drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <[email protected]>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <[email protected]>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81 #include <linux/proc_fs.h>
  82
  83 /* Uncomment to enable debugging */
  84 /* #define TUN_DEBUG 1 */
  85
  86 #ifdef TUN_DEBUG
  87 static int debug;
  88
  89 #define tun_debug(level, tun, fmt, args...)                     \
  90 do {                                                            \
  91         if (tun->debug)                                         \
  92                 netdev_printk(level, tun->dev, fmt, ##args);    \
  93 } while (0)
  94 #define DBG1(level, fmt, args...)                               \
  95 do {                                                            \
  96         if (debug == 2)                                         \
  97                 printk(level fmt, ##args);                      \
  98 } while (0)
  99 #else
 100 #define tun_debug(level, tun, fmt, args...)                     \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 netdev_printk(level, tun->dev, fmt, ##args);    \
 104 } while (0)
 105 #define DBG1(level, fmt, args...)                               \
 106 do {                                                            \
 107         if (0)                                                  \
 108                 printk(level fmt, ##args);                      \
 109 } while (0)
 110 #endif
 111
 112 #define TUN_HEADROOM 256
 113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 127
 128 #define GOODCOPY_LEN 128
 129
 130 #define FLT_EXACT_COUNT 8
 131 struct tap_filter {
 132         unsigned int    count;    /* Number of addrs. Zero means disabled */
 133         u32             mask[2];  /* Mask of the hashed addrs */
 134         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 135 };
 136
 137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 138  * to max number of VCPUs in guest. */
 139 #define MAX_TAP_QUEUES 256
 140 #define MAX_TAP_FLOWS  4096
 141
 142 #define TUN_FLOW_EXPIRE (3 * HZ)
 143
 144 struct tun_pcpu_stats {
 145         u64 rx_packets;
 146         u64 rx_bytes;
 147         u64 tx_packets;
 148         u64 tx_bytes;
 149         struct u64_stats_sync syncp;
 150         u32 rx_dropped;
 151         u32 tx_dropped;
 152         u32 rx_frame_errors;
 153 };
 154
 155 /* A tun_file connects an open character device to a tuntap netdevice. It
 156  * also contains all socket related structures (except sock_fprog and tap_filter)
 157  * to serve as one transmit queue for tuntap device. The sock_fprog and
 158  * tap_filter were kept in tun_struct since they were used for filtering for the
 159  * netdevice not for a specific queue (at least I didn't see the requirement for
 160  * this).
 161  *
 162  * RCU usage:
 163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 164  * other can only be read while rcu_read_lock or rtnl_lock is held.
 165  */
 166 struct tun_file {
 167         struct sock sk;
 168         struct socket socket;
 169         struct socket_wq wq;
 170         struct tun_struct __rcu *tun;
 171         struct fasync_struct *fasync;
 172         /* only used for fasnyc */
 173         unsigned int flags;
 174         union {
 175                 u16 queue_index;
 176                 unsigned int ifindex;
 177         };
 178         struct napi_struct napi;
 179         bool napi_enabled;
 180         struct mutex napi_mutex;        /* Protects access to the above napi */
 181         struct list_head next;
 182         struct tun_struct *detached;
 183         struct ptr_ring tx_ring;
 184         struct xdp_rxq_info xdp_rxq;
 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                 struct page_frag *pfrag = &current->task_frag;
1493                 size_t fragsz = it->iov[i].iov_len;
1494
1495                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1496                         err = -EINVAL;
1497                         goto free;
1498                 }
1499
1500                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1501                         err = -ENOMEM;
1502                         goto free;
1503                 }
1504
1505                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1506                                    pfrag->offset, fragsz);
1507                 page_ref_inc(pfrag->page);
1508                 pfrag->offset += fragsz;
1509         }
1510
1511         return skb;
1512 free:
1513         /* frees skb and all frags allocated with napi_alloc_frag() */
1514         napi_free_frags(&tfile->napi);
1515         return ERR_PTR(err);
1516 }
1517
1518 /* prepad is the amount to reserve at front.  len is length after that.
1519  * linear is a hint as to how much to copy (usually headers). */
1520 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1521                                      size_t prepad, size_t len,
1522                                      size_t linear, int noblock)
1523 {
1524         struct sock *sk = tfile->socket.sk;
1525         struct sk_buff *skb;
1526         int err;
1527
1528         /* Under a page?  Don't bother with paged skb. */
1529         if (prepad + len < PAGE_SIZE || !linear)
1530                 linear = len;
1531
1532         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1533                                    &err, 0);
1534         if (!skb)
1535                 return ERR_PTR(err);
1536
1537         skb_reserve(skb, prepad);
1538         skb_put(skb, linear);
1539         skb->data_len = len - linear;
1540         skb->len += len - linear;
1541
1542         return skb;
1543 }
1544
1545 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1546                            struct sk_buff *skb, int more)
1547 {
1548         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1549         struct sk_buff_head process_queue;
1550         u32 rx_batched = tun->rx_batched;
1551         bool rcv = false;
1552
1553         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1554                 local_bh_disable();
1555                 netif_receive_skb(skb);
1556                 local_bh_enable();
1557                 return;
1558         }
1559
1560         spin_lock(&queue->lock);
1561         if (!more || skb_queue_len(queue) == rx_batched) {
1562                 __skb_queue_head_init(&process_queue);
1563                 skb_queue_splice_tail_init(queue, &process_queue);
1564                 rcv = true;
1565         } else {
1566                 __skb_queue_tail(queue, skb);
1567         }
1568         spin_unlock(&queue->lock);
1569
1570         if (rcv) {
1571                 struct sk_buff *nskb;
1572
1573                 local_bh_disable();
1574                 while ((nskb = __skb_dequeue(&process_queue)))
1575                         netif_receive_skb(nskb);
1576                 netif_receive_skb(skb);
1577                 local_bh_enable();
1578         }
1579 }
1580
1581 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1582                               int len, int noblock, bool zerocopy)
1583 {
1584         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1585                 return false;
1586
1587         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1588                 return false;
1589
1590         if (!noblock)
1591                 return false;
1592
1593         if (zerocopy)
1594                 return false;
1595
1596         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1597             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1598                 return false;
1599
1600         return true;
1601 }
1602
1603 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1604                                      struct tun_file *tfile,
1605                                      struct iov_iter *from,
1606                                      struct virtio_net_hdr *hdr,
1607                                      int len, int *skb_xdp)
1608 {
1609         struct page_frag *alloc_frag = &current->task_frag;
1610         struct sk_buff *skb;
1611         struct bpf_prog *xdp_prog;
1612         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1613         unsigned int delta = 0;
1614         char *buf;
1615         size_t copied;
1616         int err, pad = TUN_RX_PAD;
1617
1618         rcu_read_lock();
1619         xdp_prog = rcu_dereference(tun->xdp_prog);
1620         if (xdp_prog)
1621                 pad += TUN_HEADROOM;
1622         buflen += SKB_DATA_ALIGN(len + pad);
1623         rcu_read_unlock();
1624
1625         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1626         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1627                 return ERR_PTR(-ENOMEM);
1628
1629         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1630         copied = copy_page_from_iter(alloc_frag->page,
1631                                      alloc_frag->offset + pad,
1632                                      len, from);
1633         if (copied != len)
1634                 return ERR_PTR(-EFAULT);
1635
1636         /* There's a small window that XDP may be set after the check
1637          * of xdp_prog above, this should be rare and for simplicity
1638          * we do XDP on skb in case the headroom is not enough.
