drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_HEADROOM 256
 117 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 118
 119 /* TUN device flags */
 120
 121 /* IFF_ATTACH_QUEUE is never stored in device flags,
 122  * overload it to mean fasync when stored there.
 123  */
 124 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 125 /* High bits in flags field are unused. */
 126 #define TUN_VNET_LE     0x80000000
 127 #define TUN_VNET_BE     0x40000000
 128
 129 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 130                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 131
 132 #define GOODCOPY_LEN 128
 133
 134 #define FLT_EXACT_COUNT 8
 135 struct tap_filter {
 136         unsigned int    count;    /* Number of addrs. Zero means disabled */
 137         u32             mask[2];  /* Mask of the hashed addrs */
 138         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 139 };
 140
 141 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 142  * to max number of VCPUs in guest. */
 143 #define MAX_TAP_QUEUES 256
 144 #define MAX_TAP_FLOWS  4096
 145
 146 #define TUN_FLOW_EXPIRE (3 * HZ)
 147
 148 struct tun_pcpu_stats {
 149         u64 rx_packets;
 150         u64 rx_bytes;
 151         u64 tx_packets;
 152         u64 tx_bytes;
 153         struct u64_stats_sync syncp;
 154         u32 rx_dropped;
 155         u32 tx_dropped;
 156         u32 rx_frame_errors;
 157 };
 158
 159 /* A tun_file connects an open character device to a tuntap netdevice. It
 160  * also contains all socket related structures (except sock_fprog and tap_filter)
 161  * to serve as one transmit queue for tuntap device. The sock_fprog and
 162  * tap_filter were kept in tun_struct since they were used for filtering for the
 163  * netdevice not for a specific queue (at least I didn't see the requirement for
 164  * this).
 165  *
 166  * RCU usage:
 167  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 168  * other can only be read while rcu_read_lock or rtnl_lock is held.
 169  */
 170 struct tun_file {
 171         struct sock sk;
 172         struct socket socket;
 173         struct socket_wq wq;
 174         struct tun_struct __rcu *tun;
 175         struct fasync_struct *fasync;
 176         /* only used for fasnyc */
 177         unsigned int flags;
 178         union {
 179                 u16 queue_index;
 180                 unsigned int ifindex;
 181         };
 182         struct napi_struct napi;
 183         bool napi_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_flow_entry {
 192         struct hlist_node hash_link;
 193         struct rcu_head rcu;
 194         struct tun_struct *tun;
 195
 196         u32 rxhash;
 197         u32 rps_rxhash;
 198         int queue_index;
 199         unsigned long updated;
 200 };
 201
 202 #define TUN_NUM_FLOW_ENTRIES 1024
 203 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 204
 205 struct tun_prog {
 206         struct rcu_head rcu;
 207         struct bpf_prog *prog;
 208 };
 209
 210 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 211  * device, socket filter, sndbuf and vnet header size were restore when the
 212  * file were attached to a persist device.
 213  */
 214 struct tun_struct {
 215         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 216         unsigned int            numqueues;
 217         unsigned int            flags;
 218         kuid_t                  owner;
 219         kgid_t                  group;
 220
 221         struct net_device       *dev;
 222         netdev_features_t       set_features;
 223 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 224                           NETIF_F_TSO6)
 225
 226         int                     align;
 227         int                     vnet_hdr_sz;
 228         int                     sndbuf;
 229         struct tap_filter       txflt;
 230         struct sock_fprog       fprog;
 231         /* protected by rtnl lock */
 232         bool                    filter_attached;
 233 #ifdef TUN_DEBUG
 234         int debug;
 235 #endif
 236         spinlock_t lock;
 237         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 238         struct timer_list flow_gc_timer;
 239         unsigned long ageing_time;
 240         unsigned int numdisabled;
 241         struct list_head disabled;
 242         void *security;
 243         u32 flow_count;
 244         u32 rx_batched;
 245         struct tun_pcpu_stats __percpu *pcpu_stats;
 246         struct bpf_prog __rcu *xdp_prog;
 247         struct tun_prog __rcu *steering_prog;
 248         struct tun_prog __rcu *filter_prog;
 249         struct ethtool_link_ksettings link_ksettings;
 250 };
 251
 252 struct veth {
 253         __be16 h_vlan_proto;
 254         __be16 h_vlan_TCI;
 255 };
 256
 257 bool tun_is_xdp_frame(void *ptr)
 258 {
 259         return (unsigned long)ptr & TUN_XDP_FLAG;
 260 }
 261 EXPORT_SYMBOL(tun_is_xdp_frame);
 262
 263 void *tun_xdp_to_ptr(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_xdp_to_ptr);
 268
 269 void *tun_ptr_to_xdp(void *ptr)
 270 {
 271         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 272 }
 273 EXPORT_SYMBOL(tun_ptr_to_xdp);
 274
 275 static int tun_napi_receive(struct napi_struct *napi, int budget)
 276 {
 277         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 278         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 279         struct sk_buff_head process_queue;
 280         struct sk_buff *skb;
 281         int received = 0;
 282
 283         __skb_queue_head_init(&process_queue);
 284
 285         spin_lock(&queue->lock);
 286         skb_queue_splice_tail_init(queue, &process_queue);
 287         spin_unlock(&queue->lock);
 288
 289         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 290                 napi_gro_receive(napi, skb);
 291                 ++received;
 292         }
 293
 294         if (!skb_queue_empty(&process_queue)) {
 295                 spin_lock(&queue->lock);
 296                 skb_queue_splice(&process_queue, queue);
 297                 spin_unlock(&queue->lock);
 298         }
 299
 300         return received;
 301 }
 302
 303 static int tun_napi_poll(struct napi_struct *napi, int budget)
 304 {
 305         unsigned int received;
 306
 307         received = tun_napi_receive(napi, budget);
 308
 309         if (received < budget)
 310                 napi_complete_done(napi, received);
 311
 312         return received;
 313 }
 314
 315 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 316                           bool napi_en)
 317 {
 318         tfile->napi_enabled = napi_en;
 319         if (napi_en) {
 320                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 321                                NAPI_POLL_WEIGHT);
 322                 napi_enable(&tfile->napi);
 323                 mutex_init(&tfile->napi_mutex);
 324         }
 325 }
 326
 327 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 napi_disable(&tfile->napi);
 331 }
 332
 333 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 334 {
 335         if (tfile->napi_enabled)
 336                 netif_napi_del(&tfile->napi);
 337 }
 338
 339 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 340 {
 341         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 342 }
 343
 344 #ifdef CONFIG_TUN_VNET_CROSS_LE
 345 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 346 {
 347         return tun->flags & TUN_VNET_BE ? false :
 348                 virtio_legacy_is_little_endian();
 349 }
 350
 351 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 352 {
 353         int be = !!(tun->flags & TUN_VNET_BE);
 354
 355         if (put_user(be, argp))
 356                 return -EFAULT;
 357
 358         return 0;
 359 }
 360
 361 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 362 {
 363         int be;
 364
 365         if (get_user(be, argp))
 366                 return -EFAULT;
 367
 368         if (be)
 369                 tun->flags |= TUN_VNET_BE;
 370         else
 371                 tun->flags &= ~TUN_VNET_BE;
 372
 373         return 0;
 374 }
 375 #else
 376 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 377 {
 378         return virtio_legacy_is_little_endian();
 379 }
 380
 381 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 382 {
 383         return -EINVAL;
 384 }
 385
 386 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 387 {
 388         return -EINVAL;
 389 }
 390 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 391
 392 static inline bool tun_is_little_endian(struct tun_struct *tun)
 393 {
 394         return tun->flags & TUN_VNET_LE ||
 395                 tun_legacy_is_little_endian(tun);
 396 }
 397
 398 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 399 {
 400         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 401 }
 402
 403 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 404 {
 405         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 406 }
 407
 408 static inline u32 tun_hashfn(u32 rxhash)
 409 {
 410         return rxhash & TUN_MASK_FLOW_ENTRIES;
 411 }
 412
 413 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 414 {
 415         struct tun_flow_entry *e;
 416
 417         hlist_for_each_entry_rcu(e, head, hash_link) {
 418                 if (e->rxhash == rxhash)
 419                         return e;
 420         }
 421         return NULL;
 422 }
 423
 424 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 425                                               struct hlist_head *head,
 426                                               u32 rxhash, u16 queue_index)
 427 {
 428         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 429
 430         if (e) {
 431                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 432                           rxhash, queue_index);
 433                 e->updated = jiffies;
 434                 e->rxhash = rxhash;
 435                 e->rps_rxhash = 0;
 436                 e->queue_index = queue_index;
 437                 e->tun = tun;
 438                 hlist_add_head_rcu(&e->hash_link, head);
 439                 ++tun->flow_count;
 440         }
 441         return e;
 442 }
 443
 444 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 445 {
 446         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 447                   e->rxhash, e->queue_index);
 448         hlist_del_rcu(&e->hash_link);
 449         kfree_rcu(e, rcu);
 450         --tun->flow_count;
 451 }
 452
 453 static void tun_flow_flush(struct tun_struct *tun)
 454 {
 455         int i;
 456
 457         spin_lock_bh(&tun->lock);
 458         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 459                 struct tun_flow_entry *e;
 460                 struct hlist_node *n;
 461
 462                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 463                         tun_flow_delete(tun, e);
 464         }
 465         spin_unlock_bh(&tun->lock);
 466 }
 467
 468 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 469 {
 470         int i;
 471
 472         spin_lock_bh(&tun->lock);
 473         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 474                 struct tun_flow_entry *e;
 475                 struct hlist_node *n;
 476
 477                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 478                         if (e->queue_index == queue_index)
 479                                 tun_flow_delete(tun, e);
 480                 }
 481         }
 482         spin_unlock_bh(&tun->lock);
 483 }
 484
 485 static void tun_flow_cleanup(struct timer_list *t)
 486 {
 487         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 488         unsigned long delay = tun->ageing_time;
 489         unsigned long next_timer = jiffies + delay;
 490         unsigned long count = 0;
 491         int i;
 492
 493         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 494
 495         spin_lock(&tun->lock);
 496         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 497                 struct tun_flow_entry *e;
 498                 struct hlist_node *n;
 499
 500                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 501                         unsigned long this_timer;
 502
 503                         this_timer = e->updated + delay;
 504                         if (time_before_eq(this_timer, jiffies)) {
 505                                 tun_flow_delete(tun, e);
 506                                 continue;
 507                         }
 508                         count++;
 509                         if (time_before(this_timer, next_timer))
 510                                 next_timer = this_timer;
 511                 }
 512         }
 513
 514         if (count)
 515                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 516         spin_unlock(&tun->lock);
 517 }
 518
 519 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 520                             struct tun_file *tfile)
 521 {
 522         struct hlist_head *head;
 523         struct tun_flow_entry *e;
 524         unsigned long delay = tun->ageing_time;
 525         u16 queue_index = tfile->queue_index;
 526
 527         if (!rxhash)
 528                 return;
 529         else
 530                 head = &tun->flows[tun_hashfn(rxhash)];
 531
 532         rcu_read_lock();
 533
 534         e = tun_flow_find(head, rxhash);
 535         if (likely(e)) {
 536                 /* TODO: keep queueing to old queue until it's empty? */
 537                 e->queue_index = queue_index;
 538                 e->updated = jiffies;
 539                 sock_rps_record_flow_hash(e->rps_rxhash);
 540         } else {
 541                 spin_lock_bh(&tun->lock);
 542                 if (!tun_flow_find(head, rxhash) &&
 543                     tun->flow_count < MAX_TAP_FLOWS)
 544                         tun_flow_create(tun, head, rxhash, queue_index);
 545
 546                 if (!timer_pending(&tun->flow_gc_timer))
 547                         mod_timer(&tun->flow_gc_timer,
 548                                   round_jiffies_up(jiffies + delay));
 549                 spin_unlock_bh(&tun->lock);
 550         }
 551
 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                             struct net_device *sb_dev,
 612                             select_queue_fallback_t fallback)
 613 {
 614         struct tun_struct *tun = netdev_priv(dev);
 615         u16 ret;
 616
 617         rcu_read_lock();
 618         if (rcu_dereference(tun->steering_prog))
 619                 ret = tun_ebpf_select_queue(tun, skb);
 620         else
 621                 ret = tun_automq_select_queue(tun, skb);
