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 | NETIF_F_GSO_UDP_L4)
 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         /* init args */
 213         struct file *file;
 214         struct ifreq *ifr;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static void tun_flow_init(struct tun_struct *tun);
 223 static void tun_flow_uninit(struct tun_struct *tun);
 224
 225 static int tun_napi_receive(struct napi_struct *napi, int budget)
 226 {
 227         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 228         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 229         struct sk_buff_head process_queue;
 230         struct sk_buff *skb;
 231         int received = 0;
 232
 233         __skb_queue_head_init(&process_queue);
 234
 235         spin_lock(&queue->lock);
 236         skb_queue_splice_tail_init(queue, &process_queue);
 237         spin_unlock(&queue->lock);
 238
 239         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 240                 napi_gro_receive(napi, skb);
 241                 ++received;
 242         }
 243
 244         if (!skb_queue_empty(&process_queue)) {
 245                 spin_lock(&queue->lock);
 246                 skb_queue_splice(&process_queue, queue);
 247                 spin_unlock(&queue->lock);
 248         }
 249
 250         return received;
 251 }
 252
 253 static int tun_napi_poll(struct napi_struct *napi, int budget)
 254 {
 255         unsigned int received;
 256
 257         received = tun_napi_receive(napi, budget);
 258
 259         if (received < budget)
 260                 napi_complete_done(napi, received);
 261
 262         return received;
 263 }
 264
 265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 266                           bool napi_en, bool napi_frags)
 267 {
 268         tfile->napi_enabled = napi_en;
 269         tfile->napi_frags_enabled = napi_en && napi_frags;
 270         if (napi_en) {
 271                 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 272                 napi_enable(&tfile->napi);
 273         }
 274 }
 275
 276 static void tun_napi_enable(struct tun_file *tfile)
 277 {
 278         if (tfile->napi_enabled)
 279                 napi_enable(&tfile->napi);
 280 }
 281
 282 static void tun_napi_disable(struct tun_file *tfile)
 283 {
 284         if (tfile->napi_enabled)
 285                 napi_disable(&tfile->napi);
 286 }
 287
 288 static void tun_napi_del(struct tun_file *tfile)
 289 {
 290         if (tfile->napi_enabled)
 291                 netif_napi_del(&tfile->napi);
 292 }
 293
 294 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 295 {
 296         return tfile->napi_frags_enabled;
 297 }
 298
 299 #ifdef CONFIG_TUN_VNET_CROSS_LE
 300 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 301 {
 302         return tun->flags & TUN_VNET_BE ? false :
 303                 virtio_legacy_is_little_endian();
 304 }
 305
 306 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 307 {
 308         int be = !!(tun->flags & TUN_VNET_BE);
 309
 310         if (put_user(be, argp))
 311                 return -EFAULT;
 312
 313         return 0;
 314 }
 315
 316 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 317 {
 318         int be;
 319
 320         if (get_user(be, argp))
 321                 return -EFAULT;
 322
 323         if (be)
 324                 tun->flags |= TUN_VNET_BE;
 325         else
 326                 tun->flags &= ~TUN_VNET_BE;
 327
 328         return 0;
 329 }
 330 #else
 331 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 332 {
 333         return virtio_legacy_is_little_endian();
 334 }
 335
 336 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 337 {
 338         return -EINVAL;
 339 }
 340
 341 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 342 {
 343         return -EINVAL;
 344 }
 345 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 346
 347 static inline bool tun_is_little_endian(struct tun_struct *tun)
 348 {
 349         return tun->flags & TUN_VNET_LE ||
 350                 tun_legacy_is_little_endian(tun);
 351 }
 352
 353 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 354 {
 355         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 356 }
 357
 358 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 359 {
 360         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 361 }
 362
 363 static inline u32 tun_hashfn(u32 rxhash)
 364 {
 365         return rxhash & TUN_MASK_FLOW_ENTRIES;
 366 }
 367
 368 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 369 {
 370         struct tun_flow_entry *e;
 371
 372         hlist_for_each_entry_rcu(e, head, hash_link) {
 373                 if (e->rxhash == rxhash)
 374                         return e;
 375         }
 376         return NULL;
 377 }
 378
 379 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 380                                               struct hlist_head *head,
 381                                               u32 rxhash, u16 queue_index)
 382 {
 383         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 384
 385         if (e) {
 386                 netif_info(tun, tx_queued, tun->dev,
 387                            "create flow: hash %u index %u\n",
 388                            rxhash, queue_index);
 389                 e->updated = jiffies;
 390                 e->rxhash = rxhash;
 391                 e->rps_rxhash = 0;
 392                 e->queue_index = queue_index;
 393                 e->tun = tun;
 394                 hlist_add_head_rcu(&e->hash_link, head);
 395                 ++tun->flow_count;
 396         }
 397         return e;
 398 }
 399
 400 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 401 {
 402         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 403                    e->rxhash, e->queue_index);
 404         hlist_del_rcu(&e->hash_link);
 405         kfree_rcu(e, rcu);
 406         --tun->flow_count;
 407 }
 408
 409 static void tun_flow_flush(struct tun_struct *tun)
 410 {
 411         int i;
 412
 413         spin_lock_bh(&tun->lock);
 414         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 415                 struct tun_flow_entry *e;
 416                 struct hlist_node *n;
 417
 418                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 419                         tun_flow_delete(tun, e);
 420         }
 421         spin_unlock_bh(&tun->lock);
 422 }
 423
 424 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 425 {
 426         int i;
 427
 428         spin_lock_bh(&tun->lock);
 429         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 430                 struct tun_flow_entry *e;
 431                 struct hlist_node *n;
 432
 433                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 434                         if (e->queue_index == queue_index)
 435                                 tun_flow_delete(tun, e);
 436                 }
 437         }
 438         spin_unlock_bh(&tun->lock);
 439 }
 440
 441 static void tun_flow_cleanup(struct timer_list *t)
 442 {
 443         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 444         unsigned long delay = tun->ageing_time;
 445         unsigned long next_timer = jiffies + delay;
 446         unsigned long count = 0;
 447         int i;
 448
 449         spin_lock(&tun->lock);
 450         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 451                 struct tun_flow_entry *e;
 452                 struct hlist_node *n;
 453
 454                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 455                         unsigned long this_timer;
 456
 457                         this_timer = e->updated + delay;
 458                         if (time_before_eq(this_timer, jiffies)) {
 459                                 tun_flow_delete(tun, e);
 460                                 continue;
 461                         }
 462                         count++;
 463                         if (time_before(this_timer, next_timer))
 464                                 next_timer = this_timer;
 465                 }
 466         }
 467
 468         if (count)
 469                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 470         spin_unlock(&tun->lock);
 471 }
 472
 473 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 474                             struct tun_file *tfile)
 475 {
 476         struct hlist_head *head;
 477         struct tun_flow_entry *e;
 478         unsigned long delay = tun->ageing_time;
 479         u16 queue_index = tfile->queue_index;
 480
 481         head = &tun->flows[tun_hashfn(rxhash)];
 482
 483         rcu_read_lock();
 484
 485         e = tun_flow_find(head, rxhash);
 486         if (likely(e)) {
 487                 /* TODO: keep queueing to old queue until it's empty? */
 488                 if (READ_ONCE(e->queue_index) != queue_index)
 489                         WRITE_ONCE(e->queue_index, queue_index);
 490                 if (e->updated != jiffies)
 491                         e->updated = jiffies;
 492                 sock_rps_record_flow_hash(e->rps_rxhash);
 493         } else {
 494                 spin_lock_bh(&tun->lock);
 495                 if (!tun_flow_find(head, rxhash) &&
 496                     tun->flow_count < MAX_TAP_FLOWS)
 497                         tun_flow_create(tun, head, rxhash, queue_index);
 498
 499                 if (!timer_pending(&tun->flow_gc_timer))
 500                         mod_timer(&tun->flow_gc_timer,
 501                                   round_jiffies_up(jiffies + delay));
 502                 spin_unlock_bh(&tun->lock);
 503         }
 504
 505         rcu_read_unlock();
 506 }
 507
 508 /* Save the hash received in the stack receive path and update the
 509  * flow_hash table accordingly.
 510  */
 511 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 512 {
 513         if (unlikely(e->rps_rxhash != hash))
 514                 e->rps_rxhash = hash;
 515 }
 516
 517 /* We try to identify a flow through its rxhash. The reason that
 518  * we do not check rxq no. is because some cards(e.g 82599), chooses
 519  * the rxq based on the txq where the last packet of the flow comes. As
 520  * the userspace application move between processors, we may get a
 521  * different rxq no. here.
 522  */
 523 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 524 {
 525         struct tun_flow_entry *e;
 526         u32 txq = 0;
 527         u32 numqueues = 0;
 528
 529         numqueues = READ_ONCE(tun->numqueues);
 530
 531         txq = __skb_get_hash_symmetric(skb);
 532         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 533         if (e) {
 534                 tun_flow_save_rps_rxhash(e, txq);
 535                 txq = e->queue_index;
 536         } else {
 537                 /* use multiply and shift instead of expensive divide */
 538                 txq = ((u64)txq * numqueues) >> 32;
 539         }
 540
 541         return txq;
 542 }
 543
 544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 545 {
 546         struct tun_prog *prog;
 547         u32 numqueues;
 548         u16 ret = 0;
 549
 550         numqueues = READ_ONCE(tun->numqueues);
 551         if (!numqueues)
 552                 return 0;
 553
 554         prog = rcu_dereference(tun->steering_prog);
 555         if (prog)
 556                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 557
 558         return ret % numqueues;
 559 }
 560
 561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 562                             struct net_device *sb_dev)
 563 {
 564         struct tun_struct *tun = netdev_priv(dev);
 565         u16 ret;
 566
 567         rcu_read_lock();
 568         if (rcu_dereference(tun->steering_prog))
 569                 ret = tun_ebpf_select_queue(tun, skb);
 570         else
 571                 ret = tun_automq_select_queue(tun, skb);
 572         rcu_read_unlock();
 573
 574         return ret;
 575 }
 576
 577 static inline bool tun_not_capable(struct tun_struct *tun)
 578 {
 579         const struct cred *cred = current_cred();
 580         struct net *net = dev_net(tun->dev);
 581
 582         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 583                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 584                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 585 }
 586
 587 static void tun_set_real_num_queues(struct tun_struct *tun)
 588 {
 589         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 590         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 591 }
 592
 593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 594 {
 595         tfile->detached = tun;
 596         list_add_tail(&tfile->next, &tun->disabled);
 597         ++tun->numdisabled;
 598 }
 599
 600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 601 {
 602         struct tun_struct *tun = tfile->detached;
 603
 604         tfile->detached = NULL;
 605         list_del_init(&tfile->next);
 606         --tun->numdisabled;
 607         return tun;
 608 }
 609
 610 void tun_ptr_free(void *ptr)
 611 {
 612         if (!ptr)
 613                 return;
 614         if (tun_is_xdp_frame(ptr)) {
 615                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 616
 617                 xdp_return_frame(xdpf);
 618         } else {
 619                 __skb_array_destroy_skb(ptr);
 620         }
 621 }
 622 EXPORT_SYMBOL_GPL(tun_ptr_free);
 623
 624 static void tun_queue_purge(struct tun_file *tfile)
 625 {
 626         void *ptr;
 627
 628         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 629                 tun_ptr_free(ptr);
 630
 631         skb_queue_purge(&tfile->sk.sk_write_queue);
 632         skb_queue_purge(&tfile->sk.sk_error_queue);
 633 }
 634
 635 static void __tun_detach(struct tun_file *tfile, bool clean)
 636 {
 637         struct tun_file *ntfile;
 638         struct tun_struct *tun;
 639
 640         tun = rtnl_dereference(tfile->tun);
 641
 642         if (tun && clean) {
 643                 if (!tfile->detached)
 644                         tun_napi_disable(tfile);
 645                 tun_napi_del(tfile);
 646         }
 647
 648         if (tun && !tfile->detached) {
 649                 u16 index = tfile->queue_index;
 650                 BUG_ON(index >= tun->numqueues);
 651
 652                 rcu_assign_pointer(tun->tfiles[index],
 653                                    tun->tfiles[tun->numqueues - 1]);
 654                 ntfile = rtnl_dereference(tun->tfiles[index]);
 655                 ntfile->queue_index = index;
 656                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 657                                    NULL);
 658
 659                 --tun->numqueues;
 660                 if (clean) {
 661                         RCU_INIT_POINTER(tfile->tun, NULL);
 662                         sock_put(&tfile->sk);
 663                 } else {
 664                         tun_disable_queue(tun, tfile);
 665                         tun_napi_disable(tfile);
 666                 }
 667
 668                 synchronize_net();
 669                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 670                 /* Drop read queue */
 671                 tun_queue_purge(tfile);
 672                 tun_set_real_num_queues(tun);
 673         } else if (tfile->detached && clean) {
 674                 tun = tun_enable_queue(tfile);
 675                 sock_put(&tfile->sk);
 676         }
 677
 678         if (clean) {
 679                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 680                         netif_carrier_off(tun->dev);
 681
 682                         if (!(tun->flags & IFF_PERSIST) &&
