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 netdev_queue *queue;
1014         struct tun_file *tfile;
1015         int len = skb->len;
1016
1017         rcu_read_lock();
1018         tfile = rcu_dereference(tun->tfiles[txq]);
1019
1020         /* Drop packet if interface is not attached */
1021         if (!tfile)
1022                 goto drop;
1023
1024         if (!rcu_dereference(tun->steering_prog))
1025                 tun_automq_xmit(tun, skb);
1026
1027         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1028
1029         /* Drop if the filter does not like it.
1030          * This is a noop if the filter is disabled.
1031          * Filter can be enabled only for the TAP devices. */
1032         if (!check_filter(&tun->txflt, skb))
1033                 goto drop;
1034
1035         if (tfile->socket.sk->sk_filter &&
1036             sk_filter(tfile->socket.sk, skb))
1037                 goto drop;
1038
1039         len = run_ebpf_filter(tun, skb, len);
1040         if (len == 0 || pskb_trim(skb, len))
1041                 goto drop;
1042
1043         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1044                 goto drop;
1045
1046         skb_tx_timestamp(skb);
1047
1048         /* Orphan the skb - required as we might hang on to it
1049          * for indefinite time.
1050          */
1051         skb_orphan(skb);
1052
1053         nf_reset_ct(skb);
1054
1055         if (ptr_ring_produce(&tfile->tx_ring, skb))
1056                 goto drop;
1057
1058         /* NETIF_F_LLTX requires to do our own update of trans_start */
1059         queue = netdev_get_tx_queue(dev, txq);
1060         queue->trans_start = jiffies;
1061
1062         /* Notify and wake up reader process */
1063         if (tfile->flags & TUN_FASYNC)
1064                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1065         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1066
1067         rcu_read_unlock();
1068         return NETDEV_TX_OK;
1069
1070 drop:
1071         atomic_long_inc(&dev->tx_dropped);
1072         skb_tx_error(skb);
1073         kfree_skb(skb);
1074         rcu_read_unlock();
1075         return NET_XMIT_DROP;
1076 }
1077
1078 static void tun_net_mclist(struct net_device *dev)
1079 {
1080         /*
1081          * This callback is supposed to deal with mc filter in
1082          * _rx_ path and has nothing to do with the _tx_ path.
1083          * In rx path we always accept everything userspace gives us.
1084          */
1085 }
1086
1087 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1088         netdev_features_t features)
1089 {
1090         struct tun_struct *tun = netdev_priv(dev);
1091
1092         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1093 }
1094
1095 static void tun_set_headroom(struct net_device *dev, int new_hr)
1096 {
1097         struct tun_struct *tun = netdev_priv(dev);
1098
1099         if (new_hr < NET_SKB_PAD)
1100                 new_hr = NET_SKB_PAD;
1101
1102         tun->align = new_hr;
1103 }
1104
1105 static void
1106 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1107 {
1108         struct tun_struct *tun = netdev_priv(dev);
1109
1110         dev_get_tstats64(dev, stats);
1111
1112         stats->rx_frame_errors +=
1113                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1114 }
1115
1116 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1117                        struct netlink_ext_ack *extack)
1118 {
1119         struct tun_struct *tun = netdev_priv(dev);
1120         struct tun_file *tfile;
1121         struct bpf_prog *old_prog;
1122         int i;
1123
1124         old_prog = rtnl_dereference(tun->xdp_prog);
1125         rcu_assign_pointer(tun->xdp_prog, prog);
1126         if (old_prog)
1127                 bpf_prog_put(old_prog);
1128
1129         for (i = 0; i < tun->numqueues; i++) {
1130                 tfile = rtnl_dereference(tun->tfiles[i]);
1131                 if (prog)
1132                         sock_set_flag(&tfile->sk, SOCK_XDP);
1133                 else
1134                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1135         }
1136         list_for_each_entry(tfile, &tun->disabled, next) {
1137                 if (prog)
1138                         sock_set_flag(&tfile->sk, SOCK_XDP);
1139                 else
1140                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1141         }
1142
1143         return 0;
1144 }
1145
1146 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1147 {
1148         switch (xdp->command) {
1149         case XDP_SETUP_PROG:
1150                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1151         default:
1152                 return -EINVAL;
1153         }
1154 }
1155
1156 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1157 {
1158         if (new_carrier) {
1159                 struct tun_struct *tun = netdev_priv(dev);
1160
1161                 if (!tun->numqueues)
1162                         return -EPERM;
1163
1164                 netif_carrier_on(dev);
1165         } else {
1166                 netif_carrier_off(dev);
1167         }
1168         return 0;
1169 }
1170
1171 static const struct net_device_ops tun_netdev_ops = {
1172         .ndo_uninit             = tun_net_uninit,
1173         .ndo_open               = tun_net_open,
1174         .ndo_stop               = tun_net_close,
1175         .ndo_start_xmit         = tun_net_xmit,
1176         .ndo_fix_features       = tun_net_fix_features,
1177         .ndo_select_queue       = tun_select_queue,
1178         .ndo_set_rx_headroom    = tun_set_headroom,
1179         .ndo_get_stats64        = tun_net_get_stats64,
1180         .ndo_change_carrier     = tun_net_change_carrier,
1181 };
1182
1183 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1184 {
1185         /* Notify and wake up reader process */
1186         if (tfile->flags & TUN_FASYNC)
1187                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1188         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1189 }
1190
1191 static int tun_xdp_xmit(struct net_device *dev, int n,
1192                         struct xdp_frame **frames, u32 flags)
1193 {
1194         struct tun_struct *tun = netdev_priv(dev);
1195         struct tun_file *tfile;
1196         u32 numqueues;
1197         int nxmit = 0;
1198         int i;
1199
1200         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1201                 return -EINVAL;
1202
1203         rcu_read_lock();
1204
1205 resample:
1206         numqueues = READ_ONCE(tun->numqueues);
1207         if (!numqueues) {
1208                 rcu_read_unlock();
1209                 return -ENXIO; /* Caller will free/return all frames */
1210         }
1211
1212         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1213                                             numqueues]);
1214         if (unlikely(!tfile))
1215                 goto resample;
1216
1217         spin_lock(&tfile->tx_ring.producer_lock);
1218         for (i = 0; i < n; i++) {
1219                 struct xdp_frame *xdp = frames[i];
1220                 /* Encode the XDP flag into lowest bit for consumer to differ
1221                  * XDP buffer from sk_buff.
1222                  */
1223                 void *frame = tun_xdp_to_ptr(xdp);
1224
1225                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1226                         atomic_long_inc(&dev->tx_dropped);
1227                         break;
1228                 }
1229                 nxmit++;
1230         }
1231         spin_unlock(&tfile->tx_ring.producer_lock);
1232
1233         if (flags & XDP_XMIT_FLUSH)
1234                 __tun_xdp_flush_tfile(tfile);
1235
1236         rcu_read_unlock();
1237         return nxmit;
1238 }
1239
1240 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1241 {
1242         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1243         int nxmit;
1244
1245         if (unlikely(!frame))
1246                 return -EOVERFLOW;
1247
1248         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1249         if (!nxmit)
1250                 xdp_return_frame_rx_napi(frame);
1251         return nxmit;
1252 }
1253
1254 static const struct net_device_ops tap_netdev_ops = {
1255         .ndo_uninit             = tun_net_uninit,
1256         .ndo_open               = tun_net_open,
1257         .ndo_stop               = tun_net_close,
1258         .ndo_start_xmit         = tun_net_xmit,
1259         .ndo_fix_features       = tun_net_fix_features,
1260         .ndo_set_rx_mode        = tun_net_mclist,
1261         .ndo_set_mac_address    = eth_mac_addr,
1262         .ndo_validate_addr      = eth_validate_addr,
1263         .ndo_select_queue       = tun_select_queue,
1264         .ndo_features_check     = passthru_features_check,
1265         .ndo_set_rx_headroom    = tun_set_headroom,
1266         .ndo_get_stats64        = dev_get_tstats64,
1267         .ndo_bpf                = tun_xdp,
1268         .ndo_xdp_xmit           = tun_xdp_xmit,
1269         .ndo_change_carrier     = tun_net_change_carrier,
1270 };
1271
1272 static void tun_flow_init(struct tun_struct *tun)
1273 {
1274         int i;
1275
1276         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1277                 INIT_HLIST_HEAD(&tun->flows[i]);
1278
1279         tun->ageing_time = TUN_FLOW_EXPIRE;
1280         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1281         mod_timer(&tun->flow_gc_timer,
1282                   round_jiffies_up(jiffies + tun->ageing_time));
1283 }
1284
1285 static void tun_flow_uninit(struct tun_struct *tun)
1286 {
1287         del_timer_sync(&tun->flow_gc_timer);
1288         tun_flow_flush(tun);
1289 }
1290
1291 #define MIN_MTU 68
1292 #define MAX_MTU 65535
1293
1294 /* Initialize net device. */
1295 static void tun_net_init(struct net_device *dev)
1296 {
1297         struct tun_struct *tun = netdev_priv(dev);
1298
1299         switch (tun->flags & TUN_TYPE_MASK) {
1300         case IFF_TUN:
1301                 dev->netdev_ops = &tun_netdev_ops;
1302                 dev->header_ops = &ip_tunnel_header_ops;
1303
1304                 /* Point-to-Point TUN Device */
1305                 dev->hard_header_len = 0;
1306                 dev->addr_len = 0;
1307                 dev->mtu = 1500;
1308
1309                 /* Zero header length */
1310                 dev->type = ARPHRD_NONE;
1311                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1312                 break;
1313
1314         case IFF_TAP:
1315                 dev->netdev_ops = &tap_netdev_ops;
1316                 /* Ethernet TAP Device */
1317                 ether_setup(dev);
1318                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1319                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1320
1321                 eth_hw_addr_random(dev);
1322
1323                 break;
1324         }
1325
1326         dev->min_mtu = MIN_MTU;
1327         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1328 }
1329
1330 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1331 {
1332         struct sock *sk = tfile->socket.sk;
1333
1334         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1335 }
1336
1337 /* Character device part */
1338
1339 /* Poll */
1340 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1341 {
1342         struct tun_file *tfile = file->private_data;
1343         struct tun_struct *tun = tun_get(tfile);
1344         struct sock *sk;
1345         __poll_t mask = 0;
1346
1347         if (!tun)
1348                 return EPOLLERR;
1349
1350         sk = tfile->socket.sk;
1351
1352         poll_wait(file, sk_sleep(sk), wait);
1353
1354         if (!ptr_ring_empty(&tfile->tx_ring))
1355                 mask |= EPOLLIN | EPOLLRDNORM;
1356
1357         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1358          * guarantee EPOLLOUT to be raised by either here or
1359          * tun_sock_write_space(). Then process could get notification
1360          * after it writes to a down device and meets -EIO.