1639          */
1640         if (hdr->gso_type || !xdp_prog)
1641                 *skb_xdp = 1;
1642         else
1643                 *skb_xdp = 0;
1644
1645         preempt_disable();
1646         rcu_read_lock();
1647         xdp_prog = rcu_dereference(tun->xdp_prog);
1648         if (xdp_prog && !*skb_xdp) {
1649                 struct xdp_buff xdp;
1650                 void *orig_data;
1651                 u32 act;
1652
1653                 xdp.data_hard_start = buf;
1654                 xdp.data = buf + pad;
1655                 xdp_set_data_meta_invalid(&xdp);
1656                 xdp.data_end = xdp.data + len;
1657                 xdp.rxq = &tfile->xdp_rxq;
1658                 orig_data = xdp.data;
1659                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1660
1661                 switch (act) {
1662                 case XDP_REDIRECT:
1663                         get_page(alloc_frag->page);
1664                         alloc_frag->offset += buflen;
1665                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1666                         xdp_do_flush_map();
1667                         if (err)
1668                                 goto err_redirect;
1669                         rcu_read_unlock();
1670                         preempt_enable();
1671                         return NULL;
1672                 case XDP_TX:
1673                         get_page(alloc_frag->page);
1674                         alloc_frag->offset += buflen;
1675                         if (tun_xdp_xmit(tun->dev, &xdp))
1676                                 goto err_redirect;
1677                         tun_xdp_flush(tun->dev);
1678                         rcu_read_unlock();
1679                         preempt_enable();
1680                         return NULL;
1681                 case XDP_PASS:
1682                         delta = orig_data - xdp.data;
1683                         break;
1684                 default:
1685                         bpf_warn_invalid_xdp_action(act);
1686                         /* fall through */
1687                 case XDP_ABORTED:
1688                         trace_xdp_exception(tun->dev, xdp_prog, act);
1689                         /* fall through */
1690                 case XDP_DROP:
1691                         goto err_xdp;
1692                 }
1693         }
1694
1695         skb = build_skb(buf, buflen);
1696         if (!skb) {
1697                 rcu_read_unlock();
1698                 preempt_enable();
1699                 return ERR_PTR(-ENOMEM);
1700         }
1701
1702         skb_reserve(skb, pad - delta);
1703         skb_put(skb, len + delta);
1704         get_page(alloc_frag->page);
1705         alloc_frag->offset += buflen;
1706
1707         rcu_read_unlock();
1708         preempt_enable();
1709
1710         return skb;
1711
1712 err_redirect:
1713         put_page(alloc_frag->page);
1714 err_xdp:
1715         rcu_read_unlock();
1716         preempt_enable();
1717         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1718         return NULL;
1719 }
1720
1721 /* Get packet from user space buffer */
1722 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1723                             void *msg_control, struct iov_iter *from,
1724                             int noblock, bool more)
1725 {
1726         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1727         struct sk_buff *skb;
1728         size_t total_len = iov_iter_count(from);
1729         size_t len = total_len, align = tun->align, linear;
1730         struct virtio_net_hdr gso = { 0 };
1731         struct tun_pcpu_stats *stats;
1732         int good_linear;
1733         int copylen;
1734         bool zerocopy = false;
1735         int err;
1736         u32 rxhash = 0;
1737         int skb_xdp = 1;
1738         bool frags = tun_napi_frags_enabled(tun);
1739
1740         if (!(tun->dev->flags & IFF_UP))
1741                 return -EIO;
1742
1743         if (!(tun->flags & IFF_NO_PI)) {
1744                 if (len < sizeof(pi))
1745                         return -EINVAL;
1746                 len -= sizeof(pi);
1747
1748                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1749                         return -EFAULT;
1750         }
1751
1752         if (tun->flags & IFF_VNET_HDR) {
1753                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1754
1755                 if (len < vnet_hdr_sz)
1756                         return -EINVAL;
1757                 len -= vnet_hdr_sz;
1758
1759                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1760                         return -EFAULT;
1761
1762                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1763                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1764                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1765
1766                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1767                         return -EINVAL;
1768                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1769         }
1770
1771         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1772                 align += NET_IP_ALIGN;
1773                 if (unlikely(len < ETH_HLEN ||
1774                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1775                         return -EINVAL;
1776         }
1777
1778         good_linear = SKB_MAX_HEAD(align);
1779
1780         if (msg_control) {
1781                 struct iov_iter i = *from;
1782
1783                 /* There are 256 bytes to be copied in skb, so there is
1784                  * enough room for skb expand head in case it is used.
1785                  * The rest of the buffer is mapped from userspace.
1786                  */
1787                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1788                 if (copylen > good_linear)
1789                         copylen = good_linear;
1790                 linear = copylen;
1791                 iov_iter_advance(&i, copylen);
1792                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1793                         zerocopy = true;
1794         }
1795
1796         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1797                 /* For the packet that is not easy to be processed
1798                  * (e.g gso or jumbo packet), we will do it at after
1799                  * skb was created with generic XDP routine.
1800                  */
1801                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1802                 if (IS_ERR(skb)) {
1803                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1804                         return PTR_ERR(skb);
1805                 }
1806                 if (!skb)
1807                         return total_len;
1808         } else {
1809                 if (!zerocopy) {
1810                         copylen = len;
1811                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1812                                 linear = good_linear;
1813                         else
1814                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1815                 }
1816
1817                 if (frags) {
1818                         mutex_lock(&tfile->napi_mutex);
1819                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1820                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1821                          * If zerocopy is enabled, then this layout will be
1822                          * overwritten by zerocopy_sg_from_iter().
1823                          */
1824                         zerocopy = false;
1825                 } else {
1826                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1827                                             noblock);
1828                 }
1829
1830                 if (IS_ERR(skb)) {
1831                         if (PTR_ERR(skb) != -EAGAIN)
1832                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1833                         if (frags)
1834                                 mutex_unlock(&tfile->napi_mutex);
1835                         return PTR_ERR(skb);
1836                 }
1837
1838                 if (zerocopy)
1839                         err = zerocopy_sg_from_iter(skb, from);
1840                 else
1841                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1842
1843                 if (err) {
1844                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1845                         kfree_skb(skb);
1846                         if (frags) {
1847                                 tfile->napi.skb = NULL;
1848                                 mutex_unlock(&tfile->napi_mutex);
1849                         }
1850
1851                         return -EFAULT;
1852                 }
1853         }
1854
1855         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1856                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1857                 kfree_skb(skb);
1858                 if (frags) {
1859                         tfile->napi.skb = NULL;
1860                         mutex_unlock(&tfile->napi_mutex);
1861                 }
1862
1863                 return -EINVAL;
1864         }
1865
1866         switch (tun->flags & TUN_TYPE_MASK) {
1867         case IFF_TUN:
1868                 if (tun->flags & IFF_NO_PI) {
1869                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1870
1871                         switch (ip_version) {
1872                         case 4:
1873                                 pi.proto = htons(ETH_P_IP);
1874                                 break;
1875                         case 6:
1876                                 pi.proto = htons(ETH_P_IPV6);
1877                                 break;
1878                         default:
1879                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1880                                 kfree_skb(skb);
1881                                 return -EINVAL;
1882                         }
1883                 }
1884
1885                 skb_reset_mac_header(skb);
1886                 skb->protocol = pi.proto;
1887                 skb->dev = tun->dev;
1888                 break;
1889         case IFF_TAP:
1890                 if (!frags)
1891                         skb->protocol = eth_type_trans(skb, tun->dev);
1892                 break;
1893         }
1894
1895         /* copy skb_ubuf_info for callback when skb has no error */
1896         if (zerocopy) {
1897                 skb_shinfo(skb)->destructor_arg = msg_control;
