drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <[email protected]> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <[email protected]>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <[email protected]>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <[email protected]>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <[email protected]>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72
  73 #include <linux/uaccess.h>
  74 #include <linux/proc_fs.h>
  75
  76 static void tun_default_link_ksettings(struct net_device *dev,
  77                                        struct ethtool_link_ksettings *cmd);
  78
  79 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  80
  81 /* TUN device flags */
  82
  83 /* IFF_ATTACH_QUEUE is never stored in device flags,
  84  * overload it to mean fasync when stored there.
  85  */
  86 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  87 /* High bits in flags field are unused. */
  88 #define TUN_VNET_LE     0x80000000
  89 #define TUN_VNET_BE     0x40000000
  90
  91 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
  92                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
  93
  94 #define GOODCOPY_LEN 128
  95
  96 #define FLT_EXACT_COUNT 8
  97 struct tap_filter {
  98         unsigned int    count;    /* Number of addrs. Zero means disabled */
  99         u32             mask[2];  /* Mask of the hashed addrs */
 100         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 101 };
 102
 103 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 104  * to max number of VCPUs in guest. */
 105 #define MAX_TAP_QUEUES 256
 106 #define MAX_TAP_FLOWS  4096
 107
 108 #define TUN_FLOW_EXPIRE (3 * HZ)
 109
 110 /* A tun_file connects an open character device to a tuntap netdevice. It
 111  * also contains all socket related structures (except sock_fprog and tap_filter)
 112  * to serve as one transmit queue for tuntap device. The sock_fprog and
 113  * tap_filter were kept in tun_struct since they were used for filtering for the
 114  * netdevice not for a specific queue (at least I didn't see the requirement for
 115  * this).
 116  *
 117  * RCU usage:
 118  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 119  * other can only be read while rcu_read_lock or rtnl_lock is held.
 120  */
 121 struct tun_file {
 122         struct sock sk;
 123         struct socket socket;
 124         struct tun_struct __rcu *tun;
 125         struct fasync_struct *fasync;
 126         /* only used for fasnyc */
 127         unsigned int flags;
 128         union {
 129                 u16 queue_index;
 130                 unsigned int ifindex;
 131         };
 132         struct napi_struct napi;
 133         bool napi_enabled;
 134         bool napi_frags_enabled;
 135         struct mutex napi_mutex;        /* Protects access to the above napi */
 136         struct list_head next;
 137         struct tun_struct *detached;
 138         struct ptr_ring tx_ring;
 139         struct xdp_rxq_info xdp_rxq;
 140 };
 141
 142 struct tun_page {
 143         struct page *page;
 144         int count;
 145 };
 146
 147 struct tun_flow_entry {
 148         struct hlist_node hash_link;
 149         struct rcu_head rcu;
 150         struct tun_struct *tun;
 151
 152         u32 rxhash;
 153         u32 rps_rxhash;
 154         int queue_index;
 155         unsigned long updated ____cacheline_aligned_in_smp;
 156 };
 157
 158 #define TUN_NUM_FLOW_ENTRIES 1024
 159 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 160
 161 struct tun_prog {
 162         struct rcu_head rcu;
 163         struct bpf_prog *prog;
 164 };
 165
 166 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 167  * device, socket filter, sndbuf and vnet header size were restore when the
 168  * file were attached to a persist device.
 169  */
 170 struct tun_struct {
 171         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 172         unsigned int            numqueues;
 173         unsigned int            flags;
 174         kuid_t                  owner;
 175         kgid_t                  group;
 176
 177         struct net_device       *dev;
 178         netdev_features_t       set_features;
 179 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 180                           NETIF_F_TSO6)
 181
 182         int                     align;
 183         int                     vnet_hdr_sz;
 184         int                     sndbuf;
 185         struct tap_filter       txflt;
 186         struct sock_fprog       fprog;
 187         /* protected by rtnl lock */
 188         bool                    filter_attached;
 189         u32                     msg_enable;
 190         spinlock_t lock;
 191         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 192         struct timer_list flow_gc_timer;
 193         unsigned long ageing_time;
 194         unsigned int numdisabled;
 195         struct list_head disabled;
 196         void *security;
 197         u32 flow_count;
 198         u32 rx_batched;
 199         atomic_long_t rx_frame_errors;
 200         struct bpf_prog __rcu *xdp_prog;
 201         struct tun_prog __rcu *steering_prog;
 202         struct tun_prog __rcu *filter_prog;
 203         struct ethtool_link_ksettings link_ksettings;
 204 };
 205
 206 struct veth {
 207         __be16 h_vlan_proto;
 208         __be16 h_vlan_TCI;
 209 };
 210
 211 static int tun_napi_receive(struct napi_struct *napi, int budget)
 212 {
 213         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 214         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 215         struct sk_buff_head process_queue;
 216         struct sk_buff *skb;
 217         int received = 0;
 218
 219         __skb_queue_head_init(&process_queue);
 220
 221         spin_lock(&queue->lock);
 222         skb_queue_splice_tail_init(queue, &process_queue);
 223         spin_unlock(&queue->lock);
 224
 225         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 226                 napi_gro_receive(napi, skb);
 227                 ++received;
 228         }
 229
 230         if (!skb_queue_empty(&process_queue)) {
 231                 spin_lock(&queue->lock);
 232                 skb_queue_splice(&process_queue, queue);
 233                 spin_unlock(&queue->lock);
 234         }
 235
 236         return received;
 237 }
 238
 239 static int tun_napi_poll(struct napi_struct *napi, int budget)
 240 {
 241         unsigned int received;
 242
 243         received = tun_napi_receive(napi, budget);
 244
 245         if (received < budget)
 246                 napi_complete_done(napi, received);
 247
 248         return received;
 249 }
 250
 251 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 252                           bool napi_en, bool napi_frags)
 253 {
 254         tfile->napi_enabled = napi_en;
 255         tfile->napi_frags_enabled = napi_en && napi_frags;
 256         if (napi_en) {
 257                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 258                                   NAPI_POLL_WEIGHT);
 259                 napi_enable(&tfile->napi);
 260         }
 261 }
 262
 263 static void tun_napi_disable(struct tun_file *tfile)
 264 {
 265         if (tfile->napi_enabled)
 266                 napi_disable(&tfile->napi);
 267 }
 268
 269 static void tun_napi_del(struct tun_file *tfile)
 270 {
 271         if (tfile->napi_enabled)
 272                 netif_napi_del(&tfile->napi);
 273 }
 274
 275 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 276 {
 277         return tfile->napi_frags_enabled;
 278 }
 279
 280 #ifdef CONFIG_TUN_VNET_CROSS_LE
 281 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 282 {
 283         return tun->flags & TUN_VNET_BE ? false :
 284                 virtio_legacy_is_little_endian();
 285 }
 286
 287 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 288 {
 289         int be = !!(tun->flags & TUN_VNET_BE);
 290
 291         if (put_user(be, argp))
 292                 return -EFAULT;
 293
 294         return 0;
 295 }
 296
 297 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 298 {
 299         int be;
 300
 301         if (get_user(be, argp))
 302                 return -EFAULT;
 303
 304         if (be)
 305                 tun->flags |= TUN_VNET_BE;
 306         else
 307                 tun->flags &= ~TUN_VNET_BE;
 308
 309         return 0;
 310 }
 311 #else
 312 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 313 {
 314         return virtio_legacy_is_little_endian();
 315 }
 316
 317 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 318 {
 319         return -EINVAL;
 320 }
 321
 322 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 323 {
 324         return -EINVAL;
 325 }
 326 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 327
 328 static inline bool tun_is_little_endian(struct tun_struct *tun)
 329 {
 330         return tun->flags & TUN_VNET_LE ||
 331                 tun_legacy_is_little_endian(tun);
 332 }
 333
 334 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 335 {
 336         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 337 }
 338
 339 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 340 {
 341         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 342 }
 343
 344 static inline u32 tun_hashfn(u32 rxhash)
 345 {
 346         return rxhash & TUN_MASK_FLOW_ENTRIES;
 347 }
 348
 349 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 350 {
 351         struct tun_flow_entry *e;
 352
 353         hlist_for_each_entry_rcu(e, head, hash_link) {
 354                 if (e->rxhash == rxhash)
 355                         return e;
 356         }
 357         return NULL;
 358 }
 359
 360 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 361                                               struct hlist_head *head,
 362                                               u32 rxhash, u16 queue_index)
 363 {
 364         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 365
 366         if (e) {
 367                 netif_info(tun, tx_queued, tun->dev,
 368                            "create flow: hash %u index %u\n",
 369                            rxhash, queue_index);
 370                 e->updated = jiffies;
 371                 e->rxhash = rxhash;
 372                 e->rps_rxhash = 0;
 373                 e->queue_index = queue_index;
 374                 e->tun = tun;
 375                 hlist_add_head_rcu(&e->hash_link, head);
 376                 ++tun->flow_count;
 377         }
 378         return e;
 379 }
 380
 381 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 382 {
 383         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 384                    e->rxhash, e->queue_index);
 385         hlist_del_rcu(&e->hash_link);
 386         kfree_rcu(e, rcu);
 387         --tun->flow_count;
 388 }
 389
 390 static void tun_flow_flush(struct tun_struct *tun)
 391 {
 392         int i;
 393
 394         spin_lock_bh(&tun->lock);
 395         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 396                 struct tun_flow_entry *e;
 397                 struct hlist_node *n;
 398
 399                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 400                         tun_flow_delete(tun, e);
 401         }
 402         spin_unlock_bh(&tun->lock);
 403 }
 404
 405 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 406 {
 407         int i;
 408
 409         spin_lock_bh(&tun->lock);
 410         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 411                 struct tun_flow_entry *e;
 412                 struct hlist_node *n;
 413
 414                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 415                         if (e->queue_index == queue_index)
 416                                 tun_flow_delete(tun, e);
 417                 }
 418         }
 419         spin_unlock_bh(&tun->lock);
 420 }
 421
 422 static void tun_flow_cleanup(struct timer_list *t)
 423 {
 424         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 425         unsigned long delay = tun->ageing_time;
 426         unsigned long next_timer = jiffies + delay;
 427         unsigned long count = 0;
 428         int i;
 429
 430         spin_lock(&tun->lock);
 431         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 432                 struct tun_flow_entry *e;
 433                 struct hlist_node *n;
 434
 435                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 436                         unsigned long this_timer;
 437
 438                         this_timer = e->updated + delay;
 439                         if (time_before_eq(this_timer, jiffies)) {
 440                                 tun_flow_delete(tun, e);
 441                                 continue;
 442                         }
 443                         count++;
 444                         if (time_before(this_timer, next_timer))
 445                                 next_timer = this_timer;
 446                 }
 447         }
 448
 449         if (count)
 450                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 451         spin_unlock(&tun->lock);
 452 }
 453
 454 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 455                             struct tun_file *tfile)
 456 {
 457         struct hlist_head *head;
 458         struct tun_flow_entry *e;
 459         unsigned long delay = tun->ageing_time;
 460         u16 queue_index = tfile->queue_index;
 461
 462         head = &tun->flows[tun_hashfn(rxhash)];
 463
 464         rcu_read_lock();
 465
 466         e = tun_flow_find(head, rxhash);
 467         if (likely(e)) {
 468                 /* TODO: keep queueing to old queue until it's empty? */
 469                 if (READ_ONCE(e->queue_index) != queue_index)
 470                         WRITE_ONCE(e->queue_index, queue_index);
 471                 if (e->updated != jiffies)
 472                         e->updated = jiffies;
 473                 sock_rps_record_flow_hash(e->rps_rxhash);
 474         } else {
 475                 spin_lock_bh(&tun->lock);
 476                 if (!tun_flow_find(head, rxhash) &&
 477                     tun->flow_count < MAX_TAP_FLOWS)
 478                         tun_flow_create(tun, head, rxhash, queue_index);
 479
 480                 if (!timer_pending(&tun->flow_gc_timer))
 481                         mod_timer(&tun->flow_gc_timer,
 482                                   round_jiffies_up(jiffies + delay));
 483                 spin_unlock_bh(&tun->lock);
 484         }
 485
 486         rcu_read_unlock();
 487 }
 488
 489 /* Save the hash received in the stack receive path and update the
 490  * flow_hash table accordingly.
 491  */
 492 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 493 {
 494         if (unlikely(e->rps_rxhash != hash))
 495                 e->rps_rxhash = hash;
 496 }
 497
 498 /* We try to identify a flow through its rxhash. The reason that
 499  * we do not check rxq no. is because some cards(e.g 82599), chooses
 500  * the rxq based on the txq where the last packet of the flow comes. As
 501  * the userspace application move between processors, we may get a
 502  * different rxq no. here.