 622         rcu_read_unlock();
 623
 624         return ret;
 625 }
 626
 627 static inline bool tun_not_capable(struct tun_struct *tun)
 628 {
 629         const struct cred *cred = current_cred();
 630         struct net *net = dev_net(tun->dev);
 631
 632         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 633                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 634                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 635 }
 636
 637 static void tun_set_real_num_queues(struct tun_struct *tun)
 638 {
 639         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 640         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 641 }
 642
 643 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 644 {
 645         tfile->detached = tun;
 646         list_add_tail(&tfile->next, &tun->disabled);
 647         ++tun->numdisabled;
 648 }
 649
 650 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 651 {
 652         struct tun_struct *tun = tfile->detached;
 653
 654         tfile->detached = NULL;
 655         list_del_init(&tfile->next);
 656         --tun->numdisabled;
 657         return tun;
 658 }
 659
 660 void tun_ptr_free(void *ptr)
 661 {
 662         if (!ptr)
 663                 return;
 664         if (tun_is_xdp_frame(ptr)) {
 665                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 666
 667                 xdp_return_frame(xdpf);
 668         } else {
 669                 __skb_array_destroy_skb(ptr);
 670         }
 671 }
 672 EXPORT_SYMBOL_GPL(tun_ptr_free);
 673
 674 static void tun_queue_purge(struct tun_file *tfile)
 675 {
 676         void *ptr;
 677
 678         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 679                 tun_ptr_free(ptr);
 680
 681         skb_queue_purge(&tfile->sk.sk_write_queue);
 682         skb_queue_purge(&tfile->sk.sk_error_queue);
 683 }
 684
 685 static void __tun_detach(struct tun_file *tfile, bool clean)
 686 {
 687         struct tun_file *ntfile;
 688         struct tun_struct *tun;
 689
 690         tun = rtnl_dereference(tfile->tun);
 691
 692         if (tun && clean) {
 693                 tun_napi_disable(tun, tfile);
 694                 tun_napi_del(tun, tfile);
 695         }
 696
 697         if (tun && !tfile->detached) {
 698                 u16 index = tfile->queue_index;
 699                 BUG_ON(index >= tun->numqueues);
 700
 701                 rcu_assign_pointer(tun->tfiles[index],
 702                                    tun->tfiles[tun->numqueues - 1]);
 703                 ntfile = rtnl_dereference(tun->tfiles[index]);
 704                 ntfile->queue_index = index;
 705
 706                 --tun->numqueues;
 707                 if (clean) {
 708                         RCU_INIT_POINTER(tfile->tun, NULL);
 709                         sock_put(&tfile->sk);
 710                 } else
 711                         tun_disable_queue(tun, tfile);
 712
 713                 synchronize_net();
 714                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 715                 /* Drop read queue */
 716                 tun_queue_purge(tfile);
 717                 tun_set_real_num_queues(tun);
 718         } else if (tfile->detached && clean) {
 719                 tun = tun_enable_queue(tfile);
 720                 sock_put(&tfile->sk);
 721         }
 722
 723         if (clean) {
 724                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 725                         netif_carrier_off(tun->dev);
 726
 727                         if (!(tun->flags & IFF_PERSIST) &&
 728                             tun->dev->reg_state == NETREG_REGISTERED)
 729                                 unregister_netdevice(tun->dev);
 730                 }
 731                 if (tun)
 732                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 733                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 734                 sock_put(&tfile->sk);
 735         }
 736 }
 737
 738 static void tun_detach(struct tun_file *tfile, bool clean)
 739 {
 740         struct tun_struct *tun;
 741         struct net_device *dev;
 742
 743         rtnl_lock();
 744         tun = rtnl_dereference(tfile->tun);
 745         dev = tun ? tun->dev : NULL;
 746         __tun_detach(tfile, clean);
 747         if (dev)
 748                 netdev_state_change(dev);
 749         rtnl_unlock();
 750 }
 751
 752 static void tun_detach_all(struct net_device *dev)
 753 {
 754         struct tun_struct *tun = netdev_priv(dev);
 755         struct tun_file *tfile, *tmp;
 756         int i, n = tun->numqueues;
 757
 758         for (i = 0; i < n; i++) {
 759                 tfile = rtnl_dereference(tun->tfiles[i]);
 760                 BUG_ON(!tfile);
 761                 tun_napi_disable(tun, tfile);
 762                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 763                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 764                 RCU_INIT_POINTER(tfile->tun, NULL);
 765                 --tun->numqueues;
 766         }
 767         list_for_each_entry(tfile, &tun->disabled, next) {
 768                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 769                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 770                 RCU_INIT_POINTER(tfile->tun, NULL);
 771         }
 772         BUG_ON(tun->numqueues != 0);
 773
 774         synchronize_net();
 775         for (i = 0; i < n; i++) {
 776                 tfile = rtnl_dereference(tun->tfiles[i]);
 777                 tun_napi_del(tun, tfile);
 778                 /* Drop read queue */
 779                 tun_queue_purge(tfile);
 780                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 781                 sock_put(&tfile->sk);
 782         }
 783         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 784                 tun_enable_queue(tfile);
 785                 tun_queue_purge(tfile);
 786                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 787                 sock_put(&tfile->sk);
 788         }
 789         BUG_ON(tun->numdisabled != 0);
 790
 791         if (tun->flags & IFF_PERSIST)
 792                 module_put(THIS_MODULE);
 793 }
 794
 795 static int tun_attach(struct tun_struct *tun, struct file *file,
 796                       bool skip_filter, bool napi)
 797 {
 798         struct tun_file *tfile = file->private_data;
 799         struct net_device *dev = tun->dev;
 800         int err;
 801
 802         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 803         if (err < 0)
 804                 goto out;
 805
 806         err = -EINVAL;
 807         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 808                 goto out;
 809
 810         err = -EBUSY;
 811         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 812                 goto out;
 813
 814         err = -E2BIG;
 815         if (!tfile->detached &&
 816             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 817                 goto out;
 818
 819         err = 0;
 820
 821         /* Re-attach the filter to persist device */
 822         if (!skip_filter && (tun->filter_attached == true)) {
 823                 lock_sock(tfile->socket.sk);
 824                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 825                 release_sock(tfile->socket.sk);
 826                 if (!err)
 827                         goto out;
 828         }
 829
 830         if (!tfile->detached &&
 831             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 832                             GFP_KERNEL, tun_ptr_free)) {
 833                 err = -ENOMEM;
 834                 goto out;
 835         }
 836
 837         tfile->queue_index = tun->numqueues;
 838         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 839
 840         if (tfile->detached) {
 841                 /* Re-attach detached tfile, updating XDP queue_index */
 842                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 843
 844                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 845                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 846         } else {
 847                 /* Setup XDP RX-queue info, for new tfile getting attached */
 848                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 849                                        tun->dev, tfile->queue_index);
 850                 if (err < 0)
 851                         goto out;
 852                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 853                                                  MEM_TYPE_PAGE_SHARED, NULL);
 854                 if (err < 0) {
 855                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 856                         goto out;
 857                 }
 858                 err = 0;
 859         }
 860
 861         rcu_assign_pointer(tfile->tun, tun);
 862         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 863         tun->numqueues++;
 864
 865         if (tfile->detached) {
 866                 tun_enable_queue(tfile);
 867         } else {
 868                 sock_hold(&tfile->sk);
 869                 tun_napi_init(tun, tfile, napi);
 870         }
 871
 872         tun_set_real_num_queues(tun);
 873
 874         /* device is allowed to go away first, so no need to hold extra
 875          * refcnt.
 876          */
 877
 878 out:
 879         return err;
 880 }
 881
 882 static struct tun_struct *tun_get(struct tun_file *tfile)
 883 {
 884         struct tun_struct *tun;
 885
 886         rcu_read_lock();
 887         tun = rcu_dereference(tfile->tun);
 888         if (tun)
 889                 dev_hold(tun->dev);
 890         rcu_read_unlock();
 891
 892         return tun;
 893 }
 894
 895 static void tun_put(struct tun_struct *tun)
 896 {
 897         dev_put(tun->dev);
 898 }
 899
 900 /* TAP filtering */
 901 static void addr_hash_set(u32 *mask, const u8 *addr)
 902 {
 903         int n = ether_crc(ETH_ALEN, addr) >> 26;
 904         mask[n >> 5] |= (1 << (n & 31));
 905 }
 906
 907 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 908 {
 909         int n = ether_crc(ETH_ALEN, addr) >> 26;
 910         return mask[n >> 5] & (1 << (n & 31));
 911 }
 912
 913 static int update_filter(struct tap_filter *filter, void __user *arg)
 914 {
 915         struct { u8 u[ETH_ALEN]; } *addr;
 916         struct tun_filter uf;
 917         int err, alen, n, nexact;
 918
 919         if (copy_from_user(&uf, arg, sizeof(uf)))
 920                 return -EFAULT;
 921
 922         if (!uf.count) {
 923                 /* Disabled */
 924                 filter->count = 0;
 925                 return 0;
 926         }
 927
 928         alen = ETH_ALEN * uf.count;
 929         addr = memdup_user(arg + sizeof(uf), alen);
 930         if (IS_ERR(addr))
 931                 return PTR_ERR(addr);
 932
 933         /* The filter is updated without holding any locks. Which is
 934          * perfectly safe. We disable it first and in the worst
 935          * case we'll accept a few undesired packets. */
 936         filter->count = 0;
 937         wmb();
 938
 939         /* Use first set of addresses as an exact filter */
 940         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 941                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 942
 943         nexact = n;
 944
 945         /* Remaining multicast addresses are hashed,
 946          * unicast will leave the filter disabled. */
 947         memset(filter->mask, 0, sizeof(filter->mask));
 948         for (; n < uf.count; n++) {
 949                 if (!is_multicast_ether_addr(addr[n].u)) {
 950                         err = 0; /* no filter */
 951                         goto free_addr;
 952                 }
 953                 addr_hash_set(filter->mask, addr[n].u);
 954         }
 955
 956         /* For ALLMULTI just set the mask to all ones.
 957          * This overrides the mask populated above. */
 958         if ((uf.flags & TUN_FLT_ALLMULTI))
 959                 memset(filter->mask, ~0, sizeof(filter->mask));
 960
 961         /* Now enable the filter */
 962         wmb();
 963         filter->count = nexact;
 964
 965         /* Return the number of exact filters */
 966         err = nexact;
 967 free_addr:
 968         kfree(addr);
 969         return err;
 970 }
 971
 972 /* Returns: 0 - drop, !=0 - accept */
 973 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 974 {
 975         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 976          * at this point. */
 977         struct ethhdr *eh = (struct ethhdr *) skb->data;
 978         int i;
 979
 980         /* Exact match */
 981         for (i = 0; i < filter->count; i++)
 982                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 983                         return 1;
 984
 985         /* Inexact match (multicast only) */
 986         if (is_multicast_ether_addr(eh->h_dest))
 987                 return addr_hash_test(filter->mask, eh->h_dest);
 988
 989         return 0;
 990 }
 991
 992 /*
 993  * Checks whether the packet is accepted or not.
 994  * Returns: 0 - drop, !=0 - accept
 995  */
 996 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 997 {
 998         if (!filter->count)
 999                 return 1;
1000
1001         return run_filter(filter, skb);
1002 }
1003
1004 /* Network device part of the driver */
1005
1006 static const struct ethtool_ops tun_ethtool_ops;
1007
1008 /* Net device detach from fd. */
1009 static void tun_net_uninit(struct net_device *dev)
1010 {
1011         tun_detach_all(dev);
1012 }
1013
1014 /* Net device open. */
1015 static int tun_net_open(struct net_device *dev)
1016 {
1017         struct tun_struct *tun = netdev_priv(dev);
1018         int i;
1019
1020         netif_tx_start_all_queues(dev);
1021
1022         for (i = 0; i < tun->numqueues; i++) {
1023                 struct tun_file *tfile;
1024
1025                 tfile = rtnl_dereference(tun->tfiles[i]);
1026                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1027         }
1028
1029         return 0;
1030 }
1031
1032 /* Net device close. */
1033 static int tun_net_close(struct net_device *dev)
1034 {
1035         netif_tx_stop_all_queues(dev);
1036         return 0;
1037 }
1038
1039 /* Net device start xmit */
1040 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1041 {
1042 #ifdef CONFIG_RPS
1043         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1044                 /* Select queue was not called for the skbuff, so we extract the
1045                  * RPS hash and save it into the flow_table here.
