!(tun->flags & IFF_NO_PI)) {
1834                 if (iov_iter_count(iter) < sizeof(pi))
1835                         return -EINVAL;
1836
1837                 total += sizeof(pi);
1838                 if (iov_iter_count(iter) < total) {
1839                         /* Packet will be striped */
1840                         pi.flags |= TUN_PKT_STRIP;
1841                 }
1842
1843                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1844                         return -EFAULT;
1845         }
1846
1847         if (vnet_hdr_sz) {
1848                 struct virtio_net_hdr gso;
1849
1850                 if (iov_iter_count(iter) < vnet_hdr_sz)
1851                         return -EINVAL;
1852
1853                 if (virtio_net_hdr_from_skb(skb, &gso,
1854                                             tun_is_little_endian(tun), true)) {
1855                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1856                         pr_err("unexpected GSO type: "
1857                                "0x%x, gso_size %d, hdr_len %d\n",
1858                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1859                                tun16_to_cpu(tun, gso.hdr_len));
1860                         print_hex_dump(KERN_ERR, "tun: ",
1861                                        DUMP_PREFIX_NONE,
1862                                        16, 1, skb->head,
1863                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1864                         WARN_ON_ONCE(1);
1865                         return -EINVAL;
1866                 }
1867
1868                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1869                         return -EFAULT;
1870
1871                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1872         }
1873
1874         if (vlan_hlen) {
1875                 int ret;
1876                 struct {
1877                         __be16 h_vlan_proto;
1878                         __be16 h_vlan_TCI;
1879                 } veth;
1880
1881                 veth.h_vlan_proto = skb->vlan_proto;
1882                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1883
1884                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1885
1886                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1887                 if (ret || !iov_iter_count(iter))
1888                         goto done;
1889
1890                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1891                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1892                         goto done;
1893         }
1894
1895         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1896
1897 done:
1898         /* caller is in process context, */
1899         stats = get_cpu_ptr(tun->pcpu_stats);
1900         u64_stats_update_begin(&stats->syncp);
1901         stats->tx_packets++;
1902         stats->tx_bytes += skb->len + vlan_hlen;
1903         u64_stats_update_end(&stats->syncp);
1904         put_cpu_ptr(tun->pcpu_stats);
1905
1906         return total;
1907 }
1908
1909 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1910                                      int *err)
1911 {
1912         DECLARE_WAITQUEUE(wait, current);
1913         struct sk_buff *skb = NULL;
1914         int error = 0;
1915
1916         skb = skb_array_consume(&tfile->tx_array);
1917         if (skb)
1918                 goto out;
1919         if (noblock) {
1920                 error = -EAGAIN;
1921                 goto out;
1922         }
1923
1924         add_wait_queue(&tfile->wq.wait, &wait);
1925         current->state = TASK_INTERRUPTIBLE;
1926
1927         while (1) {
1928                 skb = skb_array_consume(&tfile->tx_array);
1929                 if (skb)
1930                         break;
1931                 if (signal_pending(current)) {
1932                         error = -ERESTARTSYS;
1933                         break;
1934                 }
1935                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1936                         error = -EFAULT;
1937                         break;
1938                 }
1939
1940                 schedule();
1941         }
1942
1943         current->state = TASK_RUNNING;
1944         remove_wait_queue(&tfile->wq.wait, &wait);
1945
1946 out:
1947         *err = error;
1948         return skb;
1949 }
1950
1951 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1952                            struct iov_iter *to,
1953                            int noblock, struct sk_buff *skb)
1954 {
1955         ssize_t ret;
1956         int err;
1957
1958         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1959
1960         if (!iov_iter_count(to)) {
1961                 if (skb)
1962                         kfree_skb(skb);
1963                 return 0;
1964         }
1965
1966         if (!skb) {
1967                 /* Read frames from ring */
1968                 skb = tun_ring_recv(tfile, noblock, &err);
1969                 if (!skb)
1970                         return err;
1971         }
1972
1973         ret = tun_put_user(tun, tfile, skb, to);
1974         if (unlikely(ret < 0))
1975                 kfree_skb(skb);
1976         else
1977                 consume_skb(skb);
1978
1979         return ret;
1980 }
1981
1982 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1983 {
1984         struct file *file = iocb->ki_filp;
1985         struct tun_file *tfile = file->private_data;
1986         struct tun_struct *tun = tun_get(tfile);
1987         ssize_t len = iov_iter_count(to), ret;
1988
1989         if (!tun)
1990                 return -EBADFD;
1991         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1992         ret = min_t(ssize_t, ret, len);
1993         if (ret > 0)
1994                 iocb->ki_pos = ret;
1995         tun_put(tun);
1996         return ret;
1997 }
1998
1999 static void tun_free_netdev(struct net_device *dev)
2000 {
2001         struct tun_struct *tun = netdev_priv(dev);
2002
2003         BUG_ON(!(list_empty(&tun->disabled)));
2004         free_percpu(tun->pcpu_stats);
2005         tun_flow_uninit(tun);
2006         security_tun_dev_free_security(tun->security);
2007 }
2008
2009 static void tun_setup(struct net_device *dev)
2010 {
2011         struct tun_struct *tun = netdev_priv(dev);
2012
2013         tun->owner = INVALID_UID;
2014         tun->group = INVALID_GID;
2015
2016         dev->ethtool_ops = &tun_ethtool_ops;
2017         dev->needs_free_netdev = true;
2018         dev->priv_destructor = tun_free_netdev;
2019         /* We prefer our own queue length */
2020         dev->tx_queue_len = TUN_READQ_SIZE;
2021 }
2022
2023 /* Trivial set of netlink ops to allow deleting tun or tap
2024  * device with netlink.
