drivers/net/tun.c

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