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 <maxk@qualcomm.com>
   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 <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  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 <underley@underley.eu.org>
  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 <maxk@qualcomm.com>"
  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                 /* Currently tun does not support XDP, only tap does. */
1405                 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1406                                     NETDEV_XDP_ACT_REDIRECT |
1407                                     NETDEV_XDP_ACT_NDO_XMIT;
1408
1409                 break;
1410         }
1411
1412         dev->min_mtu = MIN_MTU;
1413         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1414 }
1415
1416 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1417 {
1418         struct sock *sk = tfile->socket.sk;
1419
1420         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1421 }
1422
1423 /* Character device part */
1424
1425 /* Poll */
1426 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1427 {
1428         struct tun_file *tfile = file->private_data;
1429         struct tun_struct *tun = tun_get(tfile);
1430         struct sock *sk;
1431         __poll_t mask = 0;
1432
1433         if (!tun)
1434                 return EPOLLERR;
1435
1436         sk = tfile->socket.sk;
1437
1438         poll_wait(file, sk_sleep(sk), wait);
1439
1440         if (!ptr_ring_empty(&tfile->tx_ring))
1441                 mask |= EPOLLIN | EPOLLRDNORM;
1442
1443         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1444          * guarantee EPOLLOUT to be raised by either here or
1445          * tun_sock_write_space(). Then process could get notification
1446          * after it writes to a down device and meets -EIO.
1447          */
1448         if (tun_sock_writeable(tun, tfile) ||
1449             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1450              tun_sock_writeable(tun, tfile)))
1451                 mask |= EPOLLOUT | EPOLLWRNORM;
1452
1453         if (tun->dev->reg_state != NETREG_REGISTERED)
1454                 mask = EPOLLERR;
1455
1456         tun_put(tun);
1457         return mask;
1458 }
1459
1460 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1461                                             size_t len,
1462                                             const struct iov_iter *it)
1463 {
1464         struct sk_buff *skb;
1465         size_t linear;
1466         int err;
1467         int i;
1468
1469         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1470             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1471                 return ERR_PTR(-EMSGSIZE);
1472
1473         local_bh_disable();
1474         skb = napi_get_frags(&tfile->napi);
1475         local_bh_enable();
1476         if (!skb)
1477                 return ERR_PTR(-ENOMEM);
1478
1479         linear = iov_iter_single_seg_count(it);
1480         err = __skb_grow(skb, linear);
1481         if (err)
1482                 goto free;
1483
1484         skb->len = len;
1485         skb->data_len = len - linear;
1486         skb->truesize += skb->data_len;
1487
1488         for (i = 1; i < it->nr_segs; i++) {
1489                 const struct iovec *iov = iter_iov(it);
1490                 size_t fragsz = iov->iov_len;
1491                 struct page *page;
1492                 void *frag;
1493
1494                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1495                         err = -EINVAL;
1496                         goto free;
1497                 }
1498                 frag = netdev_alloc_frag(fragsz);
1499                 if (!frag) {
1500                         err = -ENOMEM;
1501                         goto free;
1502                 }
1503                 page = virt_to_head_page(frag);
1504                 skb_fill_page_desc(skb, i - 1, page,
1505                                    frag - page_address(page), fragsz);
1506         }
1507
1508         return skb;
1509 free:
1510         /* frees skb and all frags allocated with napi_alloc_frag() */
1511         napi_free_frags(&tfile->napi);
1512         return ERR_PTR(err);
1513 }
1514
1515 /* prepad is the amount to reserve at front.  len is length after that.
1516  * linear is a hint as to how much to copy (usually headers). */
1517 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1518                                      size_t prepad, size_t len,
1519                                      size_t linear, int noblock)
1520 {
1521         struct sock *sk = tfile->socket.sk;
1522         struct sk_buff *skb;
1523         int err;
1524
1525         /* Under a page?  Don't bother with paged skb. */
1526         if (prepad + len < PAGE_SIZE)
1527                 linear = len;
1528
1529         if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1530                 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1531         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1532                                    &err, PAGE_ALLOC_COSTLY_ORDER);
1533         if (!skb)
1534                 return ERR_PTR(err);
1535
1536         skb_reserve(skb, prepad);
1537         skb_put(skb, linear);
1538         skb->data_len = len - linear;
1539         skb->len += len - linear;
1540
1541         return skb;
1542 }
1543
1544 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1545                            struct sk_buff *skb, int more)
1546 {
1547         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1548         struct sk_buff_head process_queue;
1549         u32 rx_batched = tun->rx_batched;
1550         bool rcv = false;
1551
1552         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1553                 local_bh_disable();
1554                 skb_record_rx_queue(skb, tfile->queue_index);
1555                 netif_receive_skb(skb);
1556                 local_bh_enable();
1557                 return;
1558         }
1559
1560         spin_lock(&queue->lock);
1561         if (!more || skb_queue_len(queue) == rx_batched) {
1562                 __skb_queue_head_init(&process_queue);
1563                 skb_queue_splice_tail_init(queue, &process_queue);
1564                 rcv = true;
1565         } else {
1566                 __skb_queue_tail(queue, skb);
1567         }
1568         spin_unlock(&queue->lock);
1569
1570         if (rcv) {
1571                 struct sk_buff *nskb;
1572
1573                 local_bh_disable();
1574                 while ((nskb = __skb_dequeue(&process_queue))) {
1575                         skb_record_rx_queue(nskb, tfile->queue_index);
1576                         netif_receive_skb(nskb);
1577                 }
1578                 skb_record_rx_queue(skb, tfile->queue_index);
1579                 netif_receive_skb(skb);
1580                 local_bh_enable();
1581         }
1582 }
1583
1584 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1585                               int len, int noblock, bool zerocopy)
1586 {
1587         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1588                 return false;
1589
1590         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1591                 return false;
1592
1593         if (!noblock)
1594                 return false;
1595
1596         if (zerocopy)
1597                 return false;
1598
1599         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1600             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1601                 return false;
1602
1603         return true;
1604 }
1605
1606 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1607                                        struct page_frag *alloc_frag, char *buf,
1608                                        int buflen, int len, int pad)
1609 {
1610         struct sk_buff *skb = build_skb(buf, buflen);
1611
1612         if (!skb)
1613                 return ERR_PTR(-ENOMEM);
1614
1615         skb_reserve(skb, pad);
1616         skb_put(skb, len);
1617         skb_set_owner_w(skb, tfile->socket.sk);
1618
1619         get_page(alloc_frag->page);
1620         alloc_frag->offset += buflen;
1621
1622         return skb;
1623 }
1624
1625 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1626                        struct xdp_buff *xdp, u32 act)
1627 {
1628         int err;
1629
1630         switch (act) {
1631         case XDP_REDIRECT:
1632                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1633                 if (err)
1634                         return err;
1635                 break;
1636         case XDP_TX:
1637                 err = tun_xdp_tx(tun->dev, xdp);
1638                 if (err < 0)
1639                         return err;
1640                 break;
1641         case XDP_PASS:
1642                 break;
1643         default:
1644                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1645                 fallthrough;
1646         case XDP_ABORTED:
1647                 trace_xdp_exception(tun->dev, xdp_prog, act);
1648                 fallthrough;
1649         case XDP_DROP:
1650                 dev_core_stats_rx_dropped_inc(tun->dev);
1651                 break;
1652         }
1653
1654         return act;
1655 }
1656
1657 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1658                                      struct tun_file *tfile,
1659                                      struct iov_iter *from,
1660                                      struct virtio_net_hdr *hdr,
1661                                      int len, int *skb_xdp)
1662 {
1663         struct page_frag *alloc_frag = &current->task_frag;
1664         struct bpf_prog *xdp_prog;
1665         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1666         char *buf;
1667         size_t copied;
1668         int pad = TUN_RX_PAD;
1669         int err = 0;
1670
1671         rcu_read_lock();
1672         xdp_prog = rcu_dereference(tun->xdp_prog);
1673         if (xdp_prog)
1674                 pad += XDP_PACKET_HEADROOM;
1675         buflen += SKB_DATA_ALIGN(len + pad);
1676         rcu_read_unlock();
1677
1678         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1679         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1680                 return ERR_PTR(-ENOMEM);
1681
1682         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1683         copied = copy_page_from_iter(alloc_frag->page,
1684                                      alloc_frag->offset + pad,
1685                                      len, from);
1686         if (copied != len)
1687                 return ERR_PTR(-EFAULT);
1688
1689         /* There's a small window that XDP may be set after the check
1690          * of xdp_prog above, this should be rare and for simplicity
1691          * we do XDP on skb in case the headroom is not enough.
1692          */
1693         if (hdr->gso_type || !xdp_prog) {
1694                 *skb_xdp = 1;
1695                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1696                                        pad);
1697         }
1698
1699         *skb_xdp = 0;
1700
1701         local_bh_disable();
1702         rcu_read_lock();
1703         xdp_prog = rcu_dereference(tun->xdp_prog);
1704         if (xdp_prog) {
1705                 struct xdp_buff xdp;
1706                 u32 act;
1707
1708                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1709                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1710
1711                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1712                 if (act == XDP_REDIRECT || act == XDP_TX) {
1713                         get_page(alloc_frag->page);
1714                         alloc_frag->offset += buflen;
1715                 }
1716                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1717                 if (err < 0) {
1718                         if (act == XDP_REDIRECT || act == XDP_TX)
1719                                 put_page(alloc_frag->page);
1720                         goto out;
1721                 }
1722
1723                 if (err == XDP_REDIRECT)
1724                         xdp_do_flush();
1725                 if (err != XDP_PASS)
1726                         goto out;
1727
1728                 pad = xdp.data - xdp.data_hard_start;
1729                 len = xdp.data_end - xdp.data;
1730         }
1731         rcu_read_unlock();
1732         local_bh_enable();
1733
1734         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1735
1736 out:
1737         rcu_read_unlock();
1738         local_bh_enable();
1739         return NULL;
1740 }
1741
1742 /* Get packet from user space buffer */
1743 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1744                             void *msg_control, struct iov_iter *from,
1745                             int noblock, bool more)
1746 {
1747         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1748         struct sk_buff *skb;
1749         size_t total_len = iov_iter_count(from);
1750         size_t len = total_len, align = tun->align, linear;
1751         struct virtio_net_hdr gso = { 0 };
1752         int good_linear;
1753         int copylen;
1754         bool zerocopy = false;
1755         int err;
1756         u32 rxhash = 0;
1757         int skb_xdp = 1;
1758         bool frags = tun_napi_frags_enabled(tfile);
1759         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1760
1761         if (!(tun->flags & IFF_NO_PI)) {
1762                 if (len < sizeof(pi))
1763                         return -EINVAL;
1764                 len -= sizeof(pi);
1765
1766                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1767                         return -EFAULT;
1768         }
1769
1770         if (tun->flags & IFF_VNET_HDR) {
1771                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1772
1773                 if (len < vnet_hdr_sz)
1774                         return -EINVAL;
1775                 len -= vnet_hdr_sz;
1776
1777                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1778                         return -EFAULT;
1779
1780                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1781                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1782                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1783
1784                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1785                         return -EINVAL;
1786                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1787         }
1788
1789         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1790                 align += NET_IP_ALIGN;
1791                 if (unlikely(len < ETH_HLEN ||
1792                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1793                         return -EINVAL;
1794         }
1795
1796         good_linear = SKB_MAX_HEAD(align);
1797
1798         if (msg_control) {
1799                 struct iov_iter i = *from;
1800
1801                 /* There are 256 bytes to be copied in skb, so there is
1802                  * enough room for skb expand head in case it is used.
1803                  * The rest of the buffer is mapped from userspace.
1804                  */
1805                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1806                 if (copylen > good_linear)
1807                         copylen = good_linear;
1808                 linear = copylen;
1809                 iov_iter_advance(&i, copylen);
1810                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1811                         zerocopy = true;
1812         }
1813
1814         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1815                 /* For the packet that is not easy to be processed
1816                  * (e.g gso or jumbo packet), we will do it at after
1817                  * skb was created with generic XDP routine.
1818                  */
1819                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1820                 err = PTR_ERR_OR_ZERO(skb);
1821                 if (err)
1822                         goto drop;
1823                 if (!skb)
1824                         return total_len;
1825         } else {
1826                 if (!zerocopy) {
1827                         copylen = len;
1828                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1829                                 linear = good_linear;
1830                         else
1831                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1832                 }
1833
1834                 if (frags) {
1835                         mutex_lock(&tfile->napi_mutex);
1836                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1837                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1838                          * If zerocopy is enabled, then this layout will be
1839                          * overwritten by zerocopy_sg_from_iter().