2025  */
2026 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2027                         struct netlink_ext_ack *extack)
2028 {
2029         return -EINVAL;
2030 }
2031
2032 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2033         .kind           = DRV_NAME,
2034         .priv_size      = sizeof(struct tun_struct),
2035         .setup          = tun_setup,
2036         .validate       = tun_validate,
2037 };
2038
2039 static void tun_sock_write_space(struct sock *sk)
2040 {
2041         struct tun_file *tfile;
2042         wait_queue_head_t *wqueue;
2043
2044         if (!sock_writeable(sk))
2045                 return;
2046
2047         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2048                 return;
2049
2050         wqueue = sk_sleep(sk);
2051         if (wqueue && waitqueue_active(wqueue))
2052                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
2053                                                 POLLWRNORM | POLLWRBAND);
2054
2055         tfile = container_of(sk, struct tun_file, sk);
2056         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2057 }
2058
2059 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2060 {
2061         int ret;
2062         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2063         struct tun_struct *tun = tun_get(tfile);
2064
2065         if (!tun)
2066                 return -EBADFD;
2067
2068         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2069                            m->msg_flags & MSG_DONTWAIT,
2070                            m->msg_flags & MSG_MORE);
2071         tun_put(tun);
2072         return ret;
2073 }
2074
2075 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2076                        int flags)
2077 {
2078         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2079         struct tun_struct *tun = tun_get(tfile);
2080         struct sk_buff *skb = m->msg_control;
2081         int ret;
2082
2083         if (!tun) {
2084                 ret = -EBADFD;
2085                 goto out_free_skb;
2086         }
2087
2088         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2089                 ret = -EINVAL;
2090                 goto out_put_tun;
2091         }
2092         if (flags & MSG_ERRQUEUE) {
2093                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2094                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2095                 goto out;
2096         }
2097         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
2098         if (ret > (ssize_t)total_len) {
2099                 m->msg_flags |= MSG_TRUNC;
2100                 ret = flags & MSG_TRUNC ? ret : total_len;
2101         }
2102 out:
2103         tun_put(tun);
2104         return ret;
2105
2106 out_put_tun:
2107         tun_put(tun);
2108 out_free_skb:
2109         if (skb)
2110                 kfree_skb(skb);
2111         return ret;
2112 }
2113
2114 static int tun_peek_len(struct socket *sock)
2115 {
2116         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2117         struct tun_struct *tun;
2118         int ret = 0;
2119
2120         tun = tun_get(tfile);
2121         if (!tun)
2122                 return 0;
2123
2124         ret = skb_array_peek_len(&tfile->tx_array);
2125         tun_put(tun);
2126
2127         return ret;
2128 }
2129
2130 /* Ops structure to mimic raw sockets with tun */
2131 static const struct proto_ops tun_socket_ops = {
2132         .peek_len = tun_peek_len,
2133         .sendmsg = tun_sendmsg,
2134         .recvmsg = tun_recvmsg,
2135 };
2136
2137 static struct proto tun_proto = {
2138         .name           = "tun",
2139         .owner          = THIS_MODULE,
2140         .obj_size       = sizeof(struct tun_file),
2141 };
2142
2143 static int tun_flags(struct tun_struct *tun)
2144 {
2145         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2146 }
2147
2148 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2149                               char *buf)
2150 {
2151         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2152         return sprintf(buf, "0x%x\n", tun_flags(tun));
2153 }
2154
2155 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2156                               char *buf)
2157 {
2158         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2159         return uid_valid(tun->owner)?
2160                 sprintf(buf, "%u\n",
2161                         from_kuid_munged(current_user_ns(), tun->owner)):
2162                 sprintf(buf, "-1\n");
2163 }
2164
2165 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2166                               char *buf)
2167 {
2168         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2169         return gid_valid(tun->group) ?