1898                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1899                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1900         } else if (msg_control) {
1901                 struct ubuf_info *uarg = msg_control;
1902                 uarg->callback(uarg, false);
1903         }
1904
1905         skb_reset_network_header(skb);
1906         skb_probe_transport_header(skb, 0);
1907
1908         if (skb_xdp) {
1909                 struct bpf_prog *xdp_prog;
1910                 int ret;
1911
1912                 rcu_read_lock();
1913                 xdp_prog = rcu_dereference(tun->xdp_prog);
1914                 if (xdp_prog) {
1915                         ret = do_xdp_generic(xdp_prog, skb);
1916                         if (ret != XDP_PASS) {
1917                                 rcu_read_unlock();
1918                                 return total_len;
1919                         }
1920                 }
1921                 rcu_read_unlock();
1922         }
1923
1924         rcu_read_lock();
1925         if (!rcu_dereference(tun->steering_prog))
1926                 rxhash = __skb_get_hash_symmetric(skb);
1927         rcu_read_unlock();
1928
1929         if (frags) {
1930                 /* Exercise flow dissector code path. */
1931                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1932
1933                 if (unlikely(headlen > skb_headlen(skb))) {
1934                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1935                         napi_free_frags(&tfile->napi);
1936                         mutex_unlock(&tfile->napi_mutex);
1937                         WARN_ON(1);
1938                         return -ENOMEM;
1939                 }
1940
1941                 local_bh_disable();
1942                 napi_gro_frags(&tfile->napi);
1943                 local_bh_enable();
1944                 mutex_unlock(&tfile->napi_mutex);
1945         } else if (tfile->napi_enabled) {
1946                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1947                 int queue_len;
1948
1949                 spin_lock_bh(&queue->lock);
1950                 __skb_queue_tail(queue, skb);
1951                 queue_len = skb_queue_len(queue);
1952                 spin_unlock(&queue->lock);
1953
1954                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1955                         napi_schedule(&tfile->napi);
1956
1957                 local_bh_enable();
1958         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1959                 tun_rx_batched(tun, tfile, skb, more);
1960         } else {
1961                 netif_rx_ni(skb);
1962         }
1963
1964         stats = get_cpu_ptr(tun->pcpu_stats);
1965         u64_stats_update_begin(&stats->syncp);
1966         stats->rx_packets++;
1967         stats->rx_bytes += len;
1968         u64_stats_update_end(&stats->syncp);
1969         put_cpu_ptr(stats);
1970
1971         if (rxhash)
1972                 tun_flow_update(tun, rxhash, tfile);
1973
1974         return total_len;
1975 }
1976
1977 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1978 {
1979         struct file *file = iocb->ki_filp;
1980         struct tun_file *tfile = file->private_data;
1981         struct tun_struct *tun = tun_get(tfile);
1982         ssize_t result;
1983
1984         if (!tun)
1985                 return -EBADFD;
1986
1987         result = tun_get_user(tun, tfile, NULL, from,
1988                               file->f_flags & O_NONBLOCK, false);
1989
1990         tun_put(tun);
1991         return result;
1992 }
1993
1994 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1995                                 struct tun_file *tfile,
1996                                 struct xdp_buff *xdp,
1997                                 struct iov_iter *iter)
1998 {
1999         int vnet_hdr_sz = 0;
2000         size_t size = xdp->data_end - xdp->data;
2001         struct tun_pcpu_stats *stats;
2002         size_t ret;
2003
2004         if (tun->flags & IFF_VNET_HDR) {
2005                 struct virtio_net_hdr gso = { 0 };
2006
2007                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2008                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2009                         return -EINVAL;
2010                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2011                              sizeof(gso)))
2012                         return -EFAULT;
2013                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2014         }
2015
2016         ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2017
2018         stats = get_cpu_ptr(tun->pcpu_stats);
2019         u64_stats_update_begin(&stats->syncp);
2020         stats->tx_packets++;
2021         stats->tx_bytes += ret;
2022         u64_stats_update_end(&stats->syncp);
2023         put_cpu_ptr(tun->pcpu_stats);
2024
2025         return ret;
2026 }
2027
2028 /* Put packet to the user space buffer */
2029 static ssize_t tun_put_user(struct tun_struct *tun,
2030                             struct tun_file *tfile,
2031                             struct sk_buff *skb,
2032                             struct iov_iter *iter)
2033 {
2034         struct tun_pi pi = { 0, skb->protocol };
2035         struct tun_pcpu_stats *stats;
2036         ssize_t total;
2037         int vlan_offset = 0;
2038         int vlan_hlen = 0;
2039         int vnet_hdr_sz = 0;
2040
2041         if (skb_vlan_tag_present(skb))
2042                 vlan_hlen = VLAN_HLEN;
2043
2044         if (tun->flags & IFF_VNET_HDR)
2045                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2046
2047         total = skb->len + vlan_hlen + vnet_hdr_sz;
2048
2049         if (!(tun->flags & IFF_NO_PI)) {
2050                 if (iov_iter_count(iter) < sizeof(pi))
2051                         return -EINVAL;
2052
2053                 total += sizeof(pi);
2054                 if (iov_iter_count(iter) < total) {
2055                         /* Packet will be striped */
2056                         pi.flags |= TUN_PKT_STRIP;
2057                 }
2058
2059                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2060                         return -EFAULT;
2061         }
2062
2063         if (vnet_hdr_sz) {
2064                 struct virtio_net_hdr gso;
2065
2066                 if (iov_iter_count(iter) < vnet_hdr_sz)
2067                         return -EINVAL;
2068
2069                 if (virtio_net_hdr_from_skb(skb, &gso,
2070                                             tun_is_little_endian(tun), true)) {
2071                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2072                         pr_err("unexpected GSO type: "
2073                                "0x%x, gso_size %d, hdr_len %d\n",
2074                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2075                                tun16_to_cpu(tun, gso.hdr_len));
2076                         print_hex_dump(KERN_ERR, "tun: ",
2077                                        DUMP_PREFIX_NONE,
2078                                        16, 1, skb->head,
2079                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2080                         WARN_ON_ONCE(1);
2081                         return -EINVAL;
2082                 }
2083
2084                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2085                         return -EFAULT;
2086
2087                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2088         }
2089
2090         if (vlan_hlen) {
2091                 int ret;
2092                 struct veth veth;
2093
2094                 veth.h_vlan_proto = skb->vlan_proto;
2095                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2096
2097                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2098
2099                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2100                 if (ret || !iov_iter_count(iter))
2101                         goto done;
2102
2103                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2104                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2105                         goto done;
2106         }
2107
2108         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2109
2110 done:
2111         /* caller is in process context, */
2112         stats = get_cpu_ptr(tun->pcpu_stats);
2113         u64_stats_update_begin(&stats->syncp);
2114         stats->tx_packets++;
2115         stats->tx_bytes += skb->len + vlan_hlen;
2116         u64_stats_update_end(&stats->syncp);
2117         put_cpu_ptr(tun->pcpu_stats);
2118
2119         return total;
2120 }
2121
2122 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2123 {
2124         DECLARE_WAITQUEUE(wait, current);
2125         void *ptr = NULL;
2126         int error = 0;
2127
2128         ptr = ptr_ring_consume(&tfile->tx_ring);
2129         if (ptr)
2130                 goto out;
2131         if (noblock) {
2132                 error = -EAGAIN;
2133                 goto out;
2134         }
2135
2136         add_wait_queue(&tfile->wq.wait, &wait);
2137         current->state = TASK_INTERRUPTIBLE;
2138
2139         while (1) {
2140                 ptr = ptr_ring_consume(&tfile->tx_ring);
2141                 if (ptr)
2142                         break;
2143                 if (signal_pending(current)) {
2144                         error = -ERESTARTSYS;
2145                         break;
2146                 }
2147                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2148                         error = -EFAULT;
2149                         break;
2150                 }
2151
2152                 schedule();
2153         }
2154
2155         current->state = TASK_RUNNING;
2156         remove_wait_queue(&tfile->wq.wait, &wait);
2157
2158 out:
2159         *err = error;
2160         return ptr;
2161 }
2162
2163 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2164                            struct iov_iter *to,
2165                            int noblock, void *ptr)
2166 {
2167         ssize_t ret;
2168         int err;
2169
2170         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2171
2172         if (!iov_iter_count(to)) {
2173                 tun_ptr_free(ptr);
2174                 return 0;
2175         }
2176
2177         if (!ptr) {
2178                 /* Read frames from ring */
2179                 ptr = tun_ring_recv(tfile, noblock, &err);
2180                 if (!ptr)
2181                         return err;
2182         }
2183