1840                          */
1841                         zerocopy = false;
1842                 } else {
1843                         if (!linear)
1844                                 linear = min_t(size_t, good_linear, copylen);
1845
1846                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1847                                             noblock);
1848                 }
1849
1850                 err = PTR_ERR_OR_ZERO(skb);
1851                 if (err)
1852                         goto drop;
1853
1854                 if (zerocopy)
1855                         err = zerocopy_sg_from_iter(skb, from);
1856                 else
1857                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1858
1859                 if (err) {
1860                         err = -EFAULT;
1861                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1862                         goto drop;
1863                 }
1864         }
1865
1866         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1867                 atomic_long_inc(&tun->rx_frame_errors);
1868                 err = -EINVAL;
1869                 goto free_skb;
1870         }
1871
1872         switch (tun->flags & TUN_TYPE_MASK) {
1873         case IFF_TUN:
1874                 if (tun->flags & IFF_NO_PI) {
1875                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1876
1877                         switch (ip_version) {
1878                         case 4:
1879                                 pi.proto = htons(ETH_P_IP);
1880                                 break;
1881                         case 6:
1882                                 pi.proto = htons(ETH_P_IPV6);
1883                                 break;
1884                         default:
1885                                 err = -EINVAL;
1886                                 goto drop;
1887                         }
1888                 }
1889
1890                 skb_reset_mac_header(skb);
1891                 skb->protocol = pi.proto;
1892                 skb->dev = tun->dev;
1893                 break;
1894         case IFF_TAP:
1895                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1896                         err = -ENOMEM;
1897                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1898                         goto drop;
1899                 }
1900                 skb->protocol = eth_type_trans(skb, tun->dev);
1901                 break;
1902         }
1903
1904         /* copy skb_ubuf_info for callback when skb has no error */
1905         if (zerocopy) {
1906                 skb_zcopy_init(skb, msg_control);
1907         } else if (msg_control) {
1908                 struct ubuf_info *uarg = msg_control;
1909                 uarg->callback(NULL, uarg, false);
1910         }
1911
1912         skb_reset_network_header(skb);
1913         skb_probe_transport_header(skb);
1914         skb_record_rx_queue(skb, tfile->queue_index);
1915
1916         if (skb_xdp) {
1917                 struct bpf_prog *xdp_prog;
1918                 int ret;
1919
1920                 local_bh_disable();
1921                 rcu_read_lock();
1922                 xdp_prog = rcu_dereference(tun->xdp_prog);
1923                 if (xdp_prog) {
1924                         ret = do_xdp_generic(xdp_prog, skb);
1925                         if (ret != XDP_PASS) {
1926                                 rcu_read_unlock();
1927                                 local_bh_enable();
1928                                 goto unlock_frags;
1929                         }
1930                 }
1931                 rcu_read_unlock();
1932                 local_bh_enable();
1933         }
1934
1935         /* Compute the costly rx hash only if needed for flow updates.
1936          * We may get a very small possibility of OOO during switching, not
1937          * worth to optimize.
1938          */
1939         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1940             !tfile->detached)
1941                 rxhash = __skb_get_hash_symmetric(skb);
1942
1943         rcu_read_lock();
1944         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1945                 err = -EIO;
1946                 rcu_read_unlock();
1947                 drop_reason = SKB_DROP_REASON_DEV_READY;
1948                 goto drop;
1949         }
1950
1951         if (frags) {
1952                 u32 headlen;
1953
1954                 /* Exercise flow dissector code path. */
1955                 skb_push(skb, ETH_HLEN);
1956                 headlen = eth_get_headlen(tun->dev, skb->data,
1957                                           skb_headlen(skb));
1958
1959                 if (unlikely(headlen > skb_headlen(skb))) {
1960                         WARN_ON_ONCE(1);
1961                         err = -ENOMEM;
1962                         dev_core_stats_rx_dropped_inc(tun->dev);
1963 napi_busy:
1964                         napi_free_frags(&tfile->napi);
1965                         rcu_read_unlock();
1966                         mutex_unlock(&tfile->napi_mutex);
1967                         return err;
1968                 }
1969
1970                 if (likely(napi_schedule_prep(&tfile->napi))) {
1971                         local_bh_disable();
1972                         napi_gro_frags(&tfile->napi);
1973                         napi_complete(&tfile->napi);
1974                         local_bh_enable();
1975                 } else {
1976                         err = -EBUSY;
1977                         goto napi_busy;
1978                 }
1979                 mutex_unlock(&tfile->napi_mutex);
1980         } else if (tfile->napi_enabled) {
1981                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1982                 int queue_len;
1983
1984                 spin_lock_bh(&queue->lock);
1985
1986                 if (unlikely(tfile->detached)) {
1987                         spin_unlock_bh(&queue->lock);
1988                         rcu_read_unlock();
1989                         err = -EBUSY;
1990                         goto free_skb;
1991                 }
1992
1993                 __skb_queue_tail(queue, skb);
1994                 queue_len = skb_queue_len(queue);
1995                 spin_unlock(&queue->lock);
1996
1997                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1998                         napi_schedule(&tfile->napi);
1999
2000                 local_bh_enable();
2001         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2002                 tun_rx_batched(tun, tfile, skb, more);
2003         } else {
2004                 netif_rx(skb);
2005         }
2006         rcu_read_unlock();
2007
2008         preempt_disable();
2009         dev_sw_netstats_rx_add(tun->dev, len);
2010         preempt_enable();
2011
2012         if (rxhash)
2013                 tun_flow_update(tun, rxhash, tfile);
2014
2015         return total_len;
2016
2017 drop:
2018         if (err != -EAGAIN)
2019                 dev_core_stats_rx_dropped_inc(tun->dev);
2020
2021 free_skb:
2022         if (!IS_ERR_OR_NULL(skb))
2023                 kfree_skb_reason(skb, drop_reason);
2024
2025 unlock_frags:
2026         if (frags) {
2027                 tfile->napi.skb = NULL;
2028                 mutex_unlock(&tfile->napi_mutex);
2029         }
2030
2031         return err ?: total_len;
2032 }
2033
2034 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2035 {
2036         struct file *file = iocb->ki_filp;
2037         struct tun_file *tfile = file->private_data;
2038         struct tun_struct *tun = tun_get(tfile);
2039         ssize_t result;
2040         int noblock = 0;
2041
2042         if (!tun)
2043                 return -EBADFD;
2044
2045         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2046                 noblock = 1;
2047
2048         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2049
2050         tun_put(tun);
2051         return result;
2052 }
2053
2054 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2055                                 struct tun_file *tfile,
2056                                 struct xdp_frame *xdp_frame,
2057                                 struct iov_iter *iter)
2058 {
2059         int vnet_hdr_sz = 0;
2060         size_t size = xdp_frame->len;
2061         size_t ret;
2062
2063         if (tun->flags & IFF_VNET_HDR) {
2064                 struct virtio_net_hdr gso = { 0 };
2065
2066                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2067                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2068                         return -EINVAL;
2069                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2070                              sizeof(gso)))
2071                         return -EFAULT;
2072                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2073         }
2074
2075         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2076
2077         preempt_disable();