1046                  */
1047                 __u32 rxhash;
1048
1049                 rxhash = __skb_get_hash_symmetric(skb);
1050                 if (rxhash) {
1051                         struct tun_flow_entry *e;
1052                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1053                                         rxhash);
1054                         if (e)
1055                                 tun_flow_save_rps_rxhash(e, rxhash);
1056                 }
1057         }
1058 #endif
1059 }
1060
1061 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1062                                     struct sk_buff *skb,
1063                                     int len)
1064 {
1065         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1066
1067         if (prog)
1068                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1069
1070         return len;
1071 }
1072
1073 /* Net device start xmit */
1074 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1075 {
1076         struct tun_struct *tun = netdev_priv(dev);
1077         int txq = skb->queue_mapping;
1078         struct tun_file *tfile;
1079         int len = skb->len;
1080
1081         rcu_read_lock();
1082         tfile = rcu_dereference(tun->tfiles[txq]);
1083
1084         /* Drop packet if interface is not attached */
1085         if (txq >= tun->numqueues)
1086                 goto drop;
1087
1088         if (!rcu_dereference(tun->steering_prog))
1089                 tun_automq_xmit(tun, skb);
1090
1091         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1092
1093         BUG_ON(!tfile);
1094
1095         /* Drop if the filter does not like it.
1096          * This is a noop if the filter is disabled.
1097          * Filter can be enabled only for the TAP devices. */
1098         if (!check_filter(&tun->txflt, skb))
1099                 goto drop;
1100
1101         if (tfile->socket.sk->sk_filter &&
1102             sk_filter(tfile->socket.sk, skb))
1103                 goto drop;
1104
1105         len = run_ebpf_filter(tun, skb, len);
1106         if (len == 0 || pskb_trim(skb, len))
1107                 goto drop;
1108
1109         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1110                 goto drop;
1111
1112         skb_tx_timestamp(skb);
1113
1114         /* Orphan the skb - required as we might hang on to it
1115          * for indefinite time.
1116          */
1117         skb_orphan(skb);
1118
1119         nf_reset(skb);
1120
1121         if (ptr_ring_produce(&tfile->tx_ring, skb))
1122                 goto drop;
1123
1124         /* Notify and wake up reader process */
1125         if (tfile->flags & TUN_FASYNC)
1126                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1127         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1128
1129         rcu_read_unlock();
1130         return NETDEV_TX_OK;
1131
1132 drop:
1133         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1134         skb_tx_error(skb);
1135         kfree_skb(skb);
1136         rcu_read_unlock();
1137         return NET_XMIT_DROP;
1138 }
1139
1140 static void tun_net_mclist(struct net_device *dev)
1141 {
1142         /*
1143          * This callback is supposed to deal with mc filter in
1144          * _rx_ path and has nothing to do with the _tx_ path.
1145          * In rx path we always accept everything userspace gives us.
1146          */
1147 }
1148
1149 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1150         netdev_features_t features)
1151 {
1152         struct tun_struct *tun = netdev_priv(dev);
1153
1154         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1155 }
1156 #ifdef CONFIG_NET_POLL_CONTROLLER
1157 static void tun_poll_controller(struct net_device *dev)
1158 {
1159         /*
1160          * Tun only receives frames when:
1161          * 1) the char device endpoint gets data from user space
1162          * 2) the tun socket gets a sendmsg call from user space
1163          * If NAPI is not enabled, since both of those are synchronous
1164          * operations, we are guaranteed never to have pending data when we poll
1165          * for it so there is nothing to do here but return.
1166          * We need this though so netpoll recognizes us as an interface that
1167          * supports polling, which enables bridge devices in virt setups to
1168          * still use netconsole
1169          * If NAPI is enabled, however, we need to schedule polling for all
1170          * queues unless we are using napi_gro_frags(), which we call in
1171          * process context and not in NAPI context.
1172          */
1173         struct tun_struct *tun = netdev_priv(dev);
1174
1175         if (tun->flags & IFF_NAPI) {
1176                 struct tun_file *tfile;
1177                 int i;
1178
1179                 if (tun_napi_frags_enabled(tun))
1180                         return;
1181
1182                 rcu_read_lock();
1183                 for (i = 0; i < tun->numqueues; i++) {
1184                         tfile = rcu_dereference(tun->tfiles[i]);
1185                         if (tfile->napi_enabled)
1186                                 napi_schedule(&tfile->napi);
1187                 }
1188                 rcu_read_unlock();
1189         }
1190         return;
1191 }
1192 #endif
1193
1194 static void tun_set_headroom(struct net_device *dev, int new_hr)
1195 {
1196         struct tun_struct *tun = netdev_priv(dev);
1197
1198         if (new_hr < NET_SKB_PAD)
1199                 new_hr = NET_SKB_PAD;
1200
1201         tun->align = new_hr;
1202 }
1203
1204 static void
1205 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1206 {
1207         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1208         struct tun_struct *tun = netdev_priv(dev);
1209         struct tun_pcpu_stats *p;
1210         int i;
1211
1212         for_each_possible_cpu(i) {
1213                 u64 rxpackets, rxbytes, txpackets, txbytes;
1214                 unsigned int start;
1215
1216                 p = per_cpu_ptr(tun->pcpu_stats, i);
1217                 do {
1218                         start = u64_stats_fetch_begin(&p->syncp);
1219                         rxpackets       = p->rx_packets;
1220                         rxbytes         = p->rx_bytes;
1221                         txpackets       = p->tx_packets;
1222                         txbytes         = p->tx_bytes;
1223                 } while (u64_stats_fetch_retry(&p->syncp, start));
1224
1225                 stats->rx_packets       += rxpackets;
1226                 stats->rx_bytes         += rxbytes;
1227                 stats->tx_packets       += txpackets;
1228                 stats->tx_bytes         += txbytes;
1229
1230                 /* u32 counters */
1231                 rx_dropped      += p->rx_dropped;
1232                 rx_frame_errors += p->rx_frame_errors;
1233                 tx_dropped      += p->tx_dropped;
1234         }
1235         stats->rx_dropped  = rx_dropped;
1236         stats->rx_frame_errors = rx_frame_errors;
1237         stats->tx_dropped = tx_dropped;
1238 }
1239
1240 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1241                        struct netlink_ext_ack *extack)
1242 {
1243         struct tun_struct *tun = netdev_priv(dev);
1244         struct bpf_prog *old_prog;
1245
1246         old_prog = rtnl_dereference(tun->xdp_prog);
1247         rcu_assign_pointer(tun->xdp_prog, prog);
1248         if (old_prog)
1249                 bpf_prog_put(old_prog);
1250
1251         return 0;
1252 }
1253
1254 static u32 tun_xdp_query(struct net_device *dev)
1255 {
1256         struct tun_struct *tun = netdev_priv(dev);
1257         const struct bpf_prog *xdp_prog;
1258
1259         xdp_prog = rtnl_dereference(tun->xdp_prog);
1260         if (xdp_prog)
1261                 return xdp_prog->aux->id;
1262
1263         return 0;
1264 }
1265
1266 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1267 {
1268         switch (xdp->command) {
1269         case XDP_SETUP_PROG:
1270                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1271         case XDP_QUERY_PROG:
1272                 xdp->prog_id = tun_xdp_query(dev);
1273                 return 0;
1274         default:
1275                 return -EINVAL;
1276         }
1277 }
1278
1279 static const struct net_device_ops tun_netdev_ops = {
1280         .ndo_uninit             = tun_net_uninit,
1281         .ndo_open               = tun_net_open,
1282         .ndo_stop               = tun_net_close,
1283         .ndo_start_xmit         = tun_net_xmit,
1284         .ndo_fix_features       = tun_net_fix_features,
1285         .ndo_select_queue       = tun_select_queue,
1286 #ifdef CONFIG_NET_POLL_CONTROLLER
1287         .ndo_poll_controller    = tun_poll_controller,
1288 #endif
1289         .ndo_set_rx_headroom    = tun_set_headroom,
1290         .ndo_get_stats64        = tun_net_get_stats64,
1291 };
1292
1293 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1294 {
1295         /* Notify and wake up reader process */
1296         if (tfile->flags & TUN_FASYNC)
1297                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1298         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1299 }
1300
1301 static int tun_xdp_xmit(struct net_device *dev, int n,
1302                         struct xdp_frame **frames, u32 flags)
1303 {
1304         struct tun_struct *tun = netdev_priv(dev);
1305         struct tun_file *tfile;
1306         u32 numqueues;
1307         int drops = 0;
1308         int cnt = n;
1309         int i;
1310
1311         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1312                 return -EINVAL;
1313
1314         rcu_read_lock();
1315
1316         numqueues = READ_ONCE(tun->numqueues);
1317         if (!numqueues) {
1318                 rcu_read_unlock();
1319                 return -ENXIO; /* Caller will free/return all frames */
1320         }
1321
1322         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1323                                             numqueues]);
1324
1325         spin_lock(&tfile->tx_ring.producer_lock);
1326         for (i = 0; i < n; i++) {
1327                 struct xdp_frame *xdp = frames[i];
1328                 /* Encode the XDP flag into lowest bit for consumer to differ
1329                  * XDP buffer from sk_buff.
1330                  */
1331                 void *frame = tun_xdp_to_ptr(xdp);
1332
1333                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1334                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1335                         xdp_return_frame_rx_napi(xdp);
1336                         drops++;
1337                 }
1338         }
1339         spin_unlock(&tfile->tx_ring.producer_lock);
1340
1341         if (flags & XDP_XMIT_FLUSH)
1342                 __tun_xdp_flush_tfile(tfile);
1343
1344         rcu_read_unlock();
1345         return cnt - drops;
1346 }
1347
1348 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1349 {
1350         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1351
1352         if (unlikely(!frame))
1353                 return -EOVERFLOW;
1354
1355         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1356 }
1357
1358 static const struct net_device_ops tap_netdev_ops = {
1359         .ndo_uninit             = tun_net_uninit,
1360         .ndo_open               = tun_net_open,
1361         .ndo_stop               = tun_net_close,
1362         .ndo_start_xmit         = tun_net_xmit,
1363         .ndo_fix_features       = tun_net_fix_features,
1364         .ndo_set_rx_mode        = tun_net_mclist,
1365         .ndo_set_mac_address    = eth_mac_addr,
1366         .ndo_validate_addr      = eth_validate_addr,
1367         .ndo_select_queue       = tun_select_queue,
1368 #ifdef CONFIG_NET_POLL_CONTROLLER
1369         .ndo_poll_controller    = tun_poll_controller,
1370 #endif
1371         .ndo_features_check     = passthru_features_check,
1372         .ndo_set_rx_headroom    = tun_set_headroom,
1373         .ndo_get_stats64        = tun_net_get_stats64,
1374         .ndo_bpf                = tun_xdp,
1375         .ndo_xdp_xmit           = tun_xdp_xmit,
1376 };
1377
1378 static void tun_flow_init(struct tun_struct *tun)
1379 {
1380         int i;
1381
1382         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1383                 INIT_HLIST_HEAD(&tun->flows[i]);
1384
1385         tun->ageing_time = TUN_FLOW_EXPIRE;
1386         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1387         mod_timer(&tun->flow_gc_timer,
1388                   round_jiffies_up(jiffies + tun->ageing_time));
1389 }
1390
1391 static void tun_flow_uninit(struct tun_struct *tun)
1392 {
1393         del_timer_sync(&tun->flow_gc_timer);
1394         tun_flow_flush(tun);
1395 }
1396
1397 #define MIN_MTU 68
1398 #define MAX_MTU 65535
1399
1400 /* Initialize net device. */
1401 static void tun_net_init(struct net_device *dev)
1402 {
1403         struct tun_struct *tun = netdev_priv(dev);
1404
1405         switch (tun->flags & TUN_TYPE_MASK) {
1406         case IFF_TUN:
1407                 dev->netdev_ops = &tun_netdev_ops;
1408
1409                 /* Point-to-Point TUN Device */
1410                 dev->hard_header_len = 0;
1411                 dev->addr_len = 0;
1412                 dev->mtu = 1500;
1413
1414                 /* Zero header length */
1415                 dev->type = ARPHRD_NONE;
1416                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1417                 break;
1418
1419         case IFF_TAP:
1420                 dev->netdev_ops = &tap_netdev_ops;
1421                 /* Ethernet TAP Device */
1422                 ether_setup(dev);
1423                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1424                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1425
1426                 eth_hw_addr_random(dev);
1427
1428                 break;
1429         }
1430
1431         dev->min_mtu = MIN_MTU;
1432         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1433 }
1434
1435 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1436 {
1437         struct sock *sk = tfile->socket.sk;
1438
1439         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1440 }
1441
1442 /* Character device part */
1443
1444 /* Poll */
1445 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1446 {
1447         struct tun_file *tfile = file->private_data;
1448         struct tun_struct *tun = tun_get(tfile);
1449         struct sock *sk;
1450         __poll_t mask = 0;
1451
1452         if (!tun)
1453                 return EPOLLERR;
1454
1455         sk = tfile->socket.sk;
1456
1457         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1458
1459         poll_wait(file, sk_sleep(sk), wait);
1460
1461         if (!ptr_ring_empty(&tfile->tx_ring))
1462                 mask |= EPOLLIN | EPOLLRDNORM;
1463
1464         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1465          * guarantee EPOLLOUT to be raised by either here or
1466          * tun_sock_write_space(). Then process could get notification
1467          * after it writes to a down device and meets -EIO.