2184         if (tun_is_xdp_buff(ptr)) {
2185                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2186
2187                 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2188                 put_page(virt_to_head_page(xdp->data));
2189         } else {
2190                 struct sk_buff *skb = ptr;
2191
2192                 ret = tun_put_user(tun, tfile, skb, to);
2193                 if (unlikely(ret < 0))
2194                         kfree_skb(skb);
2195                 else
2196                         consume_skb(skb);
2197         }
2198
2199         return ret;
2200 }
2201
2202 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2203 {
2204         struct file *file = iocb->ki_filp;
2205         struct tun_file *tfile = file->private_data;
2206         struct tun_struct *tun = tun_get(tfile);
2207         ssize_t len = iov_iter_count(to), ret;
2208
2209         if (!tun)
2210                 return -EBADFD;
2211         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2212         ret = min_t(ssize_t, ret, len);
2213         if (ret > 0)
2214                 iocb->ki_pos = ret;
2215         tun_put(tun);
2216         return ret;
2217 }
2218
2219 static void tun_prog_free(struct rcu_head *rcu)
2220 {
2221         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2222
2223         bpf_prog_destroy(prog->prog);
2224         kfree(prog);
2225 }
2226
2227 static int __tun_set_ebpf(struct tun_struct *tun,
2228                           struct tun_prog __rcu **prog_p,
2229                           struct bpf_prog *prog)
2230 {
2231         struct tun_prog *old, *new = NULL;
2232
2233         if (prog) {
2234                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2235                 if (!new)
2236                         return -ENOMEM;
2237                 new->prog = prog;
2238         }
2239
2240         spin_lock_bh(&tun->lock);
2241         old = rcu_dereference_protected(*prog_p,
2242                                         lockdep_is_held(&tun->lock));
2243         rcu_assign_pointer(*prog_p, new);
2244         spin_unlock_bh(&tun->lock);
2245
2246         if (old)
2247                 call_rcu(&old->rcu, tun_prog_free);
2248
2249         return 0;
2250 }
2251
2252 static void tun_free_netdev(struct net_device *dev)
2253 {
2254         struct tun_struct *tun = netdev_priv(dev);
2255
2256         BUG_ON(!(list_empty(&tun->disabled)));
2257         free_percpu(tun->pcpu_stats);
2258         tun_flow_uninit(tun);
2259         security_tun_dev_free_security(tun->security);
2260         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2261         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2262 }
2263
2264 static void tun_setup(struct net_device *dev)
2265 {
2266         struct tun_struct *tun = netdev_priv(dev);
2267
2268         tun->owner = INVALID_UID;
2269         tun->group = INVALID_GID;
2270
2271         dev->ethtool_ops = &tun_ethtool_ops;
2272         dev->needs_free_netdev = true;
2273         dev->priv_destructor = tun_free_netdev;
2274         /* We prefer our own queue length */
2275         dev->tx_queue_len = TUN_READQ_SIZE;
2276 }
2277
2278 /* Trivial set of netlink ops to allow deleting tun or tap
2279  * device with netlink.
2280  */
2281 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2282                         struct netlink_ext_ack *extack)
2283 {
2284         return -EINVAL;
2285 }
2286
2287 static size_t tun_get_size(const struct net_device *dev)
2288 {
2289         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2290         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2291
2292         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2293                nla_total_size(sizeof(gid_t)) + /* GROUP */
2294                nla_total_size(sizeof(u8)) + /* TYPE */
2295                nla_total_size(sizeof(u8)) + /* PI */
2296                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2297                nla_total_size(sizeof(u8)) + /* PERSIST */
2298                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2299                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2300                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2301                0;
2302 }
2303
2304 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2305 {
2306         struct tun_struct *tun = netdev_priv(dev);
2307
2308         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2309                 goto nla_put_failure;
2310         if (uid_valid(tun->owner) &&
2311             nla_put_u32(skb, IFLA_TUN_OWNER,
2312                         from_kuid_munged(current_user_ns(), tun->owner)))
2313                 goto nla_put_failure;
2314         if (gid_valid(tun->group) &&
2315             nla_put_u32(skb, IFLA_TUN_GROUP,
2316                         from_kgid_munged(current_user_ns(), tun->group)))
2317                 goto nla_put_failure;
2318         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2319                 goto nla_put_failure;
2320         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2321                 goto nla_put_failure;
2322         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2323                 goto nla_put_failure;
2324         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2325                        !!(tun->flags & IFF_MULTI_QUEUE)))
2326                 goto nla_put_failure;
2327         if (tun->flags & IFF_MULTI_QUEUE) {
2328                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2329                         goto nla_put_failure;
2330                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2331                                 tun->numdisabled))
2332                         goto nla_put_failure;
2333         }
2334
2335         return 0;
2336
2337 nla_put_failure:
2338         return -EMSGSIZE;
2339 }
2340
2341 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2342         .kind           = DRV_NAME,
2343         .priv_size      = sizeof(struct tun_struct),
2344         .setup          = tun_setup,
2345         .validate       = tun_validate,
2346         .get_size       = tun_get_size,
2347         .fill_info      = tun_fill_info,
2348 };
2349
2350 static void tun_sock_write_space(struct sock *sk)
2351 {
2352         struct tun_file *tfile;
2353         wait_queue_head_t *wqueue;
2354
2355         if (!sock_writeable(sk))
2356                 return;
2357
2358         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2359                 return;
2360
2361         wqueue = sk_sleep(sk);
2362         if (wqueue && waitqueue_active(wqueue))
2363                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2364                                                 EPOLLWRNORM | EPOLLWRBAND);
2365
2366         tfile = container_of(sk, struct tun_file, sk);
2367         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2368 }
2369
2370 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2371 {
2372         int ret;
2373         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2374         struct tun_struct *tun = tun_get(tfile);
2375
2376         if (!tun)
2377                 return -EBADFD;
2378
2379         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2380                            m->msg_flags & MSG_DONTWAIT,
2381                            m->msg_flags & MSG_MORE);
2382         tun_put(tun);
2383         return ret;
2384 }
2385
2386 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2387                        int flags)
2388 {
2389         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2390         struct tun_struct *tun = tun_get(tfile);
2391         void *ptr = m->msg_control;
2392         int ret;
2393
2394         if (!tun) {
2395                 ret = -EBADFD;
2396                 goto out_free;
2397         }
2398
2399         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2400                 ret = -EINVAL;
2401                 goto out_put_tun;
2402         }
2403         if (flags & MSG_ERRQUEUE) {
2404                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2405                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2406                 goto out;
2407         }
2408         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2409         if (ret > (ssize_t)total_len) {
2410                 m->msg_flags |= MSG_TRUNC;
2411                 ret = flags & MSG_TRUNC ? ret : total_len;
2412         }
2413 out:
2414         tun_put(tun);
2415         return ret;
2416
2417 out_put_tun:
2418         tun_put(tun);
2419 out_free:
2420         tun_ptr_free(ptr);
2421         return ret;
2422 }
2423
2424 static int tun_ptr_peek_len(void *ptr)
2425 {
2426         if (likely(ptr)) {
2427                 if (tun_is_xdp_buff(ptr)) {
2428                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2429
2430                         return xdp->data_end - xdp->data;
2431                 }
2432                 return __skb_array_len_with_tag(ptr);
2433         } else {
2434                 return 0;
2435         }
2436 }
2437
2438 static int tun_peek_len(struct socket *sock)
2439 {
2440         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2441         struct tun_struct *tun;
2442         int ret = 0;
2443
2444         tun = tun_get(tfile);
2445         if (!tun)
2446                 return 0;
2447
2448         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2449         tun_put(tun);
2450
2451         return ret;
2452 }
2453
2454 /* Ops structure to mimic raw sockets with tun */
2455 static const struct proto_ops tun_socket_ops = {
2456         .peek_len = tun_peek_len,
2457         .sendmsg = tun_sendmsg,
2458         .recvmsg = tun_recvmsg,
2459 };
2460
2461 static struct proto tun_proto = {
2462         .name           = "tun",
2463         .owner          = THIS_MODULE,
2464         .obj_size       = sizeof(struct tun_file),
2465 };
2466
2467 static int tun_flags(struct tun_struct *tun)
2468 {
2469         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2470 }
2471
2472 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2473                               char *buf)
2474 {
2475         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2476         return sprintf(buf, "0x%x\n", tun_flags(tun));
2477 }
2478
2479 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2480                               char *buf)
2481 {
2482         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2483         return uid_valid(tun->owner)?
