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 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212 };
 213
 214 struct veth {
 215         __be16 h_vlan_proto;
 216         __be16 h_vlan_TCI;
 217 };
 218
 219 static int tun_napi_receive(struct napi_struct *napi, int budget)
 220 {
 221         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 222         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 223         struct sk_buff_head process_queue;
 224         struct sk_buff *skb;
 225         int received = 0;
 226
 227         __skb_queue_head_init(&process_queue);
 228
 229         spin_lock(&queue->lock);
 230         skb_queue_splice_tail_init(queue, &process_queue);
 231         spin_unlock(&queue->lock);
 232
 233         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 234                 napi_gro_receive(napi, skb);
 235                 ++received;
 236         }
 237
 238         if (!skb_queue_empty(&process_queue)) {
 239                 spin_lock(&queue->lock);
 240                 skb_queue_splice(&process_queue, queue);
 241                 spin_unlock(&queue->lock);
 242         }
 243
 244         return received;
 245 }
 246
 247 static int tun_napi_poll(struct napi_struct *napi, int budget)
 248 {
 249         unsigned int received;
 250
 251         received = tun_napi_receive(napi, budget);
 252
 253         if (received < budget)
 254                 napi_complete_done(napi, received);
 255
 256         return received;
 257 }
 258
 259 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 260                           bool napi_en, bool napi_frags)
 261 {
 262         tfile->napi_enabled = napi_en;
 263         tfile->napi_frags_enabled = napi_en && napi_frags;
 264         if (napi_en) {
 265                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 266                                   NAPI_POLL_WEIGHT);
 267                 napi_enable(&tfile->napi);
 268         }
 269 }
 270
 271 static void tun_napi_disable(struct tun_file *tfile)
 272 {
 273         if (tfile->napi_enabled)
 274                 napi_disable(&tfile->napi);
 275 }
 276
 277 static void tun_napi_del(struct tun_file *tfile)
 278 {
 279         if (tfile->napi_enabled)
 280                 netif_napi_del(&tfile->napi);
 281 }
 282
 283 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 284 {
 285         return tfile->napi_frags_enabled;
 286 }
 287
 288 #ifdef CONFIG_TUN_VNET_CROSS_LE
 289 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 290 {
 291         return tun->flags & TUN_VNET_BE ? false :
 292                 virtio_legacy_is_little_endian();
 293 }
 294
 295 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 296 {
 297         int be = !!(tun->flags & TUN_VNET_BE);
 298
 299         if (put_user(be, argp))
 300                 return -EFAULT;
 301
 302         return 0;
 303 }
 304
 305 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 306 {
 307         int be;
 308
 309         if (get_user(be, argp))
 310                 return -EFAULT;
 311
 312         if (be)
 313                 tun->flags |= TUN_VNET_BE;
 314         else
 315                 tun->flags &= ~TUN_VNET_BE;
 316
 317         return 0;
 318 }
 319 #else
 320 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 321 {
 322         return virtio_legacy_is_little_endian();
 323 }
 324
 325 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 326 {
 327         return -EINVAL;
 328 }
 329
 330 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 331 {
 332         return -EINVAL;
 333 }
 334 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 335
 336 static inline bool tun_is_little_endian(struct tun_struct *tun)
 337 {
 338         return tun->flags & TUN_VNET_LE ||
 339                 tun_legacy_is_little_endian(tun);
 340 }
 341
 342 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 343 {
 344         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 345 }
 346
 347 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 348 {
 349         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 350 }
 351
 352 static inline u32 tun_hashfn(u32 rxhash)
 353 {
 354         return rxhash & TUN_MASK_FLOW_ENTRIES;
 355 }
 356
 357 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 358 {
 359         struct tun_flow_entry *e;
 360
 361         hlist_for_each_entry_rcu(e, head, hash_link) {
 362                 if (e->rxhash == rxhash)
 363                         return e;
 364         }
 365         return NULL;
 366 }
 367
 368 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 369                                               struct hlist_head *head,
 370                                               u32 rxhash, u16 queue_index)
 371 {
 372         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 373
 374         if (e) {
 375                 netif_info(tun, tx_queued, tun->dev,
 376                            "create flow: hash %u index %u\n",
 377                            rxhash, queue_index);
 378                 e->updated = jiffies;
 379                 e->rxhash = rxhash;
 380                 e->rps_rxhash = 0;
 381                 e->queue_index = queue_index;
 382                 e->tun = tun;
 383                 hlist_add_head_rcu(&e->hash_link, head);
 384                 ++tun->flow_count;
 385         }
 386         return e;
 387 }
 388
 389 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 390 {
 391         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 392                    e->rxhash, e->queue_index);
 393         hlist_del_rcu(&e->hash_link);
 394         kfree_rcu(e, rcu);
 395         --tun->flow_count;
 396 }
 397
 398 static void tun_flow_flush(struct tun_struct *tun)
 399 {
 400         int i;
 401
 402         spin_lock_bh(&tun->lock);
 403         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 404                 struct tun_flow_entry *e;
 405                 struct hlist_node *n;
 406
 407                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 408                         tun_flow_delete(tun, e);
 409         }
 410         spin_unlock_bh(&tun->lock);
 411 }
 412
 413 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 414 {
 415         int i;
 416
 417         spin_lock_bh(&tun->lock);
 418         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 419                 struct tun_flow_entry *e;
 420                 struct hlist_node *n;
 421
 422                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 423                         if (e->queue_index == queue_index)
 424                                 tun_flow_delete(tun, e);
 425                 }
 426         }
 427         spin_unlock_bh(&tun->lock);
 428 }
 429
 430 static void tun_flow_cleanup(struct timer_list *t)
 431 {
 432         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 433         unsigned long delay = tun->ageing_time;
 434         unsigned long next_timer = jiffies + delay;
 435         unsigned long count = 0;
 436         int i;
 437
 438         spin_lock(&tun->lock);
 439         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 440                 struct tun_flow_entry *e;
 441                 struct hlist_node *n;
 442
 443                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 444                         unsigned long this_timer;
 445
 446                         this_timer = e->updated + delay;
 447                         if (time_before_eq(this_timer, jiffies)) {
 448                                 tun_flow_delete(tun, e);
 449                                 continue;
 450                         }
 451                         count++;
 452                         if (time_before(this_timer, next_timer))
 453                                 next_timer = this_timer;
 454                 }
 455         }
 456
 457         if (count)
 458                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 459         spin_unlock(&tun->lock);
 460 }
 461
 462 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 463                             struct tun_file *tfile)
 464 {
 465         struct hlist_head *head;
 466         struct tun_flow_entry *e;
 467         unsigned long delay = tun->ageing_time;
 468         u16 queue_index = tfile->queue_index;
 469
 470         head = &tun->flows[tun_hashfn(rxhash)];
 471
 472         rcu_read_lock();
 473
 474         e = tun_flow_find(head, rxhash);
 475         if (likely(e)) {
 476                 /* TODO: keep queueing to old queue until it's empty? */
 477                 if (READ_ONCE(e->queue_index) != queue_index)
 478                         WRITE_ONCE(e->queue_index, queue_index);
 479                 if (e->updated != jiffies)
 480                         e->updated = jiffies;
 481                 sock_rps_record_flow_hash(e->rps_rxhash);
 482         } else {
 483                 spin_lock_bh(&tun->lock);
 484                 if (!tun_flow_find(head, rxhash) &&
 485                     tun->flow_count < MAX_TAP_FLOWS)
 486                         tun_flow_create(tun, head, rxhash, queue_index);
 487
 488                 if (!timer_pending(&tun->flow_gc_timer))
 489                         mod_timer(&tun->flow_gc_timer,
 490                                   round_jiffies_up(jiffies + delay));
 491                 spin_unlock_bh(&tun->lock);
 492         }
 493
 494         rcu_read_unlock();
 495 }
 496
 497 /* Save the hash received in the stack receive path and update the
 498  * flow_hash table accordingly.
 499  */
 500 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 501 {
 502         if (unlikely(e->rps_rxhash != hash))
 503                 e->rps_rxhash = hash;
 504 }
 505
 506 /* We try to identify a flow through its rxhash. The reason that
 507  * we do not check rxq no. is because some cards(e.g 82599), chooses
 508  * the rxq based on the txq where the last packet of the flow comes. As
 509  * the userspace application move between processors, we may get a
 510  * different rxq no. here.
 511  */
 512 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 513 {
 514         struct tun_flow_entry *e;
 515         u32 txq = 0;
 516         u32 numqueues = 0;
 517
 518         numqueues = READ_ONCE(tun->numqueues);
 519
 520         txq = __skb_get_hash_symmetric(skb);
 521         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 522         if (e) {
 523                 tun_flow_save_rps_rxhash(e, txq);
 524                 txq = e->queue_index;
 525         } else {
 526                 /* use multiply and shift instead of expensive divide */
 527                 txq = ((u64)txq * numqueues) >> 32;
 528         }
 529
 530         return txq;
 531 }
 532
 533 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 534 {
 535         struct tun_prog *prog;
 536         u32 numqueues;
 537         u16 ret = 0;
 538
 539         numqueues = READ_ONCE(tun->numqueues);
 540         if (!numqueues)
 541                 return 0;
 542
 543         prog = rcu_dereference(tun->steering_prog);
 544         if (prog)
 545                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 546
 547         return ret % numqueues;
 548 }
 549
 550 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 551                             struct net_device *sb_dev)
 552 {
 553         struct tun_struct *tun = netdev_priv(dev);
 554         u16 ret;
 555
 556         rcu_read_lock();
 557         if (rcu_dereference(tun->steering_prog))
 558                 ret = tun_ebpf_select_queue(tun, skb);
 559         else
 560                 ret = tun_automq_select_queue(tun, skb);
 561         rcu_read_unlock();
 562
 563         return ret;
 564 }
 565
 566 static inline bool tun_not_capable(struct tun_struct *tun)
 567 {
 568         const struct cred *cred = current_cred();
 569         struct net *net = dev_net(tun->dev);
 570
 571         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 572                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 573                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 574 }
 575
 576 static void tun_set_real_num_queues(struct tun_struct *tun)
 577 {
 578         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 579         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 580 }
 581
 582 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 583 {
 584         tfile->detached = tun;
 585         list_add_tail(&tfile->next, &tun->disabled);
 586         ++tun->numdisabled;
 587 }
 588
 589 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 590 {
 591         struct tun_struct *tun = tfile->detached;
 592
 593         tfile->detached = NULL;
 594         list_del_init(&tfile->next);
 595         --tun->numdisabled;
 596         return tun;
 597 }
 598
 599 void tun_ptr_free(void *ptr)
 600 {
 601         if (!ptr)
 602                 return;
 603         if (tun_is_xdp_frame(ptr)) {
 604                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 605
 606                 xdp_return_frame(xdpf);
 607         } else {
 608                 __skb_array_destroy_skb(ptr);
 609         }
 610 }
 611 EXPORT_SYMBOL_GPL(tun_ptr_free);
 612
 613 static void tun_queue_purge(struct tun_file *tfile)
 614 {
 615         void *ptr;
 616
 617         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 618                 tun_ptr_free(ptr);
 619
 620         skb_queue_purge(&tfile->sk.sk_write_queue);
 621         skb_queue_purge(&tfile->sk.sk_error_queue);
 622 }
 623
 624 static void __tun_detach(struct tun_file *tfile, bool clean)
 625 {
 626         struct tun_file *ntfile;
 627         struct tun_struct *tun;
 628
 629         tun = rtnl_dereference(tfile->tun);
 630
 631         if (tun && clean) {
 632                 tun_napi_disable(tfile);
 633                 tun_napi_del(tfile);
 634         }
 635
 636         if (tun && !tfile->detached) {
 637                 u16 index = tfile->queue_index;
 638                 BUG_ON(index >= tun->numqueues);
 639
 640                 rcu_assign_pointer(tun->tfiles[index],
 641                                    tun->tfiles[tun->numqueues - 1]);
 642                 ntfile = rtnl_dereference(tun->tfiles[index]);
 643                 ntfile->queue_index = index;
 644                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 645                                    NULL);
 646
 647                 --tun->numqueues;
 648                 if (clean) {
 649                         RCU_INIT_POINTER(tfile->tun, NULL);
 650                         sock_put(&tfile->sk);
 651                 } else
 652                         tun_disable_queue(tun, tfile);
 653
 654                 synchronize_net();
 655                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 656                 /* Drop read queue */