1361          */
1362         if (tun_sock_writeable(tun, tfile) ||
1363             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1364              tun_sock_writeable(tun, tfile)))
1365                 mask |= EPOLLOUT | EPOLLWRNORM;
1366
1367         if (tun->dev->reg_state != NETREG_REGISTERED)
1368                 mask = EPOLLERR;
1369
1370         tun_put(tun);
1371         return mask;
1372 }
1373
1374 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1375                                             size_t len,
1376                                             const struct iov_iter *it)
1377 {
1378         struct sk_buff *skb;
1379         size_t linear;
1380         int err;
1381         int i;
1382
1383         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1384                 return ERR_PTR(-EMSGSIZE);
1385
1386         local_bh_disable();
1387         skb = napi_get_frags(&tfile->napi);
1388         local_bh_enable();
1389         if (!skb)
1390                 return ERR_PTR(-ENOMEM);
1391
1392         linear = iov_iter_single_seg_count(it);
1393         err = __skb_grow(skb, linear);
1394         if (err)
1395                 goto free;
1396
1397         skb->len = len;
1398         skb->data_len = len - linear;
1399         skb->truesize += skb->data_len;
1400
1401         for (i = 1; i < it->nr_segs; i++) {
1402                 size_t fragsz = it->iov[i].iov_len;
1403                 struct page *page;
1404                 void *frag;
1405
1406                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1407                         err = -EINVAL;
1408                         goto free;
1409                 }
1410                 frag = netdev_alloc_frag(fragsz);
1411                 if (!frag) {
1412                         err = -ENOMEM;
1413                         goto free;
1414                 }
1415                 page = virt_to_head_page(frag);
1416                 skb_fill_page_desc(skb, i - 1, page,
1417                                    frag - page_address(page), fragsz);
1418         }
1419
1420         return skb;
1421 free:
1422         /* frees skb and all frags allocated with napi_alloc_frag() */
1423         napi_free_frags(&tfile->napi);
1424         return ERR_PTR(err);
1425 }
1426
1427 /* prepad is the amount to reserve at front.  len is length after that.
1428  * linear is a hint as to how much to copy (usually headers). */
1429 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1430                                      size_t prepad, size_t len,
1431                                      size_t linear, int noblock)
1432 {
1433         struct sock *sk = tfile->socket.sk;
1434         struct sk_buff *skb;
1435         int err;
1436
1437         /* Under a page?  Don't bother with paged skb. */
1438         if (prepad + len < PAGE_SIZE || !linear)
1439                 linear = len;
1440
1441         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1442                                    &err, 0);
1443         if (!skb)
1444                 return ERR_PTR(err);
1445
1446         skb_reserve(skb, prepad);
1447         skb_put(skb, linear);
1448         skb->data_len = len - linear;
1449         skb->len += len - linear;
1450
1451         return skb;
1452 }
1453
1454 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1455                            struct sk_buff *skb, int more)
1456 {
1457         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1458         struct sk_buff_head process_queue;
1459         u32 rx_batched = tun->rx_batched;
1460         bool rcv = false;
1461
1462         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1463                 local_bh_disable();
1464                 skb_record_rx_queue(skb, tfile->queue_index);
1465                 netif_receive_skb(skb);
1466                 local_bh_enable();
1467                 return;
1468         }
1469
1470         spin_lock(&queue->lock);
1471         if (!more || skb_queue_len(queue) == rx_batched) {
1472                 __skb_queue_head_init(&process_queue);
1473                 skb_queue_splice_tail_init(queue, &process_queue);
1474                 rcv = true;
1475         } else {
1476                 __skb_queue_tail(queue, skb);
1477         }
1478         spin_unlock(&queue->lock);
1479
1480         if (rcv) {
1481                 struct sk_buff *nskb;
1482
1483                 local_bh_disable();
1484                 while ((nskb = __skb_dequeue(&process_queue))) {
1485                         skb_record_rx_queue(nskb, tfile->queue_index);
1486                         netif_receive_skb(nskb);
1487                 }
1488                 skb_record_rx_queue(skb, tfile->queue_index);
1489                 netif_receive_skb(skb);
1490                 local_bh_enable();
1491         }
1492 }
1493
1494 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1495                               int len, int noblock, bool zerocopy)
1496 {
1497         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1498                 return false;
1499
1500         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1501                 return false;
1502
1503         if (!noblock)
1504                 return false;
1505
1506         if (zerocopy)
1507                 return false;
1508
1509         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1510             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1511                 return false;
1512
1513         return true;
1514 }
1515
1516 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1517                                        struct page_frag *alloc_frag, char *buf,
1518                                        int buflen, int len, int pad)
1519 {
1520         struct sk_buff *skb = build_skb(buf, buflen);
1521
1522         if (!skb)
1523                 return ERR_PTR(-ENOMEM);
1524
1525         skb_reserve(skb, pad);
1526         skb_put(skb, len);
1527         skb_set_owner_w(skb, tfile->socket.sk);
1528
1529         get_page(alloc_frag->page);
1530         alloc_frag->offset += buflen;
1531
1532         return skb;
1533 }
1534
1535 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1536                        struct xdp_buff *xdp, u32 act)
1537 {
1538         int err;
1539
1540         switch (act) {
1541         case XDP_REDIRECT:
1542                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1543                 if (err)
1544                         return err;
1545                 break;
1546         case XDP_TX:
1547                 err = tun_xdp_tx(tun->dev, xdp);
1548                 if (err < 0)
1549                         return err;
1550                 break;
1551         case XDP_PASS:
1552                 break;
1553         default:
1554                 bpf_warn_invalid_xdp_action(act);
1555                 fallthrough;
1556         case XDP_ABORTED:
1557                 trace_xdp_exception(tun->dev, xdp_prog, act);
1558                 fallthrough;
1559         case XDP_DROP:
1560                 atomic_long_inc(&tun->dev->rx_dropped);
1561                 break;
1562         }
1563
1564         return act;
1565 }
1566
1567 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1568                                      struct tun_file *tfile,
1569                                      struct iov_iter *from,
1570                                      struct virtio_net_hdr *hdr,
1571                                      int len, int *skb_xdp)
1572 {
1573         struct page_frag *alloc_frag = &current->task_frag;
1574         struct bpf_prog *xdp_prog;
1575         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1576         char *buf;
1577         size_t copied;
1578         int pad = TUN_RX_PAD;
1579         int err = 0;
1580
1581         rcu_read_lock();
1582         xdp_prog = rcu_dereference(tun->xdp_prog);
1583         if (xdp_prog)
1584                 pad += XDP_PACKET_HEADROOM;
1585         buflen += SKB_DATA_ALIGN(len + pad);
1586         rcu_read_unlock();
1587
1588         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1589         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1590                 return ERR_PTR(-ENOMEM);
1591
1592         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1593         copied = copy_page_from_iter(alloc_frag->page,
1594                                      alloc_frag->offset + pad,
1595                                      len, from);
1596         if (copied != len)
1597                 return ERR_PTR(-EFAULT);
1598
1599         /* There's a small window that XDP may be set after the check
1600          * of xdp_prog above, this should be rare and for simplicity
1601          * we do XDP on skb in case the headroom is not enough.
1602          */
1603         if (hdr->gso_type || !xdp_prog) {
1604                 *skb_xdp = 1;
1605                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1606                                        pad);
1607         }
1608
1609         *skb_xdp = 0;
1610
1611         local_bh_disable();
1612         rcu_read_lock();
1613         xdp_prog = rcu_dereference(tun->xdp_prog);
1614         if (xdp_prog) {
1615                 struct xdp_buff xdp;
1616                 u32 act;
1617
1618                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1619                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1620
1621                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1622                 if (act == XDP_REDIRECT || act == XDP_TX) {
1623                         get_page(alloc_frag->page);
1624                         alloc_frag->offset += buflen;
1625                 }
1626                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1627                 if (err < 0) {
1628                         if (act == XDP_REDIRECT || act == XDP_TX)
1629                                 put_page(alloc_frag->page);
1630                         goto out;
1631                 }
1632
1633                 if (err == XDP_REDIRECT)
1634                         xdp_do_flush();
1635                 if (err != XDP_PASS)
1636                         goto out;
1637
1638                 pad = xdp.data - xdp.data_hard_start;
1639                 len = xdp.data_end - xdp.data;
1640         }
1641         rcu_read_unlock();
1642         local_bh_enable();
1643
1644         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1645
1646 out:
1647         rcu_read_unlock();
1648         local_bh_enable();
1649         return NULL;
1650 }
1651
1652 /* Get packet from user space buffer */
1653 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1654                             void *msg_control, struct iov_iter *from,
1655                             int noblock, bool more)
1656 {
1657         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1658         struct sk_buff *skb;
1659         size_t total_len = iov_iter_count(from);
1660         size_t len = total_len, align = tun->align, linear;
1661         struct virtio_net_hdr gso = { 0 };
1662         int good_linear;
1663         int copylen;
1664         bool zerocopy = false;
1665         int err;
1666         u32 rxhash = 0;
1667         int skb_xdp = 1;
1668         bool frags = tun_napi_frags_enabled(tfile);
1669
1670         if (!(tun->flags & IFF_NO_PI)) {
1671                 if (len < sizeof(pi))
1672                         return -EINVAL;
1673                 len -= sizeof(pi);
1674
1675                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1676                         return -EFAULT;
1677         }
1678
1679         if (tun->flags & IFF_VNET_HDR) {
1680                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1681
1682                 if (len < vnet_hdr_sz)
1683                         return -EINVAL;
1684                 len -= vnet_hdr_sz;
1685
1686                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1687                         return -EFAULT;
1688
1689                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1690                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1691                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1692
1693                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1694                         return -EINVAL;
1695                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1696         }
1697
1698         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1699                 align += NET_IP_ALIGN;
1700                 if (unlikely(len < ETH_HLEN ||
1701                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1702                         return -EINVAL;
1703         }
1704
1705         good_linear = SKB_MAX_HEAD(align);
1706
1707         if (msg_control) {
1708                 struct iov_iter i = *from;
1709
1710                 /* There are 256 bytes to be copied in skb, so there is
1711                  * enough room for skb expand head in case it is used.