2484                 sprintf(buf, "%u\n",
2485                         from_kuid_munged(current_user_ns(), tun->owner)):
2486                 sprintf(buf, "-1\n");
2487 }
2488
2489 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2490                               char *buf)
2491 {
2492         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2493         return gid_valid(tun->group) ?
2494                 sprintf(buf, "%u\n",
2495                         from_kgid_munged(current_user_ns(), tun->group)):
2496                 sprintf(buf, "-1\n");
2497 }
2498
2499 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2500 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2501 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2502
2503 static struct attribute *tun_dev_attrs[] = {
2504         &dev_attr_tun_flags.attr,
2505         &dev_attr_owner.attr,
2506         &dev_attr_group.attr,
2507         NULL
2508 };
2509
2510 static const struct attribute_group tun_attr_group = {
2511         .attrs = tun_dev_attrs
2512 };
2513
2514 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2515 {
2516         struct tun_struct *tun;
2517         struct tun_file *tfile = file->private_data;
2518         struct net_device *dev;
2519         int err;
2520
2521         if (tfile->detached)
2522                 return -EINVAL;
2523
2524         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2525                 if (!capable(CAP_NET_ADMIN))
2526                         return -EPERM;
2527
2528                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2529                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2530                         return -EINVAL;
2531         }
2532
2533         dev = __dev_get_by_name(net, ifr->ifr_name);
2534         if (dev) {
2535                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2536                         return -EBUSY;
2537                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2538                         tun = netdev_priv(dev);
2539                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2540                         tun = netdev_priv(dev);
2541                 else
2542                         return -EINVAL;
2543
2544                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2545                     !!(tun->flags & IFF_MULTI_QUEUE))
2546                         return -EINVAL;
2547
2548                 if (tun_not_capable(tun))
2549                         return -EPERM;
2550                 err = security_tun_dev_open(tun->security);
2551                 if (err < 0)
2552                         return err;
2553
2554                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2555                                  ifr->ifr_flags & IFF_NAPI);
2556                 if (err < 0)
2557                         return err;
2558
2559                 if (tun->flags & IFF_MULTI_QUEUE &&
2560                     (tun->numqueues + tun->numdisabled > 1)) {
2561                         /* One or more queue has already been attached, no need
2562                          * to initialize the device again.
2563                          */
2564                         return 0;
2565                 }
2566         }
2567         else {
2568                 char *name;
2569                 unsigned long flags = 0;
2570                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2571                              MAX_TAP_QUEUES : 1;
2572
2573                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2574                         return -EPERM;
2575                 err = security_tun_dev_create();
2576                 if (err < 0)
2577                         return err;
2578
2579                 /* Set dev type */
2580                 if (ifr->ifr_flags & IFF_TUN) {
2581                         /* TUN device */
2582                         flags |= IFF_TUN;
2583                         name = "tun%d";
2584                 } else if (ifr->ifr_flags & IFF_TAP) {
2585                         /* TAP device */
2586                         flags |= IFF_TAP;
2587                         name = "tap%d";
2588                 } else
2589                         return -EINVAL;
2590
2591                 if (*ifr->ifr_name)
2592                         name = ifr->ifr_name;
2593
2594                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2595                                        NET_NAME_UNKNOWN, tun_setup, queues,
2596                                        queues);
2597
2598                 if (!dev)
2599                         return -ENOMEM;
2600                 err = dev_get_valid_name(net, dev, name);
2601                 if (err < 0)
2602                         goto err_free_dev;
2603
2604                 dev_net_set(dev, net);
2605                 dev->rtnl_link_ops = &tun_link_ops;
2606                 dev->ifindex = tfile->ifindex;
2607                 dev->sysfs_groups[0] = &tun_attr_group;
2608
2609                 tun = netdev_priv(dev);
2610                 tun->dev = dev;
2611                 tun->flags = flags;
2612                 tun->txflt.count = 0;
2613                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2614
2615                 tun->align = NET_SKB_PAD;
2616                 tun->filter_attached = false;
2617                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2618                 tun->rx_batched = 0;
2619                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2620
2621                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2622                 if (!tun->pcpu_stats) {
2623                         err = -ENOMEM;
2624                         goto err_free_dev;
2625                 }
2626
2627                 spin_lock_init(&tun->lock);
2628
2629                 err = security_tun_dev_alloc_security(&tun->security);
2630                 if (err < 0)
2631                         goto err_free_stat;
2632
2633                 tun_net_init(dev);
2634                 tun_flow_init(tun);
2635
2636                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2637                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2638                                    NETIF_F_HW_VLAN_STAG_TX;
2639                 dev->features = dev->hw_features | NETIF_F_LLTX;
2640                 dev->vlan_features = dev->features &
2641                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2642                                        NETIF_F_HW_VLAN_STAG_TX);
2643
2644                 INIT_LIST_HEAD(&tun->disabled);
2645                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2646                 if (err < 0)
2647                         goto err_free_flow;
2648
2649                 err = register_netdevice(tun->dev);
2650                 if (err < 0)
2651                         goto err_detach;
2652         }
2653
2654         netif_carrier_on(tun->dev);
2655
2656         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2657
2658         tun->flags = (tun->flags & ~TUN_FEATURES) |
2659                 (ifr->ifr_flags & TUN_FEATURES);
2660
2661         /* Make sure persistent devices do not get stuck in
2662          * xoff state.