 503  */
 504 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 505 {
 506         struct tun_flow_entry *e;
 507         u32 txq = 0;
 508         u32 numqueues = 0;
 509
 510         numqueues = READ_ONCE(tun->numqueues);
 511
 512         txq = __skb_get_hash_symmetric(skb);
 513         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 514         if (e) {
 515                 tun_flow_save_rps_rxhash(e, txq);
 516                 txq = e->queue_index;
 517         } else {
 518                 /* use multiply and shift instead of expensive divide */
 519                 txq = ((u64)txq * numqueues) >> 32;
 520         }
 521
 522         return txq;
 523 }
 524
 525 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 526 {
 527         struct tun_prog *prog;
 528         u32 numqueues;
 529         u16 ret = 0;
 530
 531         numqueues = READ_ONCE(tun->numqueues);
 532         if (!numqueues)
 533                 return 0;
 534
 535         prog = rcu_dereference(tun->steering_prog);
 536         if (prog)
 537                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 538
 539         return ret % numqueues;
 540 }
 541
 542 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 543                             struct net_device *sb_dev)
 544 {
 545         struct tun_struct *tun = netdev_priv(dev);
 546         u16 ret;
 547
 548         rcu_read_lock();
 549         if (rcu_dereference(tun->steering_prog))
 550                 ret = tun_ebpf_select_queue(tun, skb);
 551         else
 552                 ret = tun_automq_select_queue(tun, skb);
 553         rcu_read_unlock();
 554
 555         return ret;
 556 }
 557
 558 static inline bool tun_not_capable(struct tun_struct *tun)
 559 {
 560         const struct cred *cred = current_cred();
 561         struct net *net = dev_net(tun->dev);
 562
 563         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 564                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 565                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 566 }
 567
 568 static void tun_set_real_num_queues(struct tun_struct *tun)
 569 {
 570         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 571         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 572 }
 573
 574 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 575 {
 576         tfile->detached = tun;
 577         list_add_tail(&tfile->next, &tun->disabled);
 578         ++tun->numdisabled;
 579 }
 580
 581 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 582 {
 583         struct tun_struct *tun = tfile->detached;
 584
 585         tfile->detached = NULL;
 586         list_del_init(&tfile->next);
 587         --tun->numdisabled;
 588         return tun;
 589 }
 590
 591 void tun_ptr_free(void *ptr)
 592 {
 593         if (!ptr)
 594                 return;
 595         if (tun_is_xdp_frame(ptr)) {
 596                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 597
 598                 xdp_return_frame(xdpf);
 599         } else {
 600                 __skb_array_destroy_skb(ptr);
 601         }
 602 }
 603 EXPORT_SYMBOL_GPL(tun_ptr_free);
 604
 605 static void tun_queue_purge(struct tun_file *tfile)
 606 {
 607         void *ptr;
 608
 609         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 610                 tun_ptr_free(ptr);
 611
 612         skb_queue_purge(&tfile->sk.sk_write_queue);
 613         skb_queue_purge(&tfile->sk.sk_error_queue);
 614 }
 615
 616 static void __tun_detach(struct tun_file *tfile, bool clean)
 617 {
 618         struct tun_file *ntfile;
 619         struct tun_struct *tun;
 620
 621         tun = rtnl_dereference(tfile->tun);
 622
 623         if (tun && clean) {
 624                 tun_napi_disable(tfile);
 625                 tun_napi_del(tfile);
 626         }
 627
 628         if (tun && !tfile->detached) {
 629                 u16 index = tfile->queue_index;
 630                 BUG_ON(index >= tun->numqueues);
 631
 632                 rcu_assign_pointer(tun->tfiles[index],
 633                                    tun->tfiles[tun->numqueues - 1]);
 634                 ntfile = rtnl_dereference(tun->tfiles[index]);
 635                 ntfile->queue_index = index;
 636                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 637                                    NULL);
 638
 639                 --tun->numqueues;
 640                 if (clean) {
 641                         RCU_INIT_POINTER(tfile->tun, NULL);
 642                         sock_put(&tfile->sk);
 643                 } else
 644                         tun_disable_queue(tun, tfile);
 645
 646                 synchronize_net();
 647                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 648                 /* Drop read queue */
 649                 tun_queue_purge(tfile);
 650                 tun_set_real_num_queues(tun);
 651         } else if (tfile->detached && clean) {
 652                 tun = tun_enable_queue(tfile);
 653                 sock_put(&tfile->sk);
 654         }
 655
 656         if (clean) {
 657                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 658                         netif_carrier_off(tun->dev);
 659
 660                         if (!(tun->flags & IFF_PERSIST) &&
 661                             tun->dev->reg_state == NETREG_REGISTERED)
 662                                 unregister_netdevice(tun->dev);
 663                 }
 664                 if (tun)
 665                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 666                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 667                 sock_put(&tfile->sk);
 668         }
 669 }
 670
 671 static void tun_detach(struct tun_file *tfile, bool clean)
 672 {
 673         struct tun_struct *tun;
 674         struct net_device *dev;
 675
 676         rtnl_lock();
 677         tun = rtnl_dereference(tfile->tun);
 678         dev = tun ? tun->dev : NULL;
 679         __tun_detach(tfile, clean);
 680         if (dev)
 681                 netdev_state_change(dev);
 682         rtnl_unlock();
 683 }
 684
 685 static void tun_detach_all(struct net_device *dev)
 686 {
 687         struct tun_struct *tun = netdev_priv(dev);
 688         struct tun_file *tfile, *tmp;
 689         int i, n = tun->numqueues;
 690
 691         for (i = 0; i < n; i++) {
 692                 tfile = rtnl_dereference(tun->tfiles[i]);
 693                 BUG_ON(!tfile);
 694                 tun_napi_disable(tfile);
 695                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 696                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 697                 RCU_INIT_POINTER(tfile->tun, NULL);
 698                 --tun->numqueues;
 699         }
 700         list_for_each_entry(tfile, &tun->disabled, next) {
 701                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 702                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 703                 RCU_INIT_POINTER(tfile->tun, NULL);
 704         }
 705         BUG_ON(tun->numqueues != 0);
 706
 707         synchronize_net();
 708         for (i = 0; i < n; i++) {
 709                 tfile = rtnl_dereference(tun->tfiles[i]);
 710                 tun_napi_del(tfile);
 711                 /* Drop read queue */
 712                 tun_queue_purge(tfile);
 713                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 714                 sock_put(&tfile->sk);
 715         }
 716         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 717                 tun_enable_queue(tfile);
 718                 tun_queue_purge(tfile);
 719                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 720                 sock_put(&tfile->sk);
 721         }
 722         BUG_ON(tun->numdisabled != 0);
 723
 724         if (tun->flags & IFF_PERSIST)
 725                 module_put(THIS_MODULE);
 726 }
 727
 728 static int tun_attach(struct tun_struct *tun, struct file *file,
 729                       bool skip_filter, bool napi, bool napi_frags,
 730                       bool publish_tun)
 731 {
 732         struct tun_file *tfile = file->private_data;
 733         struct net_device *dev = tun->dev;
 734         int err;
 735
 736         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 737         if (err < 0)
 738                 goto out;
 739
 740         err = -EINVAL;
 741         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 742                 goto out;
 743
 744         err = -EBUSY;
 745         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 746                 goto out;
 747
 748         err = -E2BIG;
 749         if (!tfile->detached &&
 750             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 751                 goto out;
 752
 753         err = 0;
 754
 755         /* Re-attach the filter to persist device */
 756         if (!skip_filter && (tun->filter_attached == true)) {
 757                 lock_sock(tfile->socket.sk);
 758                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 759                 release_sock(tfile->socket.sk);
 760                 if (!err)
 761                         goto out;
 762         }
 763
 764         if (!tfile->detached &&
 765             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 766                             GFP_KERNEL, tun_ptr_free)) {
 767                 err = -ENOMEM;
 768                 goto out;
 769         }
 770
 771         tfile->queue_index = tun->numqueues;
 772         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 773
 774         if (tfile->detached) {
 775                 /* Re-attach detached tfile, updating XDP queue_index */
 776                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 777
 778                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 779                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 780         } else {
 781                 /* Setup XDP RX-queue info, for new tfile getting attached */
 782                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 783                                        tun->dev, tfile->queue_index, 0);
 784                 if (err < 0)
 785                         goto out;
 786                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 787                                                  MEM_TYPE_PAGE_SHARED, NULL);
 788                 if (err < 0) {
 789                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 790                         goto out;
 791                 }
 792                 err = 0;
 793         }
 794
 795         if (tfile->detached) {
 796                 tun_enable_queue(tfile);
 797         } else {
 798                 sock_hold(&tfile->sk);
 799                 tun_napi_init(tun, tfile, napi, napi_frags);
 800         }
 801
 802         if (rtnl_dereference(tun->xdp_prog))
 803                 sock_set_flag(&tfile->sk, SOCK_XDP);
 804
 805         /* device is allowed to go away first, so no need to hold extra
 806          * refcnt.
 807          */
 808
 809         /* Publish tfile->tun and tun->tfiles only after we've fully
 810          * initialized tfile; otherwise we risk using half-initialized
 811          * object.
 812          */
 813         if (publish_tun)
 814                 rcu_assign_pointer(tfile->tun, tun);
 815         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 816         tun->numqueues++;
 817         tun_set_real_num_queues(tun);
 818 out:
 819         return err;
 820 }
 821
 822 static struct tun_struct *tun_get(struct tun_file *tfile)
 823 {
 824         struct tun_struct *tun;
 825
 826         rcu_read_lock();
 827         tun = rcu_dereference(tfile->tun);
 828         if (tun)
 829                 dev_hold(tun->dev);
 830         rcu_read_unlock();
 831
 832         return tun;
 833 }
 834
 835 static void tun_put(struct tun_struct *tun)
 836 {
 837         dev_put(tun->dev);
 838 }
 839
 840 /* TAP filtering */
 841 static void addr_hash_set(u32 *mask, const u8 *addr)
 842 {
 843         int n = ether_crc(ETH_ALEN, addr) >> 26;
 844         mask[n >> 5] |= (1 << (n & 31));
 845 }
 846
 847 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 848 {
 849         int n = ether_crc(ETH_ALEN, addr) >> 26;
 850         return mask[n >> 5] & (1 << (n & 31));
 851 }
 852
 853 static int update_filter(struct tap_filter *filter, void __user *arg)
 854 {
 855         struct { u8 u[ETH_ALEN]; } *addr;
 856         struct tun_filter uf;
 857         int err, alen, n, nexact;
 858
 859         if (copy_from_user(&uf, arg, sizeof(uf)))
 860                 return -EFAULT;
 861
 862         if (!uf.count) {
 863                 /* Disabled */
 864                 filter->count = 0;
 865                 return 0;
 866         }
 867
 868         alen = ETH_ALEN * uf.count;
 869         addr = memdup_user(arg + sizeof(uf), alen);
 870         if (IS_ERR(addr))
 871                 return PTR_ERR(addr);
 872
 873         /* The filter is updated without holding any locks. Which is
 874          * perfectly safe. We disable it first and in the worst
 875          * case we'll accept a few undesired packets. */
 876         filter->count = 0;
 877         wmb();
 878
 879         /* Use first set of addresses as an exact filter */
 880         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 881                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 882
 883         nexact = n;
 884
 885         /* Remaining multicast addresses are hashed,
 886          * unicast will leave the filter disabled. */
 887         memset(filter->mask, 0, sizeof(filter->mask));
 888         for (; n < uf.count; n++) {
 889                 if (!is_multicast_ether_addr(addr[n].u)) {
 890                         err = 0; /* no filter */
 891                         goto free_addr;
 892                 }
 893                 addr_hash_set(filter->mask, addr[n].u);
 894         }
 895
 896         /* For ALLMULTI just set the mask to all ones.
 897          * This overrides the mask populated above. */
 898         if ((uf.flags & TUN_FLT_ALLMULTI))
 899                 memset(filter->mask, ~0, sizeof(filter->mask));
 900
 901         /* Now enable the filter */
 902         wmb();
 903         filter->count = nexact;
 904
 905         /* Return the number of exact filters */
 906         err = nexact;
 907 free_addr:
 908         kfree(addr);
 909         return err;
 910 }
 911
 912 /* Returns: 0 - drop, !=0 - accept */
 913 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 914 {
 915         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 916          * at this point. */
 917         struct ethhdr *eh = (struct ethhdr *) skb->data;
 918         int i;
 919
 920         /* Exact match */
 921         for (i = 0; i < filter->count; i++)
 922                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 923                         return 1;
 924
 925         /* Inexact match (multicast only) */
 926         if (is_multicast_ether_addr(eh->h_dest))
 927                 return addr_hash_test(filter->mask, eh->h_dest);
 928
 929         return 0;
 930 }
 931
 932 /*
 933  * Checks whether the packet is accepted or not.
 934  * Returns: 0 - drop, !=0 - accept
 935  */
 936 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 937 {
 938         if (!filter->count)
 939                 return 1;
 940
 941         return run_filter(filter, skb);
 942 }
 943
 944 /* Network device part of the driver */
 945
 946 static const struct ethtool_ops tun_ethtool_ops;
 947
 948 /* Net device detach from fd. */
 949 static void tun_net_uninit(struct net_device *dev)
 950 {
 951         tun_detach_all(dev);
 952 }
 953
 954 /* Net device open. */
 955 static int tun_net_open(struct net_device *dev)
 956 {
 957         netif_tx_start_all_queues(dev);
 958
 959         return 0;
 960 }
 961
 962 /* Net device close. */
 963 static int tun_net_close(struct net_device *dev)
 964 {
 965         netif_tx_stop_all_queues(dev);
 966         return 0;
 967 }
 968
 969 /* Net device start xmit */
 970 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
 971 {
 972 #ifdef CONFIG_RPS
 973         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
 974                 /* Select queue was not called for the skbuff, so we extract the
 975                  * RPS hash and save it into the flow_table here.