 657                 tun_queue_purge(tfile);
 658                 tun_set_real_num_queues(tun);
 659         } else if (tfile->detached && clean) {
 660                 tun = tun_enable_queue(tfile);
 661                 sock_put(&tfile->sk);
 662         }
 663
 664         if (clean) {
 665                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 666                         netif_carrier_off(tun->dev);
 667
 668                         if (!(tun->flags & IFF_PERSIST) &&
 669                             tun->dev->reg_state == NETREG_REGISTERED)
 670                                 unregister_netdevice(tun->dev);
 671                 }
 672                 if (tun)
 673                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 674                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 675                 sock_put(&tfile->sk);
 676         }
 677 }
 678
 679 static void tun_detach(struct tun_file *tfile, bool clean)
 680 {
 681         struct tun_struct *tun;
 682         struct net_device *dev;
 683
 684         rtnl_lock();
 685         tun = rtnl_dereference(tfile->tun);
 686         dev = tun ? tun->dev : NULL;
 687         __tun_detach(tfile, clean);
 688         if (dev)
 689                 netdev_state_change(dev);
 690         rtnl_unlock();
 691 }
 692
 693 static void tun_detach_all(struct net_device *dev)
 694 {
 695         struct tun_struct *tun = netdev_priv(dev);
 696         struct tun_file *tfile, *tmp;
 697         int i, n = tun->numqueues;
 698
 699         for (i = 0; i < n; i++) {
 700                 tfile = rtnl_dereference(tun->tfiles[i]);
 701                 BUG_ON(!tfile);
 702                 tun_napi_disable(tfile);
 703                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 704                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 705                 RCU_INIT_POINTER(tfile->tun, NULL);
 706                 --tun->numqueues;
 707         }
 708         list_for_each_entry(tfile, &tun->disabled, next) {
 709                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 710                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 711                 RCU_INIT_POINTER(tfile->tun, NULL);
 712         }
 713         BUG_ON(tun->numqueues != 0);
 714
 715         synchronize_net();
 716         for (i = 0; i < n; i++) {
 717                 tfile = rtnl_dereference(tun->tfiles[i]);
 718                 tun_napi_del(tfile);
 719                 /* Drop read queue */
 720                 tun_queue_purge(tfile);
 721                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 722                 sock_put(&tfile->sk);
 723         }
 724         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 725                 tun_enable_queue(tfile);
 726                 tun_queue_purge(tfile);
 727                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 728                 sock_put(&tfile->sk);
 729         }
 730         BUG_ON(tun->numdisabled != 0);
 731
 732         if (tun->flags & IFF_PERSIST)
 733                 module_put(THIS_MODULE);
 734 }
 735
 736 static int tun_attach(struct tun_struct *tun, struct file *file,
 737                       bool skip_filter, bool napi, bool napi_frags,
 738                       bool publish_tun)
 739 {
 740         struct tun_file *tfile = file->private_data;
 741         struct net_device *dev = tun->dev;
 742         int err;
 743
 744         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 745         if (err < 0)
 746                 goto out;
 747
 748         err = -EINVAL;
 749         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 750                 goto out;
 751
 752         err = -EBUSY;
 753         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 754                 goto out;
 755
 756         err = -E2BIG;
 757         if (!tfile->detached &&
 758             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 759                 goto out;
 760
 761         err = 0;
 762
 763         /* Re-attach the filter to persist device */
 764         if (!skip_filter && (tun->filter_attached == true)) {
 765                 lock_sock(tfile->socket.sk);
 766                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 767                 release_sock(tfile->socket.sk);
 768                 if (!err)
 769                         goto out;
 770         }
 771
 772         if (!tfile->detached &&
 773             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 774                             GFP_KERNEL, tun_ptr_free)) {
 775                 err = -ENOMEM;
 776                 goto out;
 777         }
 778
 779         tfile->queue_index = tun->numqueues;
 780         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 781
 782         if (tfile->detached) {
 783                 /* Re-attach detached tfile, updating XDP queue_index */
 784                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 785
 786                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 787                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 788         } else {
 789                 /* Setup XDP RX-queue info, for new tfile getting attached */
 790                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 791                                        tun->dev, tfile->queue_index, 0);
 792                 if (err < 0)
 793                         goto out;
 794                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 795                                                  MEM_TYPE_PAGE_SHARED, NULL);
 796                 if (err < 0) {
 797                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 798                         goto out;
 799                 }
 800                 err = 0;
 801         }
 802
 803         if (tfile->detached) {
 804                 tun_enable_queue(tfile);
 805         } else {
 806                 sock_hold(&tfile->sk);
 807                 tun_napi_init(tun, tfile, napi, napi_frags);
 808         }
 809
 810         if (rtnl_dereference(tun->xdp_prog))
 811                 sock_set_flag(&tfile->sk, SOCK_XDP);
 812
 813         /* device is allowed to go away first, so no need to hold extra
 814          * refcnt.
 815          */
 816
 817         /* Publish tfile->tun and tun->tfiles only after we've fully
 818          * initialized tfile; otherwise we risk using half-initialized
 819          * object.
 820          */
 821         if (publish_tun)
 822                 rcu_assign_pointer(tfile->tun, tun);
 823         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 824         tun->numqueues++;
 825         tun_set_real_num_queues(tun);
 826 out:
 827         return err;
 828 }
 829
 830 static struct tun_struct *tun_get(struct tun_file *tfile)
 831 {
 832         struct tun_struct *tun;
 833
 834         rcu_read_lock();
 835         tun = rcu_dereference(tfile->tun);
 836         if (tun)
 837                 dev_hold(tun->dev);
 838         rcu_read_unlock();
 839
 840         return tun;
 841 }
 842
 843 static void tun_put(struct tun_struct *tun)
 844 {
 845         dev_put(tun->dev);
 846 }
 847
 848 /* TAP filtering */
 849 static void addr_hash_set(u32 *mask, const u8 *addr)
 850 {
 851         int n = ether_crc(ETH_ALEN, addr) >> 26;
 852         mask[n >> 5] |= (1 << (n & 31));
 853 }
 854
 855 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 856 {
 857         int n = ether_crc(ETH_ALEN, addr) >> 26;
 858         return mask[n >> 5] & (1 << (n & 31));
 859 }
 860
 861 static int update_filter(struct tap_filter *filter, void __user *arg)
 862 {
 863         struct { u8 u[ETH_ALEN]; } *addr;
 864         struct tun_filter uf;
 865         int err, alen, n, nexact;
 866
 867         if (copy_from_user(&uf, arg, sizeof(uf)))
 868                 return -EFAULT;
 869
 870         if (!uf.count) {
 871                 /* Disabled */
 872                 filter->count = 0;
 873                 return 0;
 874         }
 875
 876         alen = ETH_ALEN * uf.count;
 877         addr = memdup_user(arg + sizeof(uf), alen);
 878         if (IS_ERR(addr))
 879                 return PTR_ERR(addr);
 880
 881         /* The filter is updated without holding any locks. Which is
 882          * perfectly safe. We disable it first and in the worst
 883          * case we'll accept a few undesired packets. */
 884         filter->count = 0;
 885         wmb();
 886
 887         /* Use first set of addresses as an exact filter */
 888         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 889                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 890
 891         nexact = n;
 892
 893         /* Remaining multicast addresses are hashed,
 894          * unicast will leave the filter disabled. */
 895         memset(filter->mask, 0, sizeof(filter->mask));
 896         for (; n < uf.count; n++) {
 897                 if (!is_multicast_ether_addr(addr[n].u)) {
 898                         err = 0; /* no filter */
 899                         goto free_addr;
 900                 }
 901                 addr_hash_set(filter->mask, addr[n].u);
 902         }
 903
 904         /* For ALLMULTI just set the mask to all ones.
 905          * This overrides the mask populated above. */
 906         if ((uf.flags & TUN_FLT_ALLMULTI))
 907                 memset(filter->mask, ~0, sizeof(filter->mask));
 908
 909         /* Now enable the filter */
 910         wmb();
 911         filter->count = nexact;
 912
 913         /* Return the number of exact filters */
 914         err = nexact;
 915 free_addr:
 916         kfree(addr);
 917         return err;
 918 }
 919
 920 /* Returns: 0 - drop, !=0 - accept */
 921 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 922 {
 923         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 924          * at this point. */
 925         struct ethhdr *eh = (struct ethhdr *) skb->data;
 926         int i;
 927
 928         /* Exact match */
 929         for (i = 0; i < filter->count; i++)
 930                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 931                         return 1;
 932
 933         /* Inexact match (multicast only) */
 934         if (is_multicast_ether_addr(eh->h_dest))
 935                 return addr_hash_test(filter->mask, eh->h_dest);
 936
 937         return 0;
 938 }
 939
 940 /*
 941  * Checks whether the packet is accepted or not.
 942  * Returns: 0 - drop, !=0 - accept
 943  */
 944 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 945 {
 946         if (!filter->count)
 947                 return 1;
 948
 949         return run_filter(filter, skb);
 950 }
 951
 952 /* Network device part of the driver */
 953
 954 static const struct ethtool_ops tun_ethtool_ops;
 955
 956 /* Net device detach from fd. */
 957 static void tun_net_uninit(struct net_device *dev)
 958 {
 959         tun_detach_all(dev);
 960 }
 961
 962 /* Net device open. */
 963 static int tun_net_open(struct net_device *dev)
 964 {
 965         netif_tx_start_all_queues(dev);
 966
 967         return 0;
 968 }
 969
 970 /* Net device close. */
 971 static int tun_net_close(struct net_device *dev)
 972 {
 973         netif_tx_stop_all_queues(dev);
 974         return 0;
 975 }
 976
 977 /* Net device start xmit */
 978 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
 979 {
 980 #ifdef CONFIG_RPS
 981         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
 982                 /* Select queue was not called for the skbuff, so we extract the
 983                  * RPS hash and save it into the flow_table here.
 984                  */
 985                 struct tun_flow_entry *e;
 986                 __u32 rxhash;
 987
 988                 rxhash = __skb_get_hash_symmetric(skb);
 989                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
 990                 if (e)
 991                         tun_flow_save_rps_rxhash(e, rxhash);
 992         }
 993 #endif
 994 }
 995
 996 static unsigned int run_ebpf_filter(struct tun_struct *tun,
 997                                     struct sk_buff *skb,
 998                                     int len)
 999 {
1000         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1001
1002         if (prog)
1003                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1004
1005         return len;
1006 }
1007
1008 /* Net device start xmit */
1009 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1010 {
1011         struct tun_struct *tun = netdev_priv(dev);
1012         int txq = skb->queue_mapping;
1013         struct tun_file *tfile;
1014         int len = skb->len;
1015
1016         rcu_read_lock();
1017         tfile = rcu_dereference(tun->tfiles[txq]);
1018
1019         /* Drop packet if interface is not attached */
1020         if (!tfile)
1021                 goto drop;
1022
1023         if (!rcu_dereference(tun->steering_prog))
1024                 tun_automq_xmit(tun, skb);
1025
1026         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1027
1028         /* Drop if the filter does not like it.
1029          * This is a noop if the filter is disabled.
1030          * Filter can be enabled only for the TAP devices. */
1031         if (!check_filter(&tun->txflt, skb))
1032                 goto drop;
1033
1034         if (tfile->socket.sk->sk_filter &&
1035             sk_filter(tfile->socket.sk, skb))
1036                 goto drop;
1037
1038         len = run_ebpf_filter(tun, skb, len);
1039         if (len == 0 || pskb_trim(skb, len))
1040                 goto drop;
1041
1042         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1043                 goto drop;
1044
1045         skb_tx_timestamp(skb);
1046
1047         /* Orphan the skb - required as we might hang on to it
1048          * for indefinite time.
1049          */
1050         skb_orphan(skb);
1051
1052         nf_reset_ct(skb);
1053
1054         if (ptr_ring_produce(&tfile->tx_ring, skb))
1055                 goto drop;
1056
1057         /* Notify and wake up reader process */
1058         if (tfile->flags & TUN_FASYNC)
1059                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1060         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1061
1062         rcu_read_unlock();
1063         return NETDEV_TX_OK;
1064
1065 drop:
1066         atomic_long_inc(&dev->tx_dropped);
1067         skb_tx_error(skb);
1068         kfree_skb(skb);
1069         rcu_read_unlock();
1070         return NET_XMIT_DROP;
1071 }
1072
1073 static void tun_net_mclist(struct net_device *dev)
1074 {
1075         /*
1076          * This callback is supposed to deal with mc filter in
1077          * _rx_ path and has nothing to do with the _tx_ path.
1078          * In rx path we always accept everything userspace gives us.