1712                  * The rest of the buffer is mapped from userspace.
1713                  */
1714                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1715                 if (copylen > good_linear)
1716                         copylen = good_linear;
1717                 linear = copylen;
1718                 iov_iter_advance(&i, copylen);
1719                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1720                         zerocopy = true;
1721         }
1722
1723         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1724                 /* For the packet that is not easy to be processed
1725                  * (e.g gso or jumbo packet), we will do it at after
1726                  * skb was created with generic XDP routine.
1727                  */
1728                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1729                 if (IS_ERR(skb)) {
1730                         atomic_long_inc(&tun->dev->rx_dropped);
1731                         return PTR_ERR(skb);
1732                 }
1733                 if (!skb)
1734                         return total_len;
1735         } else {
1736                 if (!zerocopy) {
1737                         copylen = len;
1738                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1739                                 linear = good_linear;
1740                         else
1741                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1742                 }
1743
1744                 if (frags) {
1745                         mutex_lock(&tfile->napi_mutex);
1746                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1747                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1748                          * If zerocopy is enabled, then this layout will be
1749                          * overwritten by zerocopy_sg_from_iter().
1750                          */
1751                         zerocopy = false;
1752                 } else {
1753                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1754                                             noblock);
1755                 }
1756
1757                 if (IS_ERR(skb)) {
1758                         if (PTR_ERR(skb) != -EAGAIN)
1759                                 atomic_long_inc(&tun->dev->rx_dropped);
1760                         if (frags)
1761                                 mutex_unlock(&tfile->napi_mutex);
1762                         return PTR_ERR(skb);
1763                 }
1764
1765                 if (zerocopy)
1766                         err = zerocopy_sg_from_iter(skb, from);
1767                 else
1768                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1769
1770                 if (err) {
1771                         err = -EFAULT;
1772 drop:
1773                         atomic_long_inc(&tun->dev->rx_dropped);
1774                         kfree_skb(skb);
1775                         if (frags) {
1776                                 tfile->napi.skb = NULL;
1777                                 mutex_unlock(&tfile->napi_mutex);
1778                         }
1779
1780                         return err;
1781                 }
1782         }
1783
1784         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1785                 atomic_long_inc(&tun->rx_frame_errors);
1786                 kfree_skb(skb);
1787                 if (frags) {
1788                         tfile->napi.skb = NULL;
1789                         mutex_unlock(&tfile->napi_mutex);
1790                 }
1791
1792                 return -EINVAL;
1793         }
1794
1795         switch (tun->flags & TUN_TYPE_MASK) {
1796         case IFF_TUN:
1797                 if (tun->flags & IFF_NO_PI) {
1798                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1799
1800                         switch (ip_version) {
1801                         case 4:
1802                                 pi.proto = htons(ETH_P_IP);
1803                                 break;
1804                         case 6:
1805                                 pi.proto = htons(ETH_P_IPV6);
1806                                 break;
1807                         default:
1808                                 atomic_long_inc(&tun->dev->rx_dropped);
1809                                 kfree_skb(skb);
1810                                 return -EINVAL;
1811                         }
1812                 }
1813
1814                 skb_reset_mac_header(skb);
1815                 skb->protocol = pi.proto;
1816                 skb->dev = tun->dev;
1817                 break;
1818         case IFF_TAP:
1819                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1820                         err = -ENOMEM;
1821                         goto drop;
1822                 }
1823                 skb->protocol = eth_type_trans(skb, tun->dev);
1824                 break;
1825         }
1826
1827         /* copy skb_ubuf_info for callback when skb has no error */
1828         if (zerocopy) {
1829                 skb_zcopy_init(skb, msg_control);
1830         } else if (msg_control) {
1831                 struct ubuf_info *uarg = msg_control;
1832                 uarg->callback(NULL, uarg, false);
1833         }
1834
1835         skb_reset_network_header(skb);
1836         skb_probe_transport_header(skb);
1837         skb_record_rx_queue(skb, tfile->queue_index);
1838
1839         if (skb_xdp) {
1840                 struct bpf_prog *xdp_prog;
1841                 int ret;
1842
1843                 local_bh_disable();
1844                 rcu_read_lock();
1845                 xdp_prog = rcu_dereference(tun->xdp_prog);
1846                 if (xdp_prog) {
1847                         ret = do_xdp_generic(xdp_prog, skb);
1848                         if (ret != XDP_PASS) {
1849                                 rcu_read_unlock();
1850                                 local_bh_enable();
1851                                 if (frags) {
1852                                         tfile->napi.skb = NULL;
1853                                         mutex_unlock(&tfile->napi_mutex);
1854                                 }
1855                                 return total_len;
1856                         }
1857                 }
1858                 rcu_read_unlock();
1859                 local_bh_enable();
1860         }
1861
1862         /* Compute the costly rx hash only if needed for flow updates.
1863          * We may get a very small possibility of OOO during switching, not
1864          * worth to optimize.
1865          */
1866         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1867             !tfile->detached)
1868                 rxhash = __skb_get_hash_symmetric(skb);
1869
1870         rcu_read_lock();
1871         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1872                 err = -EIO;
1873                 rcu_read_unlock();
1874                 goto drop;
1875         }
1876
1877         if (frags) {
1878                 u32 headlen;
1879
1880                 /* Exercise flow dissector code path. */
1881                 skb_push(skb, ETH_HLEN);
1882                 headlen = eth_get_headlen(tun->dev, skb->data,
1883                                           skb_headlen(skb));
1884
1885                 if (unlikely(headlen > skb_headlen(skb))) {
1886                         atomic_long_inc(&tun->dev->rx_dropped);
1887                         napi_free_frags(&tfile->napi);
1888                         rcu_read_unlock();
1889                         mutex_unlock(&tfile->napi_mutex);
1890                         WARN_ON(1);
1891                         return -ENOMEM;
1892                 }
1893
1894                 local_bh_disable();
1895                 napi_gro_frags(&tfile->napi);
1896                 local_bh_enable();
1897                 mutex_unlock(&tfile->napi_mutex);
1898         } else if (tfile->napi_enabled) {
1899                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1900                 int queue_len;
1901
1902                 spin_lock_bh(&queue->lock);
1903                 __skb_queue_tail(queue, skb);
1904                 queue_len = skb_queue_len(queue);
1905                 spin_unlock(&queue->lock);
1906
1907                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1908                         napi_schedule(&tfile->napi);
1909
1910                 local_bh_enable();
1911         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1912                 tun_rx_batched(tun, tfile, skb, more);
1913         } else {
1914                 netif_rx_ni(skb);
1915         }
1916         rcu_read_unlock();
1917
1918         preempt_disable();
1919         dev_sw_netstats_rx_add(tun->dev, len);
1920         preempt_enable();
1921
1922         if (rxhash)
1923                 tun_flow_update(tun, rxhash, tfile);
1924
1925         return total_len;
1926 }
1927
1928 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1929 {
1930         struct file *file = iocb->ki_filp;
1931         struct tun_file *tfile = file->private_data;
1932         struct tun_struct *tun = tun_get(tfile);
1933         ssize_t result;
1934         int noblock = 0;
1935
1936         if (!tun)
1937                 return -EBADFD;
1938
1939         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1940                 noblock = 1;
1941
1942         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1943
1944         tun_put(tun);
1945         return result;
1946 }
1947
1948 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1949                                 struct tun_file *tfile,
1950                                 struct xdp_frame *xdp_frame,
1951                                 struct iov_iter *iter)
1952 {
1953         int vnet_hdr_sz = 0;
1954         size_t size = xdp_frame->len;
1955         size_t ret;
1956
1957         if (tun->flags & IFF_VNET_HDR) {
1958                 struct virtio_net_hdr gso = { 0 };
1959
1960                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1961                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1962                         return -EINVAL;
1963                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1964                              sizeof(gso)))
1965                         return -EFAULT;
1966                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1967         }
1968
1969         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1970
1971         preempt_disable();
1972         dev_sw_netstats_tx_add(tun->dev, 1, ret);
1973         preempt_enable();
1974
1975         return ret;
1976 }
1977
1978 /* Put packet to the user space buffer */
1979 static ssize_t tun_put_user(struct tun_struct *tun,
1980                             struct tun_file *tfile,
1981                             struct sk_buff *skb,
1982                             struct iov_iter *iter)
1983 {
1984         struct tun_pi pi = { 0, skb->protocol };
1985         ssize_t total;
1986         int vlan_offset = 0;
1987         int vlan_hlen = 0;
1988         int vnet_hdr_sz = 0;
1989
1990         if (skb_vlan_tag_present(skb))
1991                 vlan_hlen = VLAN_HLEN;
1992
1993         if (tun->flags & IFF_VNET_HDR)
1994                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1995
1996         total = skb->len + vlan_hlen + vnet_hdr_sz;
1997
1998         if (!(tun->flags & IFF_NO_PI)) {
1999                 if (iov_iter_count(iter) < sizeof(pi))
2000                         return -EINVAL;
2001
2002                 total += sizeof(pi);
2003                 if (iov_iter_count(iter) < total) {
2004                         /* Packet will be striped */
2005                         pi.flags |= TUN_PKT_STRIP;
2006                 }
2007
2008                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2009                         return -EFAULT;
2010         }
2011
2012         if (vnet_hdr_sz) {
2013                 struct virtio_net_hdr gso;
2014
2015                 if (iov_iter_count(iter) < vnet_hdr_sz)
2016                         return -EINVAL;
2017
2018                 if (virtio_net_hdr_from_skb(skb, &gso,
2019                                             tun_is_little_endian(tun), true,
2020                                             vlan_hlen)) {
2021                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2022                         pr_err("unexpected GSO type: "
2023                                "0x%x, gso_size %d, hdr_len %d\n",
2024                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2025                                tun16_to_cpu(tun, gso.hdr_len));
2026                         print_hex_dump(KERN_ERR, "tun: ",
2027                                        DUMP_PREFIX_NONE,
2028                                        16, 1, skb->head,
2029                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2030                         WARN_ON_ONCE(1);
2031                         return -EINVAL;
2032                 }
2033
2034                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2035                         return -EFAULT;