1468          */
1469         if (tun_sock_writeable(tun, tfile) ||
1470             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1471              tun_sock_writeable(tun, tfile)))
1472                 mask |= EPOLLOUT | EPOLLWRNORM;
1473
1474         if (tun->dev->reg_state != NETREG_REGISTERED)
1475                 mask = EPOLLERR;
1476
1477         tun_put(tun);
1478         return mask;
1479 }
1480
1481 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1482                                             size_t len,
1483                                             const struct iov_iter *it)
1484 {
1485         struct sk_buff *skb;
1486         size_t linear;
1487         int err;
1488         int i;
1489
1490         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1491                 return ERR_PTR(-ENOMEM);
1492
1493         local_bh_disable();
1494         skb = napi_get_frags(&tfile->napi);
1495         local_bh_enable();
1496         if (!skb)
1497                 return ERR_PTR(-ENOMEM);
1498
1499         linear = iov_iter_single_seg_count(it);
1500         err = __skb_grow(skb, linear);
1501         if (err)
1502                 goto free;
1503
1504         skb->len = len;
1505         skb->data_len = len - linear;
1506         skb->truesize += skb->data_len;
1507
1508         for (i = 1; i < it->nr_segs; i++) {
1509                 struct page_frag *pfrag = &current->task_frag;
1510                 size_t fragsz = it->iov[i].iov_len;
1511
1512                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1513                         err = -EINVAL;
1514                         goto free;
1515                 }
1516
1517                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1518                         err = -ENOMEM;
1519                         goto free;
1520                 }
1521
1522                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1523                                    pfrag->offset, fragsz);
1524                 page_ref_inc(pfrag->page);
1525                 pfrag->offset += fragsz;
1526         }
1527
1528         return skb;
1529 free:
1530         /* frees skb and all frags allocated with napi_alloc_frag() */
1531         napi_free_frags(&tfile->napi);
1532         return ERR_PTR(err);
1533 }
1534
1535 /* prepad is the amount to reserve at front.  len is length after that.
1536  * linear is a hint as to how much to copy (usually headers). */
1537 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1538                                      size_t prepad, size_t len,
1539                                      size_t linear, int noblock)
1540 {
1541         struct sock *sk = tfile->socket.sk;
1542         struct sk_buff *skb;
1543         int err;
1544
1545         /* Under a page?  Don't bother with paged skb. */
1546         if (prepad + len < PAGE_SIZE || !linear)
1547                 linear = len;
1548
1549         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1550                                    &err, 0);
1551         if (!skb)
1552                 return ERR_PTR(err);
1553
1554         skb_reserve(skb, prepad);
1555         skb_put(skb, linear);
1556         skb->data_len = len - linear;
1557         skb->len += len - linear;
1558
1559         return skb;
1560 }
1561
1562 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1563                            struct sk_buff *skb, int more)
1564 {
1565         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1566         struct sk_buff_head process_queue;
1567         u32 rx_batched = tun->rx_batched;
1568         bool rcv = false;
1569
1570         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1571                 local_bh_disable();
1572                 netif_receive_skb(skb);
1573                 local_bh_enable();
1574                 return;
1575         }
1576
1577         spin_lock(&queue->lock);
1578         if (!more || skb_queue_len(queue) == rx_batched) {
1579                 __skb_queue_head_init(&process_queue);
1580                 skb_queue_splice_tail_init(queue, &process_queue);
1581                 rcv = true;
1582         } else {
1583                 __skb_queue_tail(queue, skb);
1584         }
1585         spin_unlock(&queue->lock);
1586
1587         if (rcv) {
1588                 struct sk_buff *nskb;
1589
1590                 local_bh_disable();
1591                 while ((nskb = __skb_dequeue(&process_queue)))
1592                         netif_receive_skb(nskb);
1593                 netif_receive_skb(skb);
1594                 local_bh_enable();
1595         }
1596 }
1597
1598 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1599                               int len, int noblock, bool zerocopy)
1600 {
1601         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1602                 return false;
1603
1604         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1605                 return false;
1606
1607         if (!noblock)
1608                 return false;
1609
1610         if (zerocopy)
1611                 return false;
1612
1613         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1614             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1615                 return false;
1616
1617         return true;
1618 }
1619
1620 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1621                                      struct tun_file *tfile,
1622                                      struct iov_iter *from,
1623                                      struct virtio_net_hdr *hdr,
1624                                      int len, int *skb_xdp)
1625 {
1626         struct page_frag *alloc_frag = &current->task_frag;
1627         struct sk_buff *skb;
1628         struct bpf_prog *xdp_prog;
1629         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1630         unsigned int delta = 0;
1631         char *buf;
1632         size_t copied;
1633         int err, pad = TUN_RX_PAD;
1634
1635         rcu_read_lock();
1636         xdp_prog = rcu_dereference(tun->xdp_prog);
1637         if (xdp_prog)
1638                 pad += TUN_HEADROOM;
1639         buflen += SKB_DATA_ALIGN(len + pad);
1640         rcu_read_unlock();
1641
1642         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1643         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1644                 return ERR_PTR(-ENOMEM);
1645
1646         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1647         copied = copy_page_from_iter(alloc_frag->page,
1648                                      alloc_frag->offset + pad,
1649                                      len, from);
1650         if (copied != len)
1651                 return ERR_PTR(-EFAULT);
1652
1653         /* There's a small window that XDP may be set after the check
1654          * of xdp_prog above, this should be rare and for simplicity
1655          * we do XDP on skb in case the headroom is not enough.
1656          */
1657         if (hdr->gso_type || !xdp_prog)
1658                 *skb_xdp = 1;
1659         else
1660                 *skb_xdp = 0;
1661
1662         local_bh_disable();
1663         rcu_read_lock();
1664         xdp_prog = rcu_dereference(tun->xdp_prog);
1665         if (xdp_prog && !*skb_xdp) {
1666                 struct xdp_buff xdp;
1667                 void *orig_data;
1668                 u32 act;
1669
1670                 xdp.data_hard_start = buf;
1671                 xdp.data = buf + pad;
1672                 xdp_set_data_meta_invalid(&xdp);
1673                 xdp.data_end = xdp.data + len;
1674                 xdp.rxq = &tfile->xdp_rxq;
1675                 orig_data = xdp.data;
1676                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1677
1678                 switch (act) {
1679                 case XDP_REDIRECT:
1680                         get_page(alloc_frag->page);
1681                         alloc_frag->offset += buflen;
1682                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1683                         xdp_do_flush_map();
1684                         if (err)
1685                                 goto err_redirect;
1686                         rcu_read_unlock();
1687                         local_bh_enable();
1688                         return NULL;
1689                 case XDP_TX:
1690                         get_page(alloc_frag->page);
1691                         alloc_frag->offset += buflen;
1692                         if (tun_xdp_tx(tun->dev, &xdp) < 0)
1693                                 goto err_redirect;
1694                         rcu_read_unlock();
1695                         local_bh_enable();
1696                         return NULL;
1697                 case XDP_PASS:
1698                         delta = orig_data - xdp.data;
1699                         len = xdp.data_end - xdp.data;
1700                         break;
1701                 default:
1702                         bpf_warn_invalid_xdp_action(act);
1703                         /* fall through */
1704                 case XDP_ABORTED:
1705                         trace_xdp_exception(tun->dev, xdp_prog, act);
1706                         /* fall through */
1707                 case XDP_DROP:
1708                         goto err_xdp;
1709                 }
1710         }
1711
1712         skb = build_skb(buf, buflen);
1713         if (!skb) {
1714                 rcu_read_unlock();
1715                 local_bh_enable();
1716                 return ERR_PTR(-ENOMEM);
1717         }
1718
1719         skb_reserve(skb, pad - delta);
1720         skb_put(skb, len);
1721         get_page(alloc_frag->page);
1722         alloc_frag->offset += buflen;
1723
1724         rcu_read_unlock();
1725         local_bh_enable();
1726
1727         return skb;
1728
1729 err_redirect:
1730         put_page(alloc_frag->page);
1731 err_xdp:
1732         rcu_read_unlock();
1733         local_bh_enable();
1734         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1735         return NULL;
1736 }
1737
1738 /* Get packet from user space buffer */
1739 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1740                             void *msg_control, struct iov_iter *from,
1741                             int noblock, bool more)
1742 {
1743         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1744         struct sk_buff *skb;
1745         size_t total_len = iov_iter_count(from);
1746         size_t len = total_len, align = tun->align, linear;
1747         struct virtio_net_hdr gso = { 0 };
1748         struct tun_pcpu_stats *stats;
1749         int good_linear;
1750         int copylen;
1751         bool zerocopy = false;
1752         int err;
1753         u32 rxhash = 0;
1754         int skb_xdp = 1;
1755         bool frags = tun_napi_frags_enabled(tun);
1756
1757         if (!(tun->dev->flags & IFF_UP))
1758                 return -EIO;
1759
1760         if (!(tun->flags & IFF_NO_PI)) {
1761                 if (len < sizeof(pi))
1762                         return -EINVAL;
1763                 len -= sizeof(pi);
1764
1765                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1766                         return -EFAULT;
1767         }
1768
1769         if (tun->flags & IFF_VNET_HDR) {
1770                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1771
1772                 if (len < vnet_hdr_sz)
1773                         return -EINVAL;
1774                 len -= vnet_hdr_sz;
1775
1776                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1777                         return -EFAULT;
1778
1779                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1780                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1781                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1782
1783                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1784                         return -EINVAL;
1785                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1786         }
1787
1788         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1789                 align += NET_IP_ALIGN;
1790                 if (unlikely(len < ETH_HLEN ||
1791                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1792                         return -EINVAL;
1793         }
1794
1795         good_linear = SKB_MAX_HEAD(align);
1796
1797         if (msg_control) {
1798                 struct iov_iter i = *from;
1799
1800                 /* There are 256 bytes to be copied in skb, so there is
1801                  * enough room for skb expand head in case it is used.
1802                  * The rest of the buffer is mapped from userspace.
1803                  */
1804                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1805                 if (copylen > good_linear)
1806                         copylen = good_linear;
1807                 linear = copylen;
1808                 iov_iter_advance(&i, copylen);
1809                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1810                         zerocopy = true;
1811         }
1812
1813         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1814                 /* For the packet that is not easy to be processed
1815                  * (e.g gso or jumbo packet), we will do it at after
1816                  * skb was created with generic XDP routine.
1817                  */
1818                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1819                 if (IS_ERR(skb)) {
1820                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1821                         return PTR_ERR(skb);
1822                 }
1823                 if (!skb)
1824                         return total_len;
1825         } else {
1826                 if (!zerocopy) {
1827                         copylen = len;
1828                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1829                                 linear = good_linear;
1830                         else
1831                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1832                 }
1833
1834                 if (frags) {
1835                         mutex_lock(&tfile->napi_mutex);
1836                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1837                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1838                          * If zerocopy is enabled, then this layout will be
1839                          * overwritten by zerocopy_sg_from_iter().