2170                 sprintf(buf, "%u\n",
2171                         from_kgid_munged(current_user_ns(), tun->group)):
2172                 sprintf(buf, "-1\n");
2173 }
2174
2175 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2176 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2177 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2178
2179 static struct attribute *tun_dev_attrs[] = {
2180         &dev_attr_tun_flags.attr,
2181         &dev_attr_owner.attr,
2182         &dev_attr_group.attr,
2183         NULL
2184 };
2185
2186 static const struct attribute_group tun_attr_group = {
2187         .attrs = tun_dev_attrs
2188 };
2189
2190 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2191 {
2192         struct tun_struct *tun;
2193         struct tun_file *tfile = file->private_data;
2194         struct net_device *dev;
2195         int err;
2196
2197         if (tfile->detached)
2198                 return -EINVAL;
2199
2200         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2201                 if (!capable(CAP_NET_ADMIN))
2202                         return -EPERM;
2203
2204                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2205                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2206                         return -EINVAL;
2207         }
2208
2209         dev = __dev_get_by_name(net, ifr->ifr_name);
2210         if (dev) {
2211                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2212                         return -EBUSY;
2213                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2214                         tun = netdev_priv(dev);
2215                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2216                         tun = netdev_priv(dev);
2217                 else
2218                         return -EINVAL;
2219
2220                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2221                     !!(tun->flags & IFF_MULTI_QUEUE))
2222                         return -EINVAL;
2223
2224                 if (tun_not_capable(tun))
2225                         return -EPERM;
2226                 err = security_tun_dev_open(tun->security);
2227                 if (err < 0)
2228                         return err;
2229
2230                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2231                                  ifr->ifr_flags & IFF_NAPI);
2232                 if (err < 0)
2233                         return err;
2234
2235                 if (tun->flags & IFF_MULTI_QUEUE &&
2236                     (tun->numqueues + tun->numdisabled > 1)) {
2237                         /* One or more queue has already been attached, no need
2238                          * to initialize the device again.
2239                          */
2240                         return 0;
2241                 }
2242         }
2243         else {
2244                 char *name;
2245                 unsigned long flags = 0;
2246                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2247                              MAX_TAP_QUEUES : 1;
2248
2249                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2250                         return -EPERM;
2251                 err = security_tun_dev_create();
2252                 if (err < 0)
2253                         return err;
2254
2255                 /* Set dev type */
2256                 if (ifr->ifr_flags & IFF_TUN) {
2257                         /* TUN device */
2258                         flags |= IFF_TUN;
2259                         name = "tun%d";
2260                 } else if (ifr->ifr_flags & IFF_TAP) {
2261                         /* TAP device */
2262                         flags |= IFF_TAP;
2263                         name = "tap%d";
2264                 } else
2265                         return -EINVAL;
2266
2267                 if (*ifr->ifr_name)
2268                         name = ifr->ifr_name;
2269
2270                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2271                                        NET_NAME_UNKNOWN, tun_setup, queues,
2272                                        queues);
2273
2274                 if (!dev)
2275                         return -ENOMEM;
2276                 err = dev_get_valid_name(net, dev, name);
2277                 if (err < 0)
2278                         goto err_free_dev;
2279
2280                 dev_net_set(dev, net);
2281                 dev->rtnl_link_ops = &tun_link_ops;
2282                 dev->ifindex = tfile->ifindex;
2283                 dev->sysfs_groups[0] = &tun_attr_group;
2284
2285                 tun = netdev_priv(dev);
2286                 tun->dev = dev;
2287                 tun->flags = flags;
2288                 tun->txflt.count = 0;
2289                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2290
2291                 tun->align = NET_SKB_PAD;
2292                 tun->filter_attached = false;
2293                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2294                 tun->rx_batched = 0;
2295
2296                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2297                 if (!tun->pcpu_stats) {
2298                         err = -ENOMEM;
2299                         goto err_free_dev;
2300                 }
2301
2302                 spin_lock_init(&tun->lock);
2303
2304                 err = security_tun_dev_alloc_security(&tun->security);
2305                 if (err < 0)
2306                         goto err_free_stat;
2307
2308                 tun_net_init(dev);
2309                 tun_flow_init(tun);
2310
2311                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2312                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2313                                    NETIF_F_HW_VLAN_STAG_TX;
2314                 dev->features = dev->hw_features | NETIF_F_LLTX;
2315                 dev->vlan_features = dev->features &
2316                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2317                                        NETIF_F_HW_VLAN_STAG_TX);
2318
2319                 INIT_LIST_HEAD(&tun->disabled);
2320                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2321                 if (err < 0)
2322                         goto err_free_flow;
2323
2324                 err = register_netdevice(tun->dev);
2325                 if (err < 0)
2326                         goto err_detach;
2327         }
2328
2329         netif_carrier_on(tun->dev);
2330
2331         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2332
2333         tun->flags = (tun->flags & ~TUN_FEATURES) |
2334                 (ifr->ifr_flags & TUN_FEATURES);
2335
2336         /* Make sure persistent devices do not get stuck in
2337          * xoff state.