2078         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2079         preempt_enable();
2080
2081         return ret;
2082 }
2083
2084 /* Put packet to the user space buffer */
2085 static ssize_t tun_put_user(struct tun_struct *tun,
2086                             struct tun_file *tfile,
2087                             struct sk_buff *skb,
2088                             struct iov_iter *iter)
2089 {
2090         struct tun_pi pi = { 0, skb->protocol };
2091         ssize_t total;
2092         int vlan_offset = 0;
2093         int vlan_hlen = 0;
2094         int vnet_hdr_sz = 0;
2095
2096         if (skb_vlan_tag_present(skb))
2097                 vlan_hlen = VLAN_HLEN;
2098
2099         if (tun->flags & IFF_VNET_HDR)
2100                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2101
2102         total = skb->len + vlan_hlen + vnet_hdr_sz;
2103
2104         if (!(tun->flags & IFF_NO_PI)) {
2105                 if (iov_iter_count(iter) < sizeof(pi))
2106                         return -EINVAL;
2107
2108                 total += sizeof(pi);
2109                 if (iov_iter_count(iter) < total) {
2110                         /* Packet will be striped */
2111                         pi.flags |= TUN_PKT_STRIP;
2112                 }
2113
2114                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2115                         return -EFAULT;
2116         }
2117
2118         if (vnet_hdr_sz) {
2119                 struct virtio_net_hdr gso;
2120
2121                 if (iov_iter_count(iter) < vnet_hdr_sz)
2122                         return -EINVAL;
2123
2124                 if (virtio_net_hdr_from_skb(skb, &gso,
2125                                             tun_is_little_endian(tun), true,
2126                                             vlan_hlen)) {
2127                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2128                         pr_err("unexpected GSO type: "
2129                                "0x%x, gso_size %d, hdr_len %d\n",
2130                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2131                                tun16_to_cpu(tun, gso.hdr_len));
2132                         print_hex_dump(KERN_ERR, "tun: ",
2133                                        DUMP_PREFIX_NONE,
2134                                        16, 1, skb->head,
2135                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2136                         WARN_ON_ONCE(1);
2137                         return -EINVAL;
2138                 }
2139
2140                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2141                         return -EFAULT;
2142
2143                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2144         }
2145
2146         if (vlan_hlen) {
2147                 int ret;
2148                 struct veth veth;
2149
2150                 veth.h_vlan_proto = skb->vlan_proto;
2151                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2152
2153                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2154
2155                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2156                 if (ret || !iov_iter_count(iter))
2157                         goto done;
2158
2159                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2160                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2161                         goto done;
2162         }
2163
2164         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2165
2166 done:
2167         /* caller is in process context, */
2168         preempt_disable();
2169         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2170         preempt_enable();
2171
2172         return total;
2173 }
2174
2175 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2176 {
2177         DECLARE_WAITQUEUE(wait, current);
2178         void *ptr = NULL;
2179         int error = 0;
2180
2181         ptr = ptr_ring_consume(&tfile->tx_ring);
2182         if (ptr)
2183                 goto out;
2184         if (noblock) {
2185                 error = -EAGAIN;
2186                 goto out;
2187         }
2188
2189         add_wait_queue(&tfile->socket.wq.wait, &wait);
2190
2191         while (1) {
2192                 set_current_state(TASK_INTERRUPTIBLE);
2193                 ptr = ptr_ring_consume(&tfile->tx_ring);
2194                 if (ptr)
2195                         break;
2196                 if (signal_pending(current)) {
2197                         error = -ERESTARTSYS;
2198                         break;
2199                 }
2200                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2201                         error = -EFAULT;
2202                         break;
2203                 }
2204
2205                 schedule();
2206         }
2207
2208         __set_current_state(TASK_RUNNING);
2209         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2210
2211 out:
2212         *err = error;
2213         return ptr;
2214 }
2215
2216 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2217                            struct iov_iter *to,
2218                            int noblock, void *ptr)
2219 {
2220         ssize_t ret;
2221         int err;
2222
2223         if (!iov_iter_count(to)) {
2224                 tun_ptr_free(ptr);
2225                 return 0;
2226         }
2227
2228         if (!ptr) {
2229                 /* Read frames from ring */
2230                 ptr = tun_ring_recv(tfile, noblock, &err);
2231                 if (!ptr)
2232                         return err;
2233         }
2234
2235         if (tun_is_xdp_frame(ptr)) {
2236                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2237
2238                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2239                 xdp_return_frame(xdpf);
2240         } else {
2241                 struct sk_buff *skb = ptr;
2242
2243                 ret = tun_put_user(tun, tfile, skb, to);
2244                 if (unlikely(ret < 0))
2245                         kfree_skb(skb);
2246                 else
2247                         consume_skb(skb);
2248         }
2249
2250         return ret;
2251 }
2252
2253 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2254 {
2255         struct file *file = iocb->ki_filp;
2256         struct tun_file *tfile = file->private_data;
2257         struct tun_struct *tun = tun_get(tfile);
2258         ssize_t len = iov_iter_count(to), ret;
2259         int noblock = 0;
2260
2261         if (!tun)
2262                 return -EBADFD;
2263
2264         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2265                 noblock = 1;
2266
2267         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2268         ret = min_t(ssize_t, ret, len);
2269         if (ret > 0)
2270                 iocb->ki_pos = ret;
2271         tun_put(tun);
2272         return ret;
2273 }
2274
2275 static void tun_prog_free(struct rcu_head *rcu)
2276 {
2277         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2278
2279         bpf_prog_destroy(prog->prog);
2280         kfree(prog);
2281 }
2282
2283 static int __tun_set_ebpf(struct tun_struct *tun,
2284                           struct tun_prog __rcu **prog_p,
2285                           struct bpf_prog *prog)
2286 {
2287         struct tun_prog *old, *new = NULL;
2288
2289         if (prog) {
2290                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2291                 if (!new)
2292                         return -ENOMEM;
2293                 new->prog = prog;
2294         }
2295
2296         spin_lock_bh(&tun->lock);
2297         old = rcu_dereference_protected(*prog_p,
2298                                         lockdep_is_held(&tun->lock));
2299         rcu_assign_pointer(*prog_p, new);
2300         spin_unlock_bh(&tun->lock);
2301
2302         if (old)
2303                 call_rcu(&old->rcu, tun_prog_free);
2304
2305         return 0;
2306 }
2307
2308 static void tun_free_netdev(struct net_device *dev)
2309 {
2310         struct tun_struct *tun = netdev_priv(dev);
2311
2312         BUG_ON(!(list_empty(&tun->disabled)));
2313
2314         free_percpu(dev->tstats);
2315         tun_flow_uninit(tun);
2316         security_tun_dev_free_security(tun->security);
2317         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2318         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2319 }
2320
2321 static void tun_setup(struct net_device *dev)
2322 {
2323         struct tun_struct *tun = netdev_priv(dev);
2324
2325         tun->owner = INVALID_UID;
2326         tun->group = INVALID_GID;
2327         tun_default_link_ksettings(dev, &tun->link_ksettings);
2328
2329         dev->ethtool_ops = &tun_ethtool_ops;
2330         dev->needs_free_netdev = true;
2331         dev->priv_destructor = tun_free_netdev;
2332         /* We prefer our own queue length */
2333         dev->tx_queue_len = TUN_READQ_SIZE;
2334 }
2335
2336 /* Trivial set of netlink ops to allow deleting tun or tap
2337  * device with netlink.