 683                             tun->dev->reg_state == NETREG_REGISTERED)
 684                                 unregister_netdevice(tun->dev);
 685                 }
 686                 if (tun)
 687                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 688                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 689         }
 690 }
 691
 692 static void tun_detach(struct tun_file *tfile, bool clean)
 693 {
 694         struct tun_struct *tun;
 695         struct net_device *dev;
 696
 697         rtnl_lock();
 698         tun = rtnl_dereference(tfile->tun);
 699         dev = tun ? tun->dev : NULL;
 700         __tun_detach(tfile, clean);
 701         if (dev)
 702                 netdev_state_change(dev);
 703         rtnl_unlock();
 704
 705         if (clean)
 706                 sock_put(&tfile->sk);
 707 }
 708
 709 static void tun_detach_all(struct net_device *dev)
 710 {
 711         struct tun_struct *tun = netdev_priv(dev);
 712         struct tun_file *tfile, *tmp;
 713         int i, n = tun->numqueues;
 714
 715         for (i = 0; i < n; i++) {
 716                 tfile = rtnl_dereference(tun->tfiles[i]);
 717                 BUG_ON(!tfile);
 718                 tun_napi_disable(tfile);
 719                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 720                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 721                 RCU_INIT_POINTER(tfile->tun, NULL);
 722                 --tun->numqueues;
 723         }
 724         list_for_each_entry(tfile, &tun->disabled, next) {
 725                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 726                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 727                 RCU_INIT_POINTER(tfile->tun, NULL);
 728         }
 729         BUG_ON(tun->numqueues != 0);
 730
 731         synchronize_net();
 732         for (i = 0; i < n; i++) {
 733                 tfile = rtnl_dereference(tun->tfiles[i]);
 734                 tun_napi_del(tfile);
 735                 /* Drop read queue */
 736                 tun_queue_purge(tfile);
 737                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 738                 sock_put(&tfile->sk);
 739         }
 740         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 741                 tun_napi_del(tfile);
 742                 tun_enable_queue(tfile);
 743                 tun_queue_purge(tfile);
 744                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 745                 sock_put(&tfile->sk);
 746         }
 747         BUG_ON(tun->numdisabled != 0);
 748
 749         if (tun->flags & IFF_PERSIST)
 750                 module_put(THIS_MODULE);
 751 }
 752
 753 static int tun_attach(struct tun_struct *tun, struct file *file,
 754                       bool skip_filter, bool napi, bool napi_frags,
 755                       bool publish_tun)
 756 {
 757         struct tun_file *tfile = file->private_data;
 758         struct net_device *dev = tun->dev;
 759         int err;
 760
 761         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 762         if (err < 0)
 763                 goto out;
 764
 765         err = -EINVAL;
 766         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 767                 goto out;
 768
 769         err = -EBUSY;
 770         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 771                 goto out;
 772
 773         err = -E2BIG;
 774         if (!tfile->detached &&
 775             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 776                 goto out;
 777
 778         err = 0;
 779
 780         /* Re-attach the filter to persist device */
 781         if (!skip_filter && (tun->filter_attached == true)) {
 782                 lock_sock(tfile->socket.sk);
 783                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 784                 release_sock(tfile->socket.sk);
 785                 if (!err)
 786                         goto out;
 787         }
 788
 789         if (!tfile->detached &&
 790             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 791                             GFP_KERNEL, tun_ptr_free)) {
 792                 err = -ENOMEM;
 793                 goto out;
 794         }
 795
 796         tfile->queue_index = tun->numqueues;
 797         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 798
 799         if (tfile->detached) {
 800                 /* Re-attach detached tfile, updating XDP queue_index */
 801                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 802
 803                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 804                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 805         } else {
 806                 /* Setup XDP RX-queue info, for new tfile getting attached */
 807                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 808                                        tun->dev, tfile->queue_index, 0);
 809                 if (err < 0)
 810                         goto out;
 811                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 812                                                  MEM_TYPE_PAGE_SHARED, NULL);
 813                 if (err < 0) {
 814                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 815                         goto out;
 816                 }
 817                 err = 0;
 818         }
 819
 820         if (tfile->detached) {
 821                 tun_enable_queue(tfile);
 822                 tun_napi_enable(tfile);
 823         } else {
 824                 sock_hold(&tfile->sk);
 825                 tun_napi_init(tun, tfile, napi, napi_frags);
 826         }
 827
 828         if (rtnl_dereference(tun->xdp_prog))
 829                 sock_set_flag(&tfile->sk, SOCK_XDP);
 830
 831         /* device is allowed to go away first, so no need to hold extra
 832          * refcnt.
 833          */
 834
 835         /* Publish tfile->tun and tun->tfiles only after we've fully
 836          * initialized tfile; otherwise we risk using half-initialized
 837          * object.
 838          */
 839         if (publish_tun)
 840                 rcu_assign_pointer(tfile->tun, tun);
 841         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 842         tun->numqueues++;
 843         tun_set_real_num_queues(tun);
 844 out:
 845         return err;
 846 }
 847
 848 static struct tun_struct *tun_get(struct tun_file *tfile)
 849 {
 850         struct tun_struct *tun;
 851
 852         rcu_read_lock();
 853         tun = rcu_dereference(tfile->tun);
 854         if (tun)
 855                 dev_hold(tun->dev);
 856         rcu_read_unlock();
 857
 858         return tun;
 859 }
 860
 861 static void tun_put(struct tun_struct *tun)
 862 {
 863         dev_put(tun->dev);
 864 }
 865
 866 /* TAP filtering */
 867 static void addr_hash_set(u32 *mask, const u8 *addr)
 868 {
 869         int n = ether_crc(ETH_ALEN, addr) >> 26;
 870         mask[n >> 5] |= (1 << (n & 31));
 871 }
 872
 873 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 874 {
 875         int n = ether_crc(ETH_ALEN, addr) >> 26;
 876         return mask[n >> 5] & (1 << (n & 31));
 877 }
 878
 879 static int update_filter(struct tap_filter *filter, void __user *arg)
 880 {
 881         struct { u8 u[ETH_ALEN]; } *addr;
 882         struct tun_filter uf;
 883         int err, alen, n, nexact;
 884
 885         if (copy_from_user(&uf, arg, sizeof(uf)))
 886                 return -EFAULT;
 887
 888         if (!uf.count) {
 889                 /* Disabled */
 890                 filter->count = 0;
 891                 return 0;
 892         }
 893
 894         alen = ETH_ALEN * uf.count;
 895         addr = memdup_user(arg + sizeof(uf), alen);
 896         if (IS_ERR(addr))
 897                 return PTR_ERR(addr);
 898
 899         /* The filter is updated without holding any locks. Which is
 900          * perfectly safe. We disable it first and in the worst
 901          * case we'll accept a few undesired packets. */
 902         filter->count = 0;
 903         wmb();
 904
 905         /* Use first set of addresses as an exact filter */
 906         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 907                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 908
 909         nexact = n;
 910
 911         /* Remaining multicast addresses are hashed,
 912          * unicast will leave the filter disabled. */
 913         memset(filter->mask, 0, sizeof(filter->mask));
 914         for (; n < uf.count; n++) {
 915                 if (!is_multicast_ether_addr(addr[n].u)) {
 916                         err = 0; /* no filter */
 917                         goto free_addr;
 918                 }
 919                 addr_hash_set(filter->mask, addr[n].u);
 920         }
 921
 922         /* For ALLMULTI just set the mask to all ones.
 923          * This overrides the mask populated above. */
 924         if ((uf.flags & TUN_FLT_ALLMULTI))
 925                 memset(filter->mask, ~0, sizeof(filter->mask));
 926
 927         /* Now enable the filter */
 928         wmb();
 929         filter->count = nexact;
 930
 931         /* Return the number of exact filters */
 932         err = nexact;
 933 free_addr:
 934         kfree(addr);
 935         return err;
 936 }
 937
 938 /* Returns: 0 - drop, !=0 - accept */
 939 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 940 {
 941         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 942          * at this point. */
 943         struct ethhdr *eh = (struct ethhdr *) skb->data;
 944         int i;
 945
 946         /* Exact match */
 947         for (i = 0; i < filter->count; i++)
 948                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 949                         return 1;
 950
 951         /* Inexact match (multicast only) */
 952         if (is_multicast_ether_addr(eh->h_dest))
 953                 return addr_hash_test(filter->mask, eh->h_dest);
 954
 955         return 0;
 956 }
 957
 958 /*
 959  * Checks whether the packet is accepted or not.
 960  * Returns: 0 - drop, !=0 - accept
 961  */
 962 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 963 {
 964         if (!filter->count)
 965                 return 1;
 966
 967         return run_filter(filter, skb);
 968 }
 969
 970 /* Network device part of the driver */
 971
 972 static const struct ethtool_ops tun_ethtool_ops;
 973
 974 static int tun_net_init(struct net_device *dev)
 975 {
 976         struct tun_struct *tun = netdev_priv(dev);
 977         struct ifreq *ifr = tun->ifr;
 978         int err;
 979
 980         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 981         if (!dev->tstats)
 982                 return -ENOMEM;
 983
 984         spin_lock_init(&tun->lock);
 985
 986         err = security_tun_dev_alloc_security(&tun->security);
 987         if (err < 0) {
 988                 free_percpu(dev->tstats);
 989                 return err;
 990         }
 991
 992         tun_flow_init(tun);
 993
 994         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 995                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 996                            NETIF_F_HW_VLAN_STAG_TX;
 997         dev->features = dev->hw_features | NETIF_F_LLTX;
 998         dev->vlan_features = dev->features &
 999                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1000                                NETIF_F_HW_VLAN_STAG_TX);
1001
1002         tun->flags = (tun->flags & ~TUN_FEATURES) |
1003                       (ifr->ifr_flags & TUN_FEATURES);
1004
1005         INIT_LIST_HEAD(&tun->disabled);
1006         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1007                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1008         if (err < 0) {
1009                 tun_flow_uninit(tun);
1010                 security_tun_dev_free_security(tun->security);
1011                 free_percpu(dev->tstats);
1012                 return err;
1013         }
1014         return 0;
1015 }
1016
1017 /* Net device detach from fd. */
1018 static void tun_net_uninit(struct net_device *dev)
1019 {
1020         tun_detach_all(dev);
1021 }
1022
1023 /* Net device open. */
1024 static int tun_net_open(struct net_device *dev)
1025 {
1026         netif_tx_start_all_queues(dev);
1027
1028         return 0;
1029 }
1030
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1033 {
1034         netif_tx_stop_all_queues(dev);
1035         return 0;
1036 }
1037
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 {
1041 #ifdef CONFIG_RPS
1042         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1043                 /* Select queue was not called for the skbuff, so we extract the
1044                  * RPS hash and save it into the flow_table here.
1045                  */
1046                 struct tun_flow_entry *e;
1047                 __u32 rxhash;
1048
1049                 rxhash = __skb_get_hash_symmetric(skb);
1050                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1051                 if (e)
1052                         tun_flow_save_rps_rxhash(e, rxhash);
1053         }
1054 #endif
1055 }
1056
1057 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1058                                     struct sk_buff *skb,
1059                                     int len)
1060 {
1061         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1062
1063         if (prog)
1064                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1065
1066         return len;
1067 }
1068
1069 /* Net device start xmit */
1070 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1071 {
1072         struct tun_struct *tun = netdev_priv(dev);
1073         enum skb_drop_reason drop_reason;
1074         int txq = skb->queue_mapping;
1075         struct netdev_queue *queue;
1076         struct tun_file *tfile;
1077         int len = skb->len;
1078
1079         rcu_read_lock();
1080         tfile = rcu_dereference(tun->tfiles[txq]);
1081
1082         /* Drop packet if interface is not attached */
1083         if (!tfile) {
1084                 drop_reason = SKB_DROP_REASON_DEV_READY;
1085                 goto drop;
1086         }
1087
1088         if (!rcu_dereference(tun->steering_prog))
1089                 tun_automq_xmit(tun, skb);
1090
1091         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1092
1093         /* Drop if the filter does not like it.
1094          * This is a noop if the filter is disabled.
1095          * Filter can be enabled only for the TAP devices. */
1096         if (!check_filter(&tun->txflt, skb)) {
1097                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1098                 goto drop;
1099         }
1100
1101         if (tfile->socket.sk->sk_filter &&
1102             sk_filter(tfile->socket.sk, skb)) {
1103                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1104                 goto drop;
1105         }
1106
1107         len = run_ebpf_filter(tun, skb, len);
1108         if (len == 0) {
1109                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1110                 goto drop;
1111         }
1112
1113         if (pskb_trim(skb, len)) {
1114                 drop_reason = SKB_DROP_REASON_NOMEM;
1115                 goto drop;
1116         }
1117
1118         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1119                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1120                 goto drop;
1121         }
1122
1123         skb_tx_timestamp(skb);
1124
1125         /* Orphan the skb - required as we might hang on to it
1126          * for indefinite time.