1079          */
1080 }
1081
1082 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1083         netdev_features_t features)
1084 {
1085         struct tun_struct *tun = netdev_priv(dev);
1086
1087         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1088 }
1089
1090 static void tun_set_headroom(struct net_device *dev, int new_hr)
1091 {
1092         struct tun_struct *tun = netdev_priv(dev);
1093
1094         if (new_hr < NET_SKB_PAD)
1095                 new_hr = NET_SKB_PAD;
1096
1097         tun->align = new_hr;
1098 }
1099
1100 static void
1101 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1102 {
1103         struct tun_struct *tun = netdev_priv(dev);
1104
1105         dev_get_tstats64(dev, stats);
1106
1107         stats->rx_frame_errors +=
1108                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1109 }
1110
1111 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1112                        struct netlink_ext_ack *extack)
1113 {
1114         struct tun_struct *tun = netdev_priv(dev);
1115         struct tun_file *tfile;
1116         struct bpf_prog *old_prog;
1117         int i;
1118
1119         old_prog = rtnl_dereference(tun->xdp_prog);
1120         rcu_assign_pointer(tun->xdp_prog, prog);
1121         if (old_prog)
1122                 bpf_prog_put(old_prog);
1123
1124         for (i = 0; i < tun->numqueues; i++) {
1125                 tfile = rtnl_dereference(tun->tfiles[i]);
1126                 if (prog)
1127                         sock_set_flag(&tfile->sk, SOCK_XDP);
1128                 else
1129                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1130         }
1131         list_for_each_entry(tfile, &tun->disabled, next) {
1132                 if (prog)
1133                         sock_set_flag(&tfile->sk, SOCK_XDP);
1134                 else
1135                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1136         }
1137
1138         return 0;
1139 }
1140
1141 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1142 {
1143         switch (xdp->command) {
1144         case XDP_SETUP_PROG:
1145                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1146         default:
1147                 return -EINVAL;
1148         }
1149 }
1150
1151 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1152 {
1153         if (new_carrier) {
1154                 struct tun_struct *tun = netdev_priv(dev);
1155
1156                 if (!tun->numqueues)
1157                         return -EPERM;
1158
1159                 netif_carrier_on(dev);
1160         } else {
1161                 netif_carrier_off(dev);
1162         }
1163         return 0;
1164 }
1165
1166 static const struct net_device_ops tun_netdev_ops = {
1167         .ndo_uninit             = tun_net_uninit,
1168         .ndo_open               = tun_net_open,
1169         .ndo_stop               = tun_net_close,
1170         .ndo_start_xmit         = tun_net_xmit,
1171         .ndo_fix_features       = tun_net_fix_features,
1172         .ndo_select_queue       = tun_select_queue,
1173         .ndo_set_rx_headroom    = tun_set_headroom,
1174         .ndo_get_stats64        = tun_net_get_stats64,
1175         .ndo_change_carrier     = tun_net_change_carrier,
1176 };
1177
1178 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1179 {
1180         /* Notify and wake up reader process */
1181         if (tfile->flags & TUN_FASYNC)
1182                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1183         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1184 }
1185
1186 static int tun_xdp_xmit(struct net_device *dev, int n,
1187                         struct xdp_frame **frames, u32 flags)
1188 {
1189         struct tun_struct *tun = netdev_priv(dev);
1190         struct tun_file *tfile;
1191         u32 numqueues;
1192         int nxmit = 0;
1193         int i;
1194
1195         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1196                 return -EINVAL;
1197
1198         rcu_read_lock();
1199
1200 resample:
1201         numqueues = READ_ONCE(tun->numqueues);
1202         if (!numqueues) {
1203                 rcu_read_unlock();
1204                 return -ENXIO; /* Caller will free/return all frames */
1205         }
1206
1207         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1208                                             numqueues]);
1209         if (unlikely(!tfile))
1210                 goto resample;
1211
1212         spin_lock(&tfile->tx_ring.producer_lock);
1213         for (i = 0; i < n; i++) {
1214                 struct xdp_frame *xdp = frames[i];
1215                 /* Encode the XDP flag into lowest bit for consumer to differ
1216                  * XDP buffer from sk_buff.
1217                  */
1218                 void *frame = tun_xdp_to_ptr(xdp);
1219
1220                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1221                         atomic_long_inc(&dev->tx_dropped);
1222                         break;
1223                 }
1224                 nxmit++;
1225         }
1226         spin_unlock(&tfile->tx_ring.producer_lock);
1227
1228         if (flags & XDP_XMIT_FLUSH)
1229                 __tun_xdp_flush_tfile(tfile);
1230
1231         rcu_read_unlock();
1232         return nxmit;
1233 }
1234
1235 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1236 {
1237         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1238         int nxmit;
1239
1240         if (unlikely(!frame))
1241                 return -EOVERFLOW;
1242
1243         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1244         if (!nxmit)
1245                 xdp_return_frame_rx_napi(frame);
1246         return nxmit;
1247 }
1248
1249 static const struct net_device_ops tap_netdev_ops = {
1250         .ndo_uninit             = tun_net_uninit,
1251         .ndo_open               = tun_net_open,
1252         .ndo_stop               = tun_net_close,
1253         .ndo_start_xmit         = tun_net_xmit,
1254         .ndo_fix_features       = tun_net_fix_features,
1255         .ndo_set_rx_mode        = tun_net_mclist,
1256         .ndo_set_mac_address    = eth_mac_addr,
1257         .ndo_validate_addr      = eth_validate_addr,
1258         .ndo_select_queue       = tun_select_queue,
1259         .ndo_features_check     = passthru_features_check,
1260         .ndo_set_rx_headroom    = tun_set_headroom,
1261         .ndo_get_stats64        = dev_get_tstats64,
1262         .ndo_bpf                = tun_xdp,
1263         .ndo_xdp_xmit           = tun_xdp_xmit,
1264         .ndo_change_carrier     = tun_net_change_carrier,
1265 };
1266
1267 static void tun_flow_init(struct tun_struct *tun)
1268 {
1269         int i;
1270
1271         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1272                 INIT_HLIST_HEAD(&tun->flows[i]);
1273
1274         tun->ageing_time = TUN_FLOW_EXPIRE;
1275         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1276         mod_timer(&tun->flow_gc_timer,
1277                   round_jiffies_up(jiffies + tun->ageing_time));
1278 }
1279
1280 static void tun_flow_uninit(struct tun_struct *tun)
1281 {
1282         del_timer_sync(&tun->flow_gc_timer);
1283         tun_flow_flush(tun);
1284 }
1285
1286 #define MIN_MTU 68
1287 #define MAX_MTU 65535
1288
1289 /* Initialize net device. */
1290 static void tun_net_init(struct net_device *dev)
1291 {
1292         struct tun_struct *tun = netdev_priv(dev);
1293
1294         switch (tun->flags & TUN_TYPE_MASK) {
1295         case IFF_TUN:
1296                 dev->netdev_ops = &tun_netdev_ops;
1297                 dev->header_ops = &ip_tunnel_header_ops;
1298
1299                 /* Point-to-Point TUN Device */
1300                 dev->hard_header_len = 0;
1301                 dev->addr_len = 0;
1302                 dev->mtu = 1500;
1303
1304                 /* Zero header length */
1305                 dev->type = ARPHRD_NONE;
1306                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1307                 break;
1308
1309         case IFF_TAP:
1310                 dev->netdev_ops = &tap_netdev_ops;
1311                 /* Ethernet TAP Device */
1312                 ether_setup(dev);
1313                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1314                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1315
1316                 eth_hw_addr_random(dev);
1317
1318                 break;
1319         }
1320
1321         dev->min_mtu = MIN_MTU;
1322         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1323 }
1324
1325 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1326 {
1327         struct sock *sk = tfile->socket.sk;
1328
1329         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1330 }
1331
1332 /* Character device part */
1333
1334 /* Poll */
1335 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1336 {
1337         struct tun_file *tfile = file->private_data;
1338         struct tun_struct *tun = tun_get(tfile);
1339         struct sock *sk;
1340         __poll_t mask = 0;
1341
1342         if (!tun)
1343                 return EPOLLERR;
1344
1345         sk = tfile->socket.sk;
1346
1347         poll_wait(file, sk_sleep(sk), wait);
1348
1349         if (!ptr_ring_empty(&tfile->tx_ring))
1350                 mask |= EPOLLIN | EPOLLRDNORM;
1351
1352         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1353          * guarantee EPOLLOUT to be raised by either here or
1354          * tun_sock_write_space(). Then process could get notification
1355          * after it writes to a down device and meets -EIO.
1356          */
1357         if (tun_sock_writeable(tun, tfile) ||
1358             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1359              tun_sock_writeable(tun, tfile)))
1360                 mask |= EPOLLOUT | EPOLLWRNORM;
1361
1362         if (tun->dev->reg_state != NETREG_REGISTERED)
1363                 mask = EPOLLERR;
1364
1365         tun_put(tun);
1366         return mask;
1367 }
1368
1369 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1370                                             size_t len,
1371                                             const struct iov_iter *it)
1372 {
1373         struct sk_buff *skb;
1374         size_t linear;
1375         int err;
1376         int i;
1377
1378         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1379                 return ERR_PTR(-EMSGSIZE);
1380
1381         local_bh_disable();
1382         skb = napi_get_frags(&tfile->napi);
1383         local_bh_enable();
1384         if (!skb)
1385                 return ERR_PTR(-ENOMEM);
1386
1387         linear = iov_iter_single_seg_count(it);
1388         err = __skb_grow(skb, linear);
1389         if (err)
1390                 goto free;
1391
1392         skb->len = len;
1393         skb->data_len = len - linear;
1394         skb->truesize += skb->data_len;
1395
1396         for (i = 1; i < it->nr_segs; i++) {
1397                 size_t fragsz = it->iov[i].iov_len;
1398                 struct page *page;
1399                 void *frag;
1400
1401                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1402                         err = -EINVAL;
1403                         goto free;
1404                 }
1405                 frag = netdev_alloc_frag(fragsz);
1406                 if (!frag) {
1407                         err = -ENOMEM;
1408                         goto free;
1409                 }
1410                 page = virt_to_head_page(frag);
1411                 skb_fill_page_desc(skb, i - 1, page,
1412                                    frag - page_address(page), fragsz);
1413         }
1414
1415         return skb;
1416 free:
1417         /* frees skb and all frags allocated with napi_alloc_frag() */
1418         napi_free_frags(&tfile->napi);
1419         return ERR_PTR(err);
1420 }
1421
1422 /* prepad is the amount to reserve at front.  len is length after that.
1423  * linear is a hint as to how much to copy (usually headers). */
1424 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1425                                      size_t prepad, size_t len,
1426                                      size_t linear, int noblock)
1427 {
1428         struct sock *sk = tfile->socket.sk;
1429         struct sk_buff *skb;
1430         int err;
1431
1432         /* Under a page?  Don't bother with paged skb. */
1433         if (prepad + len < PAGE_SIZE || !linear)
1434                 linear = len;
1435
1436         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1437                                    &err, 0);
1438         if (!skb)
1439                 return ERR_PTR(err);
1440
1441         skb_reserve(skb, prepad);
1442         skb_put(skb, linear);
1443         skb->data_len = len - linear;
1444         skb->len += len - linear;
1445
1446         return skb;
1447 }
1448
1449 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1450                            struct sk_buff *skb, int more)
1451 {
1452         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1453         struct sk_buff_head process_queue;
1454         u32 rx_batched = tun->rx_batched;
1455         bool rcv = false;
1456
1457         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1458                 local_bh_disable();
1459                 skb_record_rx_queue(skb, tfile->queue_index);
1460                 netif_receive_skb(skb);
1461                 local_bh_enable();
1462                 return;
1463         }
1464
1465         spin_lock(&queue->lock);
1466         if (!more || skb_queue_len(queue) == rx_batched) {
1467                 __skb_queue_head_init(&process_queue);
1468                 skb_queue_splice_tail_init(queue, &process_queue);
1469                 rcv = true;
1470         } else {
1471                 __skb_queue_tail(queue, skb);
1472         }
1473         spin_unlock(&queue->lock);
1474
1475         if (rcv) {
1476                 struct sk_buff *nskb;
1477
1478                 local_bh_disable();
1479                 while ((nskb = __skb_dequeue(&process_queue))) {
1480                         skb_record_rx_queue(nskb, tfile->queue_index);
1481                         netif_receive_skb(nskb);
1482                 }
1483                 skb_record_rx_queue(skb, tfile->queue_index);
1484                 netif_receive_skb(skb);
1485                 local_bh_enable();
1486         }
1487 }
1488
1489 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1490                               int len, int noblock, bool zerocopy)
1491 {
1492         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1493                 return false;
1494
1495         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1496                 return false;
1497
1498         if (!noblock)
1499                 return false;
1500
1501         if (zerocopy)
1502                 return false;
1503
1504         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1505             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1506                 return false;
1507
1508         return true;
1509 }
1510
1511 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1512                                        struct page_frag *alloc_frag, char *buf,
1513                                        int buflen, int len, int pad)
1514 {
1515         struct sk_buff *skb = build_skb(buf, buflen);
1516
1517         if (!skb)
1518                 return ERR_PTR(-ENOMEM);
1519
1520         skb_reserve(skb, pad);
1521         skb_put(skb, len);
1522         skb_set_owner_w(skb, tfile->socket.sk);
1523
1524         get_page(alloc_frag->page);
1525         alloc_frag->offset += buflen;
1526
1527         return skb;
1528 }
1529
1530 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1531                        struct xdp_buff *xdp, u32 act)
1532 {
1533         int err;
1534
1535         switch (act) {
1536         case XDP_REDIRECT:
1537                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1538                 if (err)
1539                         return err;
1540                 break;
1541         case XDP_TX:
1542                 err = tun_xdp_tx(tun->dev, xdp);
1543                 if (err < 0)
1544                         return err;
1545                 break;
1546         case XDP_PASS:
1547                 break;
1548         default:
1549                 bpf_warn_invalid_xdp_action(act);
1550                 fallthrough;
1551         case XDP_ABORTED:
1552                 trace_xdp_exception(tun->dev, xdp_prog, act);
1553                 fallthrough;
1554         case XDP_DROP:
1555                 atomic_long_inc(&tun->dev->rx_dropped);
1556                 break;
1557         }
1558
1559         return act;
1560 }
1561
1562 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1563                                      struct tun_file *tfile,
1564                                      struct iov_iter *from,
1565                                      struct virtio_net_hdr *hdr,
1566                                      int len, int *skb_xdp)
1567 {
1568         struct page_frag *alloc_frag = &current->task_frag;
1569         struct bpf_prog *xdp_prog;
1570         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1571         char *buf;
1572         size_t copied;
1573         int pad = TUN_RX_PAD;
1574         int err = 0;
1575
1576         rcu_read_lock();
1577         xdp_prog = rcu_dereference(tun->xdp_prog);
1578         if (xdp_prog)
1579                 pad += XDP_PACKET_HEADROOM;
1580         buflen += SKB_DATA_ALIGN(len + pad);
1581         rcu_read_unlock();
1582
1583         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1584         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1585                 return ERR_PTR(-ENOMEM);
1586
1587         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1588         copied = copy_page_from_iter(alloc_frag->page,
1589                                      alloc_frag->offset + pad,
1590                                      len, from);
1591         if (copied != len)
1592                 return ERR_PTR(-EFAULT);
1593
1594         /* There's a small window that XDP may be set after the check
1595          * of xdp_prog above, this should be rare and for simplicity
1596          * we do XDP on skb in case the headroom is not enough.