2338  */
2339 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2340                         struct netlink_ext_ack *extack)
2341 {
2342         NL_SET_ERR_MSG(extack,
2343                        "tun/tap creation via rtnetlink is not supported.");
2344         return -EOPNOTSUPP;
2345 }
2346
2347 static size_t tun_get_size(const struct net_device *dev)
2348 {
2349         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2350         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2351
2352         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2353                nla_total_size(sizeof(gid_t)) + /* GROUP */
2354                nla_total_size(sizeof(u8)) + /* TYPE */
2355                nla_total_size(sizeof(u8)) + /* PI */
2356                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2357                nla_total_size(sizeof(u8)) + /* PERSIST */
2358                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2359                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2360                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2361                0;
2362 }
2363
2364 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2365 {
2366         struct tun_struct *tun = netdev_priv(dev);
2367
2368         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2369                 goto nla_put_failure;
2370         if (uid_valid(tun->owner) &&
2371             nla_put_u32(skb, IFLA_TUN_OWNER,
2372                         from_kuid_munged(current_user_ns(), tun->owner)))
2373                 goto nla_put_failure;
2374         if (gid_valid(tun->group) &&
2375             nla_put_u32(skb, IFLA_TUN_GROUP,
2376                         from_kgid_munged(current_user_ns(), tun->group)))
2377                 goto nla_put_failure;
2378         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2379                 goto nla_put_failure;
2380         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2381                 goto nla_put_failure;
2382         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2383                 goto nla_put_failure;
2384         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2385                        !!(tun->flags & IFF_MULTI_QUEUE)))
2386                 goto nla_put_failure;
2387         if (tun->flags & IFF_MULTI_QUEUE) {
2388                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2389                         goto nla_put_failure;
2390                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2391                                 tun->numdisabled))
2392                         goto nla_put_failure;
2393         }
2394
2395         return 0;
2396
2397 nla_put_failure:
2398         return -EMSGSIZE;
2399 }
2400
2401 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2402         .kind           = DRV_NAME,
2403         .priv_size      = sizeof(struct tun_struct),
2404         .setup          = tun_setup,
2405         .validate       = tun_validate,
2406         .get_size       = tun_get_size,
2407         .fill_info      = tun_fill_info,
2408 };
2409
2410 static void tun_sock_write_space(struct sock *sk)
2411 {
2412         struct tun_file *tfile;
2413         wait_queue_head_t *wqueue;
2414
2415         if (!sock_writeable(sk))
2416                 return;
2417
2418         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2419                 return;
2420
2421         wqueue = sk_sleep(sk);
2422         if (wqueue && waitqueue_active(wqueue))
2423                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2424                                                 EPOLLWRNORM | EPOLLWRBAND);
2425
2426         tfile = container_of(sk, struct tun_file, sk);
2427         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2428 }
2429
2430 static void tun_put_page(struct tun_page *tpage)
2431 {
2432         if (tpage->page)
2433                 __page_frag_cache_drain(tpage->page, tpage->count);
2434 }
2435
2436 static int tun_xdp_one(struct tun_struct *tun,
2437                        struct tun_file *tfile,
2438                        struct xdp_buff *xdp, int *flush,
2439                        struct tun_page *tpage)
2440 {
2441         unsigned int datasize = xdp->data_end - xdp->data;
2442         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2443         struct virtio_net_hdr *gso = &hdr->gso;
2444         struct bpf_prog *xdp_prog;
2445         struct sk_buff *skb = NULL;
2446         struct sk_buff_head *queue;
2447         u32 rxhash = 0, act;
2448         int buflen = hdr->buflen;
2449         int ret = 0;
2450         bool skb_xdp = false;
2451         struct page *page;
2452
2453         xdp_prog = rcu_dereference(tun->xdp_prog);
2454         if (xdp_prog) {
2455                 if (gso->gso_type) {
2456                         skb_xdp = true;
2457                         goto build;
2458                 }
2459
2460                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2461                 xdp_set_data_meta_invalid(xdp);
2462
2463                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2464                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2465                 if (ret < 0) {
2466                         put_page(virt_to_head_page(xdp->data));
2467                         return ret;
2468                 }
2469
2470                 switch (ret) {
2471                 case XDP_REDIRECT:
2472                         *flush = true;
2473                         fallthrough;
2474                 case XDP_TX:
2475                         return 0;
2476                 case XDP_PASS:
2477                         break;
2478                 default:
2479                         page = virt_to_head_page(xdp->data);
2480                         if (tpage->page == page) {
2481                                 ++tpage->count;
2482                         } else {
2483                                 tun_put_page(tpage);
2484                                 tpage->page = page;
2485                                 tpage->count = 1;
2486                         }
2487                         return 0;
2488                 }
2489         }
2490
2491 build:
2492         skb = build_skb(xdp->data_hard_start, buflen);
2493         if (!skb) {
2494                 ret = -ENOMEM;
2495                 goto out;
2496         }
2497
2498         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2499         skb_put(skb, xdp->data_end - xdp->data);
2500
2501         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2502                 atomic_long_inc(&tun->rx_frame_errors);
2503                 kfree_skb(skb);
2504                 ret = -EINVAL;
2505                 goto out;
2506         }
2507
2508         skb->protocol = eth_type_trans(skb, tun->dev);
2509         skb_reset_network_header(skb);
2510         skb_probe_transport_header(skb);
2511         skb_record_rx_queue(skb, tfile->queue_index);
2512
2513         if (skb_xdp) {
2514                 ret = do_xdp_generic(xdp_prog, skb);
2515                 if (ret != XDP_PASS) {
2516                         ret = 0;
2517                         goto out;
2518                 }
2519         }
2520
2521         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2522             !tfile->detached)
2523                 rxhash = __skb_get_hash_symmetric(skb);
2524
2525         if (tfile->napi_enabled) {
2526                 queue = &tfile->sk.sk_write_queue;
2527                 spin_lock(&queue->lock);
2528
2529                 if (unlikely(tfile->detached)) {
2530                         spin_unlock(&queue->lock);
2531                         kfree_skb(skb);
2532                         return -EBUSY;
2533                 }
2534
2535                 __skb_queue_tail(queue, skb);
2536                 spin_unlock(&queue->lock);
2537                 ret = 1;
2538         } else {
2539                 netif_receive_skb(skb);
2540                 ret = 0;
2541         }
2542
2543         /* No need to disable preemption here since this function is
2544          * always called with bh disabled
2545          */
2546         dev_sw_netstats_rx_add(tun->dev, datasize);