1127          */
1128         skb_orphan(skb);
1129
1130         nf_reset_ct(skb);
1131
1132         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1133                 drop_reason = SKB_DROP_REASON_FULL_RING;
1134                 goto drop;
1135         }
1136
1137         /* NETIF_F_LLTX requires to do our own update of trans_start */
1138         queue = netdev_get_tx_queue(dev, txq);
1139         txq_trans_cond_update(queue);
1140
1141         /* Notify and wake up reader process */
1142         if (tfile->flags & TUN_FASYNC)
1143                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1144         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1145
1146         rcu_read_unlock();
1147         return NETDEV_TX_OK;
1148
1149 drop:
1150         dev_core_stats_tx_dropped_inc(dev);
1151         skb_tx_error(skb);
1152         kfree_skb_reason(skb, drop_reason);
1153         rcu_read_unlock();
1154         return NET_XMIT_DROP;
1155 }
1156
1157 static void tun_net_mclist(struct net_device *dev)
1158 {
1159         /*
1160          * This callback is supposed to deal with mc filter in
1161          * _rx_ path and has nothing to do with the _tx_ path.
1162          * In rx path we always accept everything userspace gives us.
1163          */
1164 }
1165
1166 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1167         netdev_features_t features)
1168 {
1169         struct tun_struct *tun = netdev_priv(dev);
1170
1171         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1172 }
1173
1174 static void tun_set_headroom(struct net_device *dev, int new_hr)
1175 {
1176         struct tun_struct *tun = netdev_priv(dev);
1177
1178         if (new_hr < NET_SKB_PAD)
1179                 new_hr = NET_SKB_PAD;
1180
1181         tun->align = new_hr;
1182 }
1183
1184 static void
1185 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1186 {
1187         struct tun_struct *tun = netdev_priv(dev);
1188
1189         dev_get_tstats64(dev, stats);
1190
1191         stats->rx_frame_errors +=
1192                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1193 }
1194
1195 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1196                        struct netlink_ext_ack *extack)
1197 {
1198         struct tun_struct *tun = netdev_priv(dev);
1199         struct tun_file *tfile;
1200         struct bpf_prog *old_prog;
1201         int i;
1202
1203         old_prog = rtnl_dereference(tun->xdp_prog);
1204         rcu_assign_pointer(tun->xdp_prog, prog);
1205         if (old_prog)
1206                 bpf_prog_put(old_prog);
1207
1208         for (i = 0; i < tun->numqueues; i++) {
1209                 tfile = rtnl_dereference(tun->tfiles[i]);
1210                 if (prog)
1211                         sock_set_flag(&tfile->sk, SOCK_XDP);
1212                 else
1213                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1214         }
1215         list_for_each_entry(tfile, &tun->disabled, next) {
1216                 if (prog)
1217                         sock_set_flag(&tfile->sk, SOCK_XDP);
1218                 else
1219                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1220         }
1221
1222         return 0;
1223 }
1224
1225 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1226 {
1227         switch (xdp->command) {
1228         case XDP_SETUP_PROG:
1229                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1230         default:
1231                 return -EINVAL;
1232         }
1233 }
1234
1235 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1236 {
1237         if (new_carrier) {
1238                 struct tun_struct *tun = netdev_priv(dev);
1239
1240                 if (!tun->numqueues)
1241                         return -EPERM;
1242
1243                 netif_carrier_on(dev);
1244         } else {
1245                 netif_carrier_off(dev);
1246         }
1247         return 0;
1248 }
1249
1250 static const struct net_device_ops tun_netdev_ops = {
1251         .ndo_init               = tun_net_init,
1252         .ndo_uninit             = tun_net_uninit,
1253         .ndo_open               = tun_net_open,
1254         .ndo_stop               = tun_net_close,
1255         .ndo_start_xmit         = tun_net_xmit,
1256         .ndo_fix_features       = tun_net_fix_features,
1257         .ndo_select_queue       = tun_select_queue,
1258         .ndo_set_rx_headroom    = tun_set_headroom,
1259         .ndo_get_stats64        = tun_net_get_stats64,
1260         .ndo_change_carrier     = tun_net_change_carrier,
1261 };
1262
1263 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1264 {
1265         /* Notify and wake up reader process */
1266         if (tfile->flags & TUN_FASYNC)
1267                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1268         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1269 }
1270
1271 static int tun_xdp_xmit(struct net_device *dev, int n,
1272                         struct xdp_frame **frames, u32 flags)
1273 {
1274         struct tun_struct *tun = netdev_priv(dev);
1275         struct tun_file *tfile;
1276         u32 numqueues;
1277         int nxmit = 0;
1278         int i;
1279
1280         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1281                 return -EINVAL;
1282
1283         rcu_read_lock();
1284
1285 resample:
1286         numqueues = READ_ONCE(tun->numqueues);
1287         if (!numqueues) {
1288                 rcu_read_unlock();
1289                 return -ENXIO; /* Caller will free/return all frames */
1290         }
1291
1292         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1293                                             numqueues]);
1294         if (unlikely(!tfile))
1295                 goto resample;
1296
1297         spin_lock(&tfile->tx_ring.producer_lock);
1298         for (i = 0; i < n; i++) {
1299                 struct xdp_frame *xdp = frames[i];
1300                 /* Encode the XDP flag into lowest bit for consumer to differ
1301                  * XDP buffer from sk_buff.
1302                  */
1303                 void *frame = tun_xdp_to_ptr(xdp);
1304
1305                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1306                         dev_core_stats_tx_dropped_inc(dev);
1307                         break;
1308                 }
1309                 nxmit++;
1310         }
1311         spin_unlock(&tfile->tx_ring.producer_lock);
1312
1313         if (flags & XDP_XMIT_FLUSH)
1314                 __tun_xdp_flush_tfile(tfile);
1315
1316         rcu_read_unlock();
1317         return nxmit;
1318 }
1319
1320 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1321 {
1322         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1323         int nxmit;
1324
1325         if (unlikely(!frame))
1326                 return -EOVERFLOW;
1327
1328         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1329         if (!nxmit)
1330                 xdp_return_frame_rx_napi(frame);
1331         return nxmit;
1332 }
1333
1334 static const struct net_device_ops tap_netdev_ops = {
1335         .ndo_init               = tun_net_init,
1336         .ndo_uninit             = tun_net_uninit,
1337         .ndo_open               = tun_net_open,
1338         .ndo_stop               = tun_net_close,
1339         .ndo_start_xmit         = tun_net_xmit,
1340         .ndo_fix_features       = tun_net_fix_features,
1341         .ndo_set_rx_mode        = tun_net_mclist,
1342         .ndo_set_mac_address    = eth_mac_addr,
1343         .ndo_validate_addr      = eth_validate_addr,
1344         .ndo_select_queue       = tun_select_queue,
1345         .ndo_features_check     = passthru_features_check,
1346         .ndo_set_rx_headroom    = tun_set_headroom,
1347         .ndo_get_stats64        = dev_get_tstats64,
1348         .ndo_bpf                = tun_xdp,
1349         .ndo_xdp_xmit           = tun_xdp_xmit,
1350         .ndo_change_carrier     = tun_net_change_carrier,
1351 };
1352
1353 static void tun_flow_init(struct tun_struct *tun)
1354 {
1355         int i;
1356
1357         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1358                 INIT_HLIST_HEAD(&tun->flows[i]);
1359
1360         tun->ageing_time = TUN_FLOW_EXPIRE;
1361         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1362         mod_timer(&tun->flow_gc_timer,
1363                   round_jiffies_up(jiffies + tun->ageing_time));
1364 }
1365
1366 static void tun_flow_uninit(struct tun_struct *tun)
1367 {
1368         del_timer_sync(&tun->flow_gc_timer);
1369         tun_flow_flush(tun);
1370 }
1371
1372 #define MIN_MTU 68
1373 #define MAX_MTU 65535
1374
1375 /* Initialize net device. */
1376 static void tun_net_initialize(struct net_device *dev)
1377 {
1378         struct tun_struct *tun = netdev_priv(dev);
1379
1380         switch (tun->flags & TUN_TYPE_MASK) {
1381         case IFF_TUN:
1382                 dev->netdev_ops = &tun_netdev_ops;
1383                 dev->header_ops = &ip_tunnel_header_ops;
1384
1385                 /* Point-to-Point TUN Device */
1386                 dev->hard_header_len = 0;
1387                 dev->addr_len = 0;
1388                 dev->mtu = 1500;
1389
1390                 /* Zero header length */
1391                 dev->type = ARPHRD_NONE;
1392                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1393                 break;
1394
1395         case IFF_TAP:
1396                 dev->netdev_ops = &tap_netdev_ops;
1397                 /* Ethernet TAP Device */
1398                 ether_setup(dev);
1399                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1400                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1401
1402                 eth_hw_addr_random(dev);
1403
1404                 break;
1405         }
1406
1407         dev->min_mtu = MIN_MTU;
1408         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1409 }
1410
1411 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1412 {
1413         struct sock *sk = tfile->socket.sk;
1414
1415         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1416 }
1417
1418 /* Character device part */
1419
1420 /* Poll */
1421 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1422 {
1423         struct tun_file *tfile = file->private_data;
1424         struct tun_struct *tun = tun_get(tfile);
1425         struct sock *sk;
1426         __poll_t mask = 0;
1427
1428         if (!tun)
1429                 return EPOLLERR;
1430
1431         sk = tfile->socket.sk;
1432
1433         poll_wait(file, sk_sleep(sk), wait);
1434
1435         if (!ptr_ring_empty(&tfile->tx_ring))
1436                 mask |= EPOLLIN | EPOLLRDNORM;
1437
1438         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1439          * guarantee EPOLLOUT to be raised by either here or
1440          * tun_sock_write_space(). Then process could get notification
1441          * after it writes to a down device and meets -EIO.
1442          */
1443         if (tun_sock_writeable(tun, tfile) ||
1444             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1445              tun_sock_writeable(tun, tfile)))
1446                 mask |= EPOLLOUT | EPOLLWRNORM;
1447
1448         if (tun->dev->reg_state != NETREG_REGISTERED)
1449                 mask = EPOLLERR;
1450
1451         tun_put(tun);
1452         return mask;
1453 }
1454
1455 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1456                                             size_t len,
1457                                             const struct iov_iter *it)
1458 {
1459         struct sk_buff *skb;
1460         size_t linear;
1461         int err;
1462         int i;
1463
1464         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1465             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1466                 return ERR_PTR(-EMSGSIZE);
1467
1468         local_bh_disable();
1469         skb = napi_get_frags(&tfile->napi);
1470         local_bh_enable();
1471         if (!skb)
1472                 return ERR_PTR(-ENOMEM);
1473
1474         linear = iov_iter_single_seg_count(it);
1475         err = __skb_grow(skb, linear);
1476         if (err)
1477                 goto free;
1478
1479         skb->len = len;
1480         skb->data_len = len - linear;
1481         skb->truesize += skb->data_len;
1482
1483         for (i = 1; i < it->nr_segs; i++) {
1484                 size_t fragsz = it->iov[i].iov_len;
1485                 struct page *page;
1486                 void *frag;
1487
1488                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1489                         err = -EINVAL;
1490                         goto free;
1491                 }
1492                 frag = netdev_alloc_frag(fragsz);
1493                 if (!frag) {
1494                         err = -ENOMEM;
1495                         goto free;
1496                 }
1497                 page = virt_to_head_page(frag);
1498                 skb_fill_page_desc(skb, i - 1, page,
1499                                    frag - page_address(page), fragsz);
1500         }
1501
1502         return skb;
1503 free:
1504         /* frees skb and all frags allocated with napi_alloc_frag() */
1505         napi_free_frags(&tfile->napi);
1506         return ERR_PTR(err);
1507 }
1508
1509 /* prepad is the amount to reserve at front.  len is length after that.