2663          */
2664         if (netif_running(tun->dev))
2665                 netif_tx_wake_all_queues(tun->dev);
2666
2667         strcpy(ifr->ifr_name, tun->dev->name);
2668         return 0;
2669
2670 err_detach:
2671         tun_detach_all(dev);
2672         /* register_netdevice() already called tun_free_netdev() */
2673         goto err_free_dev;
2674
2675 err_free_flow:
2676         tun_flow_uninit(tun);
2677         security_tun_dev_free_security(tun->security);
2678 err_free_stat:
2679         free_percpu(tun->pcpu_stats);
2680 err_free_dev:
2681         free_netdev(dev);
2682         return err;
2683 }
2684
2685 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2686                        struct ifreq *ifr)
2687 {
2688         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2689
2690         strcpy(ifr->ifr_name, tun->dev->name);
2691
2692         ifr->ifr_flags = tun_flags(tun);
2693
2694 }
2695
2696 /* This is like a cut-down ethtool ops, except done via tun fd so no
2697  * privs required. */
2698 static int set_offload(struct tun_struct *tun, unsigned long arg)
2699 {
2700         netdev_features_t features = 0;
2701
2702         if (arg & TUN_F_CSUM) {
2703                 features |= NETIF_F_HW_CSUM;
2704                 arg &= ~TUN_F_CSUM;
2705
2706                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2707                         if (arg & TUN_F_TSO_ECN) {
2708                                 features |= NETIF_F_TSO_ECN;
2709                                 arg &= ~TUN_F_TSO_ECN;
2710                         }
2711                         if (arg & TUN_F_TSO4)
2712                                 features |= NETIF_F_TSO;
2713                         if (arg & TUN_F_TSO6)
2714                                 features |= NETIF_F_TSO6;
2715                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2716                 }
2717
2718                 arg &= ~TUN_F_UFO;
2719         }
2720
2721         /* This gives the user a way to test for new features in future by
2722          * trying to set them. */
2723         if (arg)
2724                 return -EINVAL;
2725
2726         tun->set_features = features;
2727         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2728         tun->dev->wanted_features |= features;
2729         netdev_update_features(tun->dev);
2730
2731         return 0;
2732 }
2733
2734 static void tun_detach_filter(struct tun_struct *tun, int n)
2735 {
2736         int i;
2737         struct tun_file *tfile;
2738
2739         for (i = 0; i < n; i++) {
2740                 tfile = rtnl_dereference(tun->tfiles[i]);
2741                 lock_sock(tfile->socket.sk);
2742                 sk_detach_filter(tfile->socket.sk);
2743                 release_sock(tfile->socket.sk);
2744         }
2745
2746         tun->filter_attached = false;
2747 }
2748
2749 static int tun_attach_filter(struct tun_struct *tun)
2750 {
2751         int i, ret = 0;
2752         struct tun_file *tfile;
2753
2754         for (i = 0; i < tun->numqueues; i++) {
2755                 tfile = rtnl_dereference(tun->tfiles[i]);
2756                 lock_sock(tfile->socket.sk);
2757                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2758                 release_sock(tfile->socket.sk);
2759                 if (ret) {
2760                         tun_detach_filter(tun, i);
2761                         return ret;
2762                 }
2763         }
2764
2765         tun->filter_attached = true;
2766         return ret;
2767 }
2768
2769 static void tun_set_sndbuf(struct tun_struct *tun)
2770 {
2771         struct tun_file *tfile;
2772         int i;
2773
2774         for (i = 0; i < tun->numqueues; i++) {
2775                 tfile = rtnl_dereference(tun->tfiles[i]);
2776                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2777         }
2778 }
2779
2780 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2781 {
2782         struct tun_file *tfile = file->private_data;
2783         struct tun_struct *tun;
2784         int ret = 0;
2785
2786         rtnl_lock();
2787
2788         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2789                 tun = tfile->detached;
2790                 if (!tun) {
2791                         ret = -EINVAL;
2792                         goto unlock;
2793                 }
2794                 ret = security_tun_dev_attach_queue(tun->security);
2795                 if (ret < 0)
2796                         goto unlock;
2797                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2798         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2799                 tun = rtnl_dereference(tfile->tun);
2800                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2801                         ret = -EINVAL;
2802                 else
2803                         __tun_detach(tfile, false);
2804         } else
2805                 ret = -EINVAL;
2806
2807 unlock:
2808         rtnl_unlock();
2809         return ret;
2810 }
2811
2812 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2813                         void __user *data)
2814 {
2815         struct bpf_prog *prog;
2816         int fd;
2817
2818         if (copy_from_user(&fd, data, sizeof(fd)))
2819                 return -EFAULT;
2820
2821         if (fd == -1) {
2822                 prog = NULL;
2823         } else {
2824                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2825                 if (IS_ERR(prog))
2826                         return PTR_ERR(prog);
2827         }
2828
2829         return __tun_set_ebpf(tun, prog_p, prog);
2830 }
2831
2832 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2833                             unsigned long arg, int ifreq_len)
2834 {
2835         struct tun_file *tfile = file->private_data;
2836         struct tun_struct *tun;
2837         void __user* argp = (void __user*)arg;
2838         struct ifreq ifr;
2839         struct net *net;
2840         kuid_t owner;
2841         kgid_t group;
2842         int sndbuf;
2843         int vnet_hdr_sz;
2844         unsigned int ifindex;
2845         int le;
2846         int ret;
2847
2848         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2849             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2850                 if (copy_from_user(&ifr, argp, ifreq_len))
2851                         return -EFAULT;
2852         } else {
2853                 memset(&ifr, 0, sizeof(ifr));
2854         }
2855         if (cmd == TUNGETFEATURES) {
2856                 /* Currently this just means: "what IFF flags are valid?".
2857                  * This is needed because we never checked for invalid flags on
2858                  * TUNSETIFF.
2859                  */
2860                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2861                                 (unsigned int __user*)argp);
2862         } else if (cmd == TUNSETQUEUE)
2863                 return tun_set_queue(file, &ifr);
2864
2865         ret = 0;
2866         rtnl_lock();
2867
2868         tun = tun_get(tfile);
2869         net = sock_net(&tfile->sk);
2870         if (cmd == TUNSETIFF) {
2871                 ret = -EEXIST;
2872                 if (tun)
2873                         goto unlock;
2874
2875                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2876
2877                 ret = tun_set_iff(net, file, &ifr);
2878
2879                 if (ret)
2880                         goto unlock;
2881
2882                 if (copy_to_user(argp, &ifr, ifreq_len))
2883                         ret = -EFAULT;
2884                 goto unlock;
2885         }
2886         if (cmd == TUNSETIFINDEX) {
2887                 ret = -EPERM;
2888                 if (tun)
2889                         goto unlock;
2890
2891                 ret = -EFAULT;
2892                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2893                         goto unlock;
2894
2895                 ret = 0;
2896                 tfile->ifindex = ifindex;
2897                 goto unlock;
2898         }
2899         if (cmd == SIOCGSKNS) {
2900                 ret = -EPERM;
2901                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2902                         goto unlock;
2903
2904                 ret = open_related_ns(&net->ns, get_net_ns);
2905                 goto unlock;
2906         }
2907
2908         ret = -EBADFD;
2909         if (!tun)
2910                 goto unlock;
2911
2912         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2913
2914         ret = 0;
2915         switch (cmd) {
2916         case TUNGETIFF:
2917                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2918
2919                 if (tfile->detached)
2920                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2921                 if (!tfile->socket.sk->sk_filter)
2922                         ifr.ifr_flags |= IFF_NOFILTER;
2923
2924                 if (copy_to_user(argp, &ifr, ifreq_len))
2925                         ret = -EFAULT;
2926                 break;
2927
2928         case TUNSETNOCSUM:
2929                 /* Disable/Enable checksum */
2930
2931                 /* [unimplemented] */
2932                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2933                           arg ? "disabled" : "enabled");
2934                 break;
2935
2936         case TUNSETPERSIST:
2937                 /* Disable/Enable persist mode. Keep an extra reference to the
2938                  * module to prevent the module being unprobed.