2547
2548         if (rxhash)
2549                 tun_flow_update(tun, rxhash, tfile);
2550
2551 out:
2552         return ret;
2553 }
2554
2555 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2556 {
2557         int ret, i;
2558         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2559         struct tun_struct *tun = tun_get(tfile);
2560         struct tun_msg_ctl *ctl = m->msg_control;
2561         struct xdp_buff *xdp;
2562
2563         if (!tun)
2564                 return -EBADFD;
2565
2566         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2567             ctl && ctl->type == TUN_MSG_PTR) {
2568                 struct tun_page tpage;
2569                 int n = ctl->num;
2570                 int flush = 0, queued = 0;
2571
2572                 memset(&tpage, 0, sizeof(tpage));
2573
2574                 local_bh_disable();
2575                 rcu_read_lock();
2576
2577                 for (i = 0; i < n; i++) {
2578                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2579                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2580                         if (ret > 0)
2581                                 queued += ret;
2582                 }
2583
2584                 if (flush)
2585                         xdp_do_flush();
2586
2587                 if (tfile->napi_enabled && queued > 0)
2588                         napi_schedule(&tfile->napi);
2589
2590                 rcu_read_unlock();
2591                 local_bh_enable();
2592
2593                 tun_put_page(&tpage);
2594
2595                 ret = total_len;
2596                 goto out;
2597         }
2598
2599         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2600                            m->msg_flags & MSG_DONTWAIT,
2601                            m->msg_flags & MSG_MORE);
2602 out:
2603         tun_put(tun);
2604         return ret;
2605 }
2606
2607 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2608                        int flags)
2609 {
2610         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2611         struct tun_struct *tun = tun_get(tfile);
2612         void *ptr = m->msg_control;
2613         int ret;
2614
2615         if (!tun) {
2616                 ret = -EBADFD;
2617                 goto out_free;
2618         }
2619
2620         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2621                 ret = -EINVAL;
2622                 goto out_put_tun;
2623         }
2624         if (flags & MSG_ERRQUEUE) {
2625                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2626                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2627                 goto out;
2628         }
2629         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2630         if (ret > (ssize_t)total_len) {
2631                 m->msg_flags |= MSG_TRUNC;
2632                 ret = flags & MSG_TRUNC ? ret : total_len;
2633         }
2634 out:
2635         tun_put(tun);
2636         return ret;
2637
2638 out_put_tun:
2639         tun_put(tun);
2640 out_free:
2641         tun_ptr_free(ptr);
2642         return ret;
2643 }
2644
2645 static int tun_ptr_peek_len(void *ptr)
2646 {
2647         if (likely(ptr)) {
2648                 if (tun_is_xdp_frame(ptr)) {
2649                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2650
2651                         return xdpf->len;
2652                 }
2653                 return __skb_array_len_with_tag(ptr);
2654         } else {
2655                 return 0;
2656         }
2657 }
2658
2659 static int tun_peek_len(struct socket *sock)
2660 {
2661         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2662         struct tun_struct *tun;
2663         int ret = 0;
2664
2665         tun = tun_get(tfile);
2666         if (!tun)
2667                 return 0;
2668
2669         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2670         tun_put(tun);
2671
2672         return ret;
2673 }
2674
2675 /* Ops structure to mimic raw sockets with tun */
2676 static const struct proto_ops tun_socket_ops = {
2677         .peek_len = tun_peek_len,
2678         .sendmsg = tun_sendmsg,
2679         .recvmsg = tun_recvmsg,
2680 };
2681
2682 static struct proto tun_proto = {
2683         .name           = "tun",
2684         .owner          = THIS_MODULE,
2685         .obj_size       = sizeof(struct tun_file),
2686 };
2687
2688 static int tun_flags(struct tun_struct *tun)
2689 {
2690         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2691 }
2692
2693 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2694                               char *buf)
2695 {
2696         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2697         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2698 }
2699
2700 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2701                           char *buf)
2702 {
2703         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2704         return uid_valid(tun->owner)?
2705                 sysfs_emit(buf, "%u\n",
2706                            from_kuid_munged(current_user_ns(), tun->owner)) :
2707                 sysfs_emit(buf, "-1\n");
2708 }
2709
2710 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2711                           char *buf)
2712 {
2713         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2714         return gid_valid(tun->group) ?
2715                 sysfs_emit(buf, "%u\n",
2716                            from_kgid_munged(current_user_ns(), tun->group)) :
2717                 sysfs_emit(buf, "-1\n");
2718 }
2719
2720 static DEVICE_ATTR_RO(tun_flags);
2721 static DEVICE_ATTR_RO(owner);
2722 static DEVICE_ATTR_RO(group);
2723
2724 static struct attribute *tun_dev_attrs[] = {
2725         &dev_attr_tun_flags.attr,
2726         &dev_attr_owner.attr,
2727         &dev_attr_group.attr,
2728         NULL
2729 };
2730
2731 static const struct attribute_group tun_attr_group = {
2732         .attrs = tun_dev_attrs
2733 };
2734
2735 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2736 {
2737         struct tun_struct *tun;
2738         struct tun_file *tfile = file->private_data;
2739         struct net_device *dev;
2740         int err;
2741
2742         if (tfile->detached)
2743                 return -EINVAL;
2744
2745         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2746                 if (!capable(CAP_NET_ADMIN))
2747                         return -EPERM;
2748
2749                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2750                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2751                         return -EINVAL;
2752         }
2753
2754         dev = __dev_get_by_name(net, ifr->ifr_name);
2755         if (dev) {
2756                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2757                         return -EBUSY;
2758                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2759                         tun = netdev_priv(dev);
2760                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2761                         tun = netdev_priv(dev);
2762                 else
2763                         return -EINVAL;
2764
2765                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2766                     !!(tun->flags & IFF_MULTI_QUEUE))
2767                         return -EINVAL;
2768
2769                 if (tun_not_capable(tun))
2770                         return -EPERM;
2771                 err = security_tun_dev_open(tun->security);
2772                 if (err < 0)
2773                         return err;
2774
2775                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2776                                  ifr->ifr_flags & IFF_NAPI,
2777                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2778                 if (err < 0)
2779                         return err;
2780
2781                 if (tun->flags & IFF_MULTI_QUEUE &&
2782                     (tun->numqueues + tun->numdisabled > 1)) {
2783                         /* One or more queue has already been attached, no need
2784                          * to initialize the device again.