1510  * linear is a hint as to how much to copy (usually headers). */
1511 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1512                                      size_t prepad, size_t len,
1513                                      size_t linear, int noblock)
1514 {
1515         struct sock *sk = tfile->socket.sk;
1516         struct sk_buff *skb;
1517         int err;
1518
1519         /* Under a page?  Don't bother with paged skb. */
1520         if (prepad + len < PAGE_SIZE || !linear)
1521                 linear = len;
1522
1523         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1524                                    &err, 0);
1525         if (!skb)
1526                 return ERR_PTR(err);
1527
1528         skb_reserve(skb, prepad);
1529         skb_put(skb, linear);
1530         skb->data_len = len - linear;
1531         skb->len += len - linear;
1532
1533         return skb;
1534 }
1535
1536 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1537                            struct sk_buff *skb, int more)
1538 {
1539         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1540         struct sk_buff_head process_queue;
1541         u32 rx_batched = tun->rx_batched;
1542         bool rcv = false;
1543
1544         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1545                 local_bh_disable();
1546                 skb_record_rx_queue(skb, tfile->queue_index);
1547                 netif_receive_skb(skb);
1548                 local_bh_enable();
1549                 return;
1550         }
1551
1552         spin_lock(&queue->lock);
1553         if (!more || skb_queue_len(queue) == rx_batched) {
1554                 __skb_queue_head_init(&process_queue);
1555                 skb_queue_splice_tail_init(queue, &process_queue);
1556                 rcv = true;
1557         } else {
1558                 __skb_queue_tail(queue, skb);
1559         }
1560         spin_unlock(&queue->lock);
1561
1562         if (rcv) {
1563                 struct sk_buff *nskb;
1564
1565                 local_bh_disable();
1566                 while ((nskb = __skb_dequeue(&process_queue))) {
1567                         skb_record_rx_queue(nskb, tfile->queue_index);
1568                         netif_receive_skb(nskb);
1569                 }
1570                 skb_record_rx_queue(skb, tfile->queue_index);
1571                 netif_receive_skb(skb);
1572                 local_bh_enable();
1573         }
1574 }
1575
1576 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1577                               int len, int noblock, bool zerocopy)
1578 {
1579         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1580                 return false;
1581
1582         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1583                 return false;
1584
1585         if (!noblock)
1586                 return false;
1587
1588         if (zerocopy)
1589                 return false;
1590
1591         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1592             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1593                 return false;
1594
1595         return true;
1596 }
1597
1598 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1599                                        struct page_frag *alloc_frag, char *buf,
1600                                        int buflen, int len, int pad)
1601 {
1602         struct sk_buff *skb = build_skb(buf, buflen);
1603
1604         if (!skb)
1605                 return ERR_PTR(-ENOMEM);
1606
1607         skb_reserve(skb, pad);
1608         skb_put(skb, len);
1609         skb_set_owner_w(skb, tfile->socket.sk);
1610
1611         get_page(alloc_frag->page);
1612         alloc_frag->offset += buflen;
1613
1614         return skb;
1615 }
1616
1617 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1618                        struct xdp_buff *xdp, u32 act)
1619 {
1620         int err;
1621
1622         switch (act) {
1623         case XDP_REDIRECT:
1624                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1625                 if (err)
1626                         return err;
1627                 break;
1628         case XDP_TX:
1629                 err = tun_xdp_tx(tun->dev, xdp);
1630                 if (err < 0)
1631                         return err;
1632                 break;
1633         case XDP_PASS:
1634                 break;
1635         default:
1636                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1637                 fallthrough;
1638         case XDP_ABORTED:
1639                 trace_xdp_exception(tun->dev, xdp_prog, act);
1640                 fallthrough;
1641         case XDP_DROP:
1642                 dev_core_stats_rx_dropped_inc(tun->dev);
1643                 break;
1644         }
1645
1646         return act;
1647 }
1648
1649 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1650                                      struct tun_file *tfile,
1651                                      struct iov_iter *from,
1652                                      struct virtio_net_hdr *hdr,
1653                                      int len, int *skb_xdp)
1654 {
1655         struct page_frag *alloc_frag = &current->task_frag;
1656         struct bpf_prog *xdp_prog;
1657         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1658         char *buf;
1659         size_t copied;
1660         int pad = TUN_RX_PAD;
1661         int err = 0;
1662
1663         rcu_read_lock();
1664         xdp_prog = rcu_dereference(tun->xdp_prog);
1665         if (xdp_prog)
1666                 pad += XDP_PACKET_HEADROOM;
1667         buflen += SKB_DATA_ALIGN(len + pad);
1668         rcu_read_unlock();
1669
1670         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1671         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1672                 return ERR_PTR(-ENOMEM);
1673
1674         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1675         copied = copy_page_from_iter(alloc_frag->page,
1676                                      alloc_frag->offset + pad,
1677                                      len, from);
1678         if (copied != len)
1679                 return ERR_PTR(-EFAULT);
1680
1681         /* There's a small window that XDP may be set after the check
1682          * of xdp_prog above, this should be rare and for simplicity
1683          * we do XDP on skb in case the headroom is not enough.
1684          */
1685         if (hdr->gso_type || !xdp_prog) {
1686                 *skb_xdp = 1;
1687                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1688                                        pad);
1689         }
1690
1691         *skb_xdp = 0;
1692
1693         local_bh_disable();
1694         rcu_read_lock();
1695         xdp_prog = rcu_dereference(tun->xdp_prog);
1696         if (xdp_prog) {
1697                 struct xdp_buff xdp;
1698                 u32 act;
1699
1700                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1701                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1702
1703                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1704                 if (act == XDP_REDIRECT || act == XDP_TX) {
1705                         get_page(alloc_frag->page);
1706                         alloc_frag->offset += buflen;
1707                 }
1708                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1709                 if (err < 0) {
1710                         if (act == XDP_REDIRECT || act == XDP_TX)
1711                                 put_page(alloc_frag->page);
1712                         goto out;
1713                 }
1714
1715                 if (err == XDP_REDIRECT)
1716                         xdp_do_flush();
1717                 if (err != XDP_PASS)
1718                         goto out;
1719
1720                 pad = xdp.data - xdp.data_hard_start;
1721                 len = xdp.data_end - xdp.data;
1722         }
1723         rcu_read_unlock();
1724         local_bh_enable();
1725
1726         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1727
1728 out:
1729         rcu_read_unlock();
1730         local_bh_enable();
1731         return NULL;
1732 }
1733
1734 /* Get packet from user space buffer */
1735 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1736                             void *msg_control, struct iov_iter *from,
1737                             int noblock, bool more)
1738 {
1739         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1740         struct sk_buff *skb;
1741         size_t total_len = iov_iter_count(from);
1742         size_t len = total_len, align = tun->align, linear;
1743         struct virtio_net_hdr gso = { 0 };
1744         int good_linear;
1745         int copylen;
1746         bool zerocopy = false;
1747         int err;
1748         u32 rxhash = 0;
1749         int skb_xdp = 1;
1750         bool frags = tun_napi_frags_enabled(tfile);
1751         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1752
1753         if (!(tun->flags & IFF_NO_PI)) {
1754                 if (len < sizeof(pi))
1755                         return -EINVAL;
1756                 len -= sizeof(pi);
1757
1758                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1759                         return -EFAULT;
1760         }
1761
1762         if (tun->flags & IFF_VNET_HDR) {
1763                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1764
1765                 if (len < vnet_hdr_sz)
1766                         return -EINVAL;
1767                 len -= vnet_hdr_sz;
1768
1769                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1770                         return -EFAULT;
1771
1772                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1773                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1774                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1775
1776                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1777                         return -EINVAL;
1778                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1779         }
1780
1781         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1782                 align += NET_IP_ALIGN;
1783                 if (unlikely(len < ETH_HLEN ||
1784                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1785                         return -EINVAL;
1786         }
1787
1788         good_linear = SKB_MAX_HEAD(align);
1789
1790         if (msg_control) {
1791                 struct iov_iter i = *from;
1792
1793                 /* There are 256 bytes to be copied in skb, so there is
1794                  * enough room for skb expand head in case it is used.
1795                  * The rest of the buffer is mapped from userspace.
1796                  */
1797                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1798                 if (copylen > good_linear)
1799                         copylen = good_linear;
1800                 linear = copylen;
1801                 iov_iter_advance(&i, copylen);
1802                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1803                         zerocopy = true;
1804         }
1805
1806         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1807                 /* For the packet that is not easy to be processed
1808                  * (e.g gso or jumbo packet), we will do it at after
1809                  * skb was created with generic XDP routine.
1810                  */
1811                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1812                 err = PTR_ERR_OR_ZERO(skb);
1813                 if (err)
1814                         goto drop;
1815                 if (!skb)
1816                         return total_len;
1817         } else {
1818                 if (!zerocopy) {
1819                         copylen = len;
1820                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1821                                 linear = good_linear;
1822                         else
1823                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1824                 }
1825
1826                 if (frags) {
1827                         mutex_lock(&tfile->napi_mutex);
1828                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1829                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1830                          * If zerocopy is enabled, then this layout will be
1831                          * overwritten by zerocopy_sg_from_iter().
1832                          */
1833                         zerocopy = false;
1834                 } else {
1835                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1836                                             noblock);
1837                 }
1838
1839                 err = PTR_ERR_OR_ZERO(skb);
1840                 if (err)
1841                         goto drop;
1842
1843                 if (zerocopy)
1844                         err = zerocopy_sg_from_iter(skb, from);
1845                 else
1846                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1847
1848                 if (err) {
1849                         err = -EFAULT;
1850                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1851                         goto drop;
1852                 }
1853         }
1854
1855         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1856                 atomic_long_inc(&tun->rx_frame_errors);
1857                 err = -EINVAL;
1858                 goto free_skb;
1859         }
1860
1861         switch (tun->flags & TUN_TYPE_MASK) {
1862         case IFF_TUN:
1863                 if (tun->flags & IFF_NO_PI) {
1864                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1865
1866                         switch (ip_version) {
1867                         case 4:
1868                                 pi.proto = htons(ETH_P_IP);
1869                                 break;
1870                         case 6:
1871                                 pi.proto = htons(ETH_P_IPV6);
1872                                 break;
1873                         default:
1874                                 err = -EINVAL;
1875                                 goto drop;
1876                         }
1877                 }
1878
1879                 skb_reset_mac_header(skb);
1880                 skb->protocol = pi.proto;
1881                 skb->dev = tun->dev;
1882                 break;
1883         case IFF_TAP:
1884                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1885                         err = -ENOMEM;
1886                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1887                         goto drop;
1888                 }
1889                 skb->protocol = eth_type_trans(skb, tun->dev);
1890                 break;
1891         }
1892
1893         /* copy skb_ubuf_info for callback when skb has no error */
1894         if (zerocopy) {
1895                 skb_zcopy_init(skb, msg_control);
1896         } else if (msg_control) {
1897                 struct ubuf_info *uarg = msg_control;
1898                 uarg->callback(NULL, uarg, false);
1899         }
1900
1901         skb_reset_network_header(skb);
1902         skb_probe_transport_header(skb);
1903         skb_record_rx_queue(skb, tfile->queue_index);
1904
1905         if (skb_xdp) {
1906                 struct bpf_prog *xdp_prog;
1907                 int ret;
1908
1909                 local_bh_disable();
1910                 rcu_read_lock();
1911                 xdp_prog = rcu_dereference(tun->xdp_prog);
1912                 if (xdp_prog) {
1913                         ret = do_xdp_generic(xdp_prog, skb);
1914                         if (ret != XDP_PASS) {
1915                                 rcu_read_unlock();
1916                                 local_bh_enable();
1917                                 goto unlock_frags;
1918                         }
1919                 }
1920                 rcu_read_unlock();
1921                 local_bh_enable();
1922         }
1923
1924         /* Compute the costly rx hash only if needed for flow updates.
1925          * We may get a very small possibility of OOO during switching, not
1926          * worth to optimize.