2338          */
2339         if (netif_running(tun->dev))
2340                 netif_tx_wake_all_queues(tun->dev);
2341
2342         strcpy(ifr->ifr_name, tun->dev->name);
2343         return 0;
2344
2345 err_detach:
2346         tun_detach_all(dev);
2347         /* register_netdevice() already called tun_free_netdev() */
2348         goto err_free_dev;
2349
2350 err_free_flow:
2351         tun_flow_uninit(tun);
2352         security_tun_dev_free_security(tun->security);
2353 err_free_stat:
2354         free_percpu(tun->pcpu_stats);
2355 err_free_dev:
2356         free_netdev(dev);
2357         return err;
2358 }
2359
2360 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2361                        struct ifreq *ifr)
2362 {
2363         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2364
2365         strcpy(ifr->ifr_name, tun->dev->name);
2366
2367         ifr->ifr_flags = tun_flags(tun);
2368
2369 }
2370
2371 /* This is like a cut-down ethtool ops, except done via tun fd so no
2372  * privs required. */
2373 static int set_offload(struct tun_struct *tun, unsigned long arg)
2374 {
2375         netdev_features_t features = 0;
2376
2377         if (arg & TUN_F_CSUM) {
2378                 features |= NETIF_F_HW_CSUM;
2379                 arg &= ~TUN_F_CSUM;
2380
2381                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2382                         if (arg & TUN_F_TSO_ECN) {
2383                                 features |= NETIF_F_TSO_ECN;
2384                                 arg &= ~TUN_F_TSO_ECN;
2385                         }
2386                         if (arg & TUN_F_TSO4)
2387                                 features |= NETIF_F_TSO;
2388                         if (arg & TUN_F_TSO6)
2389                                 features |= NETIF_F_TSO6;
2390                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2391                 }
2392
2393                 arg &= ~TUN_F_UFO;
2394         }
2395
2396         /* This gives the user a way to test for new features in future by
2397          * trying to set them. */
2398         if (arg)
2399                 return -EINVAL;
2400
2401         tun->set_features = features;
2402         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2403         tun->dev->wanted_features |= features;
2404         netdev_update_features(tun->dev);
2405
2406         return 0;
2407 }
2408
2409 static void tun_detach_filter(struct tun_struct *tun, int n)
2410 {
2411         int i;
2412         struct tun_file *tfile;
2413
2414         for (i = 0; i < n; i++) {
2415                 tfile = rtnl_dereference(tun->tfiles[i]);
2416                 lock_sock(tfile->socket.sk);
2417                 sk_detach_filter(tfile->socket.sk);
2418                 release_sock(tfile->socket.sk);
2419         }
2420
2421         tun->filter_attached = false;
2422 }
2423
2424 static int tun_attach_filter(struct tun_struct *tun)
2425 {
2426         int i, ret = 0;
2427         struct tun_file *tfile;
2428
2429         for (i = 0; i < tun->numqueues; i++) {
2430                 tfile = rtnl_dereference(tun->tfiles[i]);
2431                 lock_sock(tfile->socket.sk);
2432                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2433                 release_sock(tfile->socket.sk);
2434                 if (ret) {
2435                         tun_detach_filter(tun, i);
2436                         return ret;
2437                 }
2438         }
2439
2440         tun->filter_attached = true;
2441         return ret;
2442 }
2443
2444 static void tun_set_sndbuf(struct tun_struct *tun)
2445 {
2446         struct tun_file *tfile;
2447         int i;
2448
2449         for (i = 0; i < tun->numqueues; i++) {
2450                 tfile = rtnl_dereference(tun->tfiles[i]);
2451                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2452         }
2453 }
2454
2455 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2456 {
2457         struct tun_file *tfile = file->private_data;
2458         struct tun_struct *tun;
2459         int ret = 0;
2460
2461         rtnl_lock();
2462
2463         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2464                 tun = tfile->detached;
2465                 if (!tun) {
2466                         ret = -EINVAL;
2467                         goto unlock;
2468                 }
2469                 ret = security_tun_dev_attach_queue(tun->security);
2470                 if (ret < 0)
2471                         goto unlock;
2472                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2473         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2474                 tun = rtnl_dereference(tfile->tun);
2475                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2476                         ret = -EINVAL;
2477                 else
2478                         __tun_detach(tfile, false);
2479         } else
2480                 ret = -EINVAL;
2481
2482 unlock:
2483         rtnl_unlock();
2484         return ret;
2485 }
2486
2487 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2488                             unsigned long arg, int ifreq_len)
2489 {
2490         struct tun_file *tfile = file->private_data;
2491         struct tun_struct *tun;
2492         void __user* argp = (void __user*)arg;
2493         struct ifreq ifr;
2494         kuid_t owner;
2495         kgid_t group;
2496         int sndbuf;
2497         int vnet_hdr_sz;
2498         unsigned int ifindex;
2499         int le;
2500         int ret;
2501
2502         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2503                 if (copy_from_user(&ifr, argp, ifreq_len))
2504                         return -EFAULT;
2505         } else {
2506                 memset(&ifr, 0, sizeof(ifr));
2507         }
2508         if (cmd == TUNGETFEATURES) {
2509                 /* Currently this just means: "what IFF flags are valid?".