2036
2037                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2038         }
2039
2040         if (vlan_hlen) {
2041                 int ret;
2042                 struct veth veth;
2043
2044                 veth.h_vlan_proto = skb->vlan_proto;
2045                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2046
2047                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2048
2049                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2050                 if (ret || !iov_iter_count(iter))
2051                         goto done;
2052
2053                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2054                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2055                         goto done;
2056         }
2057
2058         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2059
2060 done:
2061         /* caller is in process context, */
2062         preempt_disable();
2063         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2064         preempt_enable();
2065
2066         return total;
2067 }
2068
2069 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2070 {
2071         DECLARE_WAITQUEUE(wait, current);
2072         void *ptr = NULL;
2073         int error = 0;
2074
2075         ptr = ptr_ring_consume(&tfile->tx_ring);
2076         if (ptr)
2077                 goto out;
2078         if (noblock) {
2079                 error = -EAGAIN;
2080                 goto out;
2081         }
2082
2083         add_wait_queue(&tfile->socket.wq.wait, &wait);
2084
2085         while (1) {
2086                 set_current_state(TASK_INTERRUPTIBLE);
2087                 ptr = ptr_ring_consume(&tfile->tx_ring);
2088                 if (ptr)
2089                         break;
2090                 if (signal_pending(current)) {
2091                         error = -ERESTARTSYS;
2092                         break;
2093                 }
2094                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2095                         error = -EFAULT;
2096                         break;
2097                 }
2098
2099                 schedule();
2100         }
2101
2102         __set_current_state(TASK_RUNNING);
2103         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2104
2105 out:
2106         *err = error;
2107         return ptr;
2108 }
2109
2110 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2111                            struct iov_iter *to,
2112                            int noblock, void *ptr)
2113 {
2114         ssize_t ret;
2115         int err;
2116
2117         if (!iov_iter_count(to)) {
2118                 tun_ptr_free(ptr);
2119                 return 0;
2120         }
2121
2122         if (!ptr) {
2123                 /* Read frames from ring */
2124                 ptr = tun_ring_recv(tfile, noblock, &err);
2125                 if (!ptr)
2126                         return err;
2127         }
2128
2129         if (tun_is_xdp_frame(ptr)) {
2130                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2131
2132                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2133                 xdp_return_frame(xdpf);
2134         } else {
2135                 struct sk_buff *skb = ptr;
2136
2137                 ret = tun_put_user(tun, tfile, skb, to);
2138                 if (unlikely(ret < 0))
2139                         kfree_skb(skb);
2140                 else
2141                         consume_skb(skb);
2142         }
2143
2144         return ret;
2145 }
2146
2147 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2148 {
2149         struct file *file = iocb->ki_filp;
2150         struct tun_file *tfile = file->private_data;
2151         struct tun_struct *tun = tun_get(tfile);
2152         ssize_t len = iov_iter_count(to), ret;
2153         int noblock = 0;
2154
2155         if (!tun)
2156                 return -EBADFD;
2157
2158         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2159                 noblock = 1;
2160
2161         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2162         ret = min_t(ssize_t, ret, len);
2163         if (ret > 0)
2164                 iocb->ki_pos = ret;
2165         tun_put(tun);
2166         return ret;
2167 }
2168
2169 static void tun_prog_free(struct rcu_head *rcu)
2170 {
2171         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2172
2173         bpf_prog_destroy(prog->prog);
2174         kfree(prog);
2175 }
2176
2177 static int __tun_set_ebpf(struct tun_struct *tun,
2178                           struct tun_prog __rcu **prog_p,
2179                           struct bpf_prog *prog)
2180 {
2181         struct tun_prog *old, *new = NULL;
2182
2183         if (prog) {
2184                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2185                 if (!new)
2186                         return -ENOMEM;
2187                 new->prog = prog;
2188         }
2189
2190         spin_lock_bh(&tun->lock);
2191         old = rcu_dereference_protected(*prog_p,
2192                                         lockdep_is_held(&tun->lock));
2193         rcu_assign_pointer(*prog_p, new);
2194         spin_unlock_bh(&tun->lock);
2195
2196         if (old)
2197                 call_rcu(&old->rcu, tun_prog_free);
2198
2199         return 0;
2200 }
2201
2202 static void tun_free_netdev(struct net_device *dev)
2203 {
2204         struct tun_struct *tun = netdev_priv(dev);
2205
2206         BUG_ON(!(list_empty(&tun->disabled)));
2207
2208         free_percpu(dev->tstats);
2209         /* We clear tstats so that tun_set_iff() can tell if
2210          * tun_free_netdev() has been called from register_netdevice().
2211          */
2212         dev->tstats = NULL;
2213
2214         tun_flow_uninit(tun);
2215         security_tun_dev_free_security(tun->security);
2216         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2217         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2218 }
2219
2220 static void tun_setup(struct net_device *dev)
2221 {
2222         struct tun_struct *tun = netdev_priv(dev);
2223
2224         tun->owner = INVALID_UID;
2225         tun->group = INVALID_GID;
2226         tun_default_link_ksettings(dev, &tun->link_ksettings);
2227
2228         dev->ethtool_ops = &tun_ethtool_ops;
2229         dev->needs_free_netdev = true;
2230         dev->priv_destructor = tun_free_netdev;
2231         /* We prefer our own queue length */
2232         dev->tx_queue_len = TUN_READQ_SIZE;
2233 }
2234
2235 /* Trivial set of netlink ops to allow deleting tun or tap
2236  * device with netlink.
2237  */
2238 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2239                         struct netlink_ext_ack *extack)
2240 {
2241         NL_SET_ERR_MSG(extack,
2242                        "tun/tap creation via rtnetlink is not supported.");
2243         return -EOPNOTSUPP;
2244 }
2245
2246 static size_t tun_get_size(const struct net_device *dev)
2247 {
2248         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2249         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2250
2251         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2252                nla_total_size(sizeof(gid_t)) + /* GROUP */
2253                nla_total_size(sizeof(u8)) + /* TYPE */
2254                nla_total_size(sizeof(u8)) + /* PI */
2255                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2256                nla_total_size(sizeof(u8)) + /* PERSIST */
2257                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2258                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2259                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2260                0;
2261 }
2262
2263 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2264 {
2265         struct tun_struct *tun = netdev_priv(dev);
2266
2267         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2268                 goto nla_put_failure;
2269         if (uid_valid(tun->owner) &&
2270             nla_put_u32(skb, IFLA_TUN_OWNER,
2271                         from_kuid_munged(current_user_ns(), tun->owner)))
2272                 goto nla_put_failure;
2273         if (gid_valid(tun->group) &&
2274             nla_put_u32(skb, IFLA_TUN_GROUP,
2275                         from_kgid_munged(current_user_ns(), tun->group)))
2276                 goto nla_put_failure;
2277         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2278                 goto nla_put_failure;
2279         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2280                 goto nla_put_failure;
2281         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2282                 goto nla_put_failure;
2283         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2284                        !!(tun->flags & IFF_MULTI_QUEUE)))
2285                 goto nla_put_failure;
2286         if (tun->flags & IFF_MULTI_QUEUE) {
2287                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2288                         goto nla_put_failure;
2289                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2290                                 tun->numdisabled))
2291                         goto nla_put_failure;
2292         }
2293
2294         return 0;
2295
2296 nla_put_failure:
2297         return -EMSGSIZE;
2298 }
2299
2300 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2301         .kind           = DRV_NAME,
2302         .priv_size      = sizeof(struct tun_struct),
2303         .setup          = tun_setup,
2304         .validate       = tun_validate,
2305         .get_size       = tun_get_size,
2306         .fill_info      = tun_fill_info,
2307 };
2308
2309 static void tun_sock_write_space(struct sock *sk)
2310 {
2311         struct tun_file *tfile;
2312         wait_queue_head_t *wqueue;
2313
2314         if (!sock_writeable(sk))
2315                 return;
2316
2317         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2318                 return;
2319
2320         wqueue = sk_sleep(sk);
2321         if (wqueue && waitqueue_active(wqueue))
2322                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2323                                                 EPOLLWRNORM | EPOLLWRBAND);
2324
2325         tfile = container_of(sk, struct tun_file, sk);
2326         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2327 }
2328
2329 static void tun_put_page(struct tun_page *tpage)
2330 {
2331         if (tpage->page)
2332                 __page_frag_cache_drain(tpage->page, tpage->count);
2333 }
2334
2335 static int tun_xdp_one(struct tun_struct *tun,
2336                        struct tun_file *tfile,
2337                        struct xdp_buff *xdp, int *flush,
2338                        struct tun_page *tpage)
2339 {
2340         unsigned int datasize = xdp->data_end - xdp->data;
2341         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2342         struct virtio_net_hdr *gso = &hdr->gso;
2343         struct bpf_prog *xdp_prog;
2344         struct sk_buff *skb = NULL;
2345         u32 rxhash = 0, act;
2346         int buflen = hdr->buflen;
2347         int err = 0;
2348         bool skb_xdp = false;
2349         struct page *page;
2350
2351         xdp_prog = rcu_dereference(tun->xdp_prog);
2352         if (xdp_prog) {
2353                 if (gso->gso_type) {
2354                         skb_xdp = true;
2355                         goto build;
2356                 }
2357
2358                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2359                 xdp_set_data_meta_invalid(xdp);
2360
2361                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2362                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2363                 if (err < 0) {
2364                         put_page(virt_to_head_page(xdp->data));
2365                         return err;
2366                 }
2367
2368                 switch (err) {
2369                 case XDP_REDIRECT:
2370                         *flush = true;
2371                         fallthrough;
2372                 case XDP_TX:
2373                         return 0;
2374                 case XDP_PASS:
2375                         break;
2376                 default:
2377                         page = virt_to_head_page(xdp->data);
2378                         if (tpage->page == page) {
2379                                 ++tpage->count;
2380                         } else {
2381                                 tun_put_page(tpage);
2382                                 tpage->page = page;
2383                                 tpage->count = 1;
2384                         }
2385                         return 0;
2386                 }
2387         }
2388
2389 build:
2390         skb = build_skb(xdp->data_hard_start, buflen);
2391         if (!skb) {
2392                 err = -ENOMEM;
2393                 goto out;
2394         }
2395
2396         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2397         skb_put(skb, xdp->data_end - xdp->data);
2398