1927          */
1928         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1929             !tfile->detached)
1930                 rxhash = __skb_get_hash_symmetric(skb);
1931
1932         rcu_read_lock();
1933         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1934                 err = -EIO;
1935                 rcu_read_unlock();
1936                 drop_reason = SKB_DROP_REASON_DEV_READY;
1937                 goto drop;
1938         }
1939
1940         if (frags) {
1941                 u32 headlen;
1942
1943                 /* Exercise flow dissector code path. */
1944                 skb_push(skb, ETH_HLEN);
1945                 headlen = eth_get_headlen(tun->dev, skb->data,
1946                                           skb_headlen(skb));
1947
1948                 if (unlikely(headlen > skb_headlen(skb))) {
1949                         WARN_ON_ONCE(1);
1950                         err = -ENOMEM;
1951                         dev_core_stats_rx_dropped_inc(tun->dev);
1952 napi_busy:
1953                         napi_free_frags(&tfile->napi);
1954                         rcu_read_unlock();
1955                         mutex_unlock(&tfile->napi_mutex);
1956                         return err;
1957                 }
1958
1959                 if (likely(napi_schedule_prep(&tfile->napi))) {
1960                         local_bh_disable();
1961                         napi_gro_frags(&tfile->napi);
1962                         napi_complete(&tfile->napi);
1963                         local_bh_enable();
1964                 } else {
1965                         err = -EBUSY;
1966                         goto napi_busy;
1967                 }
1968                 mutex_unlock(&tfile->napi_mutex);
1969         } else if (tfile->napi_enabled) {
1970                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1971                 int queue_len;
1972
1973                 spin_lock_bh(&queue->lock);
1974                 __skb_queue_tail(queue, skb);
1975                 queue_len = skb_queue_len(queue);
1976                 spin_unlock(&queue->lock);
1977
1978                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1979                         napi_schedule(&tfile->napi);
1980
1981                 local_bh_enable();
1982         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1983                 tun_rx_batched(tun, tfile, skb, more);
1984         } else {
1985                 netif_rx(skb);
1986         }
1987         rcu_read_unlock();
1988
1989         preempt_disable();
1990         dev_sw_netstats_rx_add(tun->dev, len);
1991         preempt_enable();
1992
1993         if (rxhash)
1994                 tun_flow_update(tun, rxhash, tfile);
1995
1996         return total_len;
1997
1998 drop:
1999         if (err != -EAGAIN)
2000                 dev_core_stats_rx_dropped_inc(tun->dev);
2001
2002 free_skb:
2003         if (!IS_ERR_OR_NULL(skb))
2004                 kfree_skb_reason(skb, drop_reason);
2005
2006 unlock_frags:
2007         if (frags) {
2008                 tfile->napi.skb = NULL;
2009                 mutex_unlock(&tfile->napi_mutex);
2010         }
2011
2012         return err ?: total_len;
2013 }
2014
2015 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2016 {
2017         struct file *file = iocb->ki_filp;
2018         struct tun_file *tfile = file->private_data;
2019         struct tun_struct *tun = tun_get(tfile);
2020         ssize_t result;
2021         int noblock = 0;
2022
2023         if (!tun)
2024                 return -EBADFD;
2025
2026         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2027                 noblock = 1;
2028
2029         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2030
2031         tun_put(tun);
2032         return result;
2033 }
2034
2035 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2036                                 struct tun_file *tfile,
2037                                 struct xdp_frame *xdp_frame,
2038                                 struct iov_iter *iter)
2039 {
2040         int vnet_hdr_sz = 0;
2041         size_t size = xdp_frame->len;
2042         size_t ret;
2043
2044         if (tun->flags & IFF_VNET_HDR) {
2045                 struct virtio_net_hdr gso = { 0 };
2046
2047                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2048                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2049                         return -EINVAL;
2050                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2051                              sizeof(gso)))
2052                         return -EFAULT;
2053                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2054         }
2055
2056         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2057
2058         preempt_disable();
2059         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2060         preempt_enable();
2061
2062         return ret;
2063 }
2064
2065 /* Put packet to the user space buffer */
2066 static ssize_t tun_put_user(struct tun_struct *tun,
2067                             struct tun_file *tfile,
2068                             struct sk_buff *skb,
2069                             struct iov_iter *iter)
2070 {
2071         struct tun_pi pi = { 0, skb->protocol };
2072         ssize_t total;
2073         int vlan_offset = 0;
2074         int vlan_hlen = 0;
2075         int vnet_hdr_sz = 0;
2076
2077         if (skb_vlan_tag_present(skb))
2078                 vlan_hlen = VLAN_HLEN;
2079
2080         if (tun->flags & IFF_VNET_HDR)
2081                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2082
2083         total = skb->len + vlan_hlen + vnet_hdr_sz;
2084
2085         if (!(tun->flags & IFF_NO_PI)) {
2086                 if (iov_iter_count(iter) < sizeof(pi))
2087                         return -EINVAL;
2088
2089                 total += sizeof(pi);
2090                 if (iov_iter_count(iter) < total) {
2091                         /* Packet will be striped */
2092                         pi.flags |= TUN_PKT_STRIP;
2093                 }
2094
2095                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2096                         return -EFAULT;
2097         }
2098
2099         if (vnet_hdr_sz) {
2100                 struct virtio_net_hdr gso;
2101
2102                 if (iov_iter_count(iter) < vnet_hdr_sz)
2103                         return -EINVAL;
2104
2105                 if (virtio_net_hdr_from_skb(skb, &gso,
2106                                             tun_is_little_endian(tun), true,
2107                                             vlan_hlen)) {
2108                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2109                         pr_err("unexpected GSO type: "
2110                                "0x%x, gso_size %d, hdr_len %d\n",
2111                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2112                                tun16_to_cpu(tun, gso.hdr_len));
2113                         print_hex_dump(KERN_ERR, "tun: ",
2114                                        DUMP_PREFIX_NONE,
2115                                        16, 1, skb->head,
2116                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2117                         WARN_ON_ONCE(1);
2118                         return -EINVAL;
2119                 }
2120
2121                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2122                         return -EFAULT;
2123
2124                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2125         }
2126
2127         if (vlan_hlen) {
2128                 int ret;
2129                 struct veth veth;
2130
2131                 veth.h_vlan_proto = skb->vlan_proto;
2132                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2133
2134                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2135
2136                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2137                 if (ret || !iov_iter_count(iter))
2138                         goto done;
2139
2140                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2141                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2142                         goto done;
2143         }
2144
2145         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2146
2147 done:
2148         /* caller is in process context, */
2149         preempt_disable();
2150         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2151         preempt_enable();
2152
2153         return total;
2154 }
2155
2156 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2157 {
2158         DECLARE_WAITQUEUE(wait, current);
2159         void *ptr = NULL;
2160         int error = 0;
2161
2162         ptr = ptr_ring_consume(&tfile->tx_ring);
2163         if (ptr)
2164                 goto out;
2165         if (noblock) {
2166                 error = -EAGAIN;
2167                 goto out;
2168         }
2169
2170         add_wait_queue(&tfile->socket.wq.wait, &wait);
2171
2172         while (1) {
2173                 set_current_state(TASK_INTERRUPTIBLE);
2174                 ptr = ptr_ring_consume(&tfile->tx_ring);
2175                 if (ptr)
2176                         break;
2177                 if (signal_pending(current)) {
2178                         error = -ERESTARTSYS;
2179                         break;
2180                 }
2181                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2182                         error = -EFAULT;
2183                         break;
2184                 }
2185
2186                 schedule();
2187         }
2188
2189         __set_current_state(TASK_RUNNING);
2190         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2191
2192 out:
2193         *err = error;
2194         return ptr;
2195 }
2196
2197 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2198                            struct iov_iter *to,
2199                            int noblock, void *ptr)
2200 {
2201         ssize_t ret;
2202         int err;
2203
2204         if (!iov_iter_count(to)) {
2205                 tun_ptr_free(ptr);
2206                 return 0;
2207         }
2208
2209         if (!ptr) {
2210                 /* Read frames from ring */
2211                 ptr = tun_ring_recv(tfile, noblock, &err);
2212                 if (!ptr)
2213                         return err;
2214         }
2215
2216         if (tun_is_xdp_frame(ptr)) {
2217                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2218
2219                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2220                 xdp_return_frame(xdpf);
2221         } else {
2222                 struct sk_buff *skb = ptr;
2223
2224                 ret = tun_put_user(tun, tfile, skb, to);
2225                 if (unlikely(ret < 0))
2226                         kfree_skb(skb);
2227                 else
2228                         consume_skb(skb);
2229         }
2230
2231         return ret;
2232 }
2233
2234 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2235 {
2236         struct file *file = iocb->ki_filp;
2237         struct tun_file *tfile = file->private_data;
2238         struct tun_struct *tun = tun_get(tfile);
2239         ssize_t len = iov_iter_count(to), ret;
2240         int noblock = 0;
2241
2242         if (!tun)
2243                 return -EBADFD;
2244
2245         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2246                 noblock = 1;
2247
2248         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2249         ret = min_t(ssize_t, ret, len);
2250         if (ret > 0)
2251                 iocb->ki_pos = ret;
2252         tun_put(tun);
2253         return ret;
2254 }
2255
2256 static void tun_prog_free(struct rcu_head *rcu)
2257 {
2258         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2259
2260         bpf_prog_destroy(prog->prog);
2261         kfree(prog);
2262 }
2263
2264 static int __tun_set_ebpf(struct tun_struct *tun,
2265                           struct tun_prog __rcu **prog_p,
2266                           struct bpf_prog *prog)
2267 {
2268         struct tun_prog *old, *new = NULL;
2269
2270         if (prog) {
2271                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2272                 if (!new)
2273                         return -ENOMEM;
2274                 new->prog = prog;
2275         }
2276
2277         spin_lock_bh(&tun->lock);
2278         old = rcu_dereference_protected(*prog_p,
2279                                         lockdep_is_held(&tun->lock));
2280         rcu_assign_pointer(*prog_p, new);
2281         spin_unlock_bh(&tun->lock);
2282
2283         if (old)
2284                 call_rcu(&old->rcu, tun_prog_free);
2285
2286         return 0;
2287 }
2288
2289 static void tun_free_netdev(struct net_device *dev)
2290 {
2291         struct tun_struct *tun = netdev_priv(dev);
2292
2293         BUG_ON(!(list_empty(&tun->disabled)));
2294
2295         free_percpu(dev->tstats);
2296         tun_flow_uninit(tun);
2297         security_tun_dev_free_security(tun->security);
2298         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2299         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2300 }
2301
2302 static void tun_setup(struct net_device *dev)
2303 {
2304         struct tun_struct *tun = netdev_priv(dev);
2305
2306         tun->owner = INVALID_UID;
2307         tun->group = INVALID_GID;
2308         tun_default_link_ksettings(dev, &tun->link_ksettings);
2309
2310         dev->ethtool_ops = &tun_ethtool_ops;
2311         dev->needs_free_netdev = true;
2312         dev->priv_destructor = tun_free_netdev;
2313         /* We prefer our own queue length */
2314         dev->tx_queue_len = TUN_READQ_SIZE;
2315 }
2316
2317 /* Trivial set of netlink ops to allow deleting tun or tap
2318  * device with netlink.
2319  */
2320 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2321                         struct netlink_ext_ack *extack)
2322 {
2323         NL_SET_ERR_MSG(extack,
2324                        "tun/tap creation via rtnetlink is not supported.");
2325         return -EOPNOTSUPP;
2326 }
2327
2328 static size_t tun_get_size(const struct net_device *dev)
2329 {
2330         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2331         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2332
2333         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2334                nla_total_size(sizeof(gid_t)) + /* GROUP */
2335                nla_total_size(sizeof(u8)) + /* TYPE */
2336                nla_total_size(sizeof(u8)) + /* PI */
2337                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2338                nla_total_size(sizeof(u8)) + /* PERSIST */
2339                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2340                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2341                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2342                0;
2343 }
2344
2345 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2346 {
2347         struct tun_struct *tun = netdev_priv(dev);
2348
2349         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2350                 goto nla_put_failure;
2351         if (uid_valid(tun->owner) &&
2352             nla_put_u32(skb, IFLA_TUN_OWNER,
2353                         from_kuid_munged(current_user_ns(), tun->owner)))
2354                 goto nla_put_failure;
2355         if (gid_valid(tun->group) &&
2356             nla_put_u32(skb, IFLA_TUN_GROUP,
2357                         from_kgid_munged(current_user_ns(), tun->group)))
2358                 goto nla_put_failure;
2359         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2360                 goto nla_put_failure;
2361         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2362                 goto nla_put_failure;
2363         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2364                 goto nla_put_failure;
2365         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2366                        !!(tun->flags & IFF_MULTI_QUEUE)))
2367                 goto nla_put_failure;
2368         if (tun->flags & IFF_MULTI_QUEUE) {
2369                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2370                         goto nla_put_failure;
2371                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2372                                 tun->numdisabled))
2373                         goto nla_put_failure;
2374         }
2375
2376         return 0;
2377
2378 nla_put_failure:
2379         return -EMSGSIZE;
2380 }
2381
2382 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2383         .kind           = DRV_NAME,