2785                          */
2786                         netdev_state_change(dev);
2787                         return 0;
2788                 }
2789
2790                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2791                               (ifr->ifr_flags & TUN_FEATURES);
2792
2793                 netdev_state_change(dev);
2794         } else {
2795                 char *name;
2796                 unsigned long flags = 0;
2797                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2798                              MAX_TAP_QUEUES : 1;
2799
2800                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2801                         return -EPERM;
2802                 err = security_tun_dev_create();
2803                 if (err < 0)
2804                         return err;
2805
2806                 /* Set dev type */
2807                 if (ifr->ifr_flags & IFF_TUN) {
2808                         /* TUN device */
2809                         flags |= IFF_TUN;
2810                         name = "tun%d";
2811                 } else if (ifr->ifr_flags & IFF_TAP) {
2812                         /* TAP device */
2813                         flags |= IFF_TAP;
2814                         name = "tap%d";
2815                 } else
2816                         return -EINVAL;
2817
2818                 if (*ifr->ifr_name)
2819                         name = ifr->ifr_name;
2820
2821                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2822                                        NET_NAME_UNKNOWN, tun_setup, queues,
2823                                        queues);
2824
2825                 if (!dev)
2826                         return -ENOMEM;
2827
2828                 dev_net_set(dev, net);
2829                 dev->rtnl_link_ops = &tun_link_ops;
2830                 dev->ifindex = tfile->ifindex;
2831                 dev->sysfs_groups[0] = &tun_attr_group;
2832
2833                 tun = netdev_priv(dev);
2834                 tun->dev = dev;
2835                 tun->flags = flags;
2836                 tun->txflt.count = 0;
2837                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2838
2839                 tun->align = NET_SKB_PAD;
2840                 tun->filter_attached = false;
2841                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2842                 tun->rx_batched = 0;
2843                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2844
2845                 tun->ifr = ifr;
2846                 tun->file = file;
2847
2848                 tun_net_initialize(dev);
2849
2850                 err = register_netdevice(tun->dev);
2851                 if (err < 0) {
2852                         free_netdev(dev);
2853                         return err;
2854                 }
2855                 /* free_netdev() won't check refcnt, to avoid race
2856                  * with dev_put() we need publish tun after registration.
2857                  */
2858                 rcu_assign_pointer(tfile->tun, tun);
2859         }
2860
2861         if (ifr->ifr_flags & IFF_NO_CARRIER)
2862                 netif_carrier_off(tun->dev);
2863         else
2864                 netif_carrier_on(tun->dev);
2865
2866         /* Make sure persistent devices do not get stuck in
2867          * xoff state.
2868          */
2869         if (netif_running(tun->dev))
2870                 netif_tx_wake_all_queues(tun->dev);
2871
2872         strcpy(ifr->ifr_name, tun->dev->name);
2873         return 0;
2874 }
2875
2876 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2877 {
2878         strcpy(ifr->ifr_name, tun->dev->name);
2879
2880         ifr->ifr_flags = tun_flags(tun);
2881
2882 }
2883
2884 /* This is like a cut-down ethtool ops, except done via tun fd so no
2885  * privs required. */
2886 static int set_offload(struct tun_struct *tun, unsigned long arg)
2887 {
2888         netdev_features_t features = 0;
2889
2890         if (arg & TUN_F_CSUM) {
2891                 features |= NETIF_F_HW_CSUM;
2892                 arg &= ~TUN_F_CSUM;
2893
2894                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2895                         if (arg & TUN_F_TSO_ECN) {
2896                                 features |= NETIF_F_TSO_ECN;
2897                                 arg &= ~TUN_F_TSO_ECN;
2898                         }
2899                         if (arg & TUN_F_TSO4)
2900                                 features |= NETIF_F_TSO;
2901                         if (arg & TUN_F_TSO6)
2902                                 features |= NETIF_F_TSO6;
2903                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2904                 }
2905
2906                 arg &= ~TUN_F_UFO;
2907
2908                 /* TODO: for now USO4 and USO6 should work simultaneously */
2909                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2910                         features |= NETIF_F_GSO_UDP_L4;
2911                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2912                 }
2913         }
2914
2915         /* This gives the user a way to test for new features in future by
2916          * trying to set them. */
2917         if (arg)
2918                 return -EINVAL;
2919
2920         tun->set_features = features;
2921         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2922         tun->dev->wanted_features |= features;
2923         netdev_update_features(tun->dev);
2924
2925         return 0;
2926 }
2927
2928 static void tun_detach_filter(struct tun_struct *tun, int n)
2929 {
2930         int i;
2931         struct tun_file *tfile;
2932
2933         for (i = 0; i < n; i++) {
2934                 tfile = rtnl_dereference(tun->tfiles[i]);
2935                 lock_sock(tfile->socket.sk);
2936                 sk_detach_filter(tfile->socket.sk);
2937                 release_sock(tfile->socket.sk);
2938         }
2939
2940         tun->filter_attached = false;
2941 }
2942
2943 static int tun_attach_filter(struct tun_struct *tun)
2944 {
2945         int i, ret = 0;
2946         struct tun_file *tfile;
2947
2948         for (i = 0; i < tun->numqueues; i++) {
2949                 tfile = rtnl_dereference(tun->tfiles[i]);
2950                 lock_sock(tfile->socket.sk);
2951                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2952                 release_sock(tfile->socket.sk);
2953                 if (ret) {
2954                         tun_detach_filter(tun, i);
2955                         return ret;
2956                 }
2957         }
2958
2959         tun->filter_attached = true;
2960         return ret;
2961 }
2962
2963 static void tun_set_sndbuf(struct tun_struct *tun)
2964 {
2965         struct tun_file *tfile;
2966         int i;
2967
2968         for (i = 0; i < tun->numqueues; i++) {
2969                 tfile = rtnl_dereference(tun->tfiles[i]);
2970                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2971         }
2972 }
2973
2974 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2975 {
2976         struct tun_file *tfile = file->private_data;
2977         struct tun_struct *tun;
2978         int ret = 0;
2979
2980         rtnl_lock();
2981
2982         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2983                 tun = tfile->detached;
2984                 if (!tun) {
2985                         ret = -EINVAL;
2986                         goto unlock;
2987                 }