1597          */
1598         if (hdr->gso_type || !xdp_prog) {
1599                 *skb_xdp = 1;
1600                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1601                                        pad);
1602         }
1603
1604         *skb_xdp = 0;
1605
1606         local_bh_disable();
1607         rcu_read_lock();
1608         xdp_prog = rcu_dereference(tun->xdp_prog);
1609         if (xdp_prog) {
1610                 struct xdp_buff xdp;
1611                 u32 act;
1612
1613                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1614                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1615
1616                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1617                 if (act == XDP_REDIRECT || act == XDP_TX) {
1618                         get_page(alloc_frag->page);
1619                         alloc_frag->offset += buflen;
1620                 }
1621                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1622                 if (err < 0) {
1623                         if (act == XDP_REDIRECT || act == XDP_TX)
1624                                 put_page(alloc_frag->page);
1625                         goto out;
1626                 }
1627
1628                 if (err == XDP_REDIRECT)
1629                         xdp_do_flush();
1630                 if (err != XDP_PASS)
1631                         goto out;
1632
1633                 pad = xdp.data - xdp.data_hard_start;
1634                 len = xdp.data_end - xdp.data;
1635         }
1636         rcu_read_unlock();
1637         local_bh_enable();
1638
1639         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1640
1641 out:
1642         rcu_read_unlock();
1643         local_bh_enable();
1644         return NULL;
1645 }
1646
1647 /* Get packet from user space buffer */
1648 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1649                             void *msg_control, struct iov_iter *from,
1650                             int noblock, bool more)
1651 {
1652         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1653         struct sk_buff *skb;
1654         size_t total_len = iov_iter_count(from);
1655         size_t len = total_len, align = tun->align, linear;
1656         struct virtio_net_hdr gso = { 0 };
1657         int good_linear;
1658         int copylen;
1659         bool zerocopy = false;
1660         int err;
1661         u32 rxhash = 0;
1662         int skb_xdp = 1;
1663         bool frags = tun_napi_frags_enabled(tfile);
1664
1665         if (!(tun->flags & IFF_NO_PI)) {
1666                 if (len < sizeof(pi))
1667                         return -EINVAL;
1668                 len -= sizeof(pi);
1669
1670                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1671                         return -EFAULT;
1672         }
1673
1674         if (tun->flags & IFF_VNET_HDR) {
1675                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1676
1677                 if (len < vnet_hdr_sz)
1678                         return -EINVAL;
1679                 len -= vnet_hdr_sz;
1680
1681                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1682                         return -EFAULT;
1683
1684                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1685                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1686                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1687
1688                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1689                         return -EINVAL;
1690                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1691         }
1692
1693         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1694                 align += NET_IP_ALIGN;
1695                 if (unlikely(len < ETH_HLEN ||
1696                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1697                         return -EINVAL;
1698         }
1699
1700         good_linear = SKB_MAX_HEAD(align);
1701
1702         if (msg_control) {
1703                 struct iov_iter i = *from;
1704
1705                 /* There are 256 bytes to be copied in skb, so there is
1706                  * enough room for skb expand head in case it is used.
1707                  * The rest of the buffer is mapped from userspace.
1708                  */
1709                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1710                 if (copylen > good_linear)
1711                         copylen = good_linear;
1712                 linear = copylen;
1713                 iov_iter_advance(&i, copylen);
1714                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1715                         zerocopy = true;
1716         }
1717
1718         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1719                 /* For the packet that is not easy to be processed
1720                  * (e.g gso or jumbo packet), we will do it at after
1721                  * skb was created with generic XDP routine.
1722                  */
1723                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1724                 if (IS_ERR(skb)) {
1725                         atomic_long_inc(&tun->dev->rx_dropped);
1726                         return PTR_ERR(skb);
1727                 }
1728                 if (!skb)
1729                         return total_len;
1730         } else {
1731                 if (!zerocopy) {
1732                         copylen = len;
1733                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1734                                 linear = good_linear;
1735                         else
1736                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1737                 }
1738
1739                 if (frags) {
1740                         mutex_lock(&tfile->napi_mutex);
1741                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1742                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1743                          * If zerocopy is enabled, then this layout will be
1744                          * overwritten by zerocopy_sg_from_iter().
1745                          */
1746                         zerocopy = false;
1747                 } else {
1748                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1749                                             noblock);
1750                 }
1751
1752                 if (IS_ERR(skb)) {
1753                         if (PTR_ERR(skb) != -EAGAIN)
1754                                 atomic_long_inc(&tun->dev->rx_dropped);
1755                         if (frags)
1756                                 mutex_unlock(&tfile->napi_mutex);
1757                         return PTR_ERR(skb);
1758                 }
1759
1760                 if (zerocopy)
1761                         err = zerocopy_sg_from_iter(skb, from);
1762                 else
1763                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1764
1765                 if (err) {
1766                         err = -EFAULT;
1767 drop:
1768                         atomic_long_inc(&tun->dev->rx_dropped);
1769                         kfree_skb(skb);
1770                         if (frags) {
1771                                 tfile->napi.skb = NULL;
1772                                 mutex_unlock(&tfile->napi_mutex);
1773                         }
1774
1775                         return err;
1776                 }
1777         }
1778
1779         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1780                 atomic_long_inc(&tun->rx_frame_errors);
1781                 kfree_skb(skb);
1782                 if (frags) {
1783                         tfile->napi.skb = NULL;
1784                         mutex_unlock(&tfile->napi_mutex);
1785                 }
1786
1787                 return -EINVAL;
1788         }
1789
1790         switch (tun->flags & TUN_TYPE_MASK) {
1791         case IFF_TUN:
1792                 if (tun->flags & IFF_NO_PI) {
1793                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1794
1795                         switch (ip_version) {
1796                         case 4:
1797                                 pi.proto = htons(ETH_P_IP);
1798                                 break;
1799                         case 6:
1800                                 pi.proto = htons(ETH_P_IPV6);
1801                                 break;
1802                         default:
1803                                 atomic_long_inc(&tun->dev->rx_dropped);
1804                                 kfree_skb(skb);
1805                                 return -EINVAL;
1806                         }
1807                 }
1808
1809                 skb_reset_mac_header(skb);
1810                 skb->protocol = pi.proto;
1811                 skb->dev = tun->dev;
1812                 break;
1813         case IFF_TAP:
1814                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1815                         err = -ENOMEM;
1816                         goto drop;
1817                 }
1818                 skb->protocol = eth_type_trans(skb, tun->dev);
1819                 break;
1820         }
1821
1822         /* copy skb_ubuf_info for callback when skb has no error */
1823         if (zerocopy) {
1824                 skb_zcopy_init(skb, msg_control);
1825         } else if (msg_control) {
1826                 struct ubuf_info *uarg = msg_control;
1827                 uarg->callback(NULL, uarg, false);
1828         }
1829
1830         skb_reset_network_header(skb);
1831         skb_probe_transport_header(skb);
1832         skb_record_rx_queue(skb, tfile->queue_index);
1833
1834         if (skb_xdp) {
1835                 struct bpf_prog *xdp_prog;
1836                 int ret;
1837
1838                 local_bh_disable();
1839                 rcu_read_lock();
1840                 xdp_prog = rcu_dereference(tun->xdp_prog);
1841                 if (xdp_prog) {
1842                         ret = do_xdp_generic(xdp_prog, skb);
1843                         if (ret != XDP_PASS) {
1844                                 rcu_read_unlock();
1845                                 local_bh_enable();
1846                                 if (frags) {
1847                                         tfile->napi.skb = NULL;
1848                                         mutex_unlock(&tfile->napi_mutex);
1849                                 }
1850                                 return total_len;
1851                         }
1852                 }
1853                 rcu_read_unlock();
1854                 local_bh_enable();
1855         }
1856
1857         /* Compute the costly rx hash only if needed for flow updates.
1858          * We may get a very small possibility of OOO during switching, not
1859          * worth to optimize.