2399         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2400                 atomic_long_inc(&tun->rx_frame_errors);
2401                 kfree_skb(skb);
2402                 err = -EINVAL;
2403                 goto out;
2404         }
2405
2406         skb->protocol = eth_type_trans(skb, tun->dev);
2407         skb_reset_network_header(skb);
2408         skb_probe_transport_header(skb);
2409         skb_record_rx_queue(skb, tfile->queue_index);
2410
2411         if (skb_xdp) {
2412                 err = do_xdp_generic(xdp_prog, skb);
2413                 if (err != XDP_PASS)
2414                         goto out;
2415         }
2416
2417         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2418             !tfile->detached)
2419                 rxhash = __skb_get_hash_symmetric(skb);
2420
2421         netif_receive_skb(skb);
2422
2423         /* No need to disable preemption here since this function is
2424          * always called with bh disabled
2425          */
2426         dev_sw_netstats_rx_add(tun->dev, datasize);
2427
2428         if (rxhash)
2429                 tun_flow_update(tun, rxhash, tfile);
2430
2431 out:
2432         return err;
2433 }
2434
2435 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2436 {
2437         int ret, i;
2438         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2439         struct tun_struct *tun = tun_get(tfile);
2440         struct tun_msg_ctl *ctl = m->msg_control;
2441         struct xdp_buff *xdp;
2442
2443         if (!tun)
2444                 return -EBADFD;
2445
2446         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2447                 struct tun_page tpage;
2448                 int n = ctl->num;
2449                 int flush = 0;
2450
2451                 memset(&tpage, 0, sizeof(tpage));
2452
2453                 local_bh_disable();
2454                 rcu_read_lock();
2455
2456                 for (i = 0; i < n; i++) {
2457                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2458                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2459                 }
2460
2461                 if (flush)
2462                         xdp_do_flush();
2463
2464                 rcu_read_unlock();
2465                 local_bh_enable();
2466
2467                 tun_put_page(&tpage);
2468
2469                 ret = total_len;
2470                 goto out;
2471         }
2472
2473         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2474                            m->msg_flags & MSG_DONTWAIT,
2475                            m->msg_flags & MSG_MORE);
2476 out:
2477         tun_put(tun);
2478         return ret;
2479 }
2480
2481 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2482                        int flags)
2483 {
2484         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2485         struct tun_struct *tun = tun_get(tfile);
2486         void *ptr = m->msg_control;
2487         int ret;
2488
2489         if (!tun) {
2490                 ret = -EBADFD;
2491                 goto out_free;
2492         }
2493
2494         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2495                 ret = -EINVAL;
2496                 goto out_put_tun;
2497         }
2498         if (flags & MSG_ERRQUEUE) {
2499                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2500                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2501                 goto out;
2502         }
2503         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2504         if (ret > (ssize_t)total_len) {
2505                 m->msg_flags |= MSG_TRUNC;
2506                 ret = flags & MSG_TRUNC ? ret : total_len;
2507         }
2508 out:
2509         tun_put(tun);
2510         return ret;
2511
2512 out_put_tun:
2513         tun_put(tun);
2514 out_free:
2515         tun_ptr_free(ptr);
2516         return ret;
2517 }
2518
2519 static int tun_ptr_peek_len(void *ptr)
2520 {
2521         if (likely(ptr)) {
2522                 if (tun_is_xdp_frame(ptr)) {
2523                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2524
2525                         return xdpf->len;
2526                 }
2527                 return __skb_array_len_with_tag(ptr);
2528         } else {
2529                 return 0;
2530         }
2531 }
2532
2533 static int tun_peek_len(struct socket *sock)
2534 {
2535         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2536         struct tun_struct *tun;
2537         int ret = 0;
2538
2539         tun = tun_get(tfile);
2540         if (!tun)
2541                 return 0;
2542
2543         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2544         tun_put(tun);
2545
2546         return ret;
2547 }
2548
2549 /* Ops structure to mimic raw sockets with tun */
2550 static const struct proto_ops tun_socket_ops = {
2551         .peek_len = tun_peek_len,
2552         .sendmsg = tun_sendmsg,
2553         .recvmsg = tun_recvmsg,
2554 };
2555
2556 static struct proto tun_proto = {
2557         .name           = "tun",
2558         .owner          = THIS_MODULE,
2559         .obj_size       = sizeof(struct tun_file),
2560 };
2561
2562 static int tun_flags(struct tun_struct *tun)
2563 {
2564         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2565 }
2566
2567 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2568                               char *buf)
2569 {
2570         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2571         return sprintf(buf, "0x%x\n", tun_flags(tun));
2572 }
2573
2574 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2575                           char *buf)
2576 {
2577         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2578         return uid_valid(tun->owner)?
2579                 sprintf(buf, "%u\n",
2580                         from_kuid_munged(current_user_ns(), tun->owner)):
2581                 sprintf(buf, "-1\n");
2582 }
2583
2584 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2585                           char *buf)
2586 {
2587         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2588         return gid_valid(tun->group) ?
2589                 sprintf(buf, "%u\n",
2590                         from_kgid_munged(current_user_ns(), tun->group)):
2591                 sprintf(buf, "-1\n");
2592 }
2593
2594 static DEVICE_ATTR_RO(tun_flags);
2595 static DEVICE_ATTR_RO(owner);
2596 static DEVICE_ATTR_RO(group);
2597
2598 static struct attribute *tun_dev_attrs[] = {
2599         &dev_attr_tun_flags.attr,
2600         &dev_attr_owner.attr,
2601         &dev_attr_group.attr,
2602         NULL
2603 };
2604
2605 static const struct attribute_group tun_attr_group = {
2606         .attrs = tun_dev_attrs
2607 };
2608
2609 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2610 {
2611         struct tun_struct *tun;
2612         struct tun_file *tfile = file->private_data;
2613         struct net_device *dev;
2614         int err;
2615
2616         if (tfile->detached)
2617                 return -EINVAL;
2618
2619         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2620                 if (!capable(CAP_NET_ADMIN))
2621                         return -EPERM;
2622
2623                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2624                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2625                         return -EINVAL;
2626         }
2627
2628         dev = __dev_get_by_name(net, ifr->ifr_name);
2629         if (dev) {
2630                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2631                         return -EBUSY;
2632                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2633                         tun = netdev_priv(dev);
2634                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2635                         tun = netdev_priv(dev);
2636                 else
2637                         return -EINVAL;
2638
2639                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2640                     !!(tun->flags & IFF_MULTI_QUEUE))
2641                         return -EINVAL;
2642
2643                 if (tun_not_capable(tun))
2644                         return -EPERM;
2645                 err = security_tun_dev_open(tun->security);
2646                 if (err < 0)
2647                         return err;
2648
2649                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2650                                  ifr->ifr_flags & IFF_NAPI,
2651                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2652                 if (err < 0)
2653                         return err;
2654
2655                 if (tun->flags & IFF_MULTI_QUEUE &&
2656                     (tun->numqueues + tun->numdisabled > 1)) {
2657                         /* One or more queue has already been attached, no need
2658                          * to initialize the device again.
2659                          */
2660                         netdev_state_change(dev);
2661                         return 0;
2662                 }
2663
2664                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2665                               (ifr->ifr_flags & TUN_FEATURES);
2666
2667                 netdev_state_change(dev);
2668         } else {
2669                 char *name;
2670                 unsigned long flags = 0;
2671                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2672                              MAX_TAP_QUEUES : 1;
2673
2674                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2675                         return -EPERM;
2676                 err = security_tun_dev_create();
2677                 if (err < 0)
2678                         return err;
2679
2680                 /* Set dev type */
2681                 if (ifr->ifr_flags & IFF_TUN) {
2682                         /* TUN device */
2683                         flags |= IFF_TUN;
2684                         name = "tun%d";
2685                 } else if (ifr->ifr_flags & IFF_TAP) {
2686                         /* TAP device */
2687                         flags |= IFF_TAP;
2688                         name = "tap%d";
2689                 } else
2690                         return -EINVAL;
2691
2692                 if (*ifr->ifr_name)
2693                         name = ifr->ifr_name;
2694
2695                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2696                                        NET_NAME_UNKNOWN, tun_setup, queues,
2697                                        queues);
2698
2699                 if (!dev)
2700                         return -ENOMEM;
2701
2702                 dev_net_set(dev, net);
2703                 dev->rtnl_link_ops = &tun_link_ops;
2704                 dev->ifindex = tfile->ifindex;
2705                 dev->sysfs_groups[0] = &tun_attr_group;
2706
2707                 tun = netdev_priv(dev);
2708                 tun->dev = dev;
2709                 tun->flags = flags;
2710                 tun->txflt.count = 0;
2711                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2712
2713                 tun->align = NET_SKB_PAD;
2714                 tun->filter_attached = false;
2715                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2716                 tun->rx_batched = 0;
2717                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2718
2719                 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2720                 if (!dev->tstats) {
2721                         err = -ENOMEM;
2722                         goto err_free_dev;
2723                 }
2724
2725                 spin_lock_init(&tun->lock);
2726
2727                 err = security_tun_dev_alloc_security(&tun->security);
2728                 if (err < 0)
2729                         goto err_free_stat;
2730
2731                 tun_net_init(dev);
2732                 tun_flow_init(tun);
2733
2734                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2735                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2736                                    NETIF_F_HW_VLAN_STAG_TX;
2737                 dev->features = dev->hw_features | NETIF_F_LLTX;
2738                 dev->vlan_features = dev->features &
2739                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2740                                        NETIF_F_HW_VLAN_STAG_TX);
2741
2742                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2743                               (ifr->ifr_flags & TUN_FEATURES);
2744
2745                 INIT_LIST_HEAD(&tun->disabled);
2746                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2747                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2748                 if (err < 0)
2749                         goto err_free_flow;
2750
2751                 err = register_netdevice(tun->dev);
2752                 if (err < 0)
2753                         goto err_detach;
2754                 /* free_netdev() won't check refcnt, to avoid race
2755                  * with dev_put() we need publish tun after registration.
