drivers/net/tun.c

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