1840                          */
1841                         zerocopy = false;
1842                 } else {
1843                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1844                                             noblock);
1845                 }
1846
1847                 if (IS_ERR(skb)) {
1848                         if (PTR_ERR(skb) != -EAGAIN)
1849                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1850                         if (frags)
1851                                 mutex_unlock(&tfile->napi_mutex);
1852                         return PTR_ERR(skb);
1853                 }
1854
1855                 if (zerocopy)
1856                         err = zerocopy_sg_from_iter(skb, from);
1857                 else
1858                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1859
1860                 if (err) {
1861                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1862                         kfree_skb(skb);
1863                         if (frags) {
1864                                 tfile->napi.skb = NULL;
1865                                 mutex_unlock(&tfile->napi_mutex);
1866                         }
1867
1868                         return -EFAULT;
1869                 }
1870         }
1871
1872         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1873                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1874                 kfree_skb(skb);
1875                 if (frags) {
1876                         tfile->napi.skb = NULL;
1877                         mutex_unlock(&tfile->napi_mutex);
1878                 }
1879
1880                 return -EINVAL;
1881         }
1882
1883         switch (tun->flags & TUN_TYPE_MASK) {
1884         case IFF_TUN:
1885                 if (tun->flags & IFF_NO_PI) {
1886                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1887
1888                         switch (ip_version) {
1889                         case 4:
1890                                 pi.proto = htons(ETH_P_IP);
1891                                 break;
1892                         case 6:
1893                                 pi.proto = htons(ETH_P_IPV6);
1894                                 break;
1895                         default:
1896                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1897                                 kfree_skb(skb);
1898                                 return -EINVAL;
1899                         }
1900                 }
1901
1902                 skb_reset_mac_header(skb);
1903                 skb->protocol = pi.proto;
1904                 skb->dev = tun->dev;
1905                 break;
1906         case IFF_TAP:
1907                 if (!frags)
1908                         skb->protocol = eth_type_trans(skb, tun->dev);
1909                 break;
1910         }
1911
1912         /* copy skb_ubuf_info for callback when skb has no error */
1913         if (zerocopy) {
1914                 skb_shinfo(skb)->destructor_arg = msg_control;
1915                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1916                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1917         } else if (msg_control) {
1918                 struct ubuf_info *uarg = msg_control;
1919                 uarg->callback(uarg, false);
1920         }
1921
1922         skb_reset_network_header(skb);
1923         skb_probe_transport_header(skb, 0);
1924
1925         if (skb_xdp) {
1926                 struct bpf_prog *xdp_prog;
1927                 int ret;
1928
1929                 local_bh_disable();
1930                 rcu_read_lock();
1931                 xdp_prog = rcu_dereference(tun->xdp_prog);
1932                 if (xdp_prog) {
1933                         ret = do_xdp_generic(xdp_prog, skb);
1934                         if (ret != XDP_PASS) {
1935                                 rcu_read_unlock();
1936                                 local_bh_enable();
1937                                 return total_len;
1938                         }
1939                 }
1940                 rcu_read_unlock();
1941                 local_bh_enable();
1942         }
1943
1944         /* Compute the costly rx hash only if needed for flow updates.
1945          * We may get a very small possibility of OOO during switching, not
1946          * worth to optimize.
1947          */
1948         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1949             !tfile->detached)
1950                 rxhash = __skb_get_hash_symmetric(skb);
1951
1952         if (frags) {
1953                 /* Exercise flow dissector code path. */
1954                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1955
1956                 if (unlikely(headlen > skb_headlen(skb))) {
1957                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1958                         napi_free_frags(&tfile->napi);
1959                         mutex_unlock(&tfile->napi_mutex);
1960                         WARN_ON(1);
1961                         return -ENOMEM;
1962                 }
1963
1964                 local_bh_disable();
1965                 napi_gro_frags(&tfile->napi);
1966                 local_bh_enable();
1967                 mutex_unlock(&tfile->napi_mutex);
1968         } else if (tfile->napi_enabled) {
1969                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1970                 int queue_len;
1971
1972                 spin_lock_bh(&queue->lock);
1973                 __skb_queue_tail(queue, skb);
1974                 queue_len = skb_queue_len(queue);
1975                 spin_unlock(&queue->lock);
1976
1977                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1978                         napi_schedule(&tfile->napi);
1979
1980                 local_bh_enable();
1981         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1982                 tun_rx_batched(tun, tfile, skb, more);
1983         } else {
1984                 netif_rx_ni(skb);
1985         }
1986
1987         stats = get_cpu_ptr(tun->pcpu_stats);
1988         u64_stats_update_begin(&stats->syncp);
1989         stats->rx_packets++;
1990         stats->rx_bytes += len;
1991         u64_stats_update_end(&stats->syncp);
1992         put_cpu_ptr(stats);
1993
1994         if (rxhash)
1995                 tun_flow_update(tun, rxhash, tfile);
1996
1997         return total_len;
1998 }
1999
2000 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2001 {
2002         struct file *file = iocb->ki_filp;
2003         struct tun_file *tfile = file->private_data;
2004         struct tun_struct *tun = tun_get(tfile);
2005         ssize_t result;
2006
2007         if (!tun)
2008                 return -EBADFD;
2009
2010         result = tun_get_user(tun, tfile, NULL, from,
2011                               file->f_flags & O_NONBLOCK, false);
2012
2013         tun_put(tun);
2014         return result;
2015 }
2016
2017 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2018                                 struct tun_file *tfile,
2019                                 struct xdp_frame *xdp_frame,
2020                                 struct iov_iter *iter)
2021 {
2022         int vnet_hdr_sz = 0;
2023         size_t size = xdp_frame->len;
2024         struct tun_pcpu_stats *stats;
2025         size_t ret;
2026
2027         if (tun->flags & IFF_VNET_HDR) {
2028                 struct virtio_net_hdr gso = { 0 };
2029
2030                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2031                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2032                         return -EINVAL;
2033                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2034                              sizeof(gso)))
2035                         return -EFAULT;
2036                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2037         }
2038
2039         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2040
2041         stats = get_cpu_ptr(tun->pcpu_stats);
2042         u64_stats_update_begin(&stats->syncp);
2043         stats->tx_packets++;
2044         stats->tx_bytes += ret;
2045         u64_stats_update_end(&stats->syncp);
2046         put_cpu_ptr(tun->pcpu_stats);
2047
2048         return ret;
2049 }
2050
2051 /* Put packet to the user space buffer */
2052 static ssize_t tun_put_user(struct tun_struct *tun,
2053                             struct tun_file *tfile,
2054                             struct sk_buff *skb,
2055                             struct iov_iter *iter)
2056 {
2057         struct tun_pi pi = { 0, skb->protocol };
2058         struct tun_pcpu_stats *stats;
2059         ssize_t total;
2060         int vlan_offset = 0;
2061         int vlan_hlen = 0;
2062         int vnet_hdr_sz = 0;
2063
2064         if (skb_vlan_tag_present(skb))
2065                 vlan_hlen = VLAN_HLEN;
2066
2067         if (tun->flags & IFF_VNET_HDR)
2068                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2069
2070         total = skb->len + vlan_hlen + vnet_hdr_sz;
2071
2072         if (!(tun->flags & IFF_NO_PI)) {
2073                 if (iov_iter_count(iter) < sizeof(pi))
2074                         return -EINVAL;
2075
2076                 total += sizeof(pi);
2077                 if (iov_iter_count(iter) < total) {
2078                         /* Packet will be striped */
2079                         pi.flags |= TUN_PKT_STRIP;
2080                 }
2081
2082                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2083                         return -EFAULT;
2084         }
2085
2086         if (vnet_hdr_sz) {
2087                 struct virtio_net_hdr gso;
2088
2089                 if (iov_iter_count(iter) < vnet_hdr_sz)
2090                         return -EINVAL;
2091
2092                 if (virtio_net_hdr_from_skb(skb, &gso,
2093                                             tun_is_little_endian(tun), true,
2094                                             vlan_hlen)) {
2095                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2096                         pr_err("unexpected GSO type: "
2097                                "0x%x, gso_size %d, hdr_len %d\n",
2098                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2099                                tun16_to_cpu(tun, gso.hdr_len));
2100                         print_hex_dump(KERN_ERR, "tun: ",
2101                                        DUMP_PREFIX_NONE,
2102                                        16, 1, skb->head,
2103                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2104                         WARN_ON_ONCE(1);
2105                         return -EINVAL;
2106                 }
2107
2108                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2109                         return -EFAULT;
2110
2111                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2112         }
2113
2114         if (vlan_hlen) {
2115                 int ret;
2116                 struct veth veth;
2117
2118                 veth.h_vlan_proto = skb->vlan_proto;
2119                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2120
2121                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2122
2123                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2124                 if (ret || !iov_iter_count(iter))
2125                         goto done;
2126
2127                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2128                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2129                         goto done;
2130         }
2131
2132         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2133
2134 done:
2135         /* caller is in process context, */
2136         stats = get_cpu_ptr(tun->pcpu_stats);
2137         u64_stats_update_begin(&stats->syncp);
2138         stats->tx_packets++;
2139         stats->tx_bytes += skb->len + vlan_hlen;
2140         u64_stats_update_end(&stats->syncp);
2141         put_cpu_ptr(tun->pcpu_stats);
2142
2143         return total;
2144 }
2145
2146 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2147 {
2148         DECLARE_WAITQUEUE(wait, current);
2149         void *ptr = NULL;
2150         int error = 0;
2151
2152         ptr = ptr_ring_consume(&tfile->tx_ring);
2153         if (ptr)
2154                 goto out;
2155         if (noblock) {
2156                 error = -EAGAIN;
2157                 goto out;
2158         }
2159
2160         add_wait_queue(&tfile->wq.wait, &wait);
2161         current->state = TASK_INTERRUPTIBLE;
2162
2163         while (1) {
2164                 ptr = ptr_ring_consume(&tfile->tx_ring);
2165                 if (ptr)
2166                         break;
2167                 if (signal_pending(current)) {
2168                         error = -ERESTARTSYS;
2169                         break;
2170                 }
2171                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2172                         error = -EFAULT;
2173                         break;
2174                 }
2175
2176                 schedule();
2177         }
2178
2179         current->state = TASK_RUNNING;
2180         remove_wait_queue(&tfile->wq.wait, &wait);
2181
2182 out:
2183         *err = error;
2184         return ptr;
2185 }
2186
2187 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2188                            struct iov_iter *to,
2189                            int noblock, void *ptr)
2190 {
2191         ssize_t ret;
2192         int err;
2193
2194         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2195
2196         if (!iov_iter_count(to)) {
2197                 tun_ptr_free(ptr);
2198                 return 0;
2199         }
2200
2201         if (!ptr) {
2202                 /* Read frames from ring */
2203                 ptr = tun_ring_recv(tfile, noblock, &err);
2204                 if (!ptr)
2205                         return err;
2206         }
2207
2208         if (tun_is_xdp_frame(ptr)) {
2209                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2210
2211                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2212                 xdp_return_frame(xdpf);
2213         } else {
2214                 struct sk_buff *skb = ptr;
2215
2216                 ret = tun_put_user(tun, tfile, skb, to);
2217                 if (unlikely(ret < 0))
2218                         kfree_skb(skb);
2219                 else
2220                         consume_skb(skb);
2221         }
2222
2223         return ret;
2224 }
2225
2226 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2227 {
2228         struct file *file = iocb->ki_filp;
2229         struct tun_file *tfile = file->private_data;
2230         struct tun_struct *tun = tun_get(tfile);
2231         ssize_t len = iov_iter_count(to), ret;
2232
2233         if (!tun)
2234                 return -EBADFD;
2235         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2236         ret = min_t(ssize_t, ret, len);
2237         if (ret > 0)
2238                 iocb->ki_pos = ret;
2239         tun_put(tun);
2240         return ret;
2241 }
2242
2243 static void tun_prog_free(struct rcu_head *rcu)
2244 {
2245         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2246
2247         bpf_prog_destroy(prog->prog);
2248         kfree(prog);
2249 }
2250
2251 static int __tun_set_ebpf(struct tun_struct *tun,
2252                           struct tun_prog __rcu **prog_p,
2253                           struct bpf_prog *prog)
2254 {
2255         struct tun_prog *old, *new = NULL;
2256
2257         if (prog) {
2258                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2259                 if (!new)
2260                         return -ENOMEM;
2261                 new->prog = prog;
2262         }
2263
2264         spin_lock_bh(&tun->lock);
2265         old = rcu_dereference_protected(*prog_p,
2266                                         lockdep_is_held(&tun->lock));
2267         rcu_assign_pointer(*prog_p, new);
2268         spin_unlock_bh(&tun->lock);
2269
2270         if (old)
2271                 call_rcu(&old->rcu, tun_prog_free);
2272
2273         return 0;
2274 }
2275
2276 static void tun_free_netdev(struct net_device *dev)
2277 {
2278         struct tun_struct *tun = netdev_priv(dev);
2279
2280         BUG_ON(!(list_empty(&tun->disabled)));
2281         free_percpu(tun->pcpu_stats);
2282         tun_flow_uninit(tun);
2283         security_tun_dev_free_security(tun->security);
2284         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2285         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2286 }
2287
2288 static void tun_setup(struct net_device *dev)
2289 {
2290         struct tun_struct *tun = netdev_priv(dev);
2291
2292         tun->owner = INVALID_UID;
2293         tun->group = INVALID_GID;
2294         tun_default_link_ksettings(dev, &tun->link_ksettings);
2295
2296         dev->ethtool_ops = &tun_ethtool_ops;
2297         dev->needs_free_netdev = true;
2298         dev->priv_destructor = tun_free_netdev;
2299         /* We prefer our own queue length */
2300         dev->tx_queue_len = TUN_READQ_SIZE;
2301 }
2302
2303 /* Trivial set of netlink ops to allow deleting tun or tap
2304  * device with netlink.