1860          */
1861         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1862             !tfile->detached)
1863                 rxhash = __skb_get_hash_symmetric(skb);
1864
1865         rcu_read_lock();
1866         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1867                 err = -EIO;
1868                 rcu_read_unlock();
1869                 goto drop;
1870         }
1871
1872         if (frags) {
1873                 u32 headlen;
1874
1875                 /* Exercise flow dissector code path. */
1876                 skb_push(skb, ETH_HLEN);
1877                 headlen = eth_get_headlen(tun->dev, skb->data,
1878                                           skb_headlen(skb));
1879
1880                 if (unlikely(headlen > skb_headlen(skb))) {
1881                         atomic_long_inc(&tun->dev->rx_dropped);
1882                         napi_free_frags(&tfile->napi);
1883                         rcu_read_unlock();
1884                         mutex_unlock(&tfile->napi_mutex);
1885                         WARN_ON(1);
1886                         return -ENOMEM;
1887                 }
1888
1889                 local_bh_disable();
1890                 napi_gro_frags(&tfile->napi);
1891                 local_bh_enable();
1892                 mutex_unlock(&tfile->napi_mutex);
1893         } else if (tfile->napi_enabled) {
1894                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1895                 int queue_len;
1896
1897                 spin_lock_bh(&queue->lock);
1898                 __skb_queue_tail(queue, skb);
1899                 queue_len = skb_queue_len(queue);
1900                 spin_unlock(&queue->lock);
1901
1902                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1903                         napi_schedule(&tfile->napi);
1904
1905                 local_bh_enable();
1906         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1907                 tun_rx_batched(tun, tfile, skb, more);
1908         } else {
1909                 netif_rx_ni(skb);
1910         }
1911         rcu_read_unlock();
1912
1913         preempt_disable();
1914         dev_sw_netstats_rx_add(tun->dev, len);
1915         preempt_enable();
1916
1917         if (rxhash)
1918                 tun_flow_update(tun, rxhash, tfile);
1919
1920         return total_len;
1921 }
1922
1923 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1924 {
1925         struct file *file = iocb->ki_filp;
1926         struct tun_file *tfile = file->private_data;
1927         struct tun_struct *tun = tun_get(tfile);
1928         ssize_t result;
1929         int noblock = 0;
1930
1931         if (!tun)
1932                 return -EBADFD;
1933
1934         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1935                 noblock = 1;
1936
1937         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1938
1939         tun_put(tun);
1940         return result;
1941 }
1942
1943 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1944                                 struct tun_file *tfile,
1945                                 struct xdp_frame *xdp_frame,
1946                                 struct iov_iter *iter)
1947 {
1948         int vnet_hdr_sz = 0;
1949         size_t size = xdp_frame->len;
1950         size_t ret;
1951
1952         if (tun->flags & IFF_VNET_HDR) {
1953                 struct virtio_net_hdr gso = { 0 };
1954
1955                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1956                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1957                         return -EINVAL;
1958                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1959                              sizeof(gso)))
1960                         return -EFAULT;
1961                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1962         }
1963
1964         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1965
1966         preempt_disable();
1967         dev_sw_netstats_tx_add(tun->dev, 1, ret);
1968         preempt_enable();
1969
1970         return ret;
1971 }
1972
1973 /* Put packet to the user space buffer */
1974 static ssize_t tun_put_user(struct tun_struct *tun,
1975                             struct tun_file *tfile,
1976                             struct sk_buff *skb,
1977                             struct iov_iter *iter)
1978 {
1979         struct tun_pi pi = { 0, skb->protocol };
1980         ssize_t total;
1981         int vlan_offset = 0;
1982         int vlan_hlen = 0;
1983         int vnet_hdr_sz = 0;
1984
1985         if (skb_vlan_tag_present(skb))
1986                 vlan_hlen = VLAN_HLEN;
1987
1988         if (tun->flags & IFF_VNET_HDR)
1989                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1990
1991         total = skb->len + vlan_hlen + vnet_hdr_sz;
1992
1993         if (!(tun->flags & IFF_NO_PI)) {
1994                 if (iov_iter_count(iter) < sizeof(pi))
1995                         return -EINVAL;
1996
1997                 total += sizeof(pi);
1998                 if (iov_iter_count(iter) < total) {
1999                         /* Packet will be striped */
2000                         pi.flags |= TUN_PKT_STRIP;
2001                 }
2002
2003                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2004                         return -EFAULT;
2005         }
2006
2007         if (vnet_hdr_sz) {
2008                 struct virtio_net_hdr gso;
2009
2010                 if (iov_iter_count(iter) < vnet_hdr_sz)
2011                         return -EINVAL;
2012
2013                 if (virtio_net_hdr_from_skb(skb, &gso,
2014                                             tun_is_little_endian(tun), true,
2015                                             vlan_hlen)) {
2016                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2017                         pr_err("unexpected GSO type: "
2018                                "0x%x, gso_size %d, hdr_len %d\n",
2019                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2020                                tun16_to_cpu(tun, gso.hdr_len));
2021                         print_hex_dump(KERN_ERR, "tun: ",
2022                                        DUMP_PREFIX_NONE,
2023                                        16, 1, skb->head,
2024                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2025                         WARN_ON_ONCE(1);
2026                         return -EINVAL;
2027                 }
2028
2029                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2030                         return -EFAULT;
2031
2032                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2033         }
2034
2035         if (vlan_hlen) {
2036                 int ret;
2037                 struct veth veth;
2038
2039                 veth.h_vlan_proto = skb->vlan_proto;
2040                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2041
2042                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2043
2044                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2045                 if (ret || !iov_iter_count(iter))
2046                         goto done;
2047
2048                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2049                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2050                         goto done;
2051         }
2052
2053         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2054
2055 done:
2056         /* caller is in process context, */
2057         preempt_disable();
2058         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2059         preempt_enable();
2060
2061         return total;
2062 }
2063
2064 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2065 {
2066         DECLARE_WAITQUEUE(wait, current);
2067         void *ptr = NULL;
2068         int error = 0;
2069
2070         ptr = ptr_ring_consume(&tfile->tx_ring);
2071         if (ptr)
2072                 goto out;
2073         if (noblock) {
2074                 error = -EAGAIN;
2075                 goto out;
2076         }
2077
2078         add_wait_queue(&tfile->socket.wq.wait, &wait);
2079
2080         while (1) {
2081                 set_current_state(TASK_INTERRUPTIBLE);
2082                 ptr = ptr_ring_consume(&tfile->tx_ring);
2083                 if (ptr)
2084                         break;
2085                 if (signal_pending(current)) {
2086                         error = -ERESTARTSYS;
2087                         break;
2088                 }
2089                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2090                         error = -EFAULT;
2091                         break;
2092                 }
2093
2094                 schedule();
2095         }
2096
2097         __set_current_state(TASK_RUNNING);
2098         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2099
2100 out:
2101         *err = error;
2102         return ptr;
2103 }
2104
2105 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2106                            struct iov_iter *to,
2107                            int noblock, void *ptr)
2108 {
2109         ssize_t ret;
2110         int err;
2111
2112         if (!iov_iter_count(to)) {
2113                 tun_ptr_free(ptr);
2114                 return 0;
2115         }
2116
2117         if (!ptr) {
2118                 /* Read frames from ring */
2119                 ptr = tun_ring_recv(tfile, noblock, &err);
2120                 if (!ptr)
2121                         return err;
2122         }
2123
2124         if (tun_is_xdp_frame(ptr)) {
2125                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2126
2127                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2128                 xdp_return_frame(xdpf);
2129         } else {
2130                 struct sk_buff *skb = ptr;
2131
2132                 ret = tun_put_user(tun, tfile, skb, to);
2133                 if (unlikely(ret < 0))
2134                         kfree_skb(skb);
2135                 else
2136                         consume_skb(skb);
2137         }
2138
2139         return ret;
2140 }
2141
2142 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2143 {
2144         struct file *file = iocb->ki_filp;
2145         struct tun_file *tfile = file->private_data;
2146         struct tun_struct *tun = tun_get(tfile);
2147         ssize_t len = iov_iter_count(to), ret;
2148         int noblock = 0;
2149
2150         if (!tun)
2151                 return -EBADFD;
2152
2153         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2154                 noblock = 1;
2155
2156         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2157         ret = min_t(ssize_t, ret, len);
2158         if (ret > 0)
2159                 iocb->ki_pos = ret;
2160         tun_put(tun);
2161         return ret;
2162 }
2163
2164 static void tun_prog_free(struct rcu_head *rcu)
2165 {
2166         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2167
2168         bpf_prog_destroy(prog->prog);
2169         kfree(prog);
2170 }
2171
2172 static int __tun_set_ebpf(struct tun_struct *tun,
2173                           struct tun_prog __rcu **prog_p,
2174                           struct bpf_prog *prog)
2175 {
2176         struct tun_prog *old, *new = NULL;
2177
2178         if (prog) {
2179                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2180                 if (!new)
2181                         return -ENOMEM;
2182                 new->prog = prog;
2183         }
2184
2185         spin_lock_bh(&tun->lock);
2186         old = rcu_dereference_protected(*prog_p,
2187                                         lockdep_is_held(&tun->lock));
2188         rcu_assign_pointer(*prog_p, new);
2189         spin_unlock_bh(&tun->lock);
2190
2191         if (old)
2192                 call_rcu(&old->rcu, tun_prog_free);
2193
2194         return 0;
2195 }
2196
2197 static void tun_free_netdev(struct net_device *dev)
2198 {
2199         struct tun_struct *tun = netdev_priv(dev);
2200
2201         BUG_ON(!(list_empty(&tun->disabled)));
2202
2203         free_percpu(dev->tstats);
2204         /* We clear tstats so that tun_set_iff() can tell if
2205          * tun_free_netdev() has been called from register_netdevice().
2206          */
2207         dev->tstats = NULL;
2208
2209         tun_flow_uninit(tun);
2210         security_tun_dev_free_security(tun->security);
2211         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2212         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2213 }
2214
2215 static void tun_setup(struct net_device *dev)
2216 {
2217         struct tun_struct *tun = netdev_priv(dev);
2218
2219         tun->owner = INVALID_UID;
2220         tun->group = INVALID_GID;
2221         tun_default_link_ksettings(dev, &tun->link_ksettings);
2222
2223         dev->ethtool_ops = &tun_ethtool_ops;
2224         dev->needs_free_netdev = true;
2225         dev->priv_destructor = tun_free_netdev;
2226         /* We prefer our own queue length */
2227         dev->tx_queue_len = TUN_READQ_SIZE;
2228 }
2229
2230 /* Trivial set of netlink ops to allow deleting tun or tap
2231  * device with netlink.
2232  */
2233 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2234                         struct netlink_ext_ack *extack)
2235 {
2236         NL_SET_ERR_MSG(extack,
2237                        "tun/tap creation via rtnetlink is not supported.");
2238         return -EOPNOTSUPP;
2239 }
2240
2241 static size_t tun_get_size(const struct net_device *dev)
2242 {
2243         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2244         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2245
2246         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2247                nla_total_size(sizeof(gid_t)) + /* GROUP */
2248                nla_total_size(sizeof(u8)) + /* TYPE */
2249                nla_total_size(sizeof(u8)) + /* PI */
2250                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2251                nla_total_size(sizeof(u8)) + /* PERSIST */
2252                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2253                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2254                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2255                0;
2256 }
2257
2258 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2259 {
2260         struct tun_struct *tun = netdev_priv(dev);
2261
2262         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2263                 goto nla_put_failure;
2264         if (uid_valid(tun->owner) &&
2265             nla_put_u32(skb, IFLA_TUN_OWNER,
2266                         from_kuid_munged(current_user_ns(), tun->owner)))
2267                 goto nla_put_failure;
2268         if (gid_valid(tun->group) &&
2269             nla_put_u32(skb, IFLA_TUN_GROUP,
2270                         from_kgid_munged(current_user_ns(), tun->group)))
2271                 goto nla_put_failure;
2272         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2273                 goto nla_put_failure;
2274         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2275                 goto nla_put_failure;
2276         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2277                 goto nla_put_failure;
2278         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2279                        !!(tun->flags & IFF_MULTI_QUEUE)))
2280                 goto nla_put_failure;
2281         if (tun->flags & IFF_MULTI_QUEUE) {
2282                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2283                         goto nla_put_failure;
2284                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2285                                 tun->numdisabled))
2286                         goto nla_put_failure;
2287         }
2288
2289         return 0;
2290
2291 nla_put_failure:
2292         return -EMSGSIZE;
2293 }
2294
2295 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2296         .kind           = DRV_NAME,
2297         .priv_size      = sizeof(struct tun_struct),
2298         .setup          = tun_setup,
2299         .validate       = tun_validate,
2300         .get_size       = tun_get_size,
2301         .fill_info      = tun_fill_info,
2302 };
2303
2304 static void tun_sock_write_space(struct sock *sk)
2305 {
2306         struct tun_file *tfile;
2307         wait_queue_head_t *wqueue;
2308
2309         if (!sock_writeable(sk))
2310                 return;
2311
2312         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2313                 return;
2314
2315         wqueue = sk_sleep(sk);
2316         if (wqueue && waitqueue_active(wqueue))