2384         .priv_size      = sizeof(struct tun_struct),
2385         .setup          = tun_setup,
2386         .validate       = tun_validate,
2387         .get_size       = tun_get_size,
2388         .fill_info      = tun_fill_info,
2389 };
2390
2391 static void tun_sock_write_space(struct sock *sk)
2392 {
2393         struct tun_file *tfile;
2394         wait_queue_head_t *wqueue;
2395
2396         if (!sock_writeable(sk))
2397                 return;
2398
2399         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2400                 return;
2401
2402         wqueue = sk_sleep(sk);
2403         if (wqueue && waitqueue_active(wqueue))
2404                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2405                                                 EPOLLWRNORM | EPOLLWRBAND);
2406
2407         tfile = container_of(sk, struct tun_file, sk);
2408         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2409 }
2410
2411 static void tun_put_page(struct tun_page *tpage)
2412 {
2413         if (tpage->page)
2414                 __page_frag_cache_drain(tpage->page, tpage->count);
2415 }
2416
2417 static int tun_xdp_one(struct tun_struct *tun,
2418                        struct tun_file *tfile,
2419                        struct xdp_buff *xdp, int *flush,
2420                        struct tun_page *tpage)
2421 {
2422         unsigned int datasize = xdp->data_end - xdp->data;
2423         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2424         struct virtio_net_hdr *gso = &hdr->gso;
2425         struct bpf_prog *xdp_prog;
2426         struct sk_buff *skb = NULL;
2427         struct sk_buff_head *queue;
2428         u32 rxhash = 0, act;
2429         int buflen = hdr->buflen;
2430         int ret = 0;
2431         bool skb_xdp = false;
2432         struct page *page;
2433
2434         xdp_prog = rcu_dereference(tun->xdp_prog);
2435         if (xdp_prog) {
2436                 if (gso->gso_type) {
2437                         skb_xdp = true;
2438                         goto build;
2439                 }
2440
2441                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2442                 xdp_set_data_meta_invalid(xdp);
2443
2444                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2445                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2446                 if (ret < 0) {
2447                         put_page(virt_to_head_page(xdp->data));
2448                         return ret;
2449                 }
2450
2451                 switch (ret) {
2452                 case XDP_REDIRECT:
2453                         *flush = true;
2454                         fallthrough;
2455                 case XDP_TX:
2456                         return 0;
2457                 case XDP_PASS:
2458                         break;
2459                 default:
2460                         page = virt_to_head_page(xdp->data);
2461                         if (tpage->page == page) {
2462                                 ++tpage->count;
2463                         } else {
2464                                 tun_put_page(tpage);
2465                                 tpage->page = page;
2466                                 tpage->count = 1;
2467                         }
2468                         return 0;
2469                 }
2470         }
2471
2472 build:
2473         skb = build_skb(xdp->data_hard_start, buflen);
2474         if (!skb) {
2475                 ret = -ENOMEM;
2476                 goto out;
2477         }
2478
2479         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2480         skb_put(skb, xdp->data_end - xdp->data);
2481
2482         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2483                 atomic_long_inc(&tun->rx_frame_errors);
2484                 kfree_skb(skb);
2485                 ret = -EINVAL;
2486                 goto out;
2487         }
2488
2489         skb->protocol = eth_type_trans(skb, tun->dev);
2490         skb_reset_network_header(skb);
2491         skb_probe_transport_header(skb);
2492         skb_record_rx_queue(skb, tfile->queue_index);
2493
2494         if (skb_xdp) {
2495                 ret = do_xdp_generic(xdp_prog, skb);
2496                 if (ret != XDP_PASS) {
2497                         ret = 0;
2498                         goto out;
2499                 }
2500         }
2501
2502         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2503             !tfile->detached)
2504                 rxhash = __skb_get_hash_symmetric(skb);
2505
2506         if (tfile->napi_enabled) {
2507                 queue = &tfile->sk.sk_write_queue;
2508                 spin_lock(&queue->lock);
2509                 __skb_queue_tail(queue, skb);
2510                 spin_unlock(&queue->lock);
2511                 ret = 1;
2512         } else {
2513                 netif_receive_skb(skb);
2514                 ret = 0;
2515         }
2516
2517         /* No need to disable preemption here since this function is
2518          * always called with bh disabled
2519          */
2520         dev_sw_netstats_rx_add(tun->dev, datasize);
2521
2522         if (rxhash)
2523                 tun_flow_update(tun, rxhash, tfile);
2524
2525 out:
2526         return ret;
2527 }
2528
2529 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2530 {
2531         int ret, i;
2532         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2533         struct tun_struct *tun = tun_get(tfile);
2534         struct tun_msg_ctl *ctl = m->msg_control;
2535         struct xdp_buff *xdp;
2536
2537         if (!tun)
2538                 return -EBADFD;
2539
2540         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2541             ctl && ctl->type == TUN_MSG_PTR) {
2542                 struct tun_page tpage;
2543                 int n = ctl->num;
2544                 int flush = 0, queued = 0;
2545
2546                 memset(&tpage, 0, sizeof(tpage));
2547
2548                 local_bh_disable();
2549                 rcu_read_lock();
2550
2551                 for (i = 0; i < n; i++) {
2552                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2553                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2554                         if (ret > 0)
2555                                 queued += ret;
2556                 }
2557
2558                 if (flush)
2559                         xdp_do_flush();
2560
2561                 if (tfile->napi_enabled && queued > 0)
2562                         napi_schedule(&tfile->napi);
2563
2564                 rcu_read_unlock();
2565                 local_bh_enable();
2566
2567                 tun_put_page(&tpage);
2568
2569                 ret = total_len;
2570                 goto out;
2571         }
2572
2573         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2574                            m->msg_flags & MSG_DONTWAIT,
2575                            m->msg_flags & MSG_MORE);
2576 out:
2577         tun_put(tun);
2578         return ret;
2579 }
2580
2581 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2582                        int flags)
2583 {
2584         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2585         struct tun_struct *tun = tun_get(tfile);
2586         void *ptr = m->msg_control;
2587         int ret;
2588
2589         if (!tun) {
2590                 ret = -EBADFD;
2591                 goto out_free;
2592         }
2593
2594         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2595                 ret = -EINVAL;
2596                 goto out_put_tun;
2597         }
2598         if (flags & MSG_ERRQUEUE) {
2599                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2600                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2601                 goto out;
2602         }
2603         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2604         if (ret > (ssize_t)total_len) {
2605                 m->msg_flags |= MSG_TRUNC;
2606                 ret = flags & MSG_TRUNC ? ret : total_len;
2607         }
2608 out:
2609         tun_put(tun);
2610         return ret;
2611
2612 out_put_tun:
2613         tun_put(tun);
2614 out_free:
2615         tun_ptr_free(ptr);
2616         return ret;
2617 }
2618
2619 static int tun_ptr_peek_len(void *ptr)
2620 {
2621         if (likely(ptr)) {
2622                 if (tun_is_xdp_frame(ptr)) {
2623                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2624
2625                         return xdpf->len;
2626                 }
2627                 return __skb_array_len_with_tag(ptr);
2628         } else {
2629                 return 0;
2630         }
2631 }
2632
2633 static int tun_peek_len(struct socket *sock)
2634 {
2635         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2636         struct tun_struct *tun;
2637         int ret = 0;
2638
2639         tun = tun_get(tfile);
2640         if (!tun)
2641                 return 0;
2642
2643         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2644         tun_put(tun);
2645
2646         return ret;
2647 }
2648
2649 /* Ops structure to mimic raw sockets with tun */
2650 static const struct proto_ops tun_socket_ops = {
2651         .peek_len = tun_peek_len,
2652         .sendmsg = tun_sendmsg,
2653         .recvmsg = tun_recvmsg,
2654 };
2655
2656 static struct proto tun_proto = {
2657         .name           = "tun",
2658         .owner          = THIS_MODULE,
2659         .obj_size       = sizeof(struct tun_file),
2660 };
2661
2662 static int tun_flags(struct tun_struct *tun)
2663 {
2664         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2665 }
2666
2667 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2668                               char *buf)
2669 {
2670         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2671         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2672 }
2673
2674 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2675                           char *buf)
2676 {
2677         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2678         return uid_valid(tun->owner)?
2679                 sysfs_emit(buf, "%u\n",
2680                            from_kuid_munged(current_user_ns(), tun->owner)) :
2681                 sysfs_emit(buf, "-1\n");
2682 }
2683
2684 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2685                           char *buf)
2686 {
2687         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2688         return gid_valid(tun->group) ?
2689                 sysfs_emit(buf, "%u\n",
2690                            from_kgid_munged(current_user_ns(), tun->group)) :
2691                 sysfs_emit(buf, "-1\n");
2692 }
2693
2694 static DEVICE_ATTR_RO(tun_flags);
2695 static DEVICE_ATTR_RO(owner);
2696 static DEVICE_ATTR_RO(group);
2697
2698 static struct attribute *tun_dev_attrs[] = {
2699         &dev_attr_tun_flags.attr,
2700         &dev_attr_owner.attr,
2701         &dev_attr_group.attr,
2702         NULL
2703 };
2704
2705 static const struct attribute_group tun_attr_group = {
2706         .attrs = tun_dev_attrs
2707 };
2708
2709 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2710 {
2711         struct tun_struct *tun;
2712         struct tun_file *tfile = file->private_data;
2713         struct net_device *dev;
2714         int err;
2715
2716         if (tfile->detached)
2717                 return -EINVAL;
2718
2719         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2720                 if (!capable(CAP_NET_ADMIN))
2721                         return -EPERM;
2722
2723                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2724                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2725                         return -EINVAL;
2726         }
2727
2728         dev = __dev_get_by_name(net, ifr->ifr_name);
2729         if (dev) {
2730                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2731                         return -EBUSY;
2732                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2733                         tun = netdev_priv(dev);
2734                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2735                         tun = netdev_priv(dev);
2736                 else
2737                         return -EINVAL;
2738
2739                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2740                     !!(tun->flags & IFF_MULTI_QUEUE))
2741                         return -EINVAL;
2742
2743                 if (tun_not_capable(tun))
2744                         return -EPERM;
2745                 err = security_tun_dev_open(tun->security);
2746                 if (err < 0)
2747                         return err;
2748
2749                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2750                                  ifr->ifr_flags & IFF_NAPI,
2751                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2752                 if (err < 0)
2753                         return err;
2754
2755                 if (tun->flags & IFF_MULTI_QUEUE &&
2756                     (tun->numqueues + tun->numdisabled > 1)) {
2757                         /* One or more queue has already been attached, no need
2758                          * to initialize the device again.
2759                          */
2760                         netdev_state_change(dev);
2761                         return 0;
2762                 }
2763
2764                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2765                               (ifr->ifr_flags & TUN_FEATURES);
2766
2767                 netdev_state_change(dev);
2768         } else {
2769                 char *name;
2770                 unsigned long flags = 0;
2771                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2772                              MAX_TAP_QUEUES : 1;
2773
2774                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2775                         return -EPERM;
2776                 err = security_tun_dev_create();
2777                 if (err < 0)
2778                         return err;
2779
2780                 /* Set dev type */
2781                 if (ifr->ifr_flags & IFF_TUN) {
2782                         /* TUN device */
2783                         flags |= IFF_TUN;
2784                         name = "tun%d";
2785                 } else if (ifr->ifr_flags & IFF_TAP) {
2786                         /* TAP device */
2787                         flags |= IFF_TAP;
2788                         name = "tap%d";
2789                 } else
2790                         return -EINVAL;
2791
2792                 if (*ifr->ifr_name)
2793                         name = ifr->ifr_name;
2794
2795                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2796                                        NET_NAME_UNKNOWN, tun_setup, queues,
2797                                        queues);
2798
2799                 if (!dev)
2800                         return -ENOMEM;
2801
2802                 dev_net_set(dev, net);
2803                 dev->rtnl_link_ops = &tun_link_ops;
2804                 dev->ifindex = tfile->ifindex;
2805                 dev->sysfs_groups[0] = &tun_attr_group;
2806
2807                 tun = netdev_priv(dev);
2808                 tun->dev = dev;
2809                 tun->flags = flags;
2810                 tun->txflt.count = 0;
2811                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2812
2813                 tun->align = NET_SKB_PAD;
2814                 tun->filter_attached = false;
2815                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2816                 tun->rx_batched = 0;
2817                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2818
2819                 tun->ifr = ifr;
2820                 tun->file = file;
2821
2822                 tun_net_initialize(dev);
2823
2824                 err = register_netdevice(tun->dev);
2825                 if (err < 0) {
2826                         free_netdev(dev);
2827                         return err;
2828                 }
2829                 /* free_netdev() won't check refcnt, to avoid race
2830                  * with dev_put() we need publish tun after registration.
2831                  */
2832                 rcu_assign_pointer(tfile->tun, tun);
2833         }
2834
2835         if (ifr->ifr_flags & IFF_NO_CARRIER)
2836                 netif_carrier_off(tun->dev);
2837         else
2838                 netif_carrier_on(tun->dev);
2839
2840         /* Make sure persistent devices do not get stuck in
2841          * xoff state.