2305  */
2306 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2307                         struct netlink_ext_ack *extack)
2308 {
2309         return -EINVAL;
2310 }
2311
2312 static size_t tun_get_size(const struct net_device *dev)
2313 {
2314         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2315         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2316
2317         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2318                nla_total_size(sizeof(gid_t)) + /* GROUP */
2319                nla_total_size(sizeof(u8)) + /* TYPE */
2320                nla_total_size(sizeof(u8)) + /* PI */
2321                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2322                nla_total_size(sizeof(u8)) + /* PERSIST */
2323                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2324                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2325                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2326                0;
2327 }
2328
2329 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2330 {
2331         struct tun_struct *tun = netdev_priv(dev);
2332
2333         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2334                 goto nla_put_failure;
2335         if (uid_valid(tun->owner) &&
2336             nla_put_u32(skb, IFLA_TUN_OWNER,
2337                         from_kuid_munged(current_user_ns(), tun->owner)))
2338                 goto nla_put_failure;
2339         if (gid_valid(tun->group) &&
2340             nla_put_u32(skb, IFLA_TUN_GROUP,
2341                         from_kgid_munged(current_user_ns(), tun->group)))
2342                 goto nla_put_failure;
2343         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2344                 goto nla_put_failure;
2345         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2346                 goto nla_put_failure;
2347         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2348                 goto nla_put_failure;
2349         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2350                        !!(tun->flags & IFF_MULTI_QUEUE)))
2351                 goto nla_put_failure;
2352         if (tun->flags & IFF_MULTI_QUEUE) {
2353                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2354                         goto nla_put_failure;
2355                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2356                                 tun->numdisabled))
2357                         goto nla_put_failure;
2358         }
2359
2360         return 0;
2361
2362 nla_put_failure:
2363         return -EMSGSIZE;
2364 }
2365
2366 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2367         .kind           = DRV_NAME,
2368         .priv_size      = sizeof(struct tun_struct),
2369         .setup          = tun_setup,
2370         .validate       = tun_validate,
2371         .get_size       = tun_get_size,
2372         .fill_info      = tun_fill_info,
2373 };
2374
2375 static void tun_sock_write_space(struct sock *sk)
2376 {
2377         struct tun_file *tfile;
2378         wait_queue_head_t *wqueue;
2379
2380         if (!sock_writeable(sk))
2381                 return;
2382
2383         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2384                 return;
2385
2386         wqueue = sk_sleep(sk);
2387         if (wqueue && waitqueue_active(wqueue))
2388                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2389                                                 EPOLLWRNORM | EPOLLWRBAND);
2390
2391         tfile = container_of(sk, struct tun_file, sk);
2392         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2393 }
2394
2395 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2396 {
2397         int ret;
2398         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2399         struct tun_struct *tun = tun_get(tfile);
2400
2401         if (!tun)
2402                 return -EBADFD;
2403
2404         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2405                            m->msg_flags & MSG_DONTWAIT,
2406                            m->msg_flags & MSG_MORE);
2407         tun_put(tun);
2408         return ret;
2409 }
2410
2411 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2412                        int flags)
2413 {
2414         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2415         struct tun_struct *tun = tun_get(tfile);
2416         void *ptr = m->msg_control;
2417         int ret;
2418
2419         if (!tun) {
2420                 ret = -EBADFD;
2421                 goto out_free;
2422         }
2423
2424         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2425                 ret = -EINVAL;
2426                 goto out_put_tun;
2427         }
2428         if (flags & MSG_ERRQUEUE) {
2429                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2430                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2431                 goto out;
2432         }
2433         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2434         if (ret > (ssize_t)total_len) {
2435                 m->msg_flags |= MSG_TRUNC;
2436                 ret = flags & MSG_TRUNC ? ret : total_len;
2437         }
2438 out:
2439         tun_put(tun);
2440         return ret;
2441
2442 out_put_tun:
2443         tun_put(tun);
2444 out_free:
2445         tun_ptr_free(ptr);
2446         return ret;
2447 }
2448
2449 static int tun_ptr_peek_len(void *ptr)
2450 {
2451         if (likely(ptr)) {
2452                 if (tun_is_xdp_frame(ptr)) {
2453                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2454
2455                         return xdpf->len;
2456                 }
2457                 return __skb_array_len_with_tag(ptr);
2458         } else {
2459                 return 0;
2460         }
2461 }
2462
2463 static int tun_peek_len(struct socket *sock)
2464 {
2465         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2466         struct tun_struct *tun;
2467         int ret = 0;
2468
2469         tun = tun_get(tfile);
2470         if (!tun)
2471                 return 0;
2472
2473         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2474         tun_put(tun);
2475
2476         return ret;
2477 }
2478
2479 /* Ops structure to mimic raw sockets with tun */
2480 static const struct proto_ops tun_socket_ops = {
2481         .peek_len = tun_peek_len,
2482         .sendmsg = tun_sendmsg,
2483         .recvmsg = tun_recvmsg,
2484 };
2485
2486 static struct proto tun_proto = {
2487         .name           = "tun",
2488         .owner          = THIS_MODULE,
2489         .obj_size       = sizeof(struct tun_file),
2490 };
2491
2492 static int tun_flags(struct tun_struct *tun)
2493 {
2494         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2495 }
2496
2497 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2498                               char *buf)
2499 {
2500         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2501         return sprintf(buf, "0x%x\n", tun_flags(tun));
2502 }
2503
2504 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2505                               char *buf)
2506 {
2507         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2508         return uid_valid(tun->owner)?
2509                 sprintf(buf, "%u\n",
2510                         from_kuid_munged(current_user_ns(), tun->owner)):
2511                 sprintf(buf, "-1\n");
2512 }
2513
2514 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2515                               char *buf)
2516 {
2517         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2518         return gid_valid(tun->group) ?
2519                 sprintf(buf, "%u\n",
2520                         from_kgid_munged(current_user_ns(), tun->group)):
2521                 sprintf(buf, "-1\n");
2522 }
2523
2524 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2525 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2526 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2527
2528 static struct attribute *tun_dev_attrs[] = {
2529         &dev_attr_tun_flags.attr,
2530         &dev_attr_owner.attr,
2531         &dev_attr_group.attr,
2532         NULL
2533 };
2534
2535 static const struct attribute_group tun_attr_group = {
2536         .attrs = tun_dev_attrs
2537 };
2538
2539 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2540 {
2541         struct tun_struct *tun;
2542         struct tun_file *tfile = file->private_data;
2543         struct net_device *dev;
2544         int err;
2545
2546         if (tfile->detached)
2547                 return -EINVAL;
2548
2549         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2550                 if (!capable(CAP_NET_ADMIN))
2551                         return -EPERM;
2552
2553                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2554                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2555                         return -EINVAL;
2556         }
2557
2558         dev = __dev_get_by_name(net, ifr->ifr_name);
2559         if (dev) {
2560                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2561                         return -EBUSY;
2562                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2563                         tun = netdev_priv(dev);
2564                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2565                         tun = netdev_priv(dev);
2566                 else
2567                         return -EINVAL;
2568
2569                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2570                     !!(tun->flags & IFF_MULTI_QUEUE))
2571                         return -EINVAL;
2572
2573                 if (tun_not_capable(tun))
2574                         return -EPERM;
2575                 err = security_tun_dev_open(tun->security);
2576                 if (err < 0)
2577                         return err;
2578
2579                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2580                                  ifr->ifr_flags & IFF_NAPI);
2581                 if (err < 0)
2582                         return err;
2583
2584                 if (tun->flags & IFF_MULTI_QUEUE &&
2585                     (tun->numqueues + tun->numdisabled > 1)) {
2586                         /* One or more queue has already been attached, no need
2587                          * to initialize the device again.
2588                          */
2589                         netdev_state_change(dev);
2590                         return 0;
2591                 }
2592
2593                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2594                               (ifr->ifr_flags & TUN_FEATURES);
2595
2596                 netdev_state_change(dev);
2597         } else {
2598                 char *name;
2599                 unsigned long flags = 0;
2600                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2601                              MAX_TAP_QUEUES : 1;
2602
2603                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2604                         return -EPERM;
2605                 err = security_tun_dev_create();
2606                 if (err < 0)
2607                         return err;
2608
2609                 /* Set dev type */
2610                 if (ifr->ifr_flags & IFF_TUN) {
2611                         /* TUN device */
2612                         flags |= IFF_TUN;
2613                         name = "tun%d";
2614                 } else if (ifr->ifr_flags & IFF_TAP) {
2615                         /* TAP device */
2616                         flags |= IFF_TAP;
2617                         name = "tap%d";
2618                 } else
2619                         return -EINVAL;
2620
2621                 if (*ifr->ifr_name)
2622                         name = ifr->ifr_name;
2623
2624                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2625                                        NET_NAME_UNKNOWN, tun_setup, queues,
2626                                        queues);
2627
2628                 if (!dev)
2629                         return -ENOMEM;
2630                 err = dev_get_valid_name(net, dev, name);
2631                 if (err < 0)
2632                         goto err_free_dev;
2633
2634                 dev_net_set(dev, net);
2635                 dev->rtnl_link_ops = &tun_link_ops;
2636                 dev->ifindex = tfile->ifindex;
2637                 dev->sysfs_groups[0] = &tun_attr_group;
2638
2639                 tun = netdev_priv(dev);
2640                 tun->dev = dev;
2641                 tun->flags = flags;
2642                 tun->txflt.count = 0;
2643                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2644
2645                 tun->align = NET_SKB_PAD;
2646                 tun->filter_attached = false;
2647                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2648                 tun->rx_batched = 0;
2649                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2650
2651                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2652                 if (!tun->pcpu_stats) {
2653                         err = -ENOMEM;
2654                         goto err_free_dev;
2655                 }
2656
2657                 spin_lock_init(&tun->lock);
2658
2659                 err = security_tun_dev_alloc_security(&tun->security);
2660                 if (err < 0)
2661                         goto err_free_stat;
2662
2663                 tun_net_init(dev);
2664                 tun_flow_init(tun);
2665
2666                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2667                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2668                                    NETIF_F_HW_VLAN_STAG_TX;
2669                 dev->features = dev->hw_features | NETIF_F_LLTX;
2670                 dev->vlan_features = dev->features &
2671                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2672                                        NETIF_F_HW_VLAN_STAG_TX);
2673
2674                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2675                               (ifr->ifr_flags & TUN_FEATURES);
2676
2677                 INIT_LIST_HEAD(&tun->disabled);
2678                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2679                 if (err < 0)
2680                         goto err_free_flow;
2681
2682                 err = register_netdevice(tun->dev);
2683                 if (err < 0)
2684                         goto err_detach;
2685         }
2686
2687         netif_carrier_on(tun->dev);
2688
2689         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2690
2691         /* Make sure persistent devices do not get stuck in
2692          * xoff state.