2939                  */
2940                 if (arg && !(tun->flags & IFF_PERSIST)) {
2941                         tun->flags |= IFF_PERSIST;
2942                         __module_get(THIS_MODULE);
2943                 }
2944                 if (!arg && (tun->flags & IFF_PERSIST)) {
2945                         tun->flags &= ~IFF_PERSIST;
2946                         module_put(THIS_MODULE);
2947                 }
2948
2949                 tun_debug(KERN_INFO, tun, "persist %s\n",
2950                           arg ? "enabled" : "disabled");
2951                 break;
2952
2953         case TUNSETOWNER:
2954                 /* Set owner of the device */
2955                 owner = make_kuid(current_user_ns(), arg);
2956                 if (!uid_valid(owner)) {
2957                         ret = -EINVAL;
2958                         break;
2959                 }
2960                 tun->owner = owner;
2961                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2962                           from_kuid(&init_user_ns, tun->owner));
2963                 break;
2964
2965         case TUNSETGROUP:
2966                 /* Set group of the device */
2967                 group = make_kgid(current_user_ns(), arg);
2968                 if (!gid_valid(group)) {
2969                         ret = -EINVAL;
2970                         break;
2971                 }
2972                 tun->group = group;
2973                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2974                           from_kgid(&init_user_ns, tun->group));
2975                 break;
2976
2977         case TUNSETLINK:
2978                 /* Only allow setting the type when the interface is down */
2979                 if (tun->dev->flags & IFF_UP) {
2980                         tun_debug(KERN_INFO, tun,
2981                                   "Linktype set failed because interface is up\n");
2982                         ret = -EBUSY;
2983                 } else {
2984                         tun->dev->type = (int) arg;
2985                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2986                                   tun->dev->type);
2987                         ret = 0;
2988                 }
2989                 break;
2990
2991 #ifdef TUN_DEBUG
2992         case TUNSETDEBUG:
2993                 tun->debug = arg;
2994                 break;
2995 #endif
2996         case TUNSETOFFLOAD:
2997                 ret = set_offload(tun, arg);
2998                 break;
2999
3000         case TUNSETTXFILTER:
3001                 /* Can be set only for TAPs */
3002                 ret = -EINVAL;
3003                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3004                         break;
3005                 ret = update_filter(&tun->txflt, (void __user *)arg);
3006                 break;
3007
3008         case SIOCGIFHWADDR:
3009                 /* Get hw address */
3010                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3011                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3012                 if (copy_to_user(argp, &ifr, ifreq_len))
3013                         ret = -EFAULT;
3014                 break;
3015
3016         case SIOCSIFHWADDR:
3017                 /* Set hw address */
3018                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3019                           ifr.ifr_hwaddr.sa_data);
3020
3021                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3022                 break;
3023
3024         case TUNGETSNDBUF:
3025                 sndbuf = tfile->socket.sk->sk_sndbuf;
3026                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3027                         ret = -EFAULT;
3028                 break;
3029
3030         case TUNSETSNDBUF:
3031                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3032                         ret = -EFAULT;
3033                         break;
3034                 }
3035                 if (sndbuf <= 0) {
3036                         ret = -EINVAL;
3037                         break;
3038                 }
3039
3040                 tun->sndbuf = sndbuf;
3041                 tun_set_sndbuf(tun);
3042                 break;
3043
3044         case TUNGETVNETHDRSZ:
3045                 vnet_hdr_sz = tun->vnet_hdr_sz;
3046                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3047                         ret = -EFAULT;
3048                 break;
3049
3050         case TUNSETVNETHDRSZ:
3051                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3052                         ret = -EFAULT;
3053                         break;
3054                 }
3055                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3056                         ret = -EINVAL;
3057                         break;
3058                 }
3059
3060                 tun->vnet_hdr_sz = vnet_hdr_sz;
3061                 break;
3062
3063         case TUNGETVNETLE:
3064                 le = !!(tun->flags & TUN_VNET_LE);
3065                 if (put_user(le, (int __user *)argp))
3066                         ret = -EFAULT;
3067                 break;
3068
3069         case TUNSETVNETLE:
3070                 if (get_user(le, (int __user *)argp)) {
3071                         ret = -EFAULT;
3072                         break;
3073                 }
3074                 if (le)
3075                         tun->flags |= TUN_VNET_LE;
3076                 else
3077                         tun->flags &= ~TUN_VNET_LE;
3078                 break;
3079
3080         case TUNGETVNETBE:
3081                 ret = tun_get_vnet_be(tun, argp);
3082                 break;
3083
3084         case TUNSETVNETBE:
3085                 ret = tun_set_vnet_be(tun, argp);
3086                 break;
3087
3088         case TUNATTACHFILTER:
3089                 /* Can be set only for TAPs */
3090                 ret = -EINVAL;
3091                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3092                         break;
3093                 ret = -EFAULT;
3094                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3095                         break;
3096
3097                 ret = tun_attach_filter(tun);
3098                 break;
3099
3100         case TUNDETACHFILTER:
3101                 /* Can be set only for TAPs */
3102                 ret = -EINVAL;
3103                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3104                         break;
3105                 ret = 0;
3106                 tun_detach_filter(tun, tun->numqueues);
3107                 break;
3108
3109         case TUNGETFILTER:
3110                 ret = -EINVAL;
3111                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3112                         break;
3113                 ret = -EFAULT;
3114                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3115                         break;
3116                 ret = 0;
3117                 break;
3118
3119         case TUNSETSTEERINGEBPF:
3120                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3121                 break;
3122
3123         case TUNSETFILTEREBPF:
3124                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3125                 break;
3126
3127         default:
3128                 ret = -EINVAL;
3129                 break;
3130         }
3131
3132 unlock:
3133         rtnl_unlock();
3134         if (tun)
3135                 tun_put(tun);
3136         return ret;
3137 }
3138
3139 static long tun_chr_ioctl(struct file *file,
3140                           unsigned int cmd, unsigned long arg)
3141 {
3142         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3143 }
3144
3145 #ifdef CONFIG_COMPAT
3146 static long tun_chr_compat_ioctl(struct file *file,
3147                          unsigned int cmd, unsigned long arg)
3148 {
3149         switch (cmd) {
3150         case TUNSETIFF:
3151         case TUNGETIFF:
3152         case TUNSETTXFILTER:
3153         case TUNGETSNDBUF:
3154         case TUNSETSNDBUF:
3155         case SIOCGIFHWADDR:
3156         case SIOCSIFHWADDR:
3157                 arg = (unsigned long)compat_ptr(arg);
3158                 break;
3159         default:
3160                 arg = (compat_ulong_t)arg;
3161                 break;
3162         }
3163
3164         /*
3165          * compat_ifreq is shorter than ifreq, so we must not access beyond
3166          * the end of that structure. All fields that are used in this
3167          * driver are compatible though, we don't need to convert the
3168          * contents.
3169          */
3170         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3171 }
3172 #endif /* CONFIG_COMPAT */
3173
3174 static int tun_chr_fasync(int fd, struct file *file, int on)
3175 {
3176         struct tun_file *tfile = file->private_data;
3177         int ret;
3178
3179         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3180                 goto out;
3181
3182         if (on) {
3183                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3184                 tfile->flags |= TUN_FASYNC;
3185         } else
3186                 tfile->flags &= ~TUN_FASYNC;
3187         ret = 0;
3188 out:
3189         return ret;
3190 }
3191
3192 static int tun_chr_open(struct inode *inode, struct file * file)
3193 {
3194         struct net *net = current->nsproxy->net_ns;
3195         struct tun_file *tfile;
3196
3197         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3198
3199         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3200                                             &tun_proto, 0);