2510                  * This is needed because we never checked for invalid flags on
2511                  * TUNSETIFF.
2512                  */
2513                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2514                                 (unsigned int __user*)argp);
2515         } else if (cmd == TUNSETQUEUE)
2516                 return tun_set_queue(file, &ifr);
2517
2518         ret = 0;
2519         rtnl_lock();
2520
2521         tun = tun_get(tfile);
2522         if (cmd == TUNSETIFF) {
2523                 ret = -EEXIST;
2524                 if (tun)
2525                         goto unlock;
2526
2527                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2528
2529                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2530
2531                 if (ret)
2532                         goto unlock;
2533
2534                 if (copy_to_user(argp, &ifr, ifreq_len))
2535                         ret = -EFAULT;
2536                 goto unlock;
2537         }
2538         if (cmd == TUNSETIFINDEX) {
2539                 ret = -EPERM;
2540                 if (tun)
2541                         goto unlock;
2542
2543                 ret = -EFAULT;
2544                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2545                         goto unlock;
2546
2547                 ret = 0;
2548                 tfile->ifindex = ifindex;
2549                 goto unlock;
2550         }
2551
2552         ret = -EBADFD;
2553         if (!tun)
2554                 goto unlock;
2555
2556         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2557
2558         ret = 0;
2559         switch (cmd) {
2560         case TUNGETIFF:
2561                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2562
2563                 if (tfile->detached)
2564                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2565                 if (!tfile->socket.sk->sk_filter)
2566                         ifr.ifr_flags |= IFF_NOFILTER;
2567
2568                 if (copy_to_user(argp, &ifr, ifreq_len))
2569                         ret = -EFAULT;
2570                 break;
2571
2572         case TUNSETNOCSUM:
2573                 /* Disable/Enable checksum */
2574
2575                 /* [unimplemented] */
2576                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2577                           arg ? "disabled" : "enabled");
2578                 break;
2579
2580         case TUNSETPERSIST:
2581                 /* Disable/Enable persist mode. Keep an extra reference to the
2582                  * module to prevent the module being unprobed.
2583                  */
2584                 if (arg && !(tun->flags & IFF_PERSIST)) {
2585                         tun->flags |= IFF_PERSIST;
2586                         __module_get(THIS_MODULE);
2587                 }
2588                 if (!arg && (tun->flags & IFF_PERSIST)) {
2589                         tun->flags &= ~IFF_PERSIST;
2590                         module_put(THIS_MODULE);
2591                 }
2592
2593                 tun_debug(KERN_INFO, tun, "persist %s\n",
2594                           arg ? "enabled" : "disabled");
2595                 break;
2596
2597         case TUNSETOWNER:
2598                 /* Set owner of the device */
2599                 owner = make_kuid(current_user_ns(), arg);
2600                 if (!uid_valid(owner)) {
2601                         ret = -EINVAL;
2602                         break;
2603                 }
2604                 tun->owner = owner;
2605                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2606                           from_kuid(&init_user_ns, tun->owner));
2607                 break;
2608
2609         case TUNSETGROUP:
2610                 /* Set group of the device */
2611                 group = make_kgid(current_user_ns(), arg);
2612                 if (!gid_valid(group)) {
2613                         ret = -EINVAL;
2614                         break;
2615                 }
2616                 tun->group = group;
2617                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2618                           from_kgid(&init_user_ns, tun->group));
2619                 break;
2620
2621         case TUNSETLINK:
2622                 /* Only allow setting the type when the interface is down */
2623                 if (tun->dev->flags & IFF_UP) {
2624                         tun_debug(KERN_INFO, tun,
2625                                   "Linktype set failed because interface is up\n");
2626                         ret = -EBUSY;
2627                 } else {
2628                         tun->dev->type = (int) arg;
2629                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2630                                   tun->dev->type);
2631                         ret = 0;
2632                 }
2633                 break;
2634
2635 #ifdef TUN_DEBUG
2636         case TUNSETDEBUG:
2637                 tun->debug = arg;
2638                 break;
2639 #endif
2640         case TUNSETOFFLOAD:
2641                 ret = set_offload(tun, arg);
2642                 break;
2643
2644         case TUNSETTXFILTER:
2645                 /* Can be set only for TAPs */
2646                 ret = -EINVAL;
2647                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2648                         break;
2649                 ret = update_filter(&tun->txflt, (void __user *)arg);
2650                 break;
2651