 976                  */
 977                 struct tun_flow_entry *e;
 978                 __u32 rxhash;
 979
 980                 rxhash = __skb_get_hash_symmetric(skb);
 981                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
 982                 if (e)
 983                         tun_flow_save_rps_rxhash(e, rxhash);
 984         }
 985 #endif
 986 }
 987
 988 static unsigned int run_ebpf_filter(struct tun_struct *tun,
 989                                     struct sk_buff *skb,
 990                                     int len)
 991 {
 992         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
 993
 994         if (prog)
 995                 len = bpf_prog_run_clear_cb(prog->prog, skb);
 996
 997         return len;
 998 }
 999
1000 /* Net device start xmit */
1001 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1002 {
1003         struct tun_struct *tun = netdev_priv(dev);
1004         int txq = skb->queue_mapping;
1005         struct tun_file *tfile;
1006         int len = skb->len;
1007
1008         rcu_read_lock();
1009         tfile = rcu_dereference(tun->tfiles[txq]);
1010
1011         /* Drop packet if interface is not attached */
1012         if (!tfile)
1013                 goto drop;
1014
1015         if (!rcu_dereference(tun->steering_prog))
1016                 tun_automq_xmit(tun, skb);
1017
1018         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1019
1020         /* Drop if the filter does not like it.
1021          * This is a noop if the filter is disabled.
1022          * Filter can be enabled only for the TAP devices. */
1023         if (!check_filter(&tun->txflt, skb))
1024                 goto drop;
1025
1026         if (tfile->socket.sk->sk_filter &&
1027             sk_filter(tfile->socket.sk, skb))
1028                 goto drop;
1029
1030         len = run_ebpf_filter(tun, skb, len);
1031         if (len == 0 || pskb_trim(skb, len))
1032                 goto drop;
1033
1034         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1035                 goto drop;
1036
1037         skb_tx_timestamp(skb);
1038
1039         /* Orphan the skb - required as we might hang on to it
1040          * for indefinite time.
1041          */
1042         skb_orphan(skb);
1043
1044         nf_reset_ct(skb);
1045
1046         if (ptr_ring_produce(&tfile->tx_ring, skb))
1047                 goto drop;
1048
1049         /* Notify and wake up reader process */
1050         if (tfile->flags & TUN_FASYNC)
1051                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1052         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1053
1054         rcu_read_unlock();
1055         return NETDEV_TX_OK;
1056
1057 drop:
1058         atomic_long_inc(&dev->tx_dropped);
1059         skb_tx_error(skb);
1060         kfree_skb(skb);
1061         rcu_read_unlock();
1062         return NET_XMIT_DROP;
1063 }
1064
1065 static void tun_net_mclist(struct net_device *dev)
1066 {
1067         /*
1068          * This callback is supposed to deal with mc filter in
1069          * _rx_ path and has nothing to do with the _tx_ path.
1070          * In rx path we always accept everything userspace gives us.
1071          */
1072 }
1073
1074 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1075         netdev_features_t features)
1076 {
1077         struct tun_struct *tun = netdev_priv(dev);
1078
1079         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1080 }
1081
1082 static void tun_set_headroom(struct net_device *dev, int new_hr)
1083 {
1084         struct tun_struct *tun = netdev_priv(dev);
1085
1086         if (new_hr < NET_SKB_PAD)
1087                 new_hr = NET_SKB_PAD;
1088
1089         tun->align = new_hr;
1090 }
1091
1092 static void
1093 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1094 {
1095         struct tun_struct *tun = netdev_priv(dev);
1096
1097         dev_get_tstats64(dev, stats);
1098
1099         stats->rx_frame_errors +=
1100                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1101 }
1102
1103 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1104                        struct netlink_ext_ack *extack)
1105 {
1106         struct tun_struct *tun = netdev_priv(dev);
1107         struct tun_file *tfile;
1108         struct bpf_prog *old_prog;
1109         int i;
1110
1111         old_prog = rtnl_dereference(tun->xdp_prog);
1112         rcu_assign_pointer(tun->xdp_prog, prog);
1113         if (old_prog)
1114                 bpf_prog_put(old_prog);
1115
1116         for (i = 0; i < tun->numqueues; i++) {
1117                 tfile = rtnl_dereference(tun->tfiles[i]);
1118                 if (prog)
1119                         sock_set_flag(&tfile->sk, SOCK_XDP);
1120                 else
1121                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1122         }
1123         list_for_each_entry(tfile, &tun->disabled, next) {
1124                 if (prog)
1125                         sock_set_flag(&tfile->sk, SOCK_XDP);
1126                 else
1127                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1128         }
1129
1130         return 0;
1131 }
1132
1133 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1134 {
1135         switch (xdp->command) {
1136         case XDP_SETUP_PROG:
1137                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1138         default:
1139                 return -EINVAL;
1140         }
1141 }
1142
1143 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1144 {
1145         if (new_carrier) {
1146                 struct tun_struct *tun = netdev_priv(dev);
1147
1148                 if (!tun->numqueues)
1149                         return -EPERM;
1150
1151                 netif_carrier_on(dev);
1152         } else {
1153                 netif_carrier_off(dev);
1154         }
1155         return 0;
1156 }
1157
1158 static const struct net_device_ops tun_netdev_ops = {
1159         .ndo_uninit             = tun_net_uninit,
1160         .ndo_open               = tun_net_open,
1161         .ndo_stop               = tun_net_close,
1162         .ndo_start_xmit         = tun_net_xmit,
1163         .ndo_fix_features       = tun_net_fix_features,
1164         .ndo_select_queue       = tun_select_queue,
1165         .ndo_set_rx_headroom    = tun_set_headroom,
1166         .ndo_get_stats64        = tun_net_get_stats64,
1167         .ndo_change_carrier     = tun_net_change_carrier,
1168 };
1169
1170 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1171 {
1172         /* Notify and wake up reader process */
1173         if (tfile->flags & TUN_FASYNC)
1174                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1175         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1176 }
1177
1178 static int tun_xdp_xmit(struct net_device *dev, int n,
1179                         struct xdp_frame **frames, u32 flags)
1180 {
1181         struct tun_struct *tun = netdev_priv(dev);
1182         struct tun_file *tfile;
1183         u32 numqueues;
1184         int drops = 0;
1185         int cnt = n;
1186         int i;
1187
1188         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1189                 return -EINVAL;
1190
1191         rcu_read_lock();
1192
1193 resample:
1194         numqueues = READ_ONCE(tun->numqueues);
1195         if (!numqueues) {
1196                 rcu_read_unlock();
1197                 return -ENXIO; /* Caller will free/return all frames */
1198         }
1199
1200         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1201                                             numqueues]);
1202         if (unlikely(!tfile))
1203                 goto resample;
1204
1205         spin_lock(&tfile->tx_ring.producer_lock);
1206         for (i = 0; i < n; i++) {
1207                 struct xdp_frame *xdp = frames[i];
1208                 /* Encode the XDP flag into lowest bit for consumer to differ
1209                  * XDP buffer from sk_buff.
1210                  */
1211                 void *frame = tun_xdp_to_ptr(xdp);
1212
1213                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1214                         atomic_long_inc(&dev->tx_dropped);
1215                         xdp_return_frame_rx_napi(xdp);
1216                         drops++;
1217                 }
1218         }
1219         spin_unlock(&tfile->tx_ring.producer_lock);
1220
1221         if (flags & XDP_XMIT_FLUSH)
1222                 __tun_xdp_flush_tfile(tfile);
1223
1224         rcu_read_unlock();
1225         return cnt - drops;
1226 }
1227
1228 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1229 {
1230         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1231
1232         if (unlikely(!frame))
1233                 return -EOVERFLOW;
1234
1235         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1236 }
1237
1238 static const struct net_device_ops tap_netdev_ops = {
1239         .ndo_uninit             = tun_net_uninit,
1240         .ndo_open               = tun_net_open,
1241         .ndo_stop               = tun_net_close,
1242         .ndo_start_xmit         = tun_net_xmit,
1243         .ndo_fix_features       = tun_net_fix_features,
1244         .ndo_set_rx_mode        = tun_net_mclist,
1245         .ndo_set_mac_address    = eth_mac_addr,
1246         .ndo_validate_addr      = eth_validate_addr,
1247         .ndo_select_queue       = tun_select_queue,
1248         .ndo_features_check     = passthru_features_check,
1249         .ndo_set_rx_headroom    = tun_set_headroom,
1250         .ndo_get_stats64        = dev_get_tstats64,
1251         .ndo_bpf                = tun_xdp,
1252         .ndo_xdp_xmit           = tun_xdp_xmit,
1253         .ndo_change_carrier     = tun_net_change_carrier,
1254 };
1255
1256 static void tun_flow_init(struct tun_struct *tun)
1257 {
1258         int i;
1259
1260         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1261                 INIT_HLIST_HEAD(&tun->flows[i]);
1262
1263         tun->ageing_time = TUN_FLOW_EXPIRE;
1264         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1265         mod_timer(&tun->flow_gc_timer,
1266                   round_jiffies_up(jiffies + tun->ageing_time));
1267 }
1268
1269 static void tun_flow_uninit(struct tun_struct *tun)
1270 {
1271         del_timer_sync(&tun->flow_gc_timer);
1272         tun_flow_flush(tun);
1273 }
1274
1275 #define MIN_MTU 68
1276 #define MAX_MTU 65535
1277
1278 /* Initialize net device. */
1279 static void tun_net_init(struct net_device *dev)
1280 {
1281         struct tun_struct *tun = netdev_priv(dev);
1282
1283         switch (tun->flags & TUN_TYPE_MASK) {
1284         case IFF_TUN:
1285                 dev->netdev_ops = &tun_netdev_ops;
1286                 dev->header_ops = &ip_tunnel_header_ops;
1287
1288                 /* Point-to-Point TUN Device */
1289                 dev->hard_header_len = 0;
1290                 dev->addr_len = 0;
1291                 dev->mtu = 1500;
1292
1293                 /* Zero header length */
1294                 dev->type = ARPHRD_NONE;
1295                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1296                 break;
1297
1298         case IFF_TAP:
1299                 dev->netdev_ops = &tap_netdev_ops;
1300                 /* Ethernet TAP Device */
1301                 ether_setup(dev);
1302                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1303                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1304
1305                 eth_hw_addr_random(dev);
1306
1307                 break;
1308         }
1309
1310         dev->min_mtu = MIN_MTU;
1311         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1312 }
1313
1314 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1315 {
1316         struct sock *sk = tfile->socket.sk;
1317
1318         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1319 }
1320
1321 /* Character device part */
1322
1323 /* Poll */
1324 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1325 {
1326         struct tun_file *tfile = file->private_data;
1327         struct tun_struct *tun = tun_get(tfile);
1328         struct sock *sk;
1329         __poll_t mask = 0;
1330
1331         if (!tun)
1332                 return EPOLLERR;
1333
1334         sk = tfile->socket.sk;
1335
1336         poll_wait(file, sk_sleep(sk), wait);
1337
1338         if (!ptr_ring_empty(&tfile->tx_ring))
1339                 mask |= EPOLLIN | EPOLLRDNORM;
1340
1341         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1342          * guarantee EPOLLOUT to be raised by either here or
1343          * tun_sock_write_space(). Then process could get notification
1344          * after it writes to a down device and meets -EIO.
1345          */
1346         if (tun_sock_writeable(tun, tfile) ||
1347             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1348              tun_sock_writeable(tun, tfile)))
1349                 mask |= EPOLLOUT | EPOLLWRNORM;
1350
1351         if (tun->dev->reg_state != NETREG_REGISTERED)
1352                 mask = EPOLLERR;
1353
1354         tun_put(tun);
1355         return mask;
1356 }
1357
1358 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1359                                             size_t len,
1360                                             const struct iov_iter *it)
1361 {
1362         struct sk_buff *skb;
1363         size_t linear;
1364         int err;
1365         int i;
1366
1367         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1368                 return ERR_PTR(-EMSGSIZE);
1369
1370         local_bh_disable();
1371         skb = napi_get_frags(&tfile->napi);
1372         local_bh_enable();
1373         if (!skb)
1374                 return ERR_PTR(-ENOMEM);
1375
1376         linear = iov_iter_single_seg_count(it);
1377         err = __skb_grow(skb, linear);
1378         if (err)
1379                 goto free;
1380
1381         skb->len = len;
1382         skb->data_len = len - linear;
1383         skb->truesize += skb->data_len;
1384
1385         for (i = 1; i < it->nr_segs; i++) {
1386                 size_t fragsz = it->iov[i].iov_len;
1387                 struct page *page;
1388                 void *frag;
1389
1390                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1391                         err = -EINVAL;
1392                         goto free;
1393                 }
1394                 frag = netdev_alloc_frag(fragsz);
1395                 if (!frag) {
1396                         err = -ENOMEM;
1397                         goto free;
1398                 }
1399                 page = virt_to_head_page(frag);
1400                 skb_fill_page_desc(skb, i - 1, page,
1401                                    frag - page_address(page), fragsz);
1402         }
1403
1404         return skb;
1405 free:
1406         /* frees skb and all frags allocated with napi_alloc_frag() */
1407         napi_free_frags(&tfile->napi);
1408         return ERR_PTR(err);
1409 }
1410
1411 /* prepad is the amount to reserve at front.  len is length after that.