2317                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2318                                                 EPOLLWRNORM | EPOLLWRBAND);
2319
2320         tfile = container_of(sk, struct tun_file, sk);
2321         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2322 }
2323
2324 static void tun_put_page(struct tun_page *tpage)
2325 {
2326         if (tpage->page)
2327                 __page_frag_cache_drain(tpage->page, tpage->count);
2328 }
2329
2330 static int tun_xdp_one(struct tun_struct *tun,
2331                        struct tun_file *tfile,
2332                        struct xdp_buff *xdp, int *flush,
2333                        struct tun_page *tpage)
2334 {
2335         unsigned int datasize = xdp->data_end - xdp->data;
2336         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2337         struct virtio_net_hdr *gso = &hdr->gso;
2338         struct bpf_prog *xdp_prog;
2339         struct sk_buff *skb = NULL;
2340         u32 rxhash = 0, act;
2341         int buflen = hdr->buflen;
2342         int err = 0;
2343         bool skb_xdp = false;
2344         struct page *page;
2345
2346         xdp_prog = rcu_dereference(tun->xdp_prog);
2347         if (xdp_prog) {
2348                 if (gso->gso_type) {
2349                         skb_xdp = true;
2350                         goto build;
2351                 }
2352
2353                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2354                 xdp_set_data_meta_invalid(xdp);
2355
2356                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2357                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2358                 if (err < 0) {
2359                         put_page(virt_to_head_page(xdp->data));
2360                         return err;
2361                 }
2362
2363                 switch (err) {
2364                 case XDP_REDIRECT:
2365                         *flush = true;
2366                         fallthrough;
2367                 case XDP_TX:
2368                         return 0;
2369                 case XDP_PASS:
2370                         break;
2371                 default:
2372                         page = virt_to_head_page(xdp->data);
2373                         if (tpage->page == page) {
2374                                 ++tpage->count;
2375                         } else {
2376                                 tun_put_page(tpage);
2377                                 tpage->page = page;
2378                                 tpage->count = 1;
2379                         }
2380                         return 0;
2381                 }
2382         }
2383
2384 build:
2385         skb = build_skb(xdp->data_hard_start, buflen);
2386         if (!skb) {
2387                 err = -ENOMEM;
2388                 goto out;
2389         }
2390
2391         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2392         skb_put(skb, xdp->data_end - xdp->data);
2393
2394         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2395                 atomic_long_inc(&tun->rx_frame_errors);
2396                 kfree_skb(skb);
2397                 err = -EINVAL;
2398                 goto out;
2399         }
2400
2401         skb->protocol = eth_type_trans(skb, tun->dev);
2402         skb_reset_network_header(skb);
2403         skb_probe_transport_header(skb);
2404         skb_record_rx_queue(skb, tfile->queue_index);
2405
2406         if (skb_xdp) {
2407                 err = do_xdp_generic(xdp_prog, skb);
2408                 if (err != XDP_PASS)
2409                         goto out;
2410         }
2411
2412         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2413             !tfile->detached)
2414                 rxhash = __skb_get_hash_symmetric(skb);
2415
2416         netif_receive_skb(skb);
2417
2418         /* No need to disable preemption here since this function is
2419          * always called with bh disabled
2420          */
2421         dev_sw_netstats_rx_add(tun->dev, datasize);
2422
2423         if (rxhash)
2424                 tun_flow_update(tun, rxhash, tfile);
2425
2426 out:
2427         return err;
2428 }
2429
2430 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2431 {
2432         int ret, i;
2433         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2434         struct tun_struct *tun = tun_get(tfile);
2435         struct tun_msg_ctl *ctl = m->msg_control;
2436         struct xdp_buff *xdp;
2437
2438         if (!tun)
2439                 return -EBADFD;
2440
2441         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2442                 struct tun_page tpage;
2443                 int n = ctl->num;
2444                 int flush = 0;
2445
2446                 memset(&tpage, 0, sizeof(tpage));
2447
2448                 local_bh_disable();
2449                 rcu_read_lock();
2450
2451                 for (i = 0; i < n; i++) {
2452                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2453                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2454                 }
2455
2456                 if (flush)
2457                         xdp_do_flush();
2458
2459                 rcu_read_unlock();
2460                 local_bh_enable();
2461
2462                 tun_put_page(&tpage);
2463
2464                 ret = total_len;
2465                 goto out;
2466         }
2467
2468         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2469                            m->msg_flags & MSG_DONTWAIT,
2470                            m->msg_flags & MSG_MORE);
2471 out:
2472         tun_put(tun);
2473         return ret;
2474 }
2475
2476 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2477                        int flags)
2478 {
2479         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2480         struct tun_struct *tun = tun_get(tfile);
2481         void *ptr = m->msg_control;
2482         int ret;
2483
2484         if (!tun) {
2485                 ret = -EBADFD;
2486                 goto out_free;
2487         }
2488
2489         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2490                 ret = -EINVAL;
2491                 goto out_put_tun;
2492         }
2493         if (flags & MSG_ERRQUEUE) {
2494                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2495                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2496                 goto out;
2497         }
2498         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2499         if (ret > (ssize_t)total_len) {
2500                 m->msg_flags |= MSG_TRUNC;
2501                 ret = flags & MSG_TRUNC ? ret : total_len;
2502         }
2503 out:
2504         tun_put(tun);
2505         return ret;
2506
2507 out_put_tun:
2508         tun_put(tun);
2509 out_free:
2510         tun_ptr_free(ptr);
2511         return ret;
2512 }
2513
2514 static int tun_ptr_peek_len(void *ptr)
2515 {
2516         if (likely(ptr)) {
2517                 if (tun_is_xdp_frame(ptr)) {
2518                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2519
2520                         return xdpf->len;
2521                 }
2522                 return __skb_array_len_with_tag(ptr);
2523         } else {
2524                 return 0;
2525         }
2526 }
2527
2528 static int tun_peek_len(struct socket *sock)
2529 {
2530         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2531         struct tun_struct *tun;
2532         int ret = 0;
2533
2534         tun = tun_get(tfile);
2535         if (!tun)
2536                 return 0;
2537
2538         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2539         tun_put(tun);
2540
2541         return ret;
2542 }
2543
2544 /* Ops structure to mimic raw sockets with tun */
2545 static const struct proto_ops tun_socket_ops = {
2546         .peek_len = tun_peek_len,
2547         .sendmsg = tun_sendmsg,
2548         .recvmsg = tun_recvmsg,
2549 };
2550
2551 static struct proto tun_proto = {
2552         .name           = "tun",
2553         .owner          = THIS_MODULE,
2554         .obj_size       = sizeof(struct tun_file),
2555 };
2556
2557 static int tun_flags(struct tun_struct *tun)
2558 {
2559         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2560 }
2561
2562 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2563                               char *buf)
2564 {
2565         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2566         return sprintf(buf, "0x%x\n", tun_flags(tun));
2567 }
2568
2569 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2570                           char *buf)
2571 {
2572         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2573         return uid_valid(tun->owner)?
2574                 sprintf(buf, "%u\n",
2575                         from_kuid_munged(current_user_ns(), tun->owner)):
2576                 sprintf(buf, "-1\n");
2577 }
2578
2579 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2580                           char *buf)
2581 {
2582         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2583         return gid_valid(tun->group) ?
2584                 sprintf(buf, "%u\n",
2585                         from_kgid_munged(current_user_ns(), tun->group)):
2586                 sprintf(buf, "-1\n");
2587 }
2588
2589 static DEVICE_ATTR_RO(tun_flags);
2590 static DEVICE_ATTR_RO(owner);
2591 static DEVICE_ATTR_RO(group);
2592
2593 static struct attribute *tun_dev_attrs[] = {
2594         &dev_attr_tun_flags.attr,
2595         &dev_attr_owner.attr,
2596         &dev_attr_group.attr,
2597         NULL
2598 };
2599
2600 static const struct attribute_group tun_attr_group = {
2601         .attrs = tun_dev_attrs
2602 };
2603
2604 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2605 {
2606         struct tun_struct *tun;
2607         struct tun_file *tfile = file->private_data;
2608         struct net_device *dev;
2609         int err;
2610
2611         if (tfile->detached)
2612                 return -EINVAL;
2613
2614         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2615                 if (!capable(CAP_NET_ADMIN))
2616                         return -EPERM;
2617
2618                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2619                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2620                         return -EINVAL;
2621         }
2622
2623         dev = __dev_get_by_name(net, ifr->ifr_name);
2624         if (dev) {
2625                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2626                         return -EBUSY;
2627                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2628                         tun = netdev_priv(dev);
2629                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2630                         tun = netdev_priv(dev);
2631                 else
2632                         return -EINVAL;
2633
2634                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2635                     !!(tun->flags & IFF_MULTI_QUEUE))
2636                         return -EINVAL;
2637
2638                 if (tun_not_capable(tun))
2639                         return -EPERM;
2640                 err = security_tun_dev_open(tun->security);
2641                 if (err < 0)
2642                         return err;
2643
2644                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2645                                  ifr->ifr_flags & IFF_NAPI,
2646                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2647                 if (err < 0)
2648                         return err;
2649
2650                 if (tun->flags & IFF_MULTI_QUEUE &&
2651                     (tun->numqueues + tun->numdisabled > 1)) {
2652                         /* One or more queue has already been attached, no need
2653                          * to initialize the device again.
2654                          */
2655                         netdev_state_change(dev);
2656                         return 0;
2657                 }
2658
2659                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2660                               (ifr->ifr_flags & TUN_FEATURES);
2661
2662                 netdev_state_change(dev);
2663         } else {
2664                 char *name;
2665                 unsigned long flags = 0;
2666                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2667                              MAX_TAP_QUEUES : 1;
2668
2669                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2670                         return -EPERM;
2671                 err = security_tun_dev_create();
2672                 if (err < 0)
2673                         return err;
2674
2675                 /* Set dev type */
2676                 if (ifr->ifr_flags & IFF_TUN) {
2677                         /* TUN device */
2678                         flags |= IFF_TUN;
2679                         name = "tun%d";
2680                 } else if (ifr->ifr_flags & IFF_TAP) {
2681                         /* TAP device */
2682                         flags |= IFF_TAP;
2683                         name = "tap%d";
2684                 } else
2685                         return -EINVAL;
2686
2687                 if (*ifr->ifr_name)
2688                         name = ifr->ifr_name;
2689
2690                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2691                                        NET_NAME_UNKNOWN, tun_setup, queues,
2692                                        queues);
2693
2694                 if (!dev)
2695                         return -ENOMEM;
2696
2697                 dev_net_set(dev, net);
2698                 dev->rtnl_link_ops = &tun_link_ops;
2699                 dev->ifindex = tfile->ifindex;
2700                 dev->sysfs_groups[0] = &tun_attr_group;
2701
2702                 tun = netdev_priv(dev);
2703                 tun->dev = dev;
2704                 tun->flags = flags;
2705                 tun->txflt.count = 0;
2706                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2707
2708                 tun->align = NET_SKB_PAD;
2709                 tun->filter_attached = false;
2710                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2711                 tun->rx_batched = 0;
2712                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2713
2714                 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2715                 if (!dev->tstats) {
2716                         err = -ENOMEM;
2717                         goto err_free_dev;
2718                 }
2719
2720                 spin_lock_init(&tun->lock);
2721
2722                 err = security_tun_dev_alloc_security(&tun->security);
2723                 if (err < 0)
2724                         goto err_free_stat;
2725
2726                 tun_net_init(dev);
2727                 tun_flow_init(tun);
2728
2729                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2730                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2731                                    NETIF_F_HW_VLAN_STAG_TX;
2732                 dev->features = dev->hw_features | NETIF_F_LLTX;
2733                 dev->vlan_features = dev->features &
2734                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2735                                        NETIF_F_HW_VLAN_STAG_TX);
2736
2737                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2738                               (ifr->ifr_flags & TUN_FEATURES);
2739
2740                 INIT_LIST_HEAD(&tun->disabled);
2741                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2742                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2743                 if (err < 0)
2744                         goto err_free_flow;
2745
2746                 err = register_netdevice(tun->dev);
2747                 if (err < 0)
2748                         goto err_detach;
2749                 /* free_netdev() won't check refcnt, to avoid race
2750                  * with dev_put() we need publish tun after registration.
2751                  */
2752                 rcu_assign_pointer(tfile->tun, tun);
2753         }
2754
2755         netif_carrier_on(tun->dev);
2756
2757         /* Make sure persistent devices do not get stuck in
2758          * xoff state.
2759          */
2760         if (netif_running(tun->dev))
2761                 netif_tx_wake_all_queues(tun->dev);
2762
2763         strcpy(ifr->ifr_name, tun->dev->name);
2764         return 0;
2765
2766 err_detach:
2767         tun_detach_all(dev);
2768         /* We are here because register_netdevice() has failed.
2769          * If register_netdevice() already called tun_free_netdev()
2770          * while dealing with the error, dev->stats has been cleared.