2693          */
2694         if (netif_running(tun->dev))
2695                 netif_tx_wake_all_queues(tun->dev);
2696
2697         strcpy(ifr->ifr_name, tun->dev->name);
2698         return 0;
2699
2700 err_detach:
2701         tun_detach_all(dev);
2702         /* register_netdevice() already called tun_free_netdev() */
2703         goto err_free_dev;
2704
2705 err_free_flow:
2706         tun_flow_uninit(tun);
2707         security_tun_dev_free_security(tun->security);
2708 err_free_stat:
2709         free_percpu(tun->pcpu_stats);
2710 err_free_dev:
2711         free_netdev(dev);
2712         return err;
2713 }
2714
2715 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2716                        struct ifreq *ifr)
2717 {
2718         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2719
2720         strcpy(ifr->ifr_name, tun->dev->name);
2721
2722         ifr->ifr_flags = tun_flags(tun);
2723
2724 }
2725
2726 /* This is like a cut-down ethtool ops, except done via tun fd so no
2727  * privs required. */
2728 static int set_offload(struct tun_struct *tun, unsigned long arg)
2729 {
2730         netdev_features_t features = 0;
2731
2732         if (arg & TUN_F_CSUM) {
2733                 features |= NETIF_F_HW_CSUM;
2734                 arg &= ~TUN_F_CSUM;
2735
2736                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2737                         if (arg & TUN_F_TSO_ECN) {
2738                                 features |= NETIF_F_TSO_ECN;
2739                                 arg &= ~TUN_F_TSO_ECN;
2740                         }
2741                         if (arg & TUN_F_TSO4)
2742                                 features |= NETIF_F_TSO;
2743                         if (arg & TUN_F_TSO6)
2744                                 features |= NETIF_F_TSO6;
2745                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2746                 }
2747
2748                 arg &= ~TUN_F_UFO;
2749         }
2750
2751         /* This gives the user a way to test for new features in future by
2752          * trying to set them. */
2753         if (arg)
2754                 return -EINVAL;
2755
2756         tun->set_features = features;
2757         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2758         tun->dev->wanted_features |= features;
2759         netdev_update_features(tun->dev);
2760
2761         return 0;
2762 }
2763
2764 static void tun_detach_filter(struct tun_struct *tun, int n)
2765 {
2766         int i;
2767         struct tun_file *tfile;
2768
2769         for (i = 0; i < n; i++) {
2770                 tfile = rtnl_dereference(tun->tfiles[i]);
2771                 lock_sock(tfile->socket.sk);
2772                 sk_detach_filter(tfile->socket.sk);
2773                 release_sock(tfile->socket.sk);
2774         }
2775
2776         tun->filter_attached = false;
2777 }
2778
2779 static int tun_attach_filter(struct tun_struct *tun)
2780 {
2781         int i, ret = 0;
2782         struct tun_file *tfile;
2783
2784         for (i = 0; i < tun->numqueues; i++) {
2785                 tfile = rtnl_dereference(tun->tfiles[i]);
2786                 lock_sock(tfile->socket.sk);
2787                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2788                 release_sock(tfile->socket.sk);
2789                 if (ret) {
2790                         tun_detach_filter(tun, i);
2791                         return ret;
2792                 }
2793         }
2794
2795         tun->filter_attached = true;
2796         return ret;
2797 }
2798
2799 static void tun_set_sndbuf(struct tun_struct *tun)
2800 {
2801         struct tun_file *tfile;
2802         int i;
2803
2804         for (i = 0; i < tun->numqueues; i++) {
2805                 tfile = rtnl_dereference(tun->tfiles[i]);
2806                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2807         }
2808 }
2809
2810 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2811 {
2812         struct tun_file *tfile = file->private_data;
2813         struct tun_struct *tun;
2814         int ret = 0;
2815
2816         rtnl_lock();
2817
2818         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2819                 tun = tfile->detached;
2820                 if (!tun) {
2821                         ret = -EINVAL;
2822                         goto unlock;
2823                 }
2824                 ret = security_tun_dev_attach_queue(tun->security);
2825                 if (ret < 0)
2826                         goto unlock;
2827                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2828         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2829                 tun = rtnl_dereference(tfile->tun);
2830                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2831                         ret = -EINVAL;
2832                 else
2833                         __tun_detach(tfile, false);
2834         } else
2835                 ret = -EINVAL;
2836
2837         if (ret >= 0)
2838                 netdev_state_change(tun->dev);
2839
2840 unlock:
2841         rtnl_unlock();
2842         return ret;
2843 }
2844
2845 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2846                         void __user *data)
2847 {
2848         struct bpf_prog *prog;
2849         int fd;
2850
2851         if (copy_from_user(&fd, data, sizeof(fd)))
2852                 return -EFAULT;
2853
2854         if (fd == -1) {
2855                 prog = NULL;
2856         } else {
2857                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2858                 if (IS_ERR(prog))
2859                         return PTR_ERR(prog);
2860         }
2861
2862         return __tun_set_ebpf(tun, prog_p, prog);
2863 }
2864
2865 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2866                             unsigned long arg, int ifreq_len)
2867 {
2868         struct tun_file *tfile = file->private_data;
2869         struct net *net = sock_net(&tfile->sk);
2870         struct tun_struct *tun;
2871         void __user* argp = (void __user*)arg;
2872         struct ifreq ifr;
2873         kuid_t owner;
2874         kgid_t group;
2875         int sndbuf;
2876         int vnet_hdr_sz;
2877         unsigned int ifindex;
2878         int le;
2879         int ret;
2880         bool do_notify = false;
2881
2882         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2883             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2884                 if (copy_from_user(&ifr, argp, ifreq_len))
2885                         return -EFAULT;
2886         } else {
2887                 memset(&ifr, 0, sizeof(ifr));
2888         }
2889         if (cmd == TUNGETFEATURES) {
2890                 /* Currently this just means: "what IFF flags are valid?".
2891                  * This is needed because we never checked for invalid flags on
2892                  * TUNSETIFF.
2893                  */
2894                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2895                                 (unsigned int __user*)argp);
2896         } else if (cmd == TUNSETQUEUE) {
2897                 return tun_set_queue(file, &ifr);
2898         } else if (cmd == SIOCGSKNS) {
2899                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2900                         return -EPERM;
2901                 return open_related_ns(&net->ns, get_net_ns);
2902         }
2903
2904         ret = 0;
2905         rtnl_lock();
2906
2907         tun = tun_get(tfile);
2908         if (cmd == TUNSETIFF) {
2909                 ret = -EEXIST;
2910                 if (tun)
2911                         goto unlock;
2912
2913                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2914
2915                 ret = tun_set_iff(net, file, &ifr);
2916
2917                 if (ret)
2918                         goto unlock;
2919
2920                 if (copy_to_user(argp, &ifr, ifreq_len))
2921                         ret = -EFAULT;
2922                 goto unlock;
2923         }
2924         if (cmd == TUNSETIFINDEX) {
2925                 ret = -EPERM;
2926                 if (tun)
2927                         goto unlock;
2928
2929                 ret = -EFAULT;
2930                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2931                         goto unlock;
2932
2933                 ret = 0;
2934                 tfile->ifindex = ifindex;
2935                 goto unlock;
2936         }
2937
2938         ret = -EBADFD;
2939         if (!tun)
2940                 goto unlock;
2941
2942         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2943
2944         ret = 0;
2945         switch (cmd) {
2946         case TUNGETIFF:
2947                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2948
2949                 if (tfile->detached)
2950                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2951                 if (!tfile->socket.sk->sk_filter)
2952                         ifr.ifr_flags |= IFF_NOFILTER;
2953
2954                 if (copy_to_user(argp, &ifr, ifreq_len))
2955                         ret = -EFAULT;
2956                 break;
2957
2958         case TUNSETNOCSUM:
2959                 /* Disable/Enable checksum */
2960
2961                 /* [unimplemented] */
2962                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2963                           arg ? "disabled" : "enabled");
2964                 break;
2965
2966         case TUNSETPERSIST:
2967                 /* Disable/Enable persist mode. Keep an extra reference to the
2968                  * module to prevent the module being unprobed.
2969                  */
2970                 if (arg && !(tun->flags & IFF_PERSIST)) {
2971                         tun->flags |= IFF_PERSIST;
2972                         __module_get(THIS_MODULE);
2973                         do_notify = true;
2974                 }
2975                 if (!arg && (tun->flags & IFF_PERSIST)) {
2976                         tun->flags &= ~IFF_PERSIST;
2977                         module_put(THIS_MODULE);
2978                         do_notify = true;
2979                 }
2980
2981                 tun_debug(KERN_INFO, tun, "persist %s\n",
2982                           arg ? "enabled" : "disabled");
2983                 break;
2984
2985         case TUNSETOWNER:
2986                 /* Set owner of the device */
2987                 owner = make_kuid(current_user_ns(), arg);
2988                 if (!uid_valid(owner)) {
2989                         ret = -EINVAL;
2990                         break;
2991                 }
2992                 tun->owner = owner;
2993                 do_notify = true;
2994                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2995                           from_kuid(&init_user_ns, tun->owner));
2996                 break;
2997
2998         case TUNSETGROUP:
2999                 /* Set group of the device */
3000                 group = make_kgid(current_user_ns(), arg);
3001                 if (!gid_valid(group)) {
3002                         ret = -EINVAL;
3003                         break;
3004                 }
3005                 tun->group = group;
3006                 do_notify = true;
3007                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3008                           from_kgid(&init_user_ns, tun->group));
3009                 break;
3010
3011         case TUNSETLINK:
3012                 /* Only allow setting the type when the interface is down */
3013                 if (tun->dev->flags & IFF_UP) {
3014                         tun_debug(KERN_INFO, tun,
3015                                   "Linktype set failed because interface is up\n");
3016                         ret = -EBUSY;
3017                 } else {
3018                         tun->dev->type = (int) arg;
3019                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3020                                   tun->dev->type);
3021                         ret = 0;
3022                 }
3023                 break;
3024
3025 #ifdef TUN_DEBUG
3026         case TUNSETDEBUG:
3027                 tun->debug = arg;
3028                 break;
3029 #endif
3030         case TUNSETOFFLOAD:
3031                 ret = set_offload(tun, arg);
3032                 break;
3033
3034         case TUNSETTXFILTER:
3035                 /* Can be set only for TAPs */
3036                 ret = -EINVAL;
3037                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3038                         break;
3039                 ret = update_filter(&tun->txflt, (void __user *)arg);
3040                 break;
3041
3042         case SIOCGIFHWADDR:
3043                 /* Get hw address */
3044                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3045                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3046                 if (copy_to_user(argp, &ifr, ifreq_len))
3047                         ret = -EFAULT;
3048                 break;
3049
3050         case SIOCSIFHWADDR:
3051                 /* Set hw address */
3052                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3053                           ifr.ifr_hwaddr.sa_data);
3054
3055                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3056                 break;
3057
3058         case TUNGETSNDBUF:
3059                 sndbuf = tfile->socket.sk->sk_sndbuf;
3060                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3061                         ret = -EFAULT;
3062                 break;
3063
3064         case TUNSETSNDBUF:
3065                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3066                         ret = -EFAULT;
3067                         break;
3068                 }
3069                 if (sndbuf <= 0) {
3070                         ret = -EINVAL;
3071                         break;
3072                 }
3073
3074                 tun->sndbuf = sndbuf;
3075                 tun_set_sndbuf(tun);
3076                 break;
3077
3078         case TUNGETVNETHDRSZ:
3079                 vnet_hdr_sz = tun->vnet_hdr_sz;
3080                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3081                         ret = -EFAULT;
3082                 break;
3083
3084         case TUNSETVNETHDRSZ:
3085                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3086                         ret = -EFAULT;
3087                         break;
3088                 }
3089                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3090                         ret = -EINVAL;
3091                         break;
3092                 }
3093
3094                 tun->vnet_hdr_sz = vnet_hdr_sz;
3095                 break;
3096
3097         case TUNGETVNETLE:
3098                 le = !!(tun->flags & TUN_VNET_LE);
3099                 if (put_user(le, (int __user *)argp))
3100                         ret = -EFAULT;
3101                 break;
3102
3103         case TUNSETVNETLE:
3104                 if (get_user(le, (int __user *)argp)) {
3105                         ret = -EFAULT;
3106                         break;
3107                 }
3108                 if (le)
3109                         tun->flags |= TUN_VNET_LE;
3110                 else
3111                         tun->flags &= ~TUN_VNET_LE;
3112                 break;
3113
3114         case TUNGETVNETBE:
3115                 ret = tun_get_vnet_be(tun, argp);
3116                 break;
3117
3118         case TUNSETVNETBE:
3119                 ret = tun_set_vnet_be(tun, argp);
3120                 break;
3121
3122         case TUNATTACHFILTER:
3123                 /* Can be set only for TAPs */
3124                 ret = -EINVAL;
3125                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3126                         break;
3127                 ret = -EFAULT;
3128                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3129                         break;
3130
3131                 ret = tun_attach_filter(tun);
3132                 break;
3133
3134         case TUNDETACHFILTER:
3135                 /* Can be set only for TAPs */
3136                 ret = -EINVAL;
3137                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3138                         break;
3139                 ret = 0;
3140                 tun_detach_filter(tun, tun->numqueues);
3141                 break;
3142
3143         case TUNGETFILTER:
3144                 ret = -EINVAL;
3145                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3146                         break;
3147                 ret = -EFAULT;
3148                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3149                         break;
3150                 ret = 0;
3151                 break;
3152
3153         case TUNSETSTEERINGEBPF:
3154                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3155                 break;
3156
3157         case TUNSETFILTEREBPF:
3158                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3159                 break;
3160
3161         default:
3162                 ret = -EINVAL;
3163                 break;
3164         }
3165
3166         if (do_notify)
3167                 netdev_state_change(tun->dev);
3168
3169 unlock:
3170         rtnl_unlock();
3171         if (tun)
3172                 tun_put(tun);
3173         return ret;
3174 }
3175
3176 static long tun_chr_ioctl(struct file *file,
3177                           unsigned int cmd, unsigned long arg)
3178 {
3179         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3180 }
3181
3182 #ifdef CONFIG_COMPAT
3183 static long tun_chr_compat_ioctl(struct file *file,
3184                          unsigned int cmd, unsigned long arg)
3185 {
3186         switch (cmd) {
3187         case TUNSETIFF:
3188         case TUNGETIFF:
3189         case TUNSETTXFILTER:
3190         case TUNGETSNDBUF:
3191         case TUNSETSNDBUF:
3192         case SIOCGIFHWADDR:
3193         case SIOCSIFHWADDR:
3194                 arg = (unsigned long)compat_ptr(arg);
3195                 break;
3196         default:
3197                 arg = (compat_ulong_t)arg;
3198                 break;
3199         }
3200
3201         /*
3202          * compat_ifreq is shorter than ifreq, so we must not access beyond
3203          * the end of that structure. All fields that are used in this
3204          * driver are compatible though, we don't need to convert the
3205          * contents.