1412  * linear is a hint as to how much to copy (usually headers). */
1413 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1414                                      size_t prepad, size_t len,
1415                                      size_t linear, int noblock)
1416 {
1417         struct sock *sk = tfile->socket.sk;
1418         struct sk_buff *skb;
1419         int err;
1420
1421         /* Under a page?  Don't bother with paged skb. */
1422         if (prepad + len < PAGE_SIZE || !linear)
1423                 linear = len;
1424
1425         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1426                                    &err, 0);
1427         if (!skb)
1428                 return ERR_PTR(err);
1429
1430         skb_reserve(skb, prepad);
1431         skb_put(skb, linear);
1432         skb->data_len = len - linear;
1433         skb->len += len - linear;
1434
1435         return skb;
1436 }
1437
1438 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1439                            struct sk_buff *skb, int more)
1440 {
1441         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1442         struct sk_buff_head process_queue;
1443         u32 rx_batched = tun->rx_batched;
1444         bool rcv = false;
1445
1446         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1447                 local_bh_disable();
1448                 skb_record_rx_queue(skb, tfile->queue_index);
1449                 netif_receive_skb(skb);
1450                 local_bh_enable();
1451                 return;
1452         }
1453
1454         spin_lock(&queue->lock);
1455         if (!more || skb_queue_len(queue) == rx_batched) {
1456                 __skb_queue_head_init(&process_queue);
1457                 skb_queue_splice_tail_init(queue, &process_queue);
1458                 rcv = true;
1459         } else {
1460                 __skb_queue_tail(queue, skb);
1461         }
1462         spin_unlock(&queue->lock);
1463
1464         if (rcv) {
1465                 struct sk_buff *nskb;
1466
1467                 local_bh_disable();
1468                 while ((nskb = __skb_dequeue(&process_queue))) {
1469                         skb_record_rx_queue(nskb, tfile->queue_index);
1470                         netif_receive_skb(nskb);
1471                 }
1472                 skb_record_rx_queue(skb, tfile->queue_index);
1473                 netif_receive_skb(skb);
1474                 local_bh_enable();
1475         }
1476 }
1477
1478 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1479                               int len, int noblock, bool zerocopy)
1480 {
1481         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1482                 return false;
1483
1484         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1485                 return false;
1486
1487         if (!noblock)
1488                 return false;
1489
1490         if (zerocopy)
1491                 return false;
1492
1493         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1494             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1495                 return false;
1496
1497         return true;
1498 }
1499
1500 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1501                                        struct page_frag *alloc_frag, char *buf,
1502                                        int buflen, int len, int pad)
1503 {
1504         struct sk_buff *skb = build_skb(buf, buflen);
1505
1506         if (!skb)
1507                 return ERR_PTR(-ENOMEM);
1508
1509         skb_reserve(skb, pad);
1510         skb_put(skb, len);
1511         skb_set_owner_w(skb, tfile->socket.sk);
1512
1513         get_page(alloc_frag->page);
1514         alloc_frag->offset += buflen;
1515
1516         return skb;
1517 }
1518
1519 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1520                        struct xdp_buff *xdp, u32 act)
1521 {
1522         int err;
1523
1524         switch (act) {
1525         case XDP_REDIRECT:
1526                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1527                 if (err)
1528                         return err;
1529                 break;
1530         case XDP_TX:
1531                 err = tun_xdp_tx(tun->dev, xdp);
1532                 if (err < 0)
1533                         return err;
1534                 break;
1535         case XDP_PASS:
1536                 break;
1537         default:
1538                 bpf_warn_invalid_xdp_action(act);
1539                 fallthrough;
1540         case XDP_ABORTED:
1541                 trace_xdp_exception(tun->dev, xdp_prog, act);
1542                 fallthrough;
1543         case XDP_DROP:
1544                 atomic_long_inc(&tun->dev->rx_dropped);
1545                 break;
1546         }
1547
1548         return act;
1549 }
1550
1551 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1552                                      struct tun_file *tfile,
1553                                      struct iov_iter *from,
1554                                      struct virtio_net_hdr *hdr,
1555                                      int len, int *skb_xdp)
1556 {
1557         struct page_frag *alloc_frag = &current->task_frag;
1558         struct bpf_prog *xdp_prog;
1559         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1560         char *buf;
1561         size_t copied;
1562         int pad = TUN_RX_PAD;
1563         int err = 0;
1564
1565         rcu_read_lock();
1566         xdp_prog = rcu_dereference(tun->xdp_prog);
1567         if (xdp_prog)
1568                 pad += XDP_PACKET_HEADROOM;
1569         buflen += SKB_DATA_ALIGN(len + pad);
1570         rcu_read_unlock();
1571
1572         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1573         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1574                 return ERR_PTR(-ENOMEM);
1575
1576         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1577         copied = copy_page_from_iter(alloc_frag->page,
1578                                      alloc_frag->offset + pad,
1579                                      len, from);
1580         if (copied != len)
1581                 return ERR_PTR(-EFAULT);
1582
1583         /* There's a small window that XDP may be set after the check
1584          * of xdp_prog above, this should be rare and for simplicity
1585          * we do XDP on skb in case the headroom is not enough.
1586          */
1587         if (hdr->gso_type || !xdp_prog) {
1588                 *skb_xdp = 1;
1589                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1590                                        pad);
1591         }
1592
1593         *skb_xdp = 0;
1594
1595         local_bh_disable();
1596         rcu_read_lock();
1597         xdp_prog = rcu_dereference(tun->xdp_prog);
1598         if (xdp_prog) {
1599                 struct xdp_buff xdp;
1600                 u32 act;
1601
1602                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1603                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1604
1605                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1606                 if (act == XDP_REDIRECT || act == XDP_TX) {
1607                         get_page(alloc_frag->page);
1608                         alloc_frag->offset += buflen;
1609                 }
1610                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1611                 if (err < 0) {
1612                         if (act == XDP_REDIRECT || act == XDP_TX)
1613                                 put_page(alloc_frag->page);
1614                         goto out;
1615                 }
1616
1617                 if (err == XDP_REDIRECT)
1618                         xdp_do_flush();
1619                 if (err != XDP_PASS)
1620                         goto out;
1621
1622                 pad = xdp.data - xdp.data_hard_start;
1623                 len = xdp.data_end - xdp.data;
1624         }
1625         rcu_read_unlock();
1626         local_bh_enable();
1627
1628         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1629
1630 out:
1631         rcu_read_unlock();
1632         local_bh_enable();
1633         return NULL;
1634 }
1635
1636 /* Get packet from user space buffer */
1637 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1638                             void *msg_control, struct iov_iter *from,
1639                             int noblock, bool more)
1640 {
1641         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1642         struct sk_buff *skb;
1643         size_t total_len = iov_iter_count(from);
1644         size_t len = total_len, align = tun->align, linear;
1645         struct virtio_net_hdr gso = { 0 };
1646         int good_linear;
1647         int copylen;
1648         bool zerocopy = false;
1649         int err;
1650         u32 rxhash = 0;
1651         int skb_xdp = 1;
1652         bool frags = tun_napi_frags_enabled(tfile);
1653
1654         if (!(tun->flags & IFF_NO_PI)) {
1655                 if (len < sizeof(pi))
1656                         return -EINVAL;
1657                 len -= sizeof(pi);
1658
1659                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1660                         return -EFAULT;
1661         }
1662
1663         if (tun->flags & IFF_VNET_HDR) {
1664                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1665
1666                 if (len < vnet_hdr_sz)
1667                         return -EINVAL;
1668                 len -= vnet_hdr_sz;
1669
1670                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1671                         return -EFAULT;
1672
1673                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1674                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1675                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1676
1677                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1678                         return -EINVAL;
1679                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1680         }
1681
1682         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1683                 align += NET_IP_ALIGN;
1684                 if (unlikely(len < ETH_HLEN ||
1685                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1686                         return -EINVAL;
1687         }
1688
1689         good_linear = SKB_MAX_HEAD(align);
1690
1691         if (msg_control) {
1692                 struct iov_iter i = *from;
1693
1694                 /* There are 256 bytes to be copied in skb, so there is
1695                  * enough room for skb expand head in case it is used.
1696                  * The rest of the buffer is mapped from userspace.
1697                  */
1698                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1699                 if (copylen > good_linear)
1700                         copylen = good_linear;
1701                 linear = copylen;
1702                 iov_iter_advance(&i, copylen);
1703                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1704                         zerocopy = true;
1705         }
1706
1707         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1708                 /* For the packet that is not easy to be processed
1709                  * (e.g gso or jumbo packet), we will do it at after
1710                  * skb was created with generic XDP routine.
1711                  */
1712                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1713                 if (IS_ERR(skb)) {
1714                         atomic_long_inc(&tun->dev->rx_dropped);
1715                         return PTR_ERR(skb);
1716                 }
1717                 if (!skb)
1718                         return total_len;
1719         } else {
1720                 if (!zerocopy) {
1721                         copylen = len;
1722                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1723                                 linear = good_linear;
1724                         else
1725                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1726                 }
1727
1728                 if (frags) {
1729                         mutex_lock(&tfile->napi_mutex);
1730                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1731                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1732                          * If zerocopy is enabled, then this layout will be
1733                          * overwritten by zerocopy_sg_from_iter().
1734                          */
1735                         zerocopy = false;
1736                 } else {
1737                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1738                                             noblock);
1739                 }
1740
1741                 if (IS_ERR(skb)) {
1742                         if (PTR_ERR(skb) != -EAGAIN)
1743                                 atomic_long_inc(&tun->dev->rx_dropped);
1744                         if (frags)
1745                                 mutex_unlock(&tfile->napi_mutex);
1746                         return PTR_ERR(skb);
1747                 }
1748
1749                 if (zerocopy)
1750                         err = zerocopy_sg_from_iter(skb, from);
1751                 else
1752                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1753
1754                 if (err) {
1755                         err = -EFAULT;
1756 drop:
1757                         atomic_long_inc(&tun->dev->rx_dropped);
1758                         kfree_skb(skb);
1759                         if (frags) {
1760                                 tfile->napi.skb = NULL;
1761                                 mutex_unlock(&tfile->napi_mutex);
1762                         }
1763
1764                         return err;
1765                 }
1766         }
1767
1768         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1769                 atomic_long_inc(&tun->rx_frame_errors);
1770                 kfree_skb(skb);
1771                 if (frags) {
1772                         tfile->napi.skb = NULL;
1773                         mutex_unlock(&tfile->napi_mutex);
1774                 }
1775
1776                 return -EINVAL;
1777         }
1778
1779         switch (tun->flags & TUN_TYPE_MASK) {
1780         case IFF_TUN:
1781                 if (tun->flags & IFF_NO_PI) {
1782                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1783
1784                         switch (ip_version) {
1785                         case 4:
1786                                 pi.proto = htons(ETH_P_IP);
1787                                 break;
1788                         case 6:
1789                                 pi.proto = htons(ETH_P_IPV6);
1790                                 break;
1791                         default:
1792                                 atomic_long_inc(&tun->dev->rx_dropped);
1793                                 kfree_skb(skb);
1794                                 return -EINVAL;
1795                         }
1796                 }
1797
1798                 skb_reset_mac_header(skb);
1799                 skb->protocol = pi.proto;
1800                 skb->dev = tun->dev;
1801                 break;
1802         case IFF_TAP:
1803                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1804                         err = -ENOMEM;
1805                         goto drop;
1806                 }
1807                 skb->protocol = eth_type_trans(skb, tun->dev);
1808                 break;
1809         }
1810
1811         /* copy skb_ubuf_info for callback when skb has no error */
1812         if (zerocopy) {
1813                 skb_zcopy_init(skb, msg_control);
1814         } else if (msg_control) {
1815                 struct ubuf_info *uarg = msg_control;
1816                 uarg->callback(NULL, uarg, false);
1817         }
1818
1819         skb_reset_network_header(skb);
1820         skb_probe_transport_header(skb);
1821         skb_record_rx_queue(skb, tfile->queue_index);
1822
1823         if (skb_xdp) {
1824                 struct bpf_prog *xdp_prog;
1825                 int ret;
1826
1827                 local_bh_disable();
1828                 rcu_read_lock();
1829                 xdp_prog = rcu_dereference(tun->xdp_prog);
1830                 if (xdp_prog) {
1831                         ret = do_xdp_generic(xdp_prog, skb);
1832                         if (ret != XDP_PASS) {
1833                                 rcu_read_unlock();
1834                                 local_bh_enable();
1835                                 if (frags) {
1836                                         tfile->napi.skb = NULL;
1837                                         mutex_unlock(&tfile->napi_mutex);
1838                                 }
1839                                 return total_len;
1840                         }
1841                 }
1842                 rcu_read_unlock();
1843                 local_bh_enable();
1844         }
1845
1846         /* Compute the costly rx hash only if needed for flow updates.