2842          */
2843         if (netif_running(tun->dev))
2844                 netif_tx_wake_all_queues(tun->dev);
2845
2846         strcpy(ifr->ifr_name, tun->dev->name);
2847         return 0;
2848 }
2849
2850 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2851 {
2852         strcpy(ifr->ifr_name, tun->dev->name);
2853
2854         ifr->ifr_flags = tun_flags(tun);
2855
2856 }
2857
2858 /* This is like a cut-down ethtool ops, except done via tun fd so no
2859  * privs required. */
2860 static int set_offload(struct tun_struct *tun, unsigned long arg)
2861 {
2862         netdev_features_t features = 0;
2863
2864         if (arg & TUN_F_CSUM) {
2865                 features |= NETIF_F_HW_CSUM;
2866                 arg &= ~TUN_F_CSUM;
2867
2868                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2869                         if (arg & TUN_F_TSO_ECN) {
2870                                 features |= NETIF_F_TSO_ECN;
2871                                 arg &= ~TUN_F_TSO_ECN;
2872                         }
2873                         if (arg & TUN_F_TSO4)
2874                                 features |= NETIF_F_TSO;
2875                         if (arg & TUN_F_TSO6)
2876                                 features |= NETIF_F_TSO6;
2877                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2878                 }
2879
2880                 arg &= ~TUN_F_UFO;
2881
2882                 /* TODO: for now USO4 and USO6 should work simultaneously */
2883                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2884                         features |= NETIF_F_GSO_UDP_L4;
2885                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2886                 }
2887         }
2888
2889         /* This gives the user a way to test for new features in future by
2890          * trying to set them. */
2891         if (arg)
2892                 return -EINVAL;
2893
2894         tun->set_features = features;
2895         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2896         tun->dev->wanted_features |= features;
2897         netdev_update_features(tun->dev);
2898
2899         return 0;
2900 }
2901
2902 static void tun_detach_filter(struct tun_struct *tun, int n)
2903 {
2904         int i;
2905         struct tun_file *tfile;
2906
2907         for (i = 0; i < n; i++) {
2908                 tfile = rtnl_dereference(tun->tfiles[i]);
2909                 lock_sock(tfile->socket.sk);
2910                 sk_detach_filter(tfile->socket.sk);
2911                 release_sock(tfile->socket.sk);
2912         }
2913
2914         tun->filter_attached = false;
2915 }
2916
2917 static int tun_attach_filter(struct tun_struct *tun)
2918 {
2919         int i, ret = 0;
2920         struct tun_file *tfile;
2921
2922         for (i = 0; i < tun->numqueues; i++) {
2923                 tfile = rtnl_dereference(tun->tfiles[i]);
2924                 lock_sock(tfile->socket.sk);
2925                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2926                 release_sock(tfile->socket.sk);
2927                 if (ret) {
2928                         tun_detach_filter(tun, i);
2929                         return ret;
2930                 }
2931         }
2932
2933         tun->filter_attached = true;
2934         return ret;
2935 }
2936
2937 static void tun_set_sndbuf(struct tun_struct *tun)
2938 {
2939         struct tun_file *tfile;
2940         int i;
2941
2942         for (i = 0; i < tun->numqueues; i++) {
2943                 tfile = rtnl_dereference(tun->tfiles[i]);
2944                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2945         }
2946 }
2947
2948 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2949 {
2950         struct tun_file *tfile = file->private_data;
2951         struct tun_struct *tun;
2952         int ret = 0;
2953
2954         rtnl_lock();
2955
2956         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2957                 tun = tfile->detached;
2958                 if (!tun) {
2959                         ret = -EINVAL;
2960                         goto unlock;
2961                 }
2962                 ret = security_tun_dev_attach_queue(tun->security);
2963                 if (ret < 0)
2964                         goto unlock;
2965                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2966                                  tun->flags & IFF_NAPI_FRAGS, true);
2967         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2968                 tun = rtnl_dereference(tfile->tun);
2969                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2970                         ret = -EINVAL;
2971                 else
2972                         __tun_detach(tfile, false);
2973         } else
2974                 ret = -EINVAL;
2975
2976         if (ret >= 0)
2977                 netdev_state_change(tun->dev);
2978
2979 unlock:
2980         rtnl_unlock();
2981         return ret;
2982 }
2983
2984 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2985                         void __user *data)
2986 {
2987         struct bpf_prog *prog;
2988         int fd;
2989
2990         if (copy_from_user(&fd, data, sizeof(fd)))
2991                 return -EFAULT;
2992
2993         if (fd == -1) {
2994                 prog = NULL;
2995         } else {
2996                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2997                 if (IS_ERR(prog))
2998                         return PTR_ERR(prog);
2999         }
3000
3001         return __tun_set_ebpf(tun, prog_p, prog);
3002 }
3003
3004 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3005 static unsigned char tun_get_addr_len(unsigned short type)
3006 {
3007         switch (type) {
3008         case ARPHRD_IP6GRE:
3009         case ARPHRD_TUNNEL6:
3010                 return sizeof(struct in6_addr);
3011         case ARPHRD_IPGRE:
3012         case ARPHRD_TUNNEL:
3013         case ARPHRD_SIT:
3014                 return 4;
3015         case ARPHRD_ETHER:
3016                 return ETH_ALEN;
3017         case ARPHRD_IEEE802154:
3018         case ARPHRD_IEEE802154_MONITOR:
3019                 return IEEE802154_EXTENDED_ADDR_LEN;
3020         case ARPHRD_PHONET_PIPE:
3021         case ARPHRD_PPP:
3022         case ARPHRD_NONE:
3023                 return 0;
3024         case ARPHRD_6LOWPAN:
3025                 return EUI64_ADDR_LEN;
3026         case ARPHRD_FDDI:
3027                 return FDDI_K_ALEN;
3028         case ARPHRD_HIPPI:
3029                 return HIPPI_ALEN;
3030         case ARPHRD_IEEE802:
3031                 return FC_ALEN;
3032         case ARPHRD_ROSE:
3033                 return ROSE_ADDR_LEN;
3034         case ARPHRD_NETROM:
3035                 return AX25_ADDR_LEN;
3036         case ARPHRD_LOCALTLK:
3037                 return LTALK_ALEN;
3038         default:
3039                 return 0;
3040         }
3041 }
3042
3043 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3044                             unsigned long arg, int ifreq_len)
3045 {
3046         struct tun_file *tfile = file->private_data;
3047         struct net *net = sock_net(&tfile->sk);
3048         struct tun_struct *tun;
3049         void __user* argp = (void __user*)arg;
3050         unsigned int ifindex, carrier;
3051         struct ifreq ifr;
3052         kuid_t owner;
3053         kgid_t group;
3054         int sndbuf;
3055         int vnet_hdr_sz;
3056         int le;
3057         int ret;
3058         bool do_notify = false;
3059
3060         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3061             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3062                 if (copy_from_user(&ifr, argp, ifreq_len))
3063                         return -EFAULT;
3064         } else {
3065                 memset(&ifr, 0, sizeof(ifr));
3066         }
3067         if (cmd == TUNGETFEATURES) {
3068                 /* Currently this just means: "what IFF flags are valid?".
3069                  * This is needed because we never checked for invalid flags on
3070                  * TUNSETIFF.
3071                  */
3072                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3073                                 TUN_FEATURES, (unsigned int __user*)argp);
3074         } else if (cmd == TUNSETQUEUE) {
3075                 return tun_set_queue(file, &ifr);
3076         } else if (cmd == SIOCGSKNS) {
3077                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3078                         return -EPERM;
3079                 return open_related_ns(&net->ns, get_net_ns);
3080         }
3081
3082         rtnl_lock();
3083
3084         tun = tun_get(tfile);
3085         if (cmd == TUNSETIFF) {
3086                 ret = -EEXIST;
3087                 if (tun)
3088                         goto unlock;
3089
3090                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3091
3092                 ret = tun_set_iff(net, file, &ifr);
3093
3094                 if (ret)
3095                         goto unlock;
3096
3097                 if (copy_to_user(argp, &ifr, ifreq_len))
3098                         ret = -EFAULT;
3099                 goto unlock;
3100         }
3101         if (cmd == TUNSETIFINDEX) {
3102                 ret = -EPERM;
3103                 if (tun)
3104                         goto unlock;
3105
3106                 ret = -EFAULT;
3107                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3108                         goto unlock;
3109
3110                 ret = 0;
3111                 tfile->ifindex = ifindex;
3112                 goto unlock;
3113         }
3114
3115         ret = -EBADFD;
3116         if (!tun)
3117                 goto unlock;
3118
3119         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3120
3121         net = dev_net(tun->dev);
3122         ret = 0;
3123         switch (cmd) {
3124         case TUNGETIFF:
3125                 tun_get_iff(tun, &ifr);
3126
3127                 if (tfile->detached)
3128                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3129                 if (!tfile->socket.sk->sk_filter)
3130                         ifr.ifr_flags |= IFF_NOFILTER;
3131
3132                 if (copy_to_user(argp, &ifr, ifreq_len))
3133                         ret = -EFAULT;
3134                 break;
3135
3136         case TUNSETNOCSUM:
3137                 /* Disable/Enable checksum */
3138
3139                 /* [unimplemented] */
3140                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3141                            arg ? "disabled" : "enabled");
3142                 break;
3143
3144         case TUNSETPERSIST:
3145                 /* Disable/Enable persist mode. Keep an extra reference to the
3146                  * module to prevent the module being unprobed.
3147                  */
3148                 if (arg && !(tun->flags & IFF_PERSIST)) {
3149                         tun->flags |= IFF_PERSIST;
3150                         __module_get(THIS_MODULE);
3151                         do_notify = true;
3152                 }
3153                 if (!arg && (tun->flags & IFF_PERSIST)) {
3154                         tun->flags &= ~IFF_PERSIST;
3155                         module_put(THIS_MODULE);
3156                         do_notify = true;
3157                 }
3158
3159                 netif_info(tun, drv, tun->dev, "persist %s\n",
3160                            arg ? "enabled" : "disabled");
3161                 break;
3162
3163         case TUNSETOWNER:
3164                 /* Set owner of the device */
3165                 owner = make_kuid(current_user_ns(), arg);
3166                 if (!uid_valid(owner)) {
3167                         ret = -EINVAL;
3168                         break;
3169                 }
3170                 tun->owner = owner;
3171                 do_notify = true;
3172                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3173                            from_kuid(&init_user_ns, tun->owner));
3174                 break;
3175
3176         case TUNSETGROUP:
3177                 /* Set group of the device */
3178                 group = make_kgid(current_user_ns(), arg);
3179                 if (!gid_valid(group)) {
3180                         ret = -EINVAL;
3181                         break;
3182                 }
3183                 tun->group = group;
3184                 do_notify = true;
3185                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3186                            from_kgid(&init_user_ns, tun->group));
3187                 break;
3188
3189         case TUNSETLINK:
3190                 /* Only allow setting the type when the interface is down */
3191                 if (tun->dev->flags & IFF_UP) {
3192                         netif_info(tun, drv, tun->dev,
3193                                    "Linktype set failed because interface is up\n");
3194                         ret = -EBUSY;
3195                 } else {
3196                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3197                                                        tun->dev);
3198                         ret = notifier_to_errno(ret);
3199                         if (ret) {
3200                                 netif_info(tun, drv, tun->dev,
3201                                            "Refused to change device type\n");
3202                                 break;
3203                         }
3204                         tun->dev->type = (int) arg;
3205                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3206                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3207                                    tun->dev->type);
3208                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3209                                                  tun->dev);
3210                 }
3211                 break;
3212
3213         case TUNSETDEBUG:
3214                 tun->msg_enable = (u32)arg;
3215                 break;
3216
3217         case TUNSETOFFLOAD:
3218                 ret = set_offload(tun, arg);
3219                 break;
3220
3221         case TUNSETTXFILTER:
3222                 /* Can be set only for TAPs */
3223                 ret = -EINVAL;
3224                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3225                         break;
3226                 ret = update_filter(&tun->txflt, (void __user *)arg);
3227                 break;
3228
3229         case SIOCGIFHWADDR:
3230                 /* Get hw address */
3231                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3232                 if (copy_to_user(argp, &ifr, ifreq_len))
3233                         ret = -EFAULT;
3234                 break;
3235
3236         case SIOCSIFHWADDR:
3237                 /* Set hw address */
3238                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3239                 break;
3240
3241         case TUNGETSNDBUF:
3242                 sndbuf = tfile->socket.sk->sk_sndbuf;
3243                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3244                         ret = -EFAULT;
3245                 break;
3246
3247         case TUNSETSNDBUF:
3248                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3249                         ret = -EFAULT;
3250                         break;
3251                 }
3252                 if (sndbuf <= 0) {
3253                         ret = -EINVAL;
3254                         break;
3255                 }
3256
3257                 tun->sndbuf = sndbuf;
3258                 tun_set_sndbuf(tun);
3259                 break;
3260
3261         case TUNGETVNETHDRSZ:
3262                 vnet_hdr_sz = tun->vnet_hdr_sz;
3263                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3264                         ret = -EFAULT;
3265                 break;
3266
3267         case TUNSETVNETHDRSZ:
3268                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3269                         ret = -EFAULT;
3270                         break;
3271                 }
3272                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3273                         ret = -EINVAL;
3274                         break;
3275                 }
3276
3277                 tun->vnet_hdr_sz = vnet_hdr_sz;
3278                 break;
3279
3280         case TUNGETVNETLE:
3281                 le = !!(tun->flags & TUN_VNET_LE);
3282                 if (put_user(le, (int __user *)argp))
3283                         ret = -EFAULT;
3284                 break;
3285
3286         case TUNSETVNETLE:
3287                 if (get_user(le, (int __user *)argp)) {
3288                         ret = -EFAULT;
3289                         break;
3290                 }
3291                 if (le)
3292                         tun->flags |= TUN_VNET_LE;
3293                 else
3294                         tun->flags &= ~TUN_VNET_LE;
3295                 break;
3296
3297         case TUNGETVNETBE:
3298                 ret = tun_get_vnet_be(tun, argp);
3299                 break;
3300
3301         case TUNSETVNETBE:
3302                 ret = tun_set_vnet_be(tun, argp);
3303                 break;
3304
3305         case TUNATTACHFILTER:
3306                 /* Can be set only for TAPs */
3307                 ret = -EINVAL;
3308                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3309                         break;
3310                 ret = -EFAULT;
3311                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3312                         break;
3313
3314                 ret = tun_attach_filter(tun);
3315                 break;
3316
3317         case TUNDETACHFILTER:
3318                 /* Can be set only for TAPs */
3319                 ret = -EINVAL;
3320                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3321                         break;
3322                 ret = 0;
3323                 tun_detach_filter(tun, tun->numqueues);
3324                 break;
3325
3326         case TUNGETFILTER:
3327                 ret = -EINVAL;
3328                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3329                         break;
3330                 ret = -EFAULT;
3331                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3332                         break;
3333                 ret = 0;
3334                 break;
3335
3336         case TUNSETSTEERINGEBPF:
3337                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3338                 break;
3339
3340         case TUNSETFILTEREBPF:
3341                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3342                 break;
3343
3344         case TUNSETCARRIER:
3345                 ret = -EFAULT;
3346                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3347                         goto unlock;
3348
3349                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3350                 break;
3351
3352         case TUNGETDEVNETNS:
3353                 ret = -EPERM;
3354                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3355                         goto unlock;
3356                 ret = open_related_ns(&net->ns, get_net_ns);
3357                 break;
3358
3359         default:
3360                 ret = -EINVAL;
3361                 break;
3362         }
3363
3364         if (do_notify)
3365                 netdev_state_change(tun->dev);
3366
3367 unlock:
3368         rtnl_unlock();
3369         if (tun)
3370                 tun_put(tun);
3371         return ret;
3372 }
3373
3374 static long tun_chr_ioctl(struct file *file,
3375                           unsigned int cmd, unsigned long arg)
3376 {
3377         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3378 }
3379
3380 #ifdef CONFIG_COMPAT
3381 static long tun_chr_compat_ioctl(struct file *file,
3382                          unsigned int cmd, unsigned long arg)
3383 {
3384         switch (cmd) {
3385         case TUNSETIFF:
3386         case TUNGETIFF:
3387         case TUNSETTXFILTER:
3388         case TUNGETSNDBUF:
3389         case TUNSETSNDBUF:
3390         case SIOCGIFHWADDR:
3391         case SIOCSIFHWADDR:
3392                 arg = (unsigned long)compat_ptr(arg);
3393                 break;
3394         default:
3395                 arg = (compat_ulong_t)arg;
3396                 break;
3397         }
3398
3399         /*
3400          * compat_ifreq is shorter than ifreq, so we must not access beyond
3401          * the end of that structure. All fields that are used in this
3402          * driver are compatible though, we don't need to convert the
3403          * contents.