2988                 ret = security_tun_dev_attach_queue(tun->security);
2989                 if (ret < 0)
2990                         goto unlock;
2991                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2992                                  tun->flags & IFF_NAPI_FRAGS, true);
2993         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2994                 tun = rtnl_dereference(tfile->tun);
2995                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2996                         ret = -EINVAL;
2997                 else
2998                         __tun_detach(tfile, false);
2999         } else
3000                 ret = -EINVAL;
3001
3002         if (ret >= 0)
3003                 netdev_state_change(tun->dev);
3004
3005 unlock:
3006         rtnl_unlock();
3007         return ret;
3008 }
3009
3010 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3011                         void __user *data)
3012 {
3013         struct bpf_prog *prog;
3014         int fd;
3015
3016         if (copy_from_user(&fd, data, sizeof(fd)))
3017                 return -EFAULT;
3018
3019         if (fd == -1) {
3020                 prog = NULL;
3021         } else {
3022                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3023                 if (IS_ERR(prog))
3024                         return PTR_ERR(prog);
3025         }
3026
3027         return __tun_set_ebpf(tun, prog_p, prog);
3028 }
3029
3030 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3031 static unsigned char tun_get_addr_len(unsigned short type)
3032 {
3033         switch (type) {
3034         case ARPHRD_IP6GRE:
3035         case ARPHRD_TUNNEL6:
3036                 return sizeof(struct in6_addr);
3037         case ARPHRD_IPGRE:
3038         case ARPHRD_TUNNEL:
3039         case ARPHRD_SIT:
3040                 return 4;
3041         case ARPHRD_ETHER:
3042                 return ETH_ALEN;
3043         case ARPHRD_IEEE802154:
3044         case ARPHRD_IEEE802154_MONITOR:
3045                 return IEEE802154_EXTENDED_ADDR_LEN;
3046         case ARPHRD_PHONET_PIPE:
3047         case ARPHRD_PPP:
3048         case ARPHRD_NONE:
3049                 return 0;
3050         case ARPHRD_6LOWPAN:
3051                 return EUI64_ADDR_LEN;
3052         case ARPHRD_FDDI:
3053                 return FDDI_K_ALEN;
3054         case ARPHRD_HIPPI:
3055                 return HIPPI_ALEN;
3056         case ARPHRD_IEEE802:
3057                 return FC_ALEN;
3058         case ARPHRD_ROSE:
3059                 return ROSE_ADDR_LEN;
3060         case ARPHRD_NETROM:
3061                 return AX25_ADDR_LEN;
3062         case ARPHRD_LOCALTLK:
3063                 return LTALK_ALEN;
3064         default:
3065                 return 0;
3066         }
3067 }
3068
3069 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3070                             unsigned long arg, int ifreq_len)
3071 {
3072         struct tun_file *tfile = file->private_data;
3073         struct net *net = sock_net(&tfile->sk);
3074         struct tun_struct *tun;
3075         void __user* argp = (void __user*)arg;
3076         unsigned int ifindex, carrier;
3077         struct ifreq ifr;
3078         kuid_t owner;
3079         kgid_t group;
3080         int sndbuf;
3081         int vnet_hdr_sz;
3082         int le;
3083         int ret;
3084         bool do_notify = false;
3085
3086         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3087             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3088                 if (copy_from_user(&ifr, argp, ifreq_len))
3089                         return -EFAULT;
3090         } else {
3091                 memset(&ifr, 0, sizeof(ifr));
3092         }
3093         if (cmd == TUNGETFEATURES) {
3094                 /* Currently this just means: "what IFF flags are valid?".
3095                  * This is needed because we never checked for invalid flags on
3096                  * TUNSETIFF.
3097                  */
3098                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3099                                 TUN_FEATURES, (unsigned int __user*)argp);
3100         } else if (cmd == TUNSETQUEUE) {
3101                 return tun_set_queue(file, &ifr);
3102         } else if (cmd == SIOCGSKNS) {
3103                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3104                         return -EPERM;
3105                 return open_related_ns(&net->ns, get_net_ns);
3106         }
3107
3108         rtnl_lock();
3109
3110         tun = tun_get(tfile);
3111         if (cmd == TUNSETIFF) {
3112                 ret = -EEXIST;
3113                 if (tun)
3114                         goto unlock;
3115
3116                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3117
3118                 ret = tun_set_iff(net, file, &ifr);
3119
3120                 if (ret)
3121                         goto unlock;
3122
3123                 if (copy_to_user(argp, &ifr, ifreq_len))
3124                         ret = -EFAULT;
3125                 goto unlock;
3126         }
3127         if (cmd == TUNSETIFINDEX) {
3128                 ret = -EPERM;
3129                 if (tun)
3130                         goto unlock;
3131
3132                 ret = -EFAULT;
3133                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3134                         goto unlock;
3135
3136                 ret = 0;
3137                 tfile->ifindex = ifindex;
3138                 goto unlock;
3139         }
3140
3141         ret = -EBADFD;
3142         if (!tun)
3143                 goto unlock;
3144
3145         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3146
3147         net = dev_net(tun->dev);
3148         ret = 0;
3149         switch (cmd) {
3150         case TUNGETIFF:
3151                 tun_get_iff(tun, &ifr);
3152
3153                 if (tfile->detached)
3154                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3155                 if (!tfile->socket.sk->sk_filter)
3156                         ifr.ifr_flags |= IFF_NOFILTER;
3157
3158                 if (copy_to_user(argp, &ifr, ifreq_len))
3159                         ret = -EFAULT;
3160                 break;
3161
3162         case TUNSETNOCSUM:
3163                 /* Disable/Enable checksum */
3164
3165                 /* [unimplemented] */
3166                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3167                            arg ? "disabled" : "enabled");
3168                 break;
3169
3170         case TUNSETPERSIST:
3171                 /* Disable/Enable persist mode. Keep an extra reference to the
3172                  * module to prevent the module being unprobed.
3173                  */
3174                 if (arg && !(tun->flags & IFF_PERSIST)) {
3175                         tun->flags |= IFF_PERSIST;
3176                         __module_get(THIS_MODULE);
3177                         do_notify = true;
3178                 }
3179                 if (!arg && (tun->flags & IFF_PERSIST)) {
3180                         tun->flags &= ~IFF_PERSIST;
3181                         module_put(THIS_MODULE);
3182                         do_notify = true;
3183                 }
3184
3185                 netif_info(tun, drv, tun->dev, "persist %s\n",
3186                            arg ? "enabled" : "disabled");
3187                 break;
3188
3189         case TUNSETOWNER:
3190                 /* Set owner of the device */
3191