1847          * We may get a very small possibility of OOO during switching, not
1848          * worth to optimize.
1849          */
1850         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1851             !tfile->detached)
1852                 rxhash = __skb_get_hash_symmetric(skb);
1853
1854         rcu_read_lock();
1855         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1856                 err = -EIO;
1857                 rcu_read_unlock();
1858                 goto drop;
1859         }
1860
1861         if (frags) {
1862                 u32 headlen;
1863
1864                 /* Exercise flow dissector code path. */
1865                 skb_push(skb, ETH_HLEN);
1866                 headlen = eth_get_headlen(tun->dev, skb->data,
1867                                           skb_headlen(skb));
1868
1869                 if (unlikely(headlen > skb_headlen(skb))) {
1870                         atomic_long_inc(&tun->dev->rx_dropped);
1871                         napi_free_frags(&tfile->napi);
1872                         rcu_read_unlock();
1873                         mutex_unlock(&tfile->napi_mutex);
1874                         WARN_ON(1);
1875                         return -ENOMEM;
1876                 }
1877
1878                 local_bh_disable();
1879                 napi_gro_frags(&tfile->napi);
1880                 local_bh_enable();
1881                 mutex_unlock(&tfile->napi_mutex);
1882         } else if (tfile->napi_enabled) {
1883                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1884                 int queue_len;
1885
1886                 spin_lock_bh(&queue->lock);
1887                 __skb_queue_tail(queue, skb);
1888                 queue_len = skb_queue_len(queue);
1889                 spin_unlock(&queue->lock);
1890
1891                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1892                         napi_schedule(&tfile->napi);
1893
1894                 local_bh_enable();
1895         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1896                 tun_rx_batched(tun, tfile, skb, more);
1897         } else {
1898                 netif_rx_ni(skb);
1899         }
1900         rcu_read_unlock();
1901
1902         preempt_disable();
1903         dev_sw_netstats_rx_add(tun->dev, len);
1904         preempt_enable();
1905
1906         if (rxhash)
1907                 tun_flow_update(tun, rxhash, tfile);
1908
1909         return total_len;
1910 }
1911
1912 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1913 {
1914         struct file *file = iocb->ki_filp;
1915         struct tun_file *tfile = file->private_data;
1916         struct tun_struct *tun = tun_get(tfile);
1917         ssize_t result;
1918         int noblock = 0;
1919
1920         if (!tun)
1921                 return -EBADFD;
1922
1923         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1924                 noblock = 1;
1925
1926         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1927
1928         tun_put(tun);
1929         return result;
1930 }
1931
1932 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1933                                 struct tun_file *tfile,
1934                                 struct xdp_frame *xdp_frame,
1935                                 struct iov_iter *iter)
1936 {
1937         int vnet_hdr_sz = 0;
1938         size_t size = xdp_frame->len;
1939         size_t ret;
1940
1941         if (tun->flags & IFF_VNET_HDR) {
1942                 struct virtio_net_hdr gso = { 0 };
1943
1944                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1945                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1946                         return -EINVAL;
1947                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1948                              sizeof(gso)))
1949                         return -EFAULT;
1950                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1951         }
1952
1953         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1954
1955         preempt_disable();
1956         dev_sw_netstats_tx_add(tun->dev, 1, ret);
1957         preempt_enable();
1958
1959         return ret;
1960 }
1961
1962 /* Put packet to the user space buffer */
1963 static ssize_t tun_put_user(struct tun_struct *tun,
1964                             struct tun_file *tfile,
1965                             struct sk_buff *skb,
1966                             struct iov_iter *iter)
1967 {
1968         struct tun_pi pi = { 0, skb->protocol };
1969         ssize_t total;
1970         int vlan_offset = 0;
1971         int vlan_hlen = 0;
1972         int vnet_hdr_sz = 0;
1973
1974         if (skb_vlan_tag_present(skb))
1975                 vlan_hlen = VLAN_HLEN;
1976
1977         if (tun->flags & IFF_VNET_HDR)
1978                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1979
1980         total = skb->len + vlan_hlen + vnet_hdr_sz;
1981
1982         if (!(tun->flags & IFF_NO_PI)) {
1983                 if (iov_iter_count(iter) < sizeof(pi))
1984                         return -EINVAL;
1985
1986                 total += sizeof(pi);
1987                 if (iov_iter_count(iter) < total) {
1988                         /* Packet will be striped */
1989                         pi.flags |= TUN_PKT_STRIP;
1990                 }
1991
1992                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1993                         return -EFAULT;
1994         }
1995
1996         if (vnet_hdr_sz) {
1997                 struct virtio_net_hdr gso;
1998
1999                 if (iov_iter_count(iter) < vnet_hdr_sz)
2000                         return -EINVAL;
2001
2002                 if (virtio_net_hdr_from_skb(skb, &gso,
2003                                             tun_is_little_endian(tun), true,
2004                                             vlan_hlen)) {
2005                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2006                         pr_err("unexpected GSO type: "
2007                                "0x%x, gso_size %d, hdr_len %d\n",
2008                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2009                                tun16_to_cpu(tun, gso.hdr_len));
2010                         print_hex_dump(KERN_ERR, "tun: ",
2011                                        DUMP_PREFIX_NONE,
2012                                        16, 1, skb->head,
2013                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2014                         WARN_ON_ONCE(1);
2015                         return -EINVAL;
2016                 }
2017
2018                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2019                         return -EFAULT;
2020
2021                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2022         }
2023
2024         if (vlan_hlen) {
2025                 int ret;
2026                 struct veth veth;
2027
2028                 veth.h_vlan_proto = skb->vlan_proto;
2029                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2030
2031                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2032
2033                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2034                 if (ret || !iov_iter_count(iter))
2035                         goto done;
2036
2037                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2038                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2039                         goto done;
2040         }
2041
2042         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2043
2044 done:
2045         /* caller is in process context, */
2046         preempt_disable();
2047         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2048         preempt_enable();
2049
2050         return total;
2051 }
2052
2053 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2054 {
2055         DECLARE_WAITQUEUE(wait, current);
2056         void *ptr = NULL;
2057         int error = 0;
2058
2059         ptr = ptr_ring_consume(&tfile->tx_ring);
2060         if (ptr)
2061                 goto out;
2062         if (noblock) {
2063                 error = -EAGAIN;
2064                 goto out;
2065         }
2066
2067         add_wait_queue(&tfile->socket.wq.wait, &wait);
2068
2069         while (1) {
2070                 set_current_state(TASK_INTERRUPTIBLE);
2071                 ptr = ptr_ring_consume(&tfile->tx_ring);
2072                 if (ptr)
2073                         break;
2074                 if (signal_pending(current)) {
2075                         error = -ERESTARTSYS;
2076                         break;
2077                 }
2078                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2079                         error = -EFAULT;
2080                         break;
2081                 }
2082
2083                 schedule();
2084         }
2085
2086         __set_current_state(TASK_RUNNING);
2087         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2088
2089 out:
2090         *err = error;
2091         return ptr;
2092 }
2093
2094 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2095                            struct iov_iter *to,
2096                            int noblock, void *ptr)
2097 {
2098         ssize_t ret;
2099         int err;
2100
2101         if (!iov_iter_count(to)) {
2102                 tun_ptr_free(ptr);
2103                 return 0;
2104         }
2105
2106         if (!ptr) {
2107                 /* Read frames from ring */
2108                 ptr = tun_ring_recv(tfile, noblock, &err);
2109                 if (!ptr)
2110                         return err;
2111         }
2112
2113         if (tun_is_xdp_frame(ptr)) {
2114                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2115
2116                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2117                 xdp_return_frame(xdpf);
2118         } else {
2119                 struct sk_buff *skb = ptr;
2120
2121                 ret = tun_put_user(tun, tfile, skb, to);
2122                 if (unlikely(ret < 0))
2123                         kfree_skb(skb);
2124                 else
2125                         consume_skb(skb);
2126         }
2127
2128         return ret;
2129 }
2130
2131 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2132 {
2133         struct file *file = iocb->ki_filp;
2134         struct tun_file *tfile = file->private_data;
2135         struct tun_struct *tun = tun_get(tfile);
2136         ssize_t len = iov_iter_count(to), ret;
2137         int noblock = 0;
2138
2139         if (!tun)
2140                 return -EBADFD;
2141
2142         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2143                 noblock = 1;
2144
2145         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2146         ret = min_t(ssize_t, ret, len);
2147         if (ret > 0)
2148                 iocb->ki_pos = ret;
2149         tun_put(tun);
2150         return ret;
2151 }
2152
2153 static void tun_prog_free(struct rcu_head *rcu)
2154 {
2155         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2156
2157         bpf_prog_destroy(prog->prog);
2158         kfree(prog);
2159 }
2160
2161 static int __tun_set_ebpf(struct tun_struct *tun,
2162                           struct tun_prog __rcu **prog_p,
2163                           struct bpf_prog *prog)
2164 {
2165         struct tun_prog *old, *new = NULL;
2166
2167         if (prog) {
2168                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2169                 if (!new)
2170                         return -ENOMEM;
2171                 new->prog = prog;
2172         }
2173
2174         spin_lock_bh(&tun->lock);
2175         old = rcu_dereference_protected(*prog_p,
2176                                         lockdep_is_held(&tun->lock));
2177         rcu_assign_pointer(*prog_p, new);
2178         spin_unlock_bh(&tun->lock);
2179
2180         if (old)
2181                 call_rcu(&old->rcu, tun_prog_free);
2182
2183         return 0;
2184 }
2185
2186 static void tun_free_netdev(struct net_device *dev)
2187 {
2188         struct tun_struct *tun = netdev_priv(dev);
2189
2190         BUG_ON(!(list_empty(&tun->disabled)));
2191
2192         free_percpu(dev->tstats);
2193         /* We clear tstats so that tun_set_iff() can tell if
2194          * tun_free_netdev() has been called from register_netdevice().
2195          */
2196         dev->tstats = NULL;
2197
2198         tun_flow_uninit(tun);
2199         security_tun_dev_free_security(tun->security);
2200         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2201         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2202 }
2203
2204 static void tun_setup(struct net_device *dev)
2205 {
2206         struct tun_struct *tun = netdev_priv(dev);
2207
2208         tun->owner = INVALID_UID;
2209         tun->group = INVALID_GID;
2210         tun_default_link_ksettings(dev, &tun->link_ksettings);
2211
2212         dev->ethtool_ops = &tun_ethtool_ops;
2213         dev->needs_free_netdev = true;
2214         dev->priv_destructor = tun_free_netdev;
2215         /* We prefer our own queue length */
2216         dev->tx_queue_len = TUN_READQ_SIZE;
2217 }
2218
2219 /* Trivial set of netlink ops to allow deleting tun or tap
2220  * device with netlink.