2652         case SIOCGIFHWADDR:
2653                 /* Get hw address */
2654                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2655                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2656                 if (copy_to_user(argp, &ifr, ifreq_len))
2657                         ret = -EFAULT;
2658                 break;
2659
2660         case SIOCSIFHWADDR:
2661                 /* Set hw address */
2662                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2663                           ifr.ifr_hwaddr.sa_data);
2664
2665                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2666                 break;
2667
2668         case TUNGETSNDBUF:
2669                 sndbuf = tfile->socket.sk->sk_sndbuf;
2670                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2671                         ret = -EFAULT;
2672                 break;
2673
2674         case TUNSETSNDBUF:
2675                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2676                         ret = -EFAULT;
2677                         break;
2678                 }
2679                 if (sndbuf <= 0) {
2680                         ret = -EINVAL;
2681                         break;
2682                 }
2683
2684                 tun->sndbuf = sndbuf;
2685                 tun_set_sndbuf(tun);
2686                 break;
2687
2688         case TUNGETVNETHDRSZ:
2689                 vnet_hdr_sz = tun->vnet_hdr_sz;
2690                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2691                         ret = -EFAULT;
2692                 break;
2693
2694         case TUNSETVNETHDRSZ:
2695                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2696                         ret = -EFAULT;
2697                         break;
2698                 }
2699                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2700                         ret = -EINVAL;
2701                         break;
2702                 }
2703
2704                 tun->vnet_hdr_sz = vnet_hdr_sz;
2705                 break;
2706
2707         case TUNGETVNETLE:
2708                 le = !!(tun->flags & TUN_VNET_LE);
2709                 if (put_user(le, (int __user *)argp))
2710                         ret = -EFAULT;
2711                 break;
2712
2713         case TUNSETVNETLE:
2714                 if (get_user(le, (int __user *)argp)) {
2715                         ret = -EFAULT;
2716                         break;
2717                 }
2718                 if (le)
2719                         tun->flags |= TUN_VNET_LE;
2720                 else
2721                         tun->flags &= ~TUN_VNET_LE;
2722                 break;
2723
2724         case TUNGETVNETBE:
2725                 ret = tun_get_vnet_be(tun, argp);
2726                 break;
2727
2728         case TUNSETVNETBE:
2729                 ret = tun_set_vnet_be(tun, argp);
2730                 break;
2731
2732         case TUNATTACHFILTER:
2733                 /* Can be set only for TAPs */
2734                 ret = -EINVAL;
2735                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2736                         break;
2737                 ret = -EFAULT;
2738                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2739                         break;
2740
2741                 ret = tun_attach_filter(tun);
2742                 break;
2743
2744         case TUNDETACHFILTER:
2745                 /* Can be set only for TAPs */
2746                 ret = -EINVAL;
2747                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2748                         break;
2749                 ret = 0;
2750                 tun_detach_filter(tun, tun->numqueues);
2751                 break;
2752
2753         case TUNGETFILTER:
2754                 ret = -EINVAL;
2755                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2756                         break;
2757                 ret = -EFAULT;
2758                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2759                         break;
2760                 ret = 0;
2761                 break;
2762
2763         default:
2764                 ret = -EINVAL;
2765                 break;
2766         }
2767
2768 unlock:
2769         rtnl_unlock();
2770         if (tun)
2771                 tun_put(tun);
2772         return ret;
2773 }
2774
2775 static long tun_chr_ioctl(struct file *file,
2776                           unsigned int cmd, unsigned long arg)
2777 {
2778         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2779 }
2780
2781 #ifdef CONFIG_COMPAT
2782 static long tun_chr_compat_ioctl(struct file *file,
2783                          unsigned int cmd, unsigned long arg)
2784 {
2785         switch (cmd) {
2786         case TUNSETIFF:
2787         case TUNGETIFF:
2788         case TUNSETTXFILTER:
2789         case TUNGETSNDBUF:
2790         case TUNSETSNDBUF:
2791         case SIOCGIFHWADDR:
2792         case SIOCSIFHWADDR:
2793                 arg = (unsigned long)compat_ptr(arg);
2794                 break;
2795         default:
2796                 arg = (compat_ulong_t)arg;
2797                 break;
2798         }
2799
2800         /*
2801          * compat_ifreq is shorter than ifreq, so we must not access beyond
2802          * the end of that structure. All fields that are used in this
2803          * driver are compatible though, we don't need to convert the
2804          * contents.