2756                  */
2757                 rcu_assign_pointer(tfile->tun, tun);
2758         }
2759
2760         netif_carrier_on(tun->dev);
2761
2762         /* Make sure persistent devices do not get stuck in
2763          * xoff state.
2764          */
2765         if (netif_running(tun->dev))
2766                 netif_tx_wake_all_queues(tun->dev);
2767
2768         strcpy(ifr->ifr_name, tun->dev->name);
2769         return 0;
2770
2771 err_detach:
2772         tun_detach_all(dev);
2773         /* We are here because register_netdevice() has failed.
2774          * If register_netdevice() already called tun_free_netdev()
2775          * while dealing with the error, dev->stats has been cleared.
2776          */
2777         if (!dev->tstats)
2778                 goto err_free_dev;
2779
2780 err_free_flow:
2781         tun_flow_uninit(tun);
2782         security_tun_dev_free_security(tun->security);
2783 err_free_stat:
2784         free_percpu(dev->tstats);
2785 err_free_dev:
2786         free_netdev(dev);
2787         return err;
2788 }
2789
2790 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2791 {
2792         strcpy(ifr->ifr_name, tun->dev->name);
2793
2794         ifr->ifr_flags = tun_flags(tun);
2795
2796 }
2797
2798 /* This is like a cut-down ethtool ops, except done via tun fd so no
2799  * privs required. */
2800 static int set_offload(struct tun_struct *tun, unsigned long arg)
2801 {
2802         netdev_features_t features = 0;
2803
2804         if (arg & TUN_F_CSUM) {
2805                 features |= NETIF_F_HW_CSUM;
2806                 arg &= ~TUN_F_CSUM;
2807
2808                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2809                         if (arg & TUN_F_TSO_ECN) {
2810                                 features |= NETIF_F_TSO_ECN;
2811                                 arg &= ~TUN_F_TSO_ECN;
2812                         }
2813                         if (arg & TUN_F_TSO4)
2814                                 features |= NETIF_F_TSO;
2815                         if (arg & TUN_F_TSO6)
2816                                 features |= NETIF_F_TSO6;
2817                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2818                 }
2819
2820                 arg &= ~TUN_F_UFO;
2821         }
2822
2823         /* This gives the user a way to test for new features in future by
2824          * trying to set them. */
2825         if (arg)
2826                 return -EINVAL;
2827
2828         tun->set_features = features;
2829         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2830         tun->dev->wanted_features |= features;
2831         netdev_update_features(tun->dev);
2832
2833         return 0;
2834 }
2835
2836 static void tun_detach_filter(struct tun_struct *tun, int n)
2837 {
2838         int i;
2839         struct tun_file *tfile;
2840
2841         for (i = 0; i < n; i++) {
2842                 tfile = rtnl_dereference(tun->tfiles[i]);
2843                 lock_sock(tfile->socket.sk);
2844                 sk_detach_filter(tfile->socket.sk);
2845                 release_sock(tfile->socket.sk);
2846         }
2847
2848         tun->filter_attached = false;
2849 }
2850
2851 static int tun_attach_filter(struct tun_struct *tun)
2852 {
2853         int i, ret = 0;
2854         struct tun_file *tfile;
2855
2856         for (i = 0; i < tun->numqueues; i++) {
2857                 tfile = rtnl_dereference(tun->tfiles[i]);
2858                 lock_sock(tfile->socket.sk);
2859                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2860                 release_sock(tfile->socket.sk);
2861                 if (ret) {
2862                         tun_detach_filter(tun, i);
2863                         return ret;
2864                 }
2865         }
2866
2867         tun->filter_attached = true;
2868         return ret;
2869 }
2870
2871 static void tun_set_sndbuf(struct tun_struct *tun)
2872 {
2873         struct tun_file *tfile;
2874         int i;
2875
2876         for (i = 0; i < tun->numqueues; i++) {
2877                 tfile = rtnl_dereference(tun->tfiles[i]);
2878                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2879         }
2880 }
2881
2882 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2883 {
2884         struct tun_file *tfile = file->private_data;
2885         struct tun_struct *tun;
2886         int ret = 0;
2887
2888         rtnl_lock();
2889
2890         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2891                 tun = tfile->detached;
2892                 if (!tun) {
2893                         ret = -EINVAL;
2894                         goto unlock;
2895                 }
2896                 ret = security_tun_dev_attach_queue(tun->security);
2897                 if (ret < 0)
2898                         goto unlock;
2899                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2900                                  tun->flags & IFF_NAPI_FRAGS, true);
2901         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2902                 tun = rtnl_dereference(tfile->tun);
2903                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2904                         ret = -EINVAL;
2905                 else
2906                         __tun_detach(tfile, false);
2907         } else
2908                 ret = -EINVAL;
2909
2910         if (ret >= 0)
2911                 netdev_state_change(tun->dev);
2912
2913 unlock:
2914         rtnl_unlock();
2915         return ret;
2916 }
2917
2918 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2919                         void __user *data)
2920 {
2921         struct bpf_prog *prog;
2922         int fd;
2923
2924         if (copy_from_user(&fd, data, sizeof(fd)))
2925                 return -EFAULT;
2926
2927         if (fd == -1) {
2928                 prog = NULL;
2929         } else {
2930                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2931                 if (IS_ERR(prog))
2932                         return PTR_ERR(prog);
2933         }
2934
2935         return __tun_set_ebpf(tun, prog_p, prog);
2936 }
2937
2938 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2939 static unsigned char tun_get_addr_len(unsigned short type)
2940 {
2941         switch (type) {
2942         case ARPHRD_IP6GRE:
2943         case ARPHRD_TUNNEL6:
2944                 return sizeof(struct in6_addr);
2945         case ARPHRD_IPGRE:
2946         case ARPHRD_TUNNEL:
2947         case ARPHRD_SIT:
2948                 return 4;
2949         case ARPHRD_ETHER:
2950                 return ETH_ALEN;
2951         case ARPHRD_IEEE802154:
2952         case ARPHRD_IEEE802154_MONITOR:
2953                 return IEEE802154_EXTENDED_ADDR_LEN;
2954         case ARPHRD_PHONET_PIPE:
2955         case ARPHRD_PPP:
2956         case ARPHRD_NONE:
2957                 return 0;
2958         case ARPHRD_6LOWPAN:
2959                 return EUI64_ADDR_LEN;
2960         case ARPHRD_FDDI:
2961                 return FDDI_K_ALEN;
2962         case ARPHRD_HIPPI:
2963                 return HIPPI_ALEN;
2964         case ARPHRD_IEEE802:
2965                 return FC_ALEN;
2966         case ARPHRD_ROSE:
2967                 return ROSE_ADDR_LEN;
2968         case ARPHRD_NETROM:
2969                 return AX25_ADDR_LEN;
2970         case ARPHRD_LOCALTLK:
2971                 return LTALK_ALEN;
2972         default:
2973                 return 0;
2974         }
2975 }
2976
2977 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2978                             unsigned long arg, int ifreq_len)
2979 {
2980         struct tun_file *tfile = file->private_data;
2981         struct net *net = sock_net(&tfile->sk);
2982         struct tun_struct *tun;
2983         void __user* argp = (void __user*)arg;
2984         unsigned int ifindex, carrier;
2985         struct ifreq ifr;
2986         kuid_t owner;
2987         kgid_t group;
2988         int sndbuf;
2989         int vnet_hdr_sz;
2990         int le;
2991         int ret;
2992         bool do_notify = false;
2993
2994         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2995             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2996                 if (copy_from_user(&ifr, argp, ifreq_len))
2997                         return -EFAULT;
2998         } else {
2999                 memset(&ifr, 0, sizeof(ifr));
3000         }
3001         if (cmd == TUNGETFEATURES) {
3002                 /* Currently this just means: "what IFF flags are valid?".
3003                  * This is needed because we never checked for invalid flags on
3004                  * TUNSETIFF.