2221  */
2222 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2223                         struct netlink_ext_ack *extack)
2224 {
2225         NL_SET_ERR_MSG(extack,
2226                        "tun/tap creation via rtnetlink is not supported.");
2227         return -EOPNOTSUPP;
2228 }
2229
2230 static size_t tun_get_size(const struct net_device *dev)
2231 {
2232         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2233         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2234
2235         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2236                nla_total_size(sizeof(gid_t)) + /* GROUP */
2237                nla_total_size(sizeof(u8)) + /* TYPE */
2238                nla_total_size(sizeof(u8)) + /* PI */
2239                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2240                nla_total_size(sizeof(u8)) + /* PERSIST */
2241                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2242                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2243                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2244                0;
2245 }
2246
2247 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2248 {
2249         struct tun_struct *tun = netdev_priv(dev);
2250
2251         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2252                 goto nla_put_failure;
2253         if (uid_valid(tun->owner) &&
2254             nla_put_u32(skb, IFLA_TUN_OWNER,
2255                         from_kuid_munged(current_user_ns(), tun->owner)))
2256                 goto nla_put_failure;
2257         if (gid_valid(tun->group) &&
2258             nla_put_u32(skb, IFLA_TUN_GROUP,
2259                         from_kgid_munged(current_user_ns(), tun->group)))
2260                 goto nla_put_failure;
2261         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2262                 goto nla_put_failure;
2263         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2264                 goto nla_put_failure;
2265         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2266                 goto nla_put_failure;
2267         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2268                        !!(tun->flags & IFF_MULTI_QUEUE)))
2269                 goto nla_put_failure;
2270         if (tun->flags & IFF_MULTI_QUEUE) {
2271                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2272                         goto nla_put_failure;
2273                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2274                                 tun->numdisabled))
2275                         goto nla_put_failure;
2276         }
2277
2278         return 0;
2279
2280 nla_put_failure:
2281         return -EMSGSIZE;
2282 }
2283
2284 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2285         .kind           = DRV_NAME,
2286         .priv_size      = sizeof(struct tun_struct),
2287         .setup          = tun_setup,
2288         .validate       = tun_validate,
2289         .get_size       = tun_get_size,
2290         .fill_info      = tun_fill_info,
2291 };
2292
2293 static void tun_sock_write_space(struct sock *sk)
2294 {
2295         struct tun_file *tfile;
2296         wait_queue_head_t *wqueue;
2297
2298         if (!sock_writeable(sk))
2299                 return;
2300
2301         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2302                 return;
2303
2304         wqueue = sk_sleep(sk);
2305         if (wqueue && waitqueue_active(wqueue))
2306                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2307                                                 EPOLLWRNORM | EPOLLWRBAND);
2308
2309         tfile = container_of(sk, struct tun_file, sk);
2310         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2311 }
2312
2313 static void tun_put_page(struct tun_page *tpage)
2314 {
2315         if (tpage->page)
2316                 __page_frag_cache_drain(tpage->page, tpage->count);
2317 }
2318
2319 static int tun_xdp_one(struct tun_struct *tun,
2320                        struct tun_file *tfile,
2321                        struct xdp_buff *xdp, int *flush,
2322                        struct tun_page *tpage)
2323 {
2324         unsigned int datasize = xdp->data_end - xdp->data;
2325         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2326         struct virtio_net_hdr *gso = &hdr->gso;
2327         struct bpf_prog *xdp_prog;
2328         struct sk_buff *skb = NULL;
2329         u32 rxhash = 0, act;
2330         int buflen = hdr->buflen;
2331         int err = 0;
2332         bool skb_xdp = false;
2333         struct page *page;
2334
2335         xdp_prog = rcu_dereference(tun->xdp_prog);
2336         if (xdp_prog) {
2337                 if (gso->gso_type) {
2338                         skb_xdp = true;
2339                         goto build;
2340                 }
2341
2342                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2343                 xdp_set_data_meta_invalid(xdp);
2344
2345                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2346                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2347                 if (err < 0) {
2348                         put_page(virt_to_head_page(xdp->data));
2349                         return err;
2350                 }
2351
2352                 switch (err) {
2353                 case XDP_REDIRECT:
2354                         *flush = true;
2355                         fallthrough;
2356                 case XDP_TX:
2357                         return 0;
2358                 case XDP_PASS:
2359                         break;
2360                 default:
2361                         page = virt_to_head_page(xdp->data);
2362                         if (tpage->page == page) {
2363                                 ++tpage->count;
2364                         } else {
2365                                 tun_put_page(tpage);
2366                                 tpage->page = page;
2367                                 tpage->count = 1;
2368                         }
2369                         return 0;
2370                 }
2371         }
2372
2373 build:
2374         skb = build_skb(xdp->data_hard_start, buflen);
2375         if (!skb) {
2376                 err = -ENOMEM;
2377                 goto out;
2378         }
2379
2380         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2381         skb_put(skb, xdp->data_end - xdp->data);
2382
2383         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2384                 atomic_long_inc(&tun->rx_frame_errors);
2385                 kfree_skb(skb);
2386                 err = -EINVAL;
2387                 goto out;
2388         }
2389
2390         skb->protocol = eth_type_trans(skb, tun->dev);
2391         skb_reset_network_header(skb);
2392         skb_probe_transport_header(skb);
2393         skb_record_rx_queue(skb, tfile->queue_index);
2394
2395         if (skb_xdp) {
2396                 err = do_xdp_generic(xdp_prog, skb);
2397                 if (err != XDP_PASS)
2398                         goto out;
2399         }
2400
2401         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2402             !tfile->detached)
2403                 rxhash = __skb_get_hash_symmetric(skb);
2404
2405         netif_receive_skb(skb);
2406
2407         /* No need to disable preemption here since this function is
2408          * always called with bh disabled
2409          */
2410         dev_sw_netstats_rx_add(tun->dev, datasize);
2411
2412         if (rxhash)
2413                 tun_flow_update(tun, rxhash, tfile);
2414
2415 out:
2416         return err;
2417 }
2418
2419 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2420 {
2421         int ret, i;
2422         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2423         struct tun_struct *tun = tun_get(tfile);
2424         struct tun_msg_ctl *ctl = m->msg_control;
2425         struct xdp_buff *xdp;
2426
2427         if (!tun)
2428                 return -EBADFD;
2429
2430         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2431                 struct tun_page tpage;
2432                 int n = ctl->num;
2433                 int flush = 0;
2434
2435                 memset(&tpage, 0, sizeof(tpage));
2436
2437                 local_bh_disable();
2438                 rcu_read_lock();
2439
2440                 for (i = 0; i < n; i++) {
2441                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2442                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2443                 }
2444
2445                 if (flush)
2446                         xdp_do_flush();
2447
2448                 rcu_read_unlock();
2449                 local_bh_enable();
2450
2451                 tun_put_page(&tpage);
2452
2453                 ret = total_len;
2454                 goto out;
2455         }
2456
2457         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2458                            m->msg_flags & MSG_DONTWAIT,
2459                            m->msg_flags & MSG_MORE);
2460 out:
2461         tun_put(tun);
2462         return ret;
2463 }
2464
2465 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2466                        int flags)
2467 {
2468         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2469         struct tun_struct *tun = tun_get(tfile);
2470         void *ptr = m->msg_control;
2471         int ret;
2472
2473         if (!tun) {
2474                 ret = -EBADFD;
2475                 goto out_free;
2476         }
2477
2478         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2479                 ret = -EINVAL;
2480                 goto out_put_tun;
2481         }
2482         if (flags & MSG_ERRQUEUE) {
2483                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2484                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2485                 goto out;
2486         }
2487         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2488         if (ret > (ssize_t)total_len) {
2489                 m->msg_flags |= MSG_TRUNC;
2490                 ret = flags & MSG_TRUNC ? ret : total_len;
2491         }
2492 out:
2493         tun_put(tun);
2494         return ret;
2495
2496 out_put_tun:
2497         tun_put(tun);
2498 out_free:
2499         tun_ptr_free(ptr);
2500         return ret;
2501 }
2502
2503 static int tun_ptr_peek_len(void *ptr)
2504 {
2505         if (likely(ptr)) {
2506                 if (tun_is_xdp_frame(ptr)) {
2507                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2508
2509                         return xdpf->len;
2510                 }
2511                 return __skb_array_len_with_tag(ptr);
2512         } else {
2513                 return 0;
2514         }
2515 }
2516
2517 static int tun_peek_len(struct socket *sock)
2518 {
2519         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2520         struct tun_struct *tun;
2521         int ret = 0;
2522
2523         tun = tun_get(tfile);
2524         if (!tun)
2525                 return 0;
2526
2527         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2528         tun_put(tun);
2529
2530         return ret;
2531 }
2532
2533 /* Ops structure to mimic raw sockets with tun */
2534 static const struct proto_ops tun_socket_ops = {
2535         .peek_len = tun_peek_len,
2536         .sendmsg = tun_sendmsg,
2537         .recvmsg = tun_recvmsg,
2538 };
2539
2540 static struct proto tun_proto = {
2541         .name           = "tun",
2542         .owner          = THIS_MODULE,
2543         .obj_size       = sizeof(struct tun_file),
2544 };
2545
2546 static int tun_flags(struct tun_struct *tun)
2547 {
2548         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2549 }
2550
2551 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2552                               char *buf)
2553 {
2554         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2555         return sprintf(buf, "0x%x\n", tun_flags(tun));
2556 }
2557
2558 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2559                               char *buf)
2560 {
2561         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2562         return uid_valid(tun->owner)?
2563                 sprintf(buf, "%u\n",
2564                         from_kuid_munged(current_user_ns(), tun->owner)):
2565                 sprintf(buf, "-1\n");
2566 }
2567
2568 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2569                               char *buf)
2570 {
2571         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2572         return gid_valid(tun->group) ?
2573                 sprintf(buf, "%u\n",
2574                         from_kgid_munged(current_user_ns(), tun->group)):
2575                 sprintf(buf, "-1\n");
2576 }
2577
2578 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2579 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2580 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2581
2582 static struct attribute *tun_dev_attrs[] = {
2583         &dev_attr_tun_flags.attr,
2584         &dev_attr_owner.attr,
2585         &dev_attr_group.attr,
2586         NULL
2587 };
2588
2589 static const struct attribute_group tun_attr_group = {
2590         .attrs = tun_dev_attrs
2591 };
2592
2593 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2594 {
2595         struct tun_struct *tun;
2596         struct tun_file *tfile = file->private_data;
2597         struct net_device *dev;
2598         int err;
2599
2600         if (tfile->detached)
2601                 return -EINVAL;
2602
2603         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2604                 if (!capable(CAP_NET_ADMIN))
2605                         return -EPERM;
2606
2607                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2608                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2609                         return -EINVAL;
2610         }
2611
2612         dev = __dev_get_by_name(net, ifr->ifr_name);
2613         if (dev) {
2614                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2615                         return -EBUSY;
2616                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2617                         tun = netdev_priv(dev);
2618                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2619                         tun = netdev_priv(dev);
2620                 else
2621                         return -EINVAL;
2622
2623                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2624                     !!(tun->flags & IFF_MULTI_QUEUE))
2625                         return -EINVAL;
2626
2627                 if (tun_not_capable(tun))
2628                         return -EPERM;
2629                 err = security_tun_dev_open(tun->security);
2630                 if (err < 0)
2631                         return err;
2632
2633                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2634                                  ifr->ifr_flags & IFF_NAPI,
2635                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2636                 if (err < 0)
2637                         return err;
2638
2639                 if (tun->flags & IFF_MULTI_QUEUE &&
2640                     (tun->numqueues + tun->numdisabled > 1)) {
2641                         /* One or more queue has already been attached, no need
2642                          * to initialize the device again.
2643                          */
2644                         netdev_state_change(dev);
2645                         return 0;
2646                 }
2647
2648                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2649                               (ifr->ifr_flags & TUN_FEATURES);
2650
2651                 netdev_state_change(dev);
2652         } else {
2653                 char *name;
2654                 unsigned long flags = 0;
2655                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2656                              MAX_TAP_QUEUES : 1;
2657
2658                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2659                         return -EPERM;
2660                 err = security_tun_dev_create();
2661                 if (err < 0)
2662                         return err;
2663
2664                 /* Set dev type */
2665                 if (ifr->ifr_flags & IFF_TUN) {
2666                         /* TUN device */
2667                         flags |= IFF_TUN;
2668                         name = "tun%d";
2669                 } else if (ifr->ifr_flags & IFF_TAP) {
2670                         /* TAP device */
2671                         flags |= IFF_TAP;
2672                         name = "tap%d";
2673                 } else
2674                         return -EINVAL;
2675
2676                 if (*ifr->ifr_name)
2677                         name = ifr->ifr_name;
2678
2679                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2680                                        NET_NAME_UNKNOWN, tun_setup, queues,
2681                                        queues);
2682
2683                 if (!dev)
2684                         return -ENOMEM;
2685
2686                 dev_net_set(dev, net);
2687                 dev->rtnl_link_ops = &tun_link_ops;
2688                 dev->ifindex = tfile->ifindex;
2689                 dev->sysfs_groups[0] = &tun_attr_group;
2690
2691                 tun = netdev_priv(dev);
2692                 tun->dev = dev;
2693                 tun->flags = flags;
2694                 tun->txflt.count = 0;
2695                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2696
2697                 tun->align = NET_SKB_PAD;
2698                 tun->filter_attached = false;
2699                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2700                 tun->rx_batched = 0;
2701                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2702
2703                 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2704                 if (!dev->tstats) {
2705                         err = -ENOMEM;
2706                         goto err_free_dev;
2707                 }
2708
2709                 spin_lock_init(&tun->lock);
2710
2711                 err = security_tun_dev_alloc_security(&tun->security);
2712                 if (err < 0)
2713                         goto err_free_stat;
2714
2715                 tun_net_init(dev);
2716                 tun_flow_init(tun);
2717
2718                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2719                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2720                                    NETIF_F_HW_VLAN_STAG_TX;
2721                 dev->features = dev->hw_features | NETIF_F_LLTX;
2722                 dev->vlan_features = dev->features &
2723                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2724                                        NETIF_F_HW_VLAN_STAG_TX);
2725
2726                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2727                               (ifr->ifr_flags & TUN_FEATURES);
2728
2729                 INIT_LIST_HEAD(&tun->disabled);
2730                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2731                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2732                 if (err < 0)
2733                         goto err_free_flow;
2734
2735                 err = register_netdevice(tun->dev);
2736                 if (err < 0)
2737                         goto err_detach;
2738                 /* free_netdev() won't check refcnt, to avoid race
2739                  * with dev_put() we need publish tun after registration.
2740                  */
2741                 rcu_assign_pointer(tfile->tun, tun);
2742         }
2743
2744         netif_carrier_on(tun->dev);
2745
2746         /* Make sure persistent devices do not get stuck in
2747          * xoff state.
2748          */
2749         if (netif_running(tun->dev))
2750                 netif_tx_wake_all_queues(tun->dev);
2751
2752         strcpy(ifr->ifr_name, tun->dev->name);
2753         return 0;
2754
2755 err_detach:
2756         tun_detach_all(dev);
2757         /* We are here because register_netdevice() has failed.
2758          * If register_netdevice() already called tun_free_netdev()
2759          * while dealing with the error, dev->stats has been cleared.