2771          */
2772         if (!dev->tstats)
2773                 goto err_free_dev;
2774
2775 err_free_flow:
2776         tun_flow_uninit(tun);
2777         security_tun_dev_free_security(tun->security);
2778 err_free_stat:
2779         free_percpu(dev->tstats);
2780 err_free_dev:
2781         free_netdev(dev);
2782         return err;
2783 }
2784
2785 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2786 {
2787         strcpy(ifr->ifr_name, tun->dev->name);
2788
2789         ifr->ifr_flags = tun_flags(tun);
2790
2791 }
2792
2793 /* This is like a cut-down ethtool ops, except done via tun fd so no
2794  * privs required. */
2795 static int set_offload(struct tun_struct *tun, unsigned long arg)
2796 {
2797         netdev_features_t features = 0;
2798
2799         if (arg & TUN_F_CSUM) {
2800                 features |= NETIF_F_HW_CSUM;
2801                 arg &= ~TUN_F_CSUM;
2802
2803                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2804                         if (arg & TUN_F_TSO_ECN) {
2805                                 features |= NETIF_F_TSO_ECN;
2806                                 arg &= ~TUN_F_TSO_ECN;
2807                         }
2808                         if (arg & TUN_F_TSO4)
2809                                 features |= NETIF_F_TSO;
2810                         if (arg & TUN_F_TSO6)
2811                                 features |= NETIF_F_TSO6;
2812                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2813                 }
2814
2815                 arg &= ~TUN_F_UFO;
2816         }
2817
2818         /* This gives the user a way to test for new features in future by
2819          * trying to set them. */
2820         if (arg)
2821                 return -EINVAL;
2822
2823         tun->set_features = features;
2824         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2825         tun->dev->wanted_features |= features;
2826         netdev_update_features(tun->dev);
2827
2828         return 0;
2829 }
2830
2831 static void tun_detach_filter(struct tun_struct *tun, int n)
2832 {
2833         int i;
2834         struct tun_file *tfile;
2835
2836         for (i = 0; i < n; i++) {
2837                 tfile = rtnl_dereference(tun->tfiles[i]);
2838                 lock_sock(tfile->socket.sk);
2839                 sk_detach_filter(tfile->socket.sk);
2840                 release_sock(tfile->socket.sk);
2841         }
2842
2843         tun->filter_attached = false;
2844 }
2845
2846 static int tun_attach_filter(struct tun_struct *tun)
2847 {
2848         int i, ret = 0;
2849         struct tun_file *tfile;
2850
2851         for (i = 0; i < tun->numqueues; i++) {
2852                 tfile = rtnl_dereference(tun->tfiles[i]);
2853                 lock_sock(tfile->socket.sk);
2854                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2855                 release_sock(tfile->socket.sk);
2856                 if (ret) {
2857                         tun_detach_filter(tun, i);
2858                         return ret;
2859                 }
2860         }
2861
2862         tun->filter_attached = true;
2863         return ret;
2864 }
2865
2866 static void tun_set_sndbuf(struct tun_struct *tun)
2867 {
2868         struct tun_file *tfile;
2869         int i;
2870
2871         for (i = 0; i < tun->numqueues; i++) {
2872                 tfile = rtnl_dereference(tun->tfiles[i]);
2873                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2874         }
2875 }
2876
2877 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2878 {
2879         struct tun_file *tfile = file->private_data;
2880         struct tun_struct *tun;
2881         int ret = 0;
2882
2883         rtnl_lock();
2884
2885         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2886                 tun = tfile->detached;
2887                 if (!tun) {
2888                         ret = -EINVAL;
2889                         goto unlock;
2890                 }
2891                 ret = security_tun_dev_attach_queue(tun->security);
2892                 if (ret < 0)
2893                         goto unlock;
2894                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2895                                  tun->flags & IFF_NAPI_FRAGS, true);
2896         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2897                 tun = rtnl_dereference(tfile->tun);
2898                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2899                         ret = -EINVAL;
2900                 else
2901                         __tun_detach(tfile, false);
2902         } else
2903                 ret = -EINVAL;
2904
2905         if (ret >= 0)
2906                 netdev_state_change(tun->dev);
2907
2908 unlock:
2909         rtnl_unlock();
2910         return ret;
2911 }
2912
2913 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2914                         void __user *data)
2915 {
2916         struct bpf_prog *prog;
2917         int fd;
2918
2919         if (copy_from_user(&fd, data, sizeof(fd)))
2920                 return -EFAULT;
2921
2922         if (fd == -1) {
2923                 prog = NULL;
2924         } else {
2925                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2926                 if (IS_ERR(prog))
2927                         return PTR_ERR(prog);
2928         }
2929
2930         return __tun_set_ebpf(tun, prog_p, prog);
2931 }
2932
2933 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2934 static unsigned char tun_get_addr_len(unsigned short type)
2935 {
2936         switch (type) {
2937         case ARPHRD_IP6GRE:
2938         case ARPHRD_TUNNEL6:
2939                 return sizeof(struct in6_addr);
2940         case ARPHRD_IPGRE:
2941         case ARPHRD_TUNNEL:
2942         case ARPHRD_SIT:
2943                 return 4;
2944         case ARPHRD_ETHER:
2945                 return ETH_ALEN;
2946         case ARPHRD_IEEE802154:
2947         case ARPHRD_IEEE802154_MONITOR:
2948                 return IEEE802154_EXTENDED_ADDR_LEN;
2949         case ARPHRD_PHONET_PIPE:
2950         case ARPHRD_PPP:
2951         case ARPHRD_NONE:
2952                 return 0;
2953         case ARPHRD_6LOWPAN:
2954                 return EUI64_ADDR_LEN;
2955         case ARPHRD_FDDI:
2956                 return FDDI_K_ALEN;
2957         case ARPHRD_HIPPI:
2958                 return HIPPI_ALEN;
2959         case ARPHRD_IEEE802:
2960                 return FC_ALEN;
2961         case ARPHRD_ROSE:
2962                 return ROSE_ADDR_LEN;
2963         case ARPHRD_NETROM:
2964                 return AX25_ADDR_LEN;
2965         case ARPHRD_LOCALTLK:
2966                 return LTALK_ALEN;
2967         default:
2968                 return 0;
2969         }
2970 }
2971
2972 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2973                             unsigned long arg, int ifreq_len)
2974 {
2975         struct tun_file *tfile = file->private_data;
2976         struct net *net = sock_net(&tfile->sk);
2977         struct tun_struct *tun;
2978         void __user* argp = (void __user*)arg;
2979         unsigned int ifindex, carrier;
2980         struct ifreq ifr;
2981         kuid_t owner;
2982         kgid_t group;
2983         int sndbuf;
2984         int vnet_hdr_sz;
2985         int le;
2986         int ret;
2987         bool do_notify = false;
2988
2989         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2990             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2991                 if (copy_from_user(&ifr, argp, ifreq_len))
2992                         return -EFAULT;
2993         } else {
2994                 memset(&ifr, 0, sizeof(ifr));
2995         }
2996         if (cmd == TUNGETFEATURES) {
2997                 /* Currently this just means: "what IFF flags are valid?".
2998                  * This is needed because we never checked for invalid flags on
2999                  * TUNSETIFF.
3000                  */
3001                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3002                                 (unsigned int __user*)argp);
3003         } else if (cmd == TUNSETQUEUE) {
3004                 return tun_set_queue(file, &ifr);
3005         } else if (cmd == SIOCGSKNS) {
3006                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3007                         return -EPERM;
3008                 return open_related_ns(&net->ns, get_net_ns);
3009         }
3010
3011         rtnl_lock();
3012
3013         tun = tun_get(tfile);
3014         if (cmd == TUNSETIFF) {
3015                 ret = -EEXIST;
3016                 if (tun)
3017                         goto unlock;
3018
3019                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3020
3021                 ret = tun_set_iff(net, file, &ifr);
3022
3023                 if (ret)
3024                         goto unlock;
3025
3026                 if (copy_to_user(argp, &ifr, ifreq_len))
3027                         ret = -EFAULT;
3028                 goto unlock;
3029         }
3030         if (cmd == TUNSETIFINDEX) {
3031                 ret = -EPERM;
3032                 if (tun)
3033                         goto unlock;
3034
3035                 ret = -EFAULT;
3036                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3037                         goto unlock;
3038
3039                 ret = 0;
3040                 tfile->ifindex = ifindex;
3041                 goto unlock;
3042         }
3043
3044         ret = -EBADFD;
3045         if (!tun)
3046                 goto unlock;
3047
3048         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3049
3050         net = dev_net(tun->dev);
3051         ret = 0;
3052         switch (cmd) {
3053         case TUNGETIFF:
3054                 tun_get_iff(tun, &ifr);
3055
3056                 if (tfile->detached)
3057                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3058                 if (!tfile->socket.sk->sk_filter)
3059                         ifr.ifr_flags |= IFF_NOFILTER;
3060
3061                 if (copy_to_user(argp, &ifr, ifreq_len))
3062                         ret = -EFAULT;
3063                 break;
3064
3065         case TUNSETNOCSUM:
3066                 /* Disable/Enable checksum */
3067
3068                 /* [unimplemented] */
3069                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3070                            arg ? "disabled" : "enabled");
3071                 break;
3072
3073         case TUNSETPERSIST:
3074                 /* Disable/Enable persist mode. Keep an extra reference to the
3075                  * module to prevent the module being unprobed.
3076                  */
3077                 if (arg && !(tun->flags & IFF_PERSIST)) {
3078                         tun->flags |= IFF_PERSIST;
3079                         __module_get(THIS_MODULE);
3080                         do_notify = true;
3081                 }
3082                 if (!arg && (tun->flags & IFF_PERSIST)) {
3083                         tun->flags &= ~IFF_PERSIST;
3084                         module_put(THIS_MODULE);
3085                         do_notify = true;
3086                 }
3087
3088                 netif_info(tun, drv, tun->dev, "persist %s\n",
3089                            arg ? "enabled" : "disabled");
3090                 break;
3091
3092         case TUNSETOWNER:
3093                 /* Set owner of the device */
3094                 owner = make_kuid(current_user_ns(), arg);
3095                 if (!uid_valid(owner)) {
3096                         ret = -EINVAL;
3097                         break;
3098                 }
3099                 tun->owner = owner;
3100                 do_notify = true;
3101                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3102                            from_kuid(&init_user_ns, tun->owner));
3103                 break;
3104
3105         case TUNSETGROUP:
3106                 /* Set group of the device */
3107                 group = make_kgid(current_user_ns(), arg);
3108                 if (!gid_valid(group)) {
3109                         ret = -EINVAL;
3110                         break;
3111                 }
3112                 tun->group = group;
3113                 do_notify = true;
3114                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3115                            from_kgid(&init_user_ns, tun->group));
3116                 break;
3117
3118         case TUNSETLINK:
3119                 /* Only allow setting the type when the interface is down */
3120                 if (tun->dev->flags & IFF_UP) {
3121                         netif_info(tun, drv, tun->dev,
3122                                    "Linktype set failed because interface is up\n");
3123                         ret = -EBUSY;
3124                 } else {
3125                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3126                                                        tun->dev);
3127                         ret = notifier_to_errno(ret);
3128                         if (ret) {
3129                                 netif_info(tun, drv, tun->dev,
3130                                            "Refused to change device type\n");
3131                                 break;
3132                         }
3133                         tun->dev->type = (int) arg;
3134                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3135                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3136                                    tun->dev->type);
3137                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3138                                                  tun->dev);
3139                 }
3140                 break;
3141
3142         case TUNSETDEBUG:
3143                 tun->msg_enable = (u32)arg;
3144                 break;
3145
3146         case TUNSETOFFLOAD:
3147                 ret = set_offload(tun, arg);
3148                 break;
3149
3150         case TUNSETTXFILTER:
3151                 /* Can be set only for TAPs */
3152                 ret = -EINVAL;
3153                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3154                         break;
3155                 ret = update_filter(&tun->txflt, (void __user *)arg);
3156                 break;
3157
3158         case SIOCGIFHWADDR:
3159                 /* Get hw address */
3160                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3161                 if (copy_to_user(argp, &ifr, ifreq_len))
3162                         ret = -EFAULT;
3163                 break;
3164
3165         case SIOCSIFHWADDR:
3166                 /* Set hw address */
3167                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3168                 break;
3169
3170         case TUNGETSNDBUF:
3171                 sndbuf = tfile->socket.sk->sk_sndbuf;
3172                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3173                         ret = -EFAULT;
3174                 break;
3175
3176         case TUNSETSNDBUF:
3177                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3178                         ret = -EFAULT;
3179                         break;
3180                 }
3181                 if (sndbuf <= 0) {
3182                         ret = -EINVAL;
3183                         break;
3184                 }
3185
3186                 tun->sndbuf = sndbuf;
3187                 tun_set_sndbuf(tun);
3188                 break;
3189
3190         case TUNGETVNETHDRSZ:
3191                 vnet_hdr_sz = tun->vnet_hdr_sz;
3192                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3193                         ret = -EFAULT;
3194                 break;
3195
3196         case TUNSETVNETHDRSZ:
3197                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3198                         ret = -EFAULT;
3199                         break;
3200                 }
3201                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3202                         ret = -EINVAL;
3203                         break;
3204                 }
3205
3206                 tun->vnet_hdr_sz = vnet_hdr_sz;
3207                 break;
3208
3209         case TUNGETVNETLE:
3210                 le = !!(tun->flags & TUN_VNET_LE);
3211                 if (put_user(le, (int __user *)argp))
3212                         ret = -EFAULT;
3213                 break;
3214
3215         case TUNSETVNETLE:
3216                 if (get_user(le, (int __user *)argp)) {
3217                         ret = -EFAULT;
3218                         break;
3219                 }
3220                 if (le)
3221                         tun->flags |= TUN_VNET_LE;
3222                 else
3223                         tun->flags &= ~TUN_VNET_LE;
3224                 break;
3225
3226         case TUNGETVNETBE:
3227                 ret = tun_get_vnet_be(tun, argp);
3228                 break;
3229
3230         case TUNSETVNETBE:
3231                 ret = tun_set_vnet_be(tun, argp);
3232                 break;
3233
3234         case TUNATTACHFILTER:
3235                 /* Can be set only for TAPs */
3236                 ret = -EINVAL;
3237                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3238                         break;
3239                 ret = -EFAULT;
3240                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3241                         break;
3242
3243                 ret = tun_attach_filter(tun);
3244                 break;
3245
3246         case TUNDETACHFILTER:
3247                 /* Can be set only for TAPs */
3248                 ret = -EINVAL;
3249                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3250                         break;
3251                 ret = 0;
3252                 tun_detach_filter(tun, tun->numqueues);
3253                 break;
3254
3255         case TUNGETFILTER:
3256                 ret = -EINVAL;
3257                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3258                         break;
3259                 ret = -EFAULT;
3260                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3261                         break;
3262                 ret = 0;
3263                 break;
3264
3265         case TUNSETSTEERINGEBPF:
3266                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3267                 break;
3268
3269         case TUNSETFILTEREBPF:
3270                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3271                 break;
3272
3273         case TUNSETCARRIER:
3274                 ret = -EFAULT;
3275                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3276                         goto unlock;
3277
3278                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3279                 break;
3280
3281         case TUNGETDEVNETNS:
3282                 ret = -EPERM;
3283                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3284                         goto unlock;
3285                 ret = open_related_ns(&net->ns, get_net_ns);
3286                 break;
3287
3288         default:
3289                 ret = -EINVAL;
3290                 break;
3291         }
3292
3293         if (do_notify)
3294                 netdev_state_change(tun->dev);
3295
3296 unlock:
3297         rtnl_unlock();
3298         if (tun)
3299                 tun_put(tun);
3300         return ret;
3301 }
3302
3303 static long tun_chr_ioctl(struct file *file,
3304                           unsigned int cmd, unsigned long arg)
3305 {
3306         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3307 }
3308
3309 #ifdef CONFIG_COMPAT
3310 static long tun_chr_compat_ioctl(struct file *file,
3311                          unsigned int cmd, unsigned long arg)
3312 {
3313         switch (cmd) {
3314         case TUNSETIFF:
3315         case TUNGETIFF:
3316         case TUNSETTXFILTER:
3317         case TUNGETSNDBUF:
3318         case TUNSETSNDBUF:
3319         case SIOCGIFHWADDR:
3320         case SIOCSIFHWADDR:
3321                 arg = (unsigned long)compat_ptr(arg);
3322                 break;
3323         default:
3324                 arg = (compat_ulong_t)arg;
3325                 break;
3326         }
3327
3328         /*
3329          * compat_ifreq is shorter than ifreq, so we must not access beyond
3330          * the end of that structure. All fields that are used in this
3331          * driver are compatible though, we don't need to convert the
3332          * contents.