3206          */
3207         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3208 }
3209 #endif /* CONFIG_COMPAT */
3210
3211 static int tun_chr_fasync(int fd, struct file *file, int on)
3212 {
3213         struct tun_file *tfile = file->private_data;
3214         int ret;
3215
3216         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3217                 goto out;
3218
3219         if (on) {
3220                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3221                 tfile->flags |= TUN_FASYNC;
3222         } else
3223                 tfile->flags &= ~TUN_FASYNC;
3224         ret = 0;
3225 out:
3226         return ret;
3227 }
3228
3229 static int tun_chr_open(struct inode *inode, struct file * file)
3230 {
3231         struct net *net = current->nsproxy->net_ns;
3232         struct tun_file *tfile;
3233
3234         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3235
3236         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3237                                             &tun_proto, 0);
3238         if (!tfile)
3239                 return -ENOMEM;
3240         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3241                 sk_free(&tfile->sk);
3242                 return -ENOMEM;
3243         }
3244
3245         RCU_INIT_POINTER(tfile->tun, NULL);
3246         tfile->flags = 0;
3247         tfile->ifindex = 0;
3248
3249         init_waitqueue_head(&tfile->wq.wait);
3250         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3251
3252         tfile->socket.file = file;
3253         tfile->socket.ops = &tun_socket_ops;
3254
3255         sock_init_data(&tfile->socket, &tfile->sk);
3256
3257         tfile->sk.sk_write_space = tun_sock_write_space;
3258         tfile->sk.sk_sndbuf = INT_MAX;
3259
3260         file->private_data = tfile;
3261         INIT_LIST_HEAD(&tfile->next);
3262
3263         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3264
3265         return 0;
3266 }
3267
3268 static int tun_chr_close(struct inode *inode, struct file *file)
3269 {
3270         struct tun_file *tfile = file->private_data;
3271
3272         tun_detach(tfile, true);
3273
3274         return 0;
3275 }
3276
3277 #ifdef CONFIG_PROC_FS
3278 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3279 {
3280         struct tun_file *tfile = file->private_data;
3281         struct tun_struct *tun;
3282         struct ifreq ifr;
3283
3284         memset(&ifr, 0, sizeof(ifr));
3285
3286         rtnl_lock();
3287         tun = tun_get(tfile);
3288         if (tun)
3289                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3290         rtnl_unlock();
3291
3292         if (tun)
3293                 tun_put(tun);
3294
3295         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3296 }
3297 #endif
3298
3299 static const struct file_operations tun_fops = {
3300         .owner  = THIS_MODULE,
3301         .llseek = no_llseek,
3302         .read_iter  = tun_chr_read_iter,
3303         .write_iter = tun_chr_write_iter,
3304         .poll   = tun_chr_poll,
3305         .unlocked_ioctl = tun_chr_ioctl,
3306 #ifdef CONFIG_COMPAT
3307         .compat_ioctl = tun_chr_compat_ioctl,
3308 #endif
3309         .open   = tun_chr_open,
3310         .release = tun_chr_close,
3311         .fasync = tun_chr_fasync,
3312 #ifdef CONFIG_PROC_FS
3313         .show_fdinfo = tun_chr_show_fdinfo,
3314 #endif
3315 };
3316
3317 static struct miscdevice tun_miscdev = {
3318         .minor = TUN_MINOR,
3319         .name = "tun",
3320         .nodename = "net/tun",
3321         .fops = &tun_fops,
3322 };
3323
3324 /* ethtool interface */
3325
3326 static void tun_default_link_ksettings(struct net_device *dev,
3327                                        struct ethtool_link_ksettings *cmd)
3328 {
3329         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3330         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3331         cmd->base.speed         = SPEED_10;
3332         cmd->base.duplex        = DUPLEX_FULL;
3333         cmd->base.port          = PORT_TP;
3334         cmd->base.phy_address   = 0;
3335         cmd->base.autoneg       = AUTONEG_DISABLE;
3336 }
3337
3338 static int tun_get_link_ksettings(struct net_device *dev,
3339                                   struct ethtool_link_ksettings *cmd)
3340 {
3341         struct tun_struct *tun = netdev_priv(dev);
3342
3343         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3344         return 0;
3345 }
3346
3347 static int tun_set_link_ksettings(struct net_device *dev,
3348                                   const struct ethtool_link_ksettings *cmd)
3349 {
3350         struct tun_struct *tun = netdev_priv(dev);
3351
3352         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3353         return 0;
3354 }
3355
3356 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3357 {
3358         struct tun_struct *tun = netdev_priv(dev);
3359
3360         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3361         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3362
3363         switch (tun->flags & TUN_TYPE_MASK) {
3364         case IFF_TUN:
3365                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3366                 break;
3367         case IFF_TAP:
3368                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3369                 break;
3370         }
3371 }
3372
3373 static u32 tun_get_msglevel(struct net_device *dev)
3374 {
3375 #ifdef TUN_DEBUG
3376         struct tun_struct *tun = netdev_priv(dev);
3377         return tun->debug;
3378 #else
3379         return -EOPNOTSUPP;
3380 #endif
3381 }
3382
3383 static void tun_set_msglevel(struct net_device *dev, u32 value)
3384 {
3385 #ifdef TUN_DEBUG
3386         struct tun_struct *tun = netdev_priv(dev);
3387         tun->debug = value;
3388 #endif
3389 }
3390
3391 static int tun_get_coalesce(struct net_device *dev,
3392                             struct ethtool_coalesce *ec)
3393 {
3394         struct tun_struct *tun = netdev_priv(dev);
3395
3396         ec->rx_max_coalesced_frames = tun->rx_batched;
3397
3398         return 0;
3399 }
3400
3401 static int tun_set_coalesce(struct net_device *dev,
3402                             struct ethtool_coalesce *ec)
3403 {
3404         struct tun_struct *tun = netdev_priv(dev);
3405
3406         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3407                 tun->rx_batched = NAPI_POLL_WEIGHT;
3408         else
3409                 tun->rx_batched = ec->rx_max_coalesced_frames;
3410
3411         return 0;
3412 }
3413
3414 static const struct ethtool_ops tun_ethtool_ops = {
3415         .get_drvinfo    = tun_get_drvinfo,
3416         .get_msglevel   = tun_get_msglevel,
3417         .set_msglevel   = tun_set_msglevel,
3418         .get_link       = ethtool_op_get_link,
3419         .get_ts_info    = ethtool_op_get_ts_info,
3420         .get_coalesce   = tun_get_coalesce,
3421         .set_coalesce   = tun_set_coalesce,
3422         .get_link_ksettings = tun_get_link_ksettings,
3423         .set_link_ksettings = tun_set_link_ksettings,
3424 };
3425
3426 static int tun_queue_resize(struct tun_struct *tun)
3427 {
3428         struct net_device *dev = tun->dev;
3429         struct tun_file *tfile;
3430         struct ptr_ring **rings;
3431         int n = tun->numqueues + tun->numdisabled;
3432         int ret, i;
3433
3434         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3435         if (!rings)
3436                 return -ENOMEM;
3437
3438         for (i = 0; i < tun->numqueues; i++) {
3439                 tfile = rtnl_dereference(tun->tfiles[i]);
3440                 rings[i] = &tfile->tx_ring;
3441         }
3442         list_for_each_entry(tfile, &tun->disabled, next)
3443                 rings[i++] = &tfile->tx_ring;
3444
3445         ret = ptr_ring_resize_multiple(rings, n,
3446                                        dev->tx_queue_len, GFP_KERNEL,
3447                                        tun_ptr_free);
3448
3449         kfree(rings);
3450         return ret;
3451 }
3452
3453 static int tun_device_event(struct notifier_block *unused,
3454                             unsigned long event, void *ptr)
3455 {
3456         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3457         struct tun_struct *tun = netdev_priv(dev);
3458
3459         if (dev->rtnl_link_ops != &tun_link_ops)
3460                 return NOTIFY_DONE;
3461
3462         switch (event) {
3463         case NETDEV_CHANGE_TX_QUEUE_LEN:
3464                 if (tun_queue_resize(tun))
3465                         return NOTIFY_BAD;
3466                 break;
3467         default:
3468                 break;
3469         }
3470
3471         return NOTIFY_DONE;
3472 }
3473
3474 static struct notifier_block tun_notifier_block __read_mostly = {
3475         .notifier_call  = tun_device_event,
3476 };
3477
3478 static int __init tun_init(void)
3479 {
3480         int ret = 0;
3481
3482         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3483
3484         ret = rtnl_link_register(&tun_link_ops);
3485         if (ret) {
3486                 pr_err("Can't register link_ops\n");
3487                 goto err_linkops;
3488         }
3489
3490         ret = misc_register(&tun_miscdev);
3491         if (ret) {
3492                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3493                 goto err_misc;
3494         }
3495
3496         ret = register_netdevice_notifier(&tun_notifier_block);
3497         if (ret) {
3498                 pr_err("Can't register netdevice notifier\n");
3499                 goto err_notifier;
3500         }
3501
3502         return  0;
3503
3504 err_notifier:
3505         misc_deregister(&tun_miscdev);
3506 err_misc:
3507         rtnl_link_unregister(&tun_link_ops);
3508 err_linkops:
3509         return ret;
3510 }
3511
3512 static void tun_cleanup(void)
3513 {
3514         misc_deregister(&tun_miscdev);
3515         rtnl_link_unregister(&tun_link_ops);
3516         unregister_netdevice_notifier(&tun_notifier_block);
3517 }
3518
3519 /* Get an underlying socket object from tun file.  Returns error unless file is
3520  * attached to a device.  The returned object works like a packet socket, it
3521  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3522  * holding a reference to the file for as long as the socket is in use. */
3523 struct socket *tun_get_socket(struct file *file)
3524 {
3525         struct tun_file *tfile;
3526         if (file->f_op != &tun_fops)
3527                 return ERR_PTR(-EINVAL);
3528         tfile = file->private_data;
3529         if (!tfile)
3530                 return ERR_PTR(-EBADFD);
3531         return &tfile->socket;
3532 }
3533 EXPORT_SYMBOL_GPL(tun_get_socket);
3534
3535 struct ptr_ring *tun_get_tx_ring(struct file *file)
3536 {
3537         struct tun_file *tfile;
3538
3539         if (file->f_op != &tun_fops)
3540                 return ERR_PTR(-EINVAL);
3541         tfile = file->private_data;
3542         if (!tfile)
3543                 return ERR_PTR(-EBADFD);
3544         return &tfile->tx_ring;
3545 }
3546 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3547
3548 module_init(tun_init);
3549 module_exit(tun_cleanup);
3550 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3551 MODULE_AUTHOR(DRV_COPYRIGHT);
3552 MODULE_LICENSE("GPL");
3553 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3554 MODULE_ALIAS("devname:net/tun");