2760          */
2761         if (!dev->tstats)
2762                 goto err_free_dev;
2763
2764 err_free_flow:
2765         tun_flow_uninit(tun);
2766         security_tun_dev_free_security(tun->security);
2767 err_free_stat:
2768         free_percpu(dev->tstats);
2769 err_free_dev:
2770         free_netdev(dev);
2771         return err;
2772 }
2773
2774 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2775 {
2776         strcpy(ifr->ifr_name, tun->dev->name);
2777
2778         ifr->ifr_flags = tun_flags(tun);
2779
2780 }
2781
2782 /* This is like a cut-down ethtool ops, except done via tun fd so no
2783  * privs required. */
2784 static int set_offload(struct tun_struct *tun, unsigned long arg)
2785 {
2786         netdev_features_t features = 0;
2787
2788         if (arg & TUN_F_CSUM) {
2789                 features |= NETIF_F_HW_CSUM;
2790                 arg &= ~TUN_F_CSUM;
2791
2792                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2793                         if (arg & TUN_F_TSO_ECN) {
2794                                 features |= NETIF_F_TSO_ECN;
2795                                 arg &= ~TUN_F_TSO_ECN;
2796                         }
2797                         if (arg & TUN_F_TSO4)
2798                                 features |= NETIF_F_TSO;
2799                         if (arg & TUN_F_TSO6)
2800                                 features |= NETIF_F_TSO6;
2801                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2802                 }
2803
2804                 arg &= ~TUN_F_UFO;
2805         }
2806
2807         /* This gives the user a way to test for new features in future by
2808          * trying to set them. */
2809         if (arg)
2810                 return -EINVAL;
2811
2812         tun->set_features = features;
2813         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2814         tun->dev->wanted_features |= features;
2815         netdev_update_features(tun->dev);
2816
2817         return 0;
2818 }
2819
2820 static void tun_detach_filter(struct tun_struct *tun, int n)
2821 {
2822         int i;
2823         struct tun_file *tfile;
2824
2825         for (i = 0; i < n; i++) {
2826                 tfile = rtnl_dereference(tun->tfiles[i]);
2827                 lock_sock(tfile->socket.sk);
2828                 sk_detach_filter(tfile->socket.sk);
2829                 release_sock(tfile->socket.sk);
2830         }
2831
2832         tun->filter_attached = false;
2833 }
2834
2835 static int tun_attach_filter(struct tun_struct *tun)
2836 {
2837         int i, ret = 0;
2838         struct tun_file *tfile;
2839
2840         for (i = 0; i < tun->numqueues; i++) {
2841                 tfile = rtnl_dereference(tun->tfiles[i]);
2842                 lock_sock(tfile->socket.sk);
2843                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2844                 release_sock(tfile->socket.sk);
2845                 if (ret) {
2846                         tun_detach_filter(tun, i);
2847                         return ret;
2848                 }
2849         }
2850
2851         tun->filter_attached = true;
2852         return ret;
2853 }
2854
2855 static void tun_set_sndbuf(struct tun_struct *tun)
2856 {
2857         struct tun_file *tfile;
2858         int i;
2859
2860         for (i = 0; i < tun->numqueues; i++) {
2861                 tfile = rtnl_dereference(tun->tfiles[i]);
2862                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2863         }
2864 }
2865
2866 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2867 {
2868         struct tun_file *tfile = file->private_data;
2869         struct tun_struct *tun;
2870         int ret = 0;
2871
2872         rtnl_lock();
2873
2874         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2875                 tun = tfile->detached;
2876                 if (!tun) {
2877                         ret = -EINVAL;
2878                         goto unlock;
2879                 }
2880                 ret = security_tun_dev_attach_queue(tun->security);
2881                 if (ret < 0)
2882                         goto unlock;
2883                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2884                                  tun->flags & IFF_NAPI_FRAGS, true);
2885         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2886                 tun = rtnl_dereference(tfile->tun);
2887                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2888                         ret = -EINVAL;
2889                 else
2890                         __tun_detach(tfile, false);
2891         } else
2892                 ret = -EINVAL;
2893
2894         if (ret >= 0)
2895                 netdev_state_change(tun->dev);
2896
2897 unlock:
2898         rtnl_unlock();
2899         return ret;
2900 }
2901
2902 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2903                         void __user *data)
2904 {
2905         struct bpf_prog *prog;
2906         int fd;
2907
2908         if (copy_from_user(&fd, data, sizeof(fd)))
2909                 return -EFAULT;
2910
2911         if (fd == -1) {
2912                 prog = NULL;
2913         } else {
2914                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2915                 if (IS_ERR(prog))
2916                         return PTR_ERR(prog);
2917         }
2918
2919         return __tun_set_ebpf(tun, prog_p, prog);
2920 }
2921
2922 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2923                             unsigned long arg, int ifreq_len)
2924 {
2925         struct tun_file *tfile = file->private_data;
2926         struct net *net = sock_net(&tfile->sk);
2927         struct tun_struct *tun;
2928         void __user* argp = (void __user*)arg;
2929         unsigned int ifindex, carrier;
2930         struct ifreq ifr;
2931         kuid_t owner;
2932         kgid_t group;
2933         int sndbuf;
2934         int vnet_hdr_sz;
2935         int le;
2936         int ret;
2937         bool do_notify = false;
2938
2939         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2940             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2941                 if (copy_from_user(&ifr, argp, ifreq_len))
2942                         return -EFAULT;
2943         } else {
2944                 memset(&ifr, 0, sizeof(ifr));
2945         }
2946         if (cmd == TUNGETFEATURES) {
2947                 /* Currently this just means: "what IFF flags are valid?".
2948                  * This is needed because we never checked for invalid flags on
2949                  * TUNSETIFF.
2950                  */
2951                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2952                                 (unsigned int __user*)argp);
2953         } else if (cmd == TUNSETQUEUE) {
2954                 return tun_set_queue(file, &ifr);
2955         } else if (cmd == SIOCGSKNS) {
2956                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2957                         return -EPERM;
2958                 return open_related_ns(&net->ns, get_net_ns);
2959         }
2960
2961         ret = 0;
2962         rtnl_lock();
2963
2964         tun = tun_get(tfile);
2965         if (cmd == TUNSETIFF) {
2966                 ret = -EEXIST;
2967                 if (tun)
2968                         goto unlock;
2969
2970                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2971
2972                 ret = tun_set_iff(net, file, &ifr);
2973
2974                 if (ret)
2975                         goto unlock;
2976
2977                 if (copy_to_user(argp, &ifr, ifreq_len))
2978                         ret = -EFAULT;
2979                 goto unlock;
2980         }
2981         if (cmd == TUNSETIFINDEX) {
2982                 ret = -EPERM;
2983                 if (tun)
2984                         goto unlock;
2985
2986                 ret = -EFAULT;
2987                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2988                         goto unlock;
2989
2990                 ret = 0;
2991                 tfile->ifindex = ifindex;
2992                 goto unlock;
2993         }
2994
2995         ret = -EBADFD;
2996         if (!tun)
2997                 goto unlock;
2998
2999         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3000
3001         net = dev_net(tun->dev);
3002         ret = 0;
3003         switch (cmd) {
3004         case TUNGETIFF:
3005                 tun_get_iff(tun, &ifr);
3006
3007                 if (tfile->detached)
3008                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3009                 if (!tfile->socket.sk->sk_filter)
3010                         ifr.ifr_flags |= IFF_NOFILTER;
3011
3012                 if (copy_to_user(argp, &ifr, ifreq_len))
3013                         ret = -EFAULT;
3014                 break;
3015
3016         case TUNSETNOCSUM:
3017                 /* Disable/Enable checksum */
3018
3019                 /* [unimplemented] */
3020                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3021                            arg ? "disabled" : "enabled");
3022                 break;
3023
3024         case TUNSETPERSIST:
3025                 /* Disable/Enable persist mode. Keep an extra reference to the
3026                  * module to prevent the module being unprobed.
3027                  */
3028                 if (arg && !(tun->flags & IFF_PERSIST)) {
3029                         tun->flags |= IFF_PERSIST;
3030                         __module_get(THIS_MODULE);
3031                         do_notify = true;
3032                 }
3033                 if (!arg && (tun->flags & IFF_PERSIST)) {
3034                         tun->flags &= ~IFF_PERSIST;
3035                         module_put(THIS_MODULE);
3036                         do_notify = true;
3037                 }
3038
3039                 netif_info(tun, drv, tun->dev, "persist %s\n",
3040                            arg ? "enabled" : "disabled");
3041                 break;
3042
3043         case TUNSETOWNER:
3044                 /* Set owner of the device */
3045                 owner = make_kuid(current_user_ns(), arg);
3046                 if (!uid_valid(owner)) {
3047                         ret = -EINVAL;
3048                         break;
3049                 }
3050                 tun->owner = owner;
3051                 do_notify = true;
3052                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3053                            from_kuid(&init_user_ns, tun->owner));
3054                 break;
3055
3056         case TUNSETGROUP:
3057                 /* Set group of the device */
3058                 group = make_kgid(current_user_ns(), arg);
3059                 if (!gid_valid(group)) {
3060                         ret = -EINVAL;
3061                         break;
3062                 }
3063                 tun->group = group;
3064                 do_notify = true;
3065                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3066                            from_kgid(&init_user_ns, tun->group));
3067                 break;
3068
3069         case TUNSETLINK:
3070                 /* Only allow setting the type when the interface is down */
3071                 if (tun->dev->flags & IFF_UP) {
3072                         netif_info(tun, drv, tun->dev,
3073                                    "Linktype set failed because interface is up\n");
3074                         ret = -EBUSY;
3075                 } else {
3076                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3077                                                        tun->dev);
3078                         ret = notifier_to_errno(ret);
3079                         if (ret) {
3080                                 netif_info(tun, drv, tun->dev,
3081                                            "Refused to change device type\n");
3082                                 break;
3083                         }
3084                         tun->dev->type = (int) arg;
3085                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3086                                    tun->dev->type);
3087                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3088                                                  tun->dev);
3089                 }
3090                 break;
3091
3092         case TUNSETDEBUG:
3093                 tun->msg_enable = (u32)arg;
3094                 break;
3095
3096         case TUNSETOFFLOAD:
3097                 ret = set_offload(tun, arg);
3098                 break;
3099
3100         case TUNSETTXFILTER:
3101                 /* Can be set only for TAPs */
3102                 ret = -EINVAL;
3103                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3104                         break;
3105                 ret = update_filter(&tun->txflt, (void __user *)arg);
3106                 break;
3107
3108         case SIOCGIFHWADDR:
3109                 /* Get hw address */
3110                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3111                 if (copy_to_user(argp, &ifr, ifreq_len))
3112                         ret = -EFAULT;
3113                 break;
3114
3115         case SIOCSIFHWADDR:
3116                 /* Set hw address */
3117                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3118                 break;
3119
3120         case TUNGETSNDBUF:
3121                 sndbuf = tfile->socket.sk->sk_sndbuf;
3122                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3123                         ret = -EFAULT;
3124                 break;
3125
3126         case TUNSETSNDBUF:
3127                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3128                         ret = -EFAULT;
3129                         break;
3130                 }
3131                 if (sndbuf <= 0) {
3132                         ret = -EINVAL;
3133                         break;
3134                 }
3135
3136                 tun->sndbuf = sndbuf;
3137                 tun_set_sndbuf(tun);
3138                 break;
3139
3140         case TUNGETVNETHDRSZ:
3141                 vnet_hdr_sz = tun->vnet_hdr_sz;
3142                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3143                         ret = -EFAULT;
3144                 break;
3145
3146         case TUNSETVNETHDRSZ:
3147                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3148                         ret = -EFAULT;
3149                         break;
3150                 }
3151                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3152                         ret = -EINVAL;
3153                         break;
3154                 }
3155
3156                 tun->vnet_hdr_sz = vnet_hdr_sz;
3157                 break;
3158
3159         case TUNGETVNETLE:
3160                 le = !!(tun->flags & TUN_VNET_LE);
3161                 if (put_user(le, (int __user *)argp))
3162                         ret = -EFAULT;
3163                 break;
3164
3165         case TUNSETVNETLE:
3166                 if (get_user(le, (int __user *)argp)) {
3167                         ret = -EFAULT;
3168                         break;
3169                 }
3170                 if (le)
3171                         tun->flags |= TUN_VNET_LE;
3172                 else
3173                         tun->flags &= ~TUN_VNET_LE;
3174                 break;
3175
3176         case TUNGETVNETBE:
3177                 ret = tun_get_vnet_be(tun, argp);
3178                 break;
3179
3180         case TUNSETVNETBE:
3181                 ret = tun_set_vnet_be(tun, argp);
3182                 break;
3183
3184         case TUNATTACHFILTER:
3185                 /* Can be set only for TAPs */
3186                 ret = -EINVAL;
3187                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3188                         break;
3189                 ret = -EFAULT;
3190                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3191                         break;
3192
3193                 ret = tun_attach_filter(tun);
3194                 break;
3195
3196         case TUNDETACHFILTER:
3197                 /* Can be set only for TAPs */
3198                 ret = -EINVAL;
3199                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3200                         break;
3201                 ret = 0;
3202                 tun_detach_filter(tun, tun->numqueues);
3203                 break;
3204
3205         case TUNGETFILTER:
3206                 ret = -EINVAL;
3207                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3208                         break;
3209                 ret = -EFAULT;
3210                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3211                         break;
3212                 ret = 0;
3213                 break;
3214
3215         case TUNSETSTEERINGEBPF:
3216                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3217                 break;
3218
3219         case TUNSETFILTEREBPF:
3220                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3221                 break;
3222
3223         case TUNSETCARRIER:
3224                 ret = -EFAULT;
3225                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3226                         goto unlock;
3227
3228                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3229                 break;
3230
3231         case TUNGETDEVNETNS:
3232                 ret = -EPERM;
3233                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3234                         goto unlock;
3235                 ret = open_related_ns(&net->ns, get_net_ns);
3236                 break;
3237
3238         default:
3239                 ret = -EINVAL;
3240                 break;
3241         }
3242
3243         if (do_notify)
3244                 netdev_state_change(tun->dev);
3245
3246 unlock:
3247         rtnl_unlock();
3248         if (tun)
3249                 tun_put(tun);
3250         return ret;
3251 }
3252
3253 static long tun_chr_ioctl(struct file *file,
3254                           unsigned int cmd, unsigned long arg)
3255 {
3256         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3257 }
3258
3259 #ifdef CONFIG_COMPAT
3260 static long tun_chr_compat_ioctl(struct file *file,
3261                          unsigned int cmd, unsigned long arg)
3262 {
3263         switch (cmd) {
3264         case TUNSETIFF:
3265         case TUNGETIFF:
3266         case TUNSETTXFILTER:
3267         case TUNGETSNDBUF:
3268         case TUNSETSNDBUF:
3269         case SIOCGIFHWADDR:
3270         case SIOCSIFHWADDR:
3271                 arg = (unsigned long)compat_ptr(arg);
3272                 break;
3273         default:
3274                 arg = (compat_ulong_t)arg;
3275                 break;
3276         }
3277
3278         /*
3279          * compat_ifreq is shorter than ifreq, so we must not access beyond
3280          * the end of that structure. All fields that are used in this
3281          * driver are compatible though, we don't need to convert the
3282          * contents.