2805          */
2806         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2807 }
2808 #endif /* CONFIG_COMPAT */
2809
2810 static int tun_chr_fasync(int fd, struct file *file, int on)
2811 {
2812         struct tun_file *tfile = file->private_data;
2813         int ret;
2814
2815         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2816                 goto out;
2817
2818         if (on) {
2819                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2820                 tfile->flags |= TUN_FASYNC;
2821         } else
2822                 tfile->flags &= ~TUN_FASYNC;
2823         ret = 0;
2824 out:
2825         return ret;
2826 }
2827
2828 static int tun_chr_open(struct inode *inode, struct file * file)
2829 {
2830         struct net *net = current->nsproxy->net_ns;
2831         struct tun_file *tfile;
2832
2833         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2834
2835         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2836                                             &tun_proto, 0);
2837         if (!tfile)
2838                 return -ENOMEM;
2839         RCU_INIT_POINTER(tfile->tun, NULL);
2840         tfile->flags = 0;
2841         tfile->ifindex = 0;
2842
2843         init_waitqueue_head(&tfile->wq.wait);
2844         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2845
2846         tfile->socket.file = file;
2847         tfile->socket.ops = &tun_socket_ops;
2848
2849         sock_init_data(&tfile->socket, &tfile->sk);
2850
2851         tfile->sk.sk_write_space = tun_sock_write_space;
2852         tfile->sk.sk_sndbuf = INT_MAX;
2853
2854         file->private_data = tfile;
2855         INIT_LIST_HEAD(&tfile->next);
2856
2857         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2858
2859         memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
2860
2861         return 0;
2862 }
2863
2864 static int tun_chr_close(struct inode *inode, struct file *file)
2865 {
2866         struct tun_file *tfile = file->private_data;
2867
2868         tun_detach(tfile, true);
2869
2870         return 0;
2871 }
2872
2873 #ifdef CONFIG_PROC_FS
2874 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
2875 {
2876         struct tun_file *tfile = file->private_data;
2877         struct tun_struct *tun;
2878         struct ifreq ifr;
2879
2880         memset(&ifr, 0, sizeof(ifr));
2881
2882         rtnl_lock();
2883         tun = tun_get(tfile);
2884         if (tun)
2885                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2886         rtnl_unlock();
2887
2888         if (tun)
2889                 tun_put(tun);
2890
2891         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2892 }
2893 #endif
2894
2895 static const struct file_operations tun_fops = {
2896         .owner  = THIS_MODULE,
2897         .llseek = no_llseek,
2898         .read_iter  = tun_chr_read_iter,
2899         .write_iter = tun_chr_write_iter,
2900         .poll   = tun_chr_poll,
2901         .unlocked_ioctl = tun_chr_ioctl,
2902 #ifdef CONFIG_COMPAT
2903         .compat_ioctl = tun_chr_compat_ioctl,
2904 #endif
2905         .open   = tun_chr_open,
2906         .release = tun_chr_close,
2907         .fasync = tun_chr_fasync,
2908 #ifdef CONFIG_PROC_FS
2909         .show_fdinfo = tun_chr_show_fdinfo,
2910 #endif
2911 };
2912
2913 static struct miscdevice tun_miscdev = {
2914         .minor = TUN_MINOR,
2915         .name = "tun",
2916         .nodename = "net/tun",
2917         .fops = &tun_fops,
2918 };
2919
2920 /* ethtool interface */
2921
2922 static int tun_get_link_ksettings(struct net_device *dev,
2923                                   struct ethtool_link_ksettings *cmd)
2924 {
2925         ethtool_link_ksettings_zero_link_mode(cmd, supported);
2926         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2927         cmd->base.speed         = SPEED_10;
2928         cmd->base.duplex        = DUPLEX_FULL;
2929         cmd->base.port          = PORT_TP;
2930         cmd->base.phy_address   = 0;
2931         cmd->base.autoneg       = AUTONEG_DISABLE;
2932         return 0;
2933 }
2934
2935 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2936 {
2937         struct tun_struct *tun = netdev_priv(dev);
2938
2939         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2940         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2941
2942         switch (tun->flags & TUN_TYPE_MASK) {
2943         case IFF_TUN:
2944                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2945                 break;
2946         case IFF_TAP:
2947                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2948                 break;
2949         }
2950 }
2951
2952 static u32 tun_get_msglevel(struct net_device *dev)
2953 {
2954 #ifdef TUN_DEBUG
2955         struct tun_struct *tun = netdev_priv(dev);
2956         return tun->debug;
2957 #else
2958         return -EOPNOTSUPP;
2959 #endif
2960 }
2961
2962 static void tun_set_msglevel(struct net_device *dev, u32 value)
2963 {
2964 #ifdef TUN_DEBUG
2965         struct tun_struct *tun = netdev_priv(dev);
2966         tun->debug = value;
2967 #endif
2968 }
2969
2970 static int tun_get_coalesce(struct net_device *dev,
2971                             struct ethtool_coalesce *ec)
2972 {
2973         struct tun_struct *tun = netdev_priv(dev);