3005                  */
3006                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3007                                 (unsigned int __user*)argp);
3008         } else if (cmd == TUNSETQUEUE) {
3009                 return tun_set_queue(file, &ifr);
3010         } else if (cmd == SIOCGSKNS) {
3011                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3012                         return -EPERM;
3013                 return open_related_ns(&net->ns, get_net_ns);
3014         }
3015
3016         rtnl_lock();
3017
3018         tun = tun_get(tfile);
3019         if (cmd == TUNSETIFF) {
3020                 ret = -EEXIST;
3021                 if (tun)
3022                         goto unlock;
3023
3024                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3025
3026                 ret = tun_set_iff(net, file, &ifr);
3027
3028                 if (ret)
3029                         goto unlock;
3030
3031                 if (copy_to_user(argp, &ifr, ifreq_len))
3032                         ret = -EFAULT;
3033                 goto unlock;
3034         }
3035         if (cmd == TUNSETIFINDEX) {
3036                 ret = -EPERM;
3037                 if (tun)
3038                         goto unlock;
3039
3040                 ret = -EFAULT;
3041                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3042                         goto unlock;
3043
3044                 ret = 0;
3045                 tfile->ifindex = ifindex;
3046                 goto unlock;
3047         }
3048
3049         ret = -EBADFD;
3050         if (!tun)
3051                 goto unlock;
3052
3053         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3054
3055         net = dev_net(tun->dev);
3056         ret = 0;
3057         switch (cmd) {
3058         case TUNGETIFF:
3059                 tun_get_iff(tun, &ifr);
3060
3061                 if (tfile->detached)
3062                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3063                 if (!tfile->socket.sk->sk_filter)
3064                         ifr.ifr_flags |= IFF_NOFILTER;
3065
3066                 if (copy_to_user(argp, &ifr, ifreq_len))
3067                         ret = -EFAULT;
3068                 break;
3069
3070         case TUNSETNOCSUM:
3071                 /* Disable/Enable checksum */
3072
3073                 /* [unimplemented] */
3074                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3075                            arg ? "disabled" : "enabled");
3076                 break;
3077
3078         case TUNSETPERSIST:
3079                 /* Disable/Enable persist mode. Keep an extra reference to the
3080                  * module to prevent the module being unprobed.
3081                  */
3082                 if (arg && !(tun->flags & IFF_PERSIST)) {
3083                         tun->flags |= IFF_PERSIST;
3084                         __module_get(THIS_MODULE);
3085                         do_notify = true;
3086                 }
3087                 if (!arg && (tun->flags & IFF_PERSIST)) {
3088                         tun->flags &= ~IFF_PERSIST;
3089                         module_put(THIS_MODULE);
3090                         do_notify = true;
3091                 }
3092
3093                 netif_info(tun, drv, tun->dev, "persist %s\n",
3094                            arg ? "enabled" : "disabled");
3095                 break;
3096
3097         case TUNSETOWNER:
3098                 /* Set owner of the device */
3099                 owner = make_kuid(current_user_ns(), arg);
3100                 if (!uid_valid(owner)) {
3101                         ret = -EINVAL;
3102                         break;
3103                 }
3104                 tun->owner = owner;
3105                 do_notify = true;
3106                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3107                            from_kuid(&init_user_ns, tun->owner));
3108                 break;
3109
3110         case TUNSETGROUP:
3111                 /* Set group of the device */
3112                 group = make_kgid(current_user_ns(), arg);
3113                 if (!gid_valid(group)) {
3114                         ret = -EINVAL;
3115                         break;
3116                 }
3117                 tun->group = group;
3118                 do_notify = true;
3119                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3120                            from_kgid(&init_user_ns, tun->group));
3121                 break;
3122
3123         case TUNSETLINK:
3124                 /* Only allow setting the type when the interface is down */
3125                 if (tun->dev->flags & IFF_UP) {
3126                         netif_info(tun, drv, tun->dev,
3127                                    "Linktype set failed because interface is up\n");
3128                         ret = -EBUSY;
3129                 } else {
3130                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3131                                                        tun->dev);
3132                         ret = notifier_to_errno(ret);
3133                         if (ret) {
3134                                 netif_info(tun, drv, tun->dev,
3135                                            "Refused to change device type\n");
3136                                 break;
3137                         }
3138                         tun->dev->type = (int) arg;
3139                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3140                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3141                                    tun->dev->type);
3142                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3143                                                  tun->dev);
3144                 }
3145                 break;
3146
3147         case TUNSETDEBUG:
3148                 tun->msg_enable = (u32)arg;
3149                 break;
3150
3151         case TUNSETOFFLOAD:
3152                 ret = set_offload(tun, arg);
3153                 break;
3154
3155         case TUNSETTXFILTER:
3156                 /* Can be set only for TAPs */
3157                 ret = -EINVAL;
3158                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3159                         break;
3160                 ret = update_filter(&tun->txflt, (void __user *)arg);
3161                 break;
3162
3163         case SIOCGIFHWADDR:
3164                 /* Get hw address */
3165                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3166                 if (copy_to_user(argp, &ifr, ifreq_len))
3167                         ret = -EFAULT;
3168                 break;
3169
3170         case SIOCSIFHWADDR:
3171                 /* Set hw address */
3172                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3173                 break;
3174
3175         case TUNGETSNDBUF:
3176                 sndbuf = tfile->socket.sk->sk_sndbuf;
3177                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3178                         ret = -EFAULT;
3179                 break;
3180
3181         case TUNSETSNDBUF:
3182                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3183                         ret = -EFAULT;
3184                         break;
3185                 }
3186                 if (sndbuf <= 0) {
3187                         ret = -EINVAL;
3188                         break;
3189                 }
3190
3191                 tun->sndbuf = sndbuf;
3192                 tun_set_sndbuf(tun);
3193                 break;
3194
3195         case TUNGETVNETHDRSZ:
3196                 vnet_hdr_sz = tun->vnet_hdr_sz;
3197                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3198                         ret = -EFAULT;
3199                 break;
3200
3201         case TUNSETVNETHDRSZ:
3202                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3203                         ret = -EFAULT;
3204                         break;
3205                 }
3206                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3207                         ret = -EINVAL;
3208                         break;
3209                 }
3210
3211                 tun->vnet_hdr_sz = vnet_hdr_sz;
3212                 break;
3213
3214         case TUNGETVNETLE:
3215                 le = !!(tun->flags & TUN_VNET_LE);
3216                 if (put_user(le, (int __user *)argp))
3217                         ret = -EFAULT;
3218                 break;
3219
3220         case TUNSETVNETLE:
3221                 if (get_user(le, (int __user *)argp)) {
3222                         ret = -EFAULT;
3223                         break;
3224                 }
3225                 if (le)
3226                         tun->flags |= TUN_VNET_LE;
3227                 else
3228                         tun->flags &= ~TUN_VNET_LE;
3229                 break;
3230
3231         case TUNGETVNETBE:
3232                 ret = tun_get_vnet_be(tun, argp);
3233                 break;
3234
3235         case TUNSETVNETBE:
3236                 ret = tun_set_vnet_be(tun, argp);
3237                 break;
3238
3239         case TUNATTACHFILTER:
3240                 /* Can be set only for TAPs */
3241                 ret = -EINVAL;
3242                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3243                         break;
3244                 ret = -EFAULT;
3245                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3246                         break;
3247
3248                 ret = tun_attach_filter(tun);
3249                 break;
3250
3251         case TUNDETACHFILTER:
3252                 /* Can be set only for TAPs */
3253                 ret = -EINVAL;
3254                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3255                         break;
3256                 ret = 0;
3257                 tun_detach_filter(tun, tun->numqueues);
3258                 break;
3259
3260         case TUNGETFILTER:
3261                 ret = -EINVAL;
3262                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3263                         break;
3264                 ret = -EFAULT;
3265                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3266                         break;
3267                 ret = 0;
3268                 break;
3269
3270         case TUNSETSTEERINGEBPF:
3271                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3272                 break;
3273
3274         case TUNSETFILTEREBPF:
3275                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3276                 break;
3277
3278         case TUNSETCARRIER:
3279                 ret = -EFAULT;
3280                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3281                         goto unlock;
3282
3283                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3284                 break;
3285
3286         case TUNGETDEVNETNS:
3287                 ret = -EPERM;
3288                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3289                         goto unlock;
3290                 ret = open_related_ns(&net->ns, get_net_ns);
3291                 break;
3292
3293         default:
3294                 ret = -EINVAL;
3295                 break;
3296         }
3297
3298         if (do_notify)
3299                 netdev_state_change(tun->dev);
3300
3301 unlock:
3302         rtnl_unlock();
3303         if (tun)
3304                 tun_put(tun);
3305         return ret;
3306 }
3307
3308 static long tun_chr_ioctl(struct file *file,
3309                           unsigned int cmd, unsigned long arg)
3310 {
3311         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3312 }
3313
3314 #ifdef CONFIG_COMPAT
3315 static long tun_chr_compat_ioctl(struct file *file,
3316                          unsigned int cmd, unsigned long arg)
3317 {
3318         switch (cmd) {
3319         case TUNSETIFF:
3320         case TUNGETIFF:
3321         case TUNSETTXFILTER:
3322         case TUNGETSNDBUF:
3323         case TUNSETSNDBUF:
3324         case SIOCGIFHWADDR:
3325         case SIOCSIFHWADDR:
3326                 arg = (unsigned long)compat_ptr(arg);
3327                 break;
3328         default:
3329                 arg = (compat_ulong_t)arg;
3330                 break;
3331         }
3332
3333         /*
3334          * compat_ifreq is shorter than ifreq, so we must not access beyond
3335          * the end of that structure. All fields that are used in this
3336          * driver are compatible though, we don't need to convert the
3337          * contents.