3201         if (!tfile)
3202                 return -ENOMEM;
3203         RCU_INIT_POINTER(tfile->tun, NULL);
3204         tfile->flags = 0;
3205         tfile->ifindex = 0;
3206
3207         init_waitqueue_head(&tfile->wq.wait);
3208         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3209
3210         tfile->socket.file = file;
3211         tfile->socket.ops = &tun_socket_ops;
3212
3213         sock_init_data(&tfile->socket, &tfile->sk);
3214
3215         tfile->sk.sk_write_space = tun_sock_write_space;
3216         tfile->sk.sk_sndbuf = INT_MAX;
3217
3218         file->private_data = tfile;
3219         INIT_LIST_HEAD(&tfile->next);
3220
3221         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3222
3223         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3224
3225         return 0;
3226 }
3227
3228 static int tun_chr_close(struct inode *inode, struct file *file)
3229 {
3230         struct tun_file *tfile = file->private_data;
3231
3232         tun_detach(tfile, true);
3233
3234         return 0;
3235 }
3236
3237 #ifdef CONFIG_PROC_FS
3238 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3239 {
3240         struct tun_file *tfile = file->private_data;
3241         struct tun_struct *tun;
3242         struct ifreq ifr;
3243
3244         memset(&ifr, 0, sizeof(ifr));
3245
3246         rtnl_lock();
3247         tun = tun_get(tfile);
3248         if (tun)
3249                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3250         rtnl_unlock();
3251
3252         if (tun)
3253                 tun_put(tun);
3254
3255         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3256 }
3257 #endif
3258
3259 static const struct file_operations tun_fops = {
3260         .owner  = THIS_MODULE,
3261         .llseek = no_llseek,
3262         .read_iter  = tun_chr_read_iter,
3263         .write_iter = tun_chr_write_iter,
3264         .poll   = tun_chr_poll,
3265         .unlocked_ioctl = tun_chr_ioctl,
3266 #ifdef CONFIG_COMPAT
3267         .compat_ioctl = tun_chr_compat_ioctl,
3268 #endif
3269         .open   = tun_chr_open,
3270         .release = tun_chr_close,
3271         .fasync = tun_chr_fasync,
3272 #ifdef CONFIG_PROC_FS
3273         .show_fdinfo = tun_chr_show_fdinfo,
3274 #endif
3275 };
3276
3277 static struct miscdevice tun_miscdev = {
3278         .minor = TUN_MINOR,
3279         .name = "tun",
3280         .nodename = "net/tun",
3281         .fops = &tun_fops,
3282 };
3283
3284 /* ethtool interface */
3285
3286 static int tun_get_link_ksettings(struct net_device *dev,
3287                                   struct ethtool_link_ksettings *cmd)
3288 {
3289         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3290         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3291         cmd->base.speed         = SPEED_10;
3292         cmd->base.duplex        = DUPLEX_FULL;
3293         cmd->base.port          = PORT_TP;
3294         cmd->base.phy_address   = 0;
3295         cmd->base.autoneg       = AUTONEG_DISABLE;
3296         return 0;
3297 }
3298
3299 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3300 {
3301         struct tun_struct *tun = netdev_priv(dev);
3302
3303         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3304         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3305
3306         switch (tun->flags & TUN_TYPE_MASK) {
3307         case IFF_TUN:
3308                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3309                 break;
3310         case IFF_TAP:
3311                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3312                 break;
3313         }
3314 }
3315
3316 static u32 tun_get_msglevel(struct net_device *dev)
3317 {
3318 #ifdef TUN_DEBUG
3319         struct tun_struct *tun = netdev_priv(dev);
3320         return tun->debug;
3321 #else
3322         return -EOPNOTSUPP;
3323 #endif
3324 }
3325
3326 static void tun_set_msglevel(struct net_device *dev, u32 value)
3327 {
3328 #ifdef TUN_DEBUG
3329         struct tun_struct *tun = netdev_priv(dev);
3330         tun->debug = value;
3331 #endif
3332 }
3333
3334 static int tun_get_coalesce(struct net_device *dev,
3335                             struct ethtool_coalesce *ec)
3336 {
3337         struct tun_struct *tun = netdev_priv(dev);
3338
3339         ec->rx_max_coalesced_frames = tun->rx_batched;
3340
3341         return 0;
3342 }
3343
3344 static int tun_set_coalesce(struct net_device *dev,
3345                             struct ethtool_coalesce *ec)
3346 {
3347         struct tun_struct *tun = netdev_priv(dev);
3348
3349         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3350                 tun->rx_batched = NAPI_POLL_WEIGHT;
3351         else
3352                 tun->rx_batched = ec->rx_max_coalesced_frames;
3353
3354         return 0;
3355 }
3356
3357 static const struct ethtool_ops tun_ethtool_ops = {
3358         .get_drvinfo    = tun_get_drvinfo,
3359         .get_msglevel   = tun_get_msglevel,
3360         .set_msglevel   = tun_set_msglevel,
3361         .get_link       = ethtool_op_get_link,
3362         .get_ts_info    = ethtool_op_get_ts_info,
3363         .get_coalesce   = tun_get_coalesce,
3364         .set_coalesce   = tun_set_coalesce,
3365         .get_link_ksettings = tun_get_link_ksettings,
3366 };
3367
3368 static int tun_queue_resize(struct tun_struct *tun)
3369 {
3370         struct net_device *dev = tun->dev;
3371         struct tun_file *tfile;
3372         struct ptr_ring **rings;
3373         int n = tun->numqueues + tun->numdisabled;
3374         int ret, i;
3375
3376         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3377         if (!rings)
3378                 return -ENOMEM;
3379
3380         for (i = 0; i < tun->numqueues; i++) {
3381                 tfile = rtnl_dereference(tun->tfiles[i]);
3382                 rings[i] = &tfile->tx_ring;
3383         }
3384         list_for_each_entry(tfile, &tun->disabled, next)
3385                 rings[i++] = &tfile->tx_ring;
3386
3387         ret = ptr_ring_resize_multiple(rings, n,
3388                                        dev->tx_queue_len, GFP_KERNEL,
3389                                        tun_ptr_free);
3390
3391         kfree(rings);
3392         return ret;
3393 }
3394
3395 static int tun_device_event(struct notifier_block *unused,
3396                             unsigned long event, void *ptr)
3397 {
3398         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3399         struct tun_struct *tun = netdev_priv(dev);
3400
3401         if (dev->rtnl_link_ops != &tun_link_ops)
3402                 return NOTIFY_DONE;
3403
3404         switch (event) {
3405         case NETDEV_CHANGE_TX_QUEUE_LEN:
3406                 if (tun_queue_resize(tun))
3407                         return NOTIFY_BAD;
3408                 break;
3409         default:
3410                 break;
3411         }
3412
3413         return NOTIFY_DONE;
3414 }
3415
3416 static struct notifier_block tun_notifier_block __read_mostly = {
3417         .notifier_call  = tun_device_event,
3418 };
3419
3420 static int __init tun_init(void)
3421 {
3422         int ret = 0;
3423
3424         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3425
3426         ret = rtnl_link_register(&tun_link_ops);
3427         if (ret) {
3428                 pr_err("Can't register link_ops\n");
3429                 goto err_linkops;
3430         }
3431
3432         ret = misc_register(&tun_miscdev);
3433         if (ret) {
3434                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3435                 goto err_misc;
3436         }
3437
3438         ret = register_netdevice_notifier(&tun_notifier_block);
3439         if (ret) {
3440                 pr_err("Can't register netdevice notifier\n");
3441                 goto err_notifier;
3442         }
3443
3444         return  0;
3445
3446 err_notifier:
3447         misc_deregister(&tun_miscdev);
3448 err_misc:
3449         rtnl_link_unregister(&tun_link_ops);
3450 err_linkops:
3451         return ret;
3452 }
3453
3454 static void tun_cleanup(void)
3455 {
3456         misc_deregister(&tun_miscdev);
3457         rtnl_link_unregister(&tun_link_ops);
3458         unregister_netdevice_notifier(&tun_notifier_block);
3459 }
3460
3461 /* Get an underlying socket object from tun file.  Returns error unless file is
3462  * attached to a device.  The returned object works like a packet socket, it
3463  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3464  * holding a reference to the file for as long as the socket is in use. */
3465 struct socket *tun_get_socket(struct file *file)
3466 {
3467         struct tun_file *tfile;
3468         if (file->f_op != &tun_fops)
3469                 return ERR_PTR(-EINVAL);
3470         tfile = file->private_data;
3471         if (!tfile)
3472                 return ERR_PTR(-EBADFD);
3473         return &tfile->socket;
3474 }
3475 EXPORT_SYMBOL_GPL(tun_get_socket);
3476
3477 struct ptr_ring *tun_get_tx_ring(struct file *file)
3478 {
3479         struct tun_file *tfile;
3480
3481         if (file->f_op != &tun_fops)
3482                 return ERR_PTR(-EINVAL);
3483         tfile = file->private_data;
3484         if (!tfile)
3485                 return ERR_PTR(-EBADFD);
3486         return &tfile->tx_ring;
3487 }
3488 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3489
3490 module_init(tun_init);
3491 module_exit(tun_cleanup);
3492 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3493 MODULE_AUTHOR(DRV_COPYRIGHT);
3494 MODULE_LICENSE("GPL");
3495 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3496 MODULE_ALIAS("devname:net/tun");