2974
2975         ec->rx_max_coalesced_frames = tun->rx_batched;
2976
2977         return 0;
2978 }
2979
2980 static int tun_set_coalesce(struct net_device *dev,
2981                             struct ethtool_coalesce *ec)
2982 {
2983         struct tun_struct *tun = netdev_priv(dev);
2984
2985         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2986                 tun->rx_batched = NAPI_POLL_WEIGHT;
2987         else
2988                 tun->rx_batched = ec->rx_max_coalesced_frames;
2989
2990         return 0;
2991 }
2992
2993 static const struct ethtool_ops tun_ethtool_ops = {
2994         .get_drvinfo    = tun_get_drvinfo,
2995         .get_msglevel   = tun_get_msglevel,
2996         .set_msglevel   = tun_set_msglevel,
2997         .get_link       = ethtool_op_get_link,
2998         .get_ts_info    = ethtool_op_get_ts_info,
2999         .get_coalesce   = tun_get_coalesce,
3000         .set_coalesce   = tun_set_coalesce,
3001         .get_link_ksettings = tun_get_link_ksettings,
3002 };
3003
3004 static int tun_queue_resize(struct tun_struct *tun)
3005 {
3006         struct net_device *dev = tun->dev;
3007         struct tun_file *tfile;
3008         struct skb_array **arrays;
3009         int n = tun->numqueues + tun->numdisabled;
3010         int ret, i;
3011
3012         arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
3013         if (!arrays)
3014                 return -ENOMEM;
3015
3016         for (i = 0; i < tun->numqueues; i++) {
3017                 tfile = rtnl_dereference(tun->tfiles[i]);
3018                 arrays[i] = &tfile->tx_array;
3019         }
3020         list_for_each_entry(tfile, &tun->disabled, next)
3021                 arrays[i++] = &tfile->tx_array;
3022
3023         ret = skb_array_resize_multiple(arrays, n,
3024                                         dev->tx_queue_len, GFP_KERNEL);
3025
3026         kfree(arrays);
3027         return ret;
3028 }
3029
3030 static int tun_device_event(struct notifier_block *unused,
3031                             unsigned long event, void *ptr)
3032 {
3033         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3034         struct tun_struct *tun = netdev_priv(dev);
3035
3036         if (dev->rtnl_link_ops != &tun_link_ops)
3037                 return NOTIFY_DONE;
3038
3039         switch (event) {
3040         case NETDEV_CHANGE_TX_QUEUE_LEN:
3041                 if (tun_queue_resize(tun))
3042                         return NOTIFY_BAD;
3043                 break;
3044         default:
3045                 break;
3046         }
3047
3048         return NOTIFY_DONE;
3049 }
3050
3051 static struct notifier_block tun_notifier_block __read_mostly = {
3052         .notifier_call  = tun_device_event,
3053 };
3054
3055 static int __init tun_init(void)
3056 {
3057         int ret = 0;
3058
3059         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3060
3061         ret = rtnl_link_register(&tun_link_ops);
3062         if (ret) {
3063                 pr_err("Can't register link_ops\n");
3064                 goto err_linkops;
3065         }
3066
3067         ret = misc_register(&tun_miscdev);
3068         if (ret) {
3069                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3070                 goto err_misc;
3071         }
3072
3073         ret = register_netdevice_notifier(&tun_notifier_block);
3074         if (ret) {
3075                 pr_err("Can't register netdevice notifier\n");
3076                 goto err_notifier;
3077         }
3078
3079         return  0;
3080
3081 err_notifier:
3082         misc_deregister(&tun_miscdev);
3083 err_misc:
3084         rtnl_link_unregister(&tun_link_ops);
3085 err_linkops:
3086         return ret;
3087 }
3088
3089 static void tun_cleanup(void)
3090 {
3091         misc_deregister(&tun_miscdev);
3092         rtnl_link_unregister(&tun_link_ops);
3093         unregister_netdevice_notifier(&tun_notifier_block);
3094 }
3095
3096 /* Get an underlying socket object from tun file.  Returns error unless file is
3097  * attached to a device.  The returned object works like a packet socket, it
3098  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3099  * holding a reference to the file for as long as the socket is in use. */
3100 struct socket *tun_get_socket(struct file *file)
3101 {
3102         struct tun_file *tfile;
3103         if (file->f_op != &tun_fops)
3104                 return ERR_PTR(-EINVAL);
3105         tfile = file->private_data;
3106         if (!tfile)
3107                 return ERR_PTR(-EBADFD);
3108         return &tfile->socket;
3109 }
3110 EXPORT_SYMBOL_GPL(tun_get_socket);
3111
3112 struct skb_array *tun_get_skb_array(struct file *file)
3113 {
3114         struct tun_file *tfile;
3115
3116         if (file->f_op != &tun_fops)
3117                 return ERR_PTR(-EINVAL);
3118         tfile = file->private_data;
3119         if (!tfile)
3120                 return ERR_PTR(-EBADFD);
3121         return &tfile->tx_array;
3122 }
3123 EXPORT_SYMBOL_GPL(tun_get_skb_array);
3124
3125 module_init(tun_init);
3126 module_exit(tun_cleanup);
3127 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3128 MODULE_AUTHOR(DRV_COPYRIGHT);
3129 MODULE_LICENSE("GPL");
3130 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3131 MODULE_ALIAS("devname:net/tun");