3333          */
3334         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3335 }
3336 #endif /* CONFIG_COMPAT */
3337
3338 static int tun_chr_fasync(int fd, struct file *file, int on)
3339 {
3340         struct tun_file *tfile = file->private_data;
3341         int ret;
3342
3343         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3344                 goto out;
3345
3346         if (on) {
3347                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3348                 tfile->flags |= TUN_FASYNC;
3349         } else
3350                 tfile->flags &= ~TUN_FASYNC;
3351         ret = 0;
3352 out:
3353         return ret;
3354 }
3355
3356 static int tun_chr_open(struct inode *inode, struct file * file)
3357 {
3358         struct net *net = current->nsproxy->net_ns;
3359         struct tun_file *tfile;
3360
3361         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3362                                             &tun_proto, 0);
3363         if (!tfile)
3364                 return -ENOMEM;
3365         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3366                 sk_free(&tfile->sk);
3367                 return -ENOMEM;
3368         }
3369
3370         mutex_init(&tfile->napi_mutex);
3371         RCU_INIT_POINTER(tfile->tun, NULL);
3372         tfile->flags = 0;
3373         tfile->ifindex = 0;
3374
3375         init_waitqueue_head(&tfile->socket.wq.wait);
3376
3377         tfile->socket.file = file;
3378         tfile->socket.ops = &tun_socket_ops;
3379
3380         sock_init_data(&tfile->socket, &tfile->sk);
3381
3382         tfile->sk.sk_write_space = tun_sock_write_space;
3383         tfile->sk.sk_sndbuf = INT_MAX;
3384
3385         file->private_data = tfile;
3386         INIT_LIST_HEAD(&tfile->next);
3387
3388         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3389
3390         return 0;
3391 }
3392
3393 static int tun_chr_close(struct inode *inode, struct file *file)
3394 {
3395         struct tun_file *tfile = file->private_data;
3396
3397         tun_detach(tfile, true);
3398
3399         return 0;
3400 }
3401
3402 #ifdef CONFIG_PROC_FS
3403 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3404 {
3405         struct tun_file *tfile = file->private_data;
3406         struct tun_struct *tun;
3407         struct ifreq ifr;
3408
3409         memset(&ifr, 0, sizeof(ifr));
3410
3411         rtnl_lock();
3412         tun = tun_get(tfile);
3413         if (tun)
3414                 tun_get_iff(tun, &ifr);
3415         rtnl_unlock();
3416
3417         if (tun)
3418                 tun_put(tun);
3419
3420         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3421 }
3422 #endif
3423
3424 static const struct file_operations tun_fops = {
3425         .owner  = THIS_MODULE,
3426         .llseek = no_llseek,
3427         .read_iter  = tun_chr_read_iter,
3428         .write_iter = tun_chr_write_iter,
3429         .poll   = tun_chr_poll,
3430         .unlocked_ioctl = tun_chr_ioctl,
3431 #ifdef CONFIG_COMPAT
3432         .compat_ioctl = tun_chr_compat_ioctl,
3433 #endif
3434         .open   = tun_chr_open,
3435         .release = tun_chr_close,
3436         .fasync = tun_chr_fasync,
3437 #ifdef CONFIG_PROC_FS
3438         .show_fdinfo = tun_chr_show_fdinfo,
3439 #endif
3440 };
3441
3442 static struct miscdevice tun_miscdev = {
3443         .minor = TUN_MINOR,
3444         .name = "tun",
3445         .nodename = "net/tun",
3446         .fops = &tun_fops,
3447 };
3448
3449 /* ethtool interface */
3450
3451 static void tun_default_link_ksettings(struct net_device *dev,
3452                                        struct ethtool_link_ksettings *cmd)
3453 {
3454         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3455         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3456         cmd->base.speed         = SPEED_10;
3457         cmd->base.duplex        = DUPLEX_FULL;
3458         cmd->base.port          = PORT_TP;
3459         cmd->base.phy_address   = 0;
3460         cmd->base.autoneg       = AUTONEG_DISABLE;
3461 }
3462
3463 static int tun_get_link_ksettings(struct net_device *dev,
3464                                   struct ethtool_link_ksettings *cmd)
3465 {
3466         struct tun_struct *tun = netdev_priv(dev);
3467
3468         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3469         return 0;
3470 }
3471
3472 static int tun_set_link_ksettings(struct net_device *dev,
3473                                   const struct ethtool_link_ksettings *cmd)
3474 {
3475         struct tun_struct *tun = netdev_priv(dev);
3476
3477         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3478         return 0;
3479 }
3480
3481 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3482 {
3483         struct tun_struct *tun = netdev_priv(dev);
3484
3485         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3486         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3487
3488         switch (tun->flags & TUN_TYPE_MASK) {
3489         case IFF_TUN:
3490                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3491                 break;
3492         case IFF_TAP:
3493                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3494                 break;
3495         }
3496 }
3497
3498 static u32 tun_get_msglevel(struct net_device *dev)
3499 {
3500         struct tun_struct *tun = netdev_priv(dev);
3501
3502         return tun->msg_enable;
3503 }
3504
3505 static void tun_set_msglevel(struct net_device *dev, u32 value)
3506 {
3507         struct tun_struct *tun = netdev_priv(dev);
3508
3509         tun->msg_enable = value;
3510 }
3511
3512 static int tun_get_coalesce(struct net_device *dev,
3513                             struct ethtool_coalesce *ec)
3514 {
3515         struct tun_struct *tun = netdev_priv(dev);
3516
3517         ec->rx_max_coalesced_frames = tun->rx_batched;
3518
3519         return 0;
3520 }
3521
3522 static int tun_set_coalesce(struct net_device *dev,
3523                             struct ethtool_coalesce *ec)
3524 {
3525         struct tun_struct *tun = netdev_priv(dev);
3526
3527         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3528                 tun->rx_batched = NAPI_POLL_WEIGHT;
3529         else
3530                 tun->rx_batched = ec->rx_max_coalesced_frames;
3531
3532         return 0;
3533 }
3534
3535 static const struct ethtool_ops tun_ethtool_ops = {
3536         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3537         .get_drvinfo    = tun_get_drvinfo,
3538         .get_msglevel   = tun_get_msglevel,
3539         .set_msglevel   = tun_set_msglevel,
3540         .get_link       = ethtool_op_get_link,
3541         .get_ts_info    = ethtool_op_get_ts_info,
3542         .get_coalesce   = tun_get_coalesce,
3543         .set_coalesce   = tun_set_coalesce,
3544         .get_link_ksettings = tun_get_link_ksettings,
3545         .set_link_ksettings = tun_set_link_ksettings,
3546 };
3547
3548 static int tun_queue_resize(struct tun_struct *tun)
3549 {
3550         struct net_device *dev = tun->dev;
3551         struct tun_file *tfile;
3552         struct ptr_ring **rings;
3553         int n = tun->numqueues + tun->numdisabled;
3554         int ret, i;
3555
3556         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3557         if (!rings)
3558                 return -ENOMEM;
3559
3560         for (i = 0; i < tun->numqueues; i++) {
3561                 tfile = rtnl_dereference(tun->tfiles[i]);
3562                 rings[i] = &tfile->tx_ring;
3563         }
3564         list_for_each_entry(tfile, &tun->disabled, next)
3565                 rings[i++] = &tfile->tx_ring;
3566
3567         ret = ptr_ring_resize_multiple(rings, n,
3568                                        dev->tx_queue_len, GFP_KERNEL,
3569                                        tun_ptr_free);
3570
3571         kfree(rings);
3572         return ret;
3573 }
3574
3575 static int tun_device_event(struct notifier_block *unused,
3576                             unsigned long event, void *ptr)
3577 {
3578         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3579         struct tun_struct *tun = netdev_priv(dev);
3580         int i;
3581
3582         if (dev->rtnl_link_ops != &tun_link_ops)
3583                 return NOTIFY_DONE;
3584
3585         switch (event) {
3586         case NETDEV_CHANGE_TX_QUEUE_LEN:
3587                 if (tun_queue_resize(tun))
3588                         return NOTIFY_BAD;
3589                 break;
3590         case NETDEV_UP:
3591                 for (i = 0; i < tun->numqueues; i++) {
3592                         struct tun_file *tfile;
3593
3594                         tfile = rtnl_dereference(tun->tfiles[i]);
3595                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3596                 }
3597                 break;
3598         default:
3599                 break;
3600         }
3601
3602         return NOTIFY_DONE;
3603 }
3604
3605 static struct notifier_block tun_notifier_block __read_mostly = {
3606         .notifier_call  = tun_device_event,
3607 };
3608
3609 static int __init tun_init(void)
3610 {
3611         int ret = 0;
3612
3613         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3614
3615         ret = rtnl_link_register(&tun_link_ops);
3616         if (ret) {
3617                 pr_err("Can't register link_ops\n");
3618                 goto err_linkops;
3619         }
3620
3621         ret = misc_register(&tun_miscdev);
3622         if (ret) {
3623                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3624                 goto err_misc;
3625         }
3626
3627         ret = register_netdevice_notifier(&tun_notifier_block);
3628         if (ret) {
3629                 pr_err("Can't register netdevice notifier\n");
3630                 goto err_notifier;
3631         }
3632
3633         return  0;
3634
3635 err_notifier:
3636         misc_deregister(&tun_miscdev);
3637 err_misc:
3638         rtnl_link_unregister(&tun_link_ops);
3639 err_linkops:
3640         return ret;
3641 }
3642
3643 static void tun_cleanup(void)
3644 {
3645         misc_deregister(&tun_miscdev);
3646         rtnl_link_unregister(&tun_link_ops);
3647         unregister_netdevice_notifier(&tun_notifier_block);
3648 }
3649
3650 /* Get an underlying socket object from tun file.  Returns error unless file is
3651  * attached to a device.  The returned object works like a packet socket, it
3652  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3653  * holding a reference to the file for as long as the socket is in use. */
3654 struct socket *tun_get_socket(struct file *file)
3655 {
3656         struct tun_file *tfile;
3657         if (file->f_op != &tun_fops)
3658                 return ERR_PTR(-EINVAL);
3659         tfile = file->private_data;
3660         if (!tfile)
3661                 return ERR_PTR(-EBADFD);
3662         return &tfile->socket;
3663 }
3664 EXPORT_SYMBOL_GPL(tun_get_socket);
3665
3666 struct ptr_ring *tun_get_tx_ring(struct file *file)
3667 {
3668         struct tun_file *tfile;
3669
3670         if (file->f_op != &tun_fops)
3671                 return ERR_PTR(-EINVAL);
3672         tfile = file->private_data;
3673         if (!tfile)
3674                 return ERR_PTR(-EBADFD);
3675         return &tfile->tx_ring;
3676 }
3677 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3678
3679 module_init(tun_init);
3680 module_exit(tun_cleanup);
3681 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3682 MODULE_AUTHOR(DRV_COPYRIGHT);
3683 MODULE_LICENSE("GPL");
3684 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3685 MODULE_ALIAS("devname:net/tun");