3338          */
3339         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3340 }
3341 #endif /* CONFIG_COMPAT */
3342
3343 static int tun_chr_fasync(int fd, struct file *file, int on)
3344 {
3345         struct tun_file *tfile = file->private_data;
3346         int ret;
3347
3348         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3349                 goto out;
3350
3351         if (on) {
3352                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3353                 tfile->flags |= TUN_FASYNC;
3354         } else
3355                 tfile->flags &= ~TUN_FASYNC;
3356         ret = 0;
3357 out:
3358         return ret;
3359 }
3360
3361 static int tun_chr_open(struct inode *inode, struct file * file)
3362 {
3363         struct net *net = current->nsproxy->net_ns;
3364         struct tun_file *tfile;
3365
3366         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3367                                             &tun_proto, 0);
3368         if (!tfile)
3369                 return -ENOMEM;
3370         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3371                 sk_free(&tfile->sk);
3372                 return -ENOMEM;
3373         }
3374
3375         mutex_init(&tfile->napi_mutex);
3376         RCU_INIT_POINTER(tfile->tun, NULL);
3377         tfile->flags = 0;
3378         tfile->ifindex = 0;
3379
3380         init_waitqueue_head(&tfile->socket.wq.wait);
3381
3382         tfile->socket.file = file;
3383         tfile->socket.ops = &tun_socket_ops;
3384
3385         sock_init_data(&tfile->socket, &tfile->sk);
3386
3387         tfile->sk.sk_write_space = tun_sock_write_space;
3388         tfile->sk.sk_sndbuf = INT_MAX;
3389
3390         file->private_data = tfile;
3391         INIT_LIST_HEAD(&tfile->next);
3392
3393         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3394
3395         return 0;
3396 }
3397
3398 static int tun_chr_close(struct inode *inode, struct file *file)
3399 {
3400         struct tun_file *tfile = file->private_data;
3401
3402         tun_detach(tfile, true);
3403
3404         return 0;
3405 }
3406
3407 #ifdef CONFIG_PROC_FS
3408 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3409 {
3410         struct tun_file *tfile = file->private_data;
3411         struct tun_struct *tun;
3412         struct ifreq ifr;
3413
3414         memset(&ifr, 0, sizeof(ifr));
3415
3416         rtnl_lock();
3417         tun = tun_get(tfile);
3418         if (tun)
3419                 tun_get_iff(tun, &ifr);
3420         rtnl_unlock();
3421
3422         if (tun)
3423                 tun_put(tun);
3424
3425         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3426 }
3427 #endif
3428
3429 static const struct file_operations tun_fops = {
3430         .owner  = THIS_MODULE,
3431         .llseek = no_llseek,
3432         .read_iter  = tun_chr_read_iter,
3433         .write_iter = tun_chr_write_iter,
3434         .poll   = tun_chr_poll,
3435         .unlocked_ioctl = tun_chr_ioctl,
3436 #ifdef CONFIG_COMPAT
3437         .compat_ioctl = tun_chr_compat_ioctl,
3438 #endif
3439         .open   = tun_chr_open,
3440         .release = tun_chr_close,
3441         .fasync = tun_chr_fasync,
3442 #ifdef CONFIG_PROC_FS
3443         .show_fdinfo = tun_chr_show_fdinfo,
3444 #endif
3445 };
3446
3447 static struct miscdevice tun_miscdev = {
3448         .minor = TUN_MINOR,
3449         .name = "tun",
3450         .nodename = "net/tun",
3451         .fops = &tun_fops,
3452 };
3453
3454 /* ethtool interface */
3455
3456 static void tun_default_link_ksettings(struct net_device *dev,
3457                                        struct ethtool_link_ksettings *cmd)
3458 {
3459         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3460         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3461         cmd->base.speed         = SPEED_10;
3462         cmd->base.duplex        = DUPLEX_FULL;
3463         cmd->base.port          = PORT_TP;
3464         cmd->base.phy_address   = 0;
3465         cmd->base.autoneg       = AUTONEG_DISABLE;
3466 }
3467
3468 static int tun_get_link_ksettings(struct net_device *dev,
3469                                   struct ethtool_link_ksettings *cmd)
3470 {
3471         struct tun_struct *tun = netdev_priv(dev);
3472
3473         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3474         return 0;
3475 }
3476
3477 static int tun_set_link_ksettings(struct net_device *dev,
3478                                   const struct ethtool_link_ksettings *cmd)
3479 {
3480         struct tun_struct *tun = netdev_priv(dev);
3481
3482         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3483         return 0;
3484 }
3485
3486 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3487 {
3488         struct tun_struct *tun = netdev_priv(dev);
3489
3490         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3491         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3492
3493         switch (tun->flags & TUN_TYPE_MASK) {
3494         case IFF_TUN:
3495                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3496                 break;
3497         case IFF_TAP:
3498                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3499                 break;
3500         }
3501 }
3502
3503 static u32 tun_get_msglevel(struct net_device *dev)
3504 {
3505         struct tun_struct *tun = netdev_priv(dev);
3506
3507         return tun->msg_enable;
3508 }
3509
3510 static void tun_set_msglevel(struct net_device *dev, u32 value)
3511 {
3512         struct tun_struct *tun = netdev_priv(dev);
3513
3514         tun->msg_enable = value;
3515 }
3516
3517 static int tun_get_coalesce(struct net_device *dev,
3518                             struct ethtool_coalesce *ec,
3519                             struct kernel_ethtool_coalesce *kernel_coal,
3520                             struct netlink_ext_ack *extack)
3521 {
3522         struct tun_struct *tun = netdev_priv(dev);
3523
3524         ec->rx_max_coalesced_frames = tun->rx_batched;
3525
3526         return 0;
3527 }
3528
3529 static int tun_set_coalesce(struct net_device *dev,
3530                             struct ethtool_coalesce *ec,
3531                             struct kernel_ethtool_coalesce *kernel_coal,
3532                             struct netlink_ext_ack *extack)
3533 {
3534         struct tun_struct *tun = netdev_priv(dev);
3535
3536         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3537                 tun->rx_batched = NAPI_POLL_WEIGHT;
3538         else
3539                 tun->rx_batched = ec->rx_max_coalesced_frames;
3540
3541         return 0;
3542 }
3543
3544 static const struct ethtool_ops tun_ethtool_ops = {
3545         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3546         .get_drvinfo    = tun_get_drvinfo,
3547         .get_msglevel   = tun_get_msglevel,
3548         .set_msglevel   = tun_set_msglevel,
3549         .get_link       = ethtool_op_get_link,
3550         .get_ts_info    = ethtool_op_get_ts_info,
3551         .get_coalesce   = tun_get_coalesce,
3552         .set_coalesce   = tun_set_coalesce,
3553         .get_link_ksettings = tun_get_link_ksettings,
3554         .set_link_ksettings = tun_set_link_ksettings,
3555 };
3556
3557 static int tun_queue_resize(struct tun_struct *tun)
3558 {
3559         struct net_device *dev = tun->dev;
3560         struct tun_file *tfile;
3561         struct ptr_ring **rings;
3562         int n = tun->numqueues + tun->numdisabled;
3563         int ret, i;
3564
3565         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3566         if (!rings)
3567                 return -ENOMEM;
3568
3569         for (i = 0; i < tun->numqueues; i++) {
3570                 tfile = rtnl_dereference(tun->tfiles[i]);
3571                 rings[i] = &tfile->tx_ring;
3572         }
3573         list_for_each_entry(tfile, &tun->disabled, next)
3574                 rings[i++] = &tfile->tx_ring;
3575
3576         ret = ptr_ring_resize_multiple(rings, n,
3577                                        dev->tx_queue_len, GFP_KERNEL,
3578                                        tun_ptr_free);
3579
3580         kfree(rings);
3581         return ret;
3582 }
3583
3584 static int tun_device_event(struct notifier_block *unused,
3585                             unsigned long event, void *ptr)
3586 {
3587         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3588         struct tun_struct *tun = netdev_priv(dev);
3589         int i;
3590
3591         if (dev->rtnl_link_ops != &tun_link_ops)
3592                 return NOTIFY_DONE;
3593
3594         switch (event) {
3595         case NETDEV_CHANGE_TX_QUEUE_LEN:
3596                 if (tun_queue_resize(tun))
3597                         return NOTIFY_BAD;
3598                 break;
3599         case NETDEV_UP:
3600                 for (i = 0; i < tun->numqueues; i++) {
3601                         struct tun_file *tfile;
3602
3603                         tfile = rtnl_dereference(tun->tfiles[i]);
3604                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3605                 }
3606                 break;
3607         default:
3608                 break;
3609         }
3610
3611         return NOTIFY_DONE;
3612 }
3613
3614 static struct notifier_block tun_notifier_block __read_mostly = {
3615         .notifier_call  = tun_device_event,
3616 };
3617
3618 static int __init tun_init(void)
3619 {
3620         int ret = 0;
3621
3622         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3623
3624         ret = rtnl_link_register(&tun_link_ops);
3625         if (ret) {
3626                 pr_err("Can't register link_ops\n");
3627                 goto err_linkops;
3628         }
3629
3630         ret = misc_register(&tun_miscdev);
3631         if (ret) {
3632                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3633                 goto err_misc;
3634         }
3635
3636         ret = register_netdevice_notifier(&tun_notifier_block);
3637         if (ret) {
3638                 pr_err("Can't register netdevice notifier\n");
3639                 goto err_notifier;
3640         }
3641
3642         return  0;
3643
3644 err_notifier:
3645         misc_deregister(&tun_miscdev);
3646 err_misc:
3647         rtnl_link_unregister(&tun_link_ops);
3648 err_linkops:
3649         return ret;
3650 }
3651
3652 static void tun_cleanup(void)
3653 {
3654         misc_deregister(&tun_miscdev);
3655         rtnl_link_unregister(&tun_link_ops);
3656         unregister_netdevice_notifier(&tun_notifier_block);
3657 }
3658
3659 /* Get an underlying socket object from tun file.  Returns error unless file is
3660  * attached to a device.  The returned object works like a packet socket, it
3661  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3662  * holding a reference to the file for as long as the socket is in use. */
3663 struct socket *tun_get_socket(struct file *file)
3664 {
3665         struct tun_file *tfile;
3666         if (file->f_op != &tun_fops)
3667                 return ERR_PTR(-EINVAL);
3668         tfile = file->private_data;
3669         if (!tfile)
3670                 return ERR_PTR(-EBADFD);
3671         return &tfile->socket;
3672 }
3673 EXPORT_SYMBOL_GPL(tun_get_socket);
3674
3675 struct ptr_ring *tun_get_tx_ring(struct file *file)
3676 {
3677         struct tun_file *tfile;
3678
3679         if (file->f_op != &tun_fops)
3680                 return ERR_PTR(-EINVAL);
3681         tfile = file->private_data;
3682         if (!tfile)
3683                 return ERR_PTR(-EBADFD);
3684         return &tfile->tx_ring;
3685 }
3686 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3687
3688 module_init(tun_init);
3689 module_exit(tun_cleanup);
3690 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3691 MODULE_AUTHOR(DRV_COPYRIGHT);
3692 MODULE_LICENSE("GPL");
3693 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3694 MODULE_ALIAS("devname:net/tun");