3404          */
3405         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3406 }
3407 #endif /* CONFIG_COMPAT */
3408
3409 static int tun_chr_fasync(int fd, struct file *file, int on)
3410 {
3411         struct tun_file *tfile = file->private_data;
3412         int ret;
3413
3414         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3415                 goto out;
3416
3417         if (on) {
3418                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3419                 tfile->flags |= TUN_FASYNC;
3420         } else
3421                 tfile->flags &= ~TUN_FASYNC;
3422         ret = 0;
3423 out:
3424         return ret;
3425 }
3426
3427 static int tun_chr_open(struct inode *inode, struct file * file)
3428 {
3429         struct net *net = current->nsproxy->net_ns;
3430         struct tun_file *tfile;
3431
3432         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3433                                             &tun_proto, 0);
3434         if (!tfile)
3435                 return -ENOMEM;
3436         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3437                 sk_free(&tfile->sk);
3438                 return -ENOMEM;
3439         }
3440
3441         mutex_init(&tfile->napi_mutex);
3442         RCU_INIT_POINTER(tfile->tun, NULL);
3443         tfile->flags = 0;
3444         tfile->ifindex = 0;
3445
3446         init_waitqueue_head(&tfile->socket.wq.wait);
3447
3448         tfile->socket.file = file;
3449         tfile->socket.ops = &tun_socket_ops;
3450
3451         sock_init_data(&tfile->socket, &tfile->sk);
3452
3453         tfile->sk.sk_write_space = tun_sock_write_space;
3454         tfile->sk.sk_sndbuf = INT_MAX;
3455
3456         file->private_data = tfile;
3457         INIT_LIST_HEAD(&tfile->next);
3458
3459         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3460
3461         return 0;
3462 }
3463
3464 static int tun_chr_close(struct inode *inode, struct file *file)
3465 {
3466         struct tun_file *tfile = file->private_data;
3467
3468         tun_detach(tfile, true);
3469
3470         return 0;
3471 }
3472
3473 #ifdef CONFIG_PROC_FS
3474 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3475 {
3476         struct tun_file *tfile = file->private_data;
3477         struct tun_struct *tun;
3478         struct ifreq ifr;
3479
3480         memset(&ifr, 0, sizeof(ifr));
3481
3482         rtnl_lock();
3483         tun = tun_get(tfile);
3484         if (tun)
3485                 tun_get_iff(tun, &ifr);
3486         rtnl_unlock();
3487
3488         if (tun)
3489                 tun_put(tun);
3490
3491         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3492 }
3493 #endif
3494
3495 static const struct file_operations tun_fops = {
3496         .owner  = THIS_MODULE,
3497         .llseek = no_llseek,
3498         .read_iter  = tun_chr_read_iter,
3499         .write_iter = tun_chr_write_iter,
3500         .poll   = tun_chr_poll,
3501         .unlocked_ioctl = tun_chr_ioctl,
3502 #ifdef CONFIG_COMPAT
3503         .compat_ioctl = tun_chr_compat_ioctl,
3504 #endif
3505         .open   = tun_chr_open,
3506         .release = tun_chr_close,
3507         .fasync = tun_chr_fasync,
3508 #ifdef CONFIG_PROC_FS
3509         .show_fdinfo = tun_chr_show_fdinfo,
3510 #endif
3511 };
3512
3513 static struct miscdevice tun_miscdev = {
3514         .minor = TUN_MINOR,
3515         .name = "tun",
3516         .nodename = "net/tun",
3517         .fops = &tun_fops,
3518 };
3519
3520 /* ethtool interface */
3521
3522 static void tun_default_link_ksettings(struct net_device *dev,
3523                                        struct ethtool_link_ksettings *cmd)
3524 {
3525         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3526         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3527         cmd->base.speed         = SPEED_10000;
3528         cmd->base.duplex        = DUPLEX_FULL;
3529         cmd->base.port          = PORT_TP;
3530         cmd->base.phy_address   = 0;
3531         cmd->base.autoneg       = AUTONEG_DISABLE;
3532 }
3533
3534 static int tun_get_link_ksettings(struct net_device *dev,
3535                                   struct ethtool_link_ksettings *cmd)
3536 {
3537         struct tun_struct *tun = netdev_priv(dev);
3538
3539         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3540         return 0;
3541 }
3542
3543 static int tun_set_link_ksettings(struct net_device *dev,
3544                                   const struct ethtool_link_ksettings *cmd)
3545 {
3546         struct tun_struct *tun = netdev_priv(dev);
3547
3548         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3549         return 0;
3550 }
3551
3552 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3553 {
3554         struct tun_struct *tun = netdev_priv(dev);
3555
3556         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3557         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3558
3559         switch (tun->flags & TUN_TYPE_MASK) {
3560         case IFF_TUN:
3561                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3562                 break;
3563         case IFF_TAP:
3564                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3565                 break;
3566         }
3567 }
3568
3569 static u32 tun_get_msglevel(struct net_device *dev)
3570 {
3571         struct tun_struct *tun = netdev_priv(dev);
3572
3573         return tun->msg_enable;
3574 }
3575
3576 static void tun_set_msglevel(struct net_device *dev, u32 value)
3577 {
3578         struct tun_struct *tun = netdev_priv(dev);
3579
3580         tun->msg_enable = value;
3581 }
3582
3583 static int tun_get_coalesce(struct net_device *dev,
3584                             struct ethtool_coalesce *ec,
3585                             struct kernel_ethtool_coalesce *kernel_coal,
3586                             struct netlink_ext_ack *extack)
3587 {
3588         struct tun_struct *tun = netdev_priv(dev);
3589
3590         ec->rx_max_coalesced_frames = tun->rx_batched;
3591
3592         return 0;
3593 }
3594
3595 static int tun_set_coalesce(struct net_device *dev,
3596                             struct ethtool_coalesce *ec,
3597                             struct kernel_ethtool_coalesce *kernel_coal,
3598                             struct netlink_ext_ack *extack)
3599 {
3600         struct tun_struct *tun = netdev_priv(dev);
3601
3602         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3603                 tun->rx_batched = NAPI_POLL_WEIGHT;
3604         else
3605                 tun->rx_batched = ec->rx_max_coalesced_frames;
3606
3607         return 0;
3608 }
3609
3610 static const struct ethtool_ops tun_ethtool_ops = {
3611         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3612         .get_drvinfo    = tun_get_drvinfo,
3613         .get_msglevel   = tun_get_msglevel,
3614         .set_msglevel   = tun_set_msglevel,
3615         .get_link       = ethtool_op_get_link,
3616         .get_ts_info    = ethtool_op_get_ts_info,
3617         .get_coalesce   = tun_get_coalesce,
3618         .set_coalesce   = tun_set_coalesce,
3619         .get_link_ksettings = tun_get_link_ksettings,
3620         .set_link_ksettings = tun_set_link_ksettings,
3621 };
3622
3623 static int tun_queue_resize(struct tun_struct *tun)
3624 {
3625         struct net_device *dev = tun->dev;
3626         struct tun_file *tfile;
3627         struct ptr_ring **rings;
3628         int n = tun->numqueues + tun->numdisabled;
3629         int ret, i;
3630
3631         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3632         if (!rings)
3633                 return -ENOMEM;
3634
3635         for (i = 0; i < tun->numqueues; i++) {
3636                 tfile = rtnl_dereference(tun->tfiles[i]);
3637                 rings[i] = &tfile->tx_ring;
3638         }
3639         list_for_each_entry(tfile, &tun->disabled, next)
3640                 rings[i++] = &tfile->tx_ring;
3641
3642         ret = ptr_ring_resize_multiple(rings, n,
3643                                        dev->tx_queue_len, GFP_KERNEL,
3644                                        tun_ptr_free);
3645
3646         kfree(rings);
3647         return ret;
3648 }
3649
3650 static int tun_device_event(struct notifier_block *unused,
3651                             unsigned long event, void *ptr)
3652 {
3653         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3654         struct tun_struct *tun = netdev_priv(dev);
3655         int i;
3656
3657         if (dev->rtnl_link_ops != &tun_link_ops)
3658                 return NOTIFY_DONE;
3659
3660         switch (event) {
3661         case NETDEV_CHANGE_TX_QUEUE_LEN:
3662                 if (tun_queue_resize(tun))
3663                         return NOTIFY_BAD;
3664                 break;
3665         case NETDEV_UP:
3666                 for (i = 0; i < tun->numqueues; i++) {
3667                         struct tun_file *tfile;
3668
3669                         tfile = rtnl_dereference(tun->tfiles[i]);
3670                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3671                 }
3672                 break;
3673         default:
3674                 break;
3675         }
3676
3677         return NOTIFY_DONE;
3678 }
3679
3680 static struct notifier_block tun_notifier_block __read_mostly = {
3681         .notifier_call  = tun_device_event,
3682 };
3683
3684 static int __init tun_init(void)
3685 {
3686         int ret = 0;
3687
3688         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3689
3690         ret = rtnl_link_register(&tun_link_ops);
3691         if (ret) {
3692                 pr_err("Can't register link_ops\n");
3693                 goto err_linkops;
3694         }
3695
3696         ret = misc_register(&tun_miscdev);
3697         if (ret) {
3698                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3699                 goto err_misc;
3700         }
3701
3702         ret = register_netdevice_notifier(&tun_notifier_block);
3703         if (ret) {
3704                 pr_err("Can't register netdevice notifier\n");
3705                 goto err_notifier;
3706         }
3707
3708         return  0;
3709
3710 err_notifier:
3711         misc_deregister(&tun_miscdev);
3712 err_misc:
3713         rtnl_link_unregister(&tun_link_ops);
3714 err_linkops:
3715         return ret;
3716 }
3717
3718 static void tun_cleanup(void)
3719 {
3720         misc_deregister(&tun_miscdev);
3721         rtnl_link_unregister(&tun_link_ops);
3722         unregister_netdevice_notifier(&tun_notifier_block);
3723 }
3724
3725 /* Get an underlying socket object from tun file.  Returns error unless file is
3726  * attached to a device.  The returned object works like a packet socket, it
3727  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3728  * holding a reference to the file for as long as the socket is in use. */
3729 struct socket *tun_get_socket(struct file *file)
3730 {
3731         struct tun_file *tfile;
3732         if (file->f_op != &tun_fops)
3733                 return ERR_PTR(-EINVAL);
3734         tfile = file->private_data;
3735         if (!tfile)
3736                 return ERR_PTR(-EBADFD);
3737         return &tfile->socket;
3738 }
3739 EXPORT_SYMBOL_GPL(tun_get_socket);
3740
3741 struct ptr_ring *tun_get_tx_ring(struct file *file)
3742 {
3743         struct tun_file *tfile;
3744
3745         if (file->f_op != &tun_fops)
3746                 return ERR_PTR(-EINVAL);
3747         tfile = file->private_data;
3748         if (!tfile)
3749                 return ERR_PTR(-EBADFD);
3750         return &tfile->tx_ring;
3751 }
3752 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3753
3754 module_init(tun_init);
3755 module_exit(tun_cleanup);
3756 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3757 MODULE_AUTHOR(DRV_COPYRIGHT);
3758 MODULE_LICENSE("GPL");
3759 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3760 MODULE_ALIAS("devname:net/tun");