3283          */
3284         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3285 }
3286 #endif /* CONFIG_COMPAT */
3287
3288 static int tun_chr_fasync(int fd, struct file *file, int on)
3289 {
3290         struct tun_file *tfile = file->private_data;
3291         int ret;
3292
3293         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3294                 goto out;
3295
3296         if (on) {
3297                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3298                 tfile->flags |= TUN_FASYNC;
3299         } else
3300                 tfile->flags &= ~TUN_FASYNC;
3301         ret = 0;
3302 out:
3303         return ret;
3304 }
3305
3306 static int tun_chr_open(struct inode *inode, struct file * file)
3307 {
3308         struct net *net = current->nsproxy->net_ns;
3309         struct tun_file *tfile;
3310
3311         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3312                                             &tun_proto, 0);
3313         if (!tfile)
3314                 return -ENOMEM;
3315         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3316                 sk_free(&tfile->sk);
3317                 return -ENOMEM;
3318         }
3319
3320         mutex_init(&tfile->napi_mutex);
3321         RCU_INIT_POINTER(tfile->tun, NULL);
3322         tfile->flags = 0;
3323         tfile->ifindex = 0;
3324
3325         init_waitqueue_head(&tfile->socket.wq.wait);
3326
3327         tfile->socket.file = file;
3328         tfile->socket.ops = &tun_socket_ops;
3329
3330         sock_init_data(&tfile->socket, &tfile->sk);
3331
3332         tfile->sk.sk_write_space = tun_sock_write_space;
3333         tfile->sk.sk_sndbuf = INT_MAX;
3334
3335         file->private_data = tfile;
3336         INIT_LIST_HEAD(&tfile->next);
3337
3338         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3339
3340         return 0;
3341 }
3342
3343 static int tun_chr_close(struct inode *inode, struct file *file)
3344 {
3345         struct tun_file *tfile = file->private_data;
3346
3347         tun_detach(tfile, true);
3348
3349         return 0;
3350 }
3351
3352 #ifdef CONFIG_PROC_FS
3353 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3354 {
3355         struct tun_file *tfile = file->private_data;
3356         struct tun_struct *tun;
3357         struct ifreq ifr;
3358
3359         memset(&ifr, 0, sizeof(ifr));
3360
3361         rtnl_lock();
3362         tun = tun_get(tfile);
3363         if (tun)
3364                 tun_get_iff(tun, &ifr);
3365         rtnl_unlock();
3366
3367         if (tun)
3368                 tun_put(tun);
3369
3370         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3371 }
3372 #endif
3373
3374 static const struct file_operations tun_fops = {
3375         .owner  = THIS_MODULE,
3376         .llseek = no_llseek,
3377         .read_iter  = tun_chr_read_iter,
3378         .write_iter = tun_chr_write_iter,
3379         .poll   = tun_chr_poll,
3380         .unlocked_ioctl = tun_chr_ioctl,
3381 #ifdef CONFIG_COMPAT
3382         .compat_ioctl = tun_chr_compat_ioctl,
3383 #endif
3384         .open   = tun_chr_open,
3385         .release = tun_chr_close,
3386         .fasync = tun_chr_fasync,
3387 #ifdef CONFIG_PROC_FS
3388         .show_fdinfo = tun_chr_show_fdinfo,
3389 #endif
3390 };
3391
3392 static struct miscdevice tun_miscdev = {
3393         .minor = TUN_MINOR,
3394         .name = "tun",
3395         .nodename = "net/tun",
3396         .fops = &tun_fops,
3397 };
3398
3399 /* ethtool interface */
3400
3401 static void tun_default_link_ksettings(struct net_device *dev,
3402                                        struct ethtool_link_ksettings *cmd)
3403 {
3404         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3405         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3406         cmd->base.speed         = SPEED_10;
3407         cmd->base.duplex        = DUPLEX_FULL;
3408         cmd->base.port          = PORT_TP;
3409         cmd->base.phy_address   = 0;
3410         cmd->base.autoneg       = AUTONEG_DISABLE;
3411 }
3412
3413 static int tun_get_link_ksettings(struct net_device *dev,
3414                                   struct ethtool_link_ksettings *cmd)
3415 {
3416         struct tun_struct *tun = netdev_priv(dev);
3417
3418         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3419         return 0;
3420 }
3421
3422 static int tun_set_link_ksettings(struct net_device *dev,
3423                                   const struct ethtool_link_ksettings *cmd)
3424 {
3425         struct tun_struct *tun = netdev_priv(dev);
3426
3427         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3428         return 0;
3429 }
3430
3431 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3432 {
3433         struct tun_struct *tun = netdev_priv(dev);
3434
3435         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3436         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3437
3438         switch (tun->flags & TUN_TYPE_MASK) {
3439         case IFF_TUN:
3440                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3441                 break;
3442         case IFF_TAP:
3443                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3444                 break;
3445         }
3446 }
3447
3448 static u32 tun_get_msglevel(struct net_device *dev)
3449 {
3450         struct tun_struct *tun = netdev_priv(dev);
3451
3452         return tun->msg_enable;
3453 }
3454
3455 static void tun_set_msglevel(struct net_device *dev, u32 value)
3456 {
3457         struct tun_struct *tun = netdev_priv(dev);
3458
3459         tun->msg_enable = value;
3460 }
3461
3462 static int tun_get_coalesce(struct net_device *dev,
3463                             struct ethtool_coalesce *ec)
3464 {
3465         struct tun_struct *tun = netdev_priv(dev);
3466
3467         ec->rx_max_coalesced_frames = tun->rx_batched;
3468
3469         return 0;
3470 }
3471
3472 static int tun_set_coalesce(struct net_device *dev,
3473                             struct ethtool_coalesce *ec)
3474 {
3475         struct tun_struct *tun = netdev_priv(dev);
3476
3477         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3478                 tun->rx_batched = NAPI_POLL_WEIGHT;
3479         else
3480                 tun->rx_batched = ec->rx_max_coalesced_frames;
3481
3482         return 0;
3483 }
3484
3485 static const struct ethtool_ops tun_ethtool_ops = {
3486         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3487         .get_drvinfo    = tun_get_drvinfo,
3488         .get_msglevel   = tun_get_msglevel,
3489         .set_msglevel   = tun_set_msglevel,
3490         .get_link       = ethtool_op_get_link,
3491         .get_ts_info    = ethtool_op_get_ts_info,
3492         .get_coalesce   = tun_get_coalesce,
3493         .set_coalesce   = tun_set_coalesce,
3494         .get_link_ksettings = tun_get_link_ksettings,
3495         .set_link_ksettings = tun_set_link_ksettings,
3496 };
3497
3498 static int tun_queue_resize(struct tun_struct *tun)
3499 {
3500         struct net_device *dev = tun->dev;
3501         struct tun_file *tfile;
3502         struct ptr_ring **rings;
3503         int n = tun->numqueues + tun->numdisabled;
3504         int ret, i;
3505
3506         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3507         if (!rings)
3508                 return -ENOMEM;
3509
3510         for (i = 0; i < tun->numqueues; i++) {
3511                 tfile = rtnl_dereference(tun->tfiles[i]);
3512                 rings[i] = &tfile->tx_ring;
3513         }
3514         list_for_each_entry(tfile, &tun->disabled, next)
3515                 rings[i++] = &tfile->tx_ring;
3516
3517         ret = ptr_ring_resize_multiple(rings, n,
3518                                        dev->tx_queue_len, GFP_KERNEL,
3519                                        tun_ptr_free);
3520
3521         kfree(rings);
3522         return ret;
3523 }
3524
3525 static int tun_device_event(struct notifier_block *unused,
3526                             unsigned long event, void *ptr)
3527 {
3528         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3529         struct tun_struct *tun = netdev_priv(dev);
3530         int i;
3531
3532         if (dev->rtnl_link_ops != &tun_link_ops)
3533                 return NOTIFY_DONE;
3534
3535         switch (event) {
3536         case NETDEV_CHANGE_TX_QUEUE_LEN:
3537                 if (tun_queue_resize(tun))
3538                         return NOTIFY_BAD;
3539                 break;
3540         case NETDEV_UP:
3541                 for (i = 0; i < tun->numqueues; i++) {
3542                         struct tun_file *tfile;
3543
3544                         tfile = rtnl_dereference(tun->tfiles[i]);
3545                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3546                 }
3547                 break;
3548         default:
3549                 break;
3550         }
3551
3552         return NOTIFY_DONE;
3553 }
3554
3555 static struct notifier_block tun_notifier_block __read_mostly = {
3556         .notifier_call  = tun_device_event,
3557 };
3558
3559 static int __init tun_init(void)
3560 {
3561         int ret = 0;
3562
3563         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3564
3565         ret = rtnl_link_register(&tun_link_ops);
3566         if (ret) {
3567                 pr_err("Can't register link_ops\n");
3568                 goto err_linkops;
3569         }
3570
3571         ret = misc_register(&tun_miscdev);
3572         if (ret) {
3573                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3574                 goto err_misc;
3575         }
3576
3577         ret = register_netdevice_notifier(&tun_notifier_block);
3578         if (ret) {
3579                 pr_err("Can't register netdevice notifier\n");
3580                 goto err_notifier;
3581         }
3582
3583         return  0;
3584
3585 err_notifier:
3586         misc_deregister(&tun_miscdev);
3587 err_misc:
3588         rtnl_link_unregister(&tun_link_ops);
3589 err_linkops:
3590         return ret;
3591 }
3592
3593 static void tun_cleanup(void)
3594 {
3595         misc_deregister(&tun_miscdev);
3596         rtnl_link_unregister(&tun_link_ops);
3597         unregister_netdevice_notifier(&tun_notifier_block);
3598 }
3599
3600 /* Get an underlying socket object from tun file.  Returns error unless file is
3601  * attached to a device.  The returned object works like a packet socket, it
3602  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3603  * holding a reference to the file for as long as the socket is in use. */
3604 struct socket *tun_get_socket(struct file *file)
3605 {
3606         struct tun_file *tfile;
3607         if (file->f_op != &tun_fops)
3608                 return ERR_PTR(-EINVAL);
3609         tfile = file->private_data;
3610         if (!tfile)
3611                 return ERR_PTR(-EBADFD);
3612         return &tfile->socket;
3613 }
3614 EXPORT_SYMBOL_GPL(tun_get_socket);
3615
3616 struct ptr_ring *tun_get_tx_ring(struct file *file)
3617 {
3618         struct tun_file *tfile;
3619
3620         if (file->f_op != &tun_fops)
3621                 return ERR_PTR(-EINVAL);
3622         tfile = file->private_data;
3623         if (!tfile)
3624                 return ERR_PTR(-EBADFD);
3625         return &tfile->tx_ring;
3626 }
3627 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3628
3629 module_init(tun_init);
3630 module_exit(tun_cleanup);
3631 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3632 MODULE_AUTHOR(DRV_COPYRIGHT);
3633 MODULE_LICENSE("GPL");
3634 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3635 MODULE_ALIAS("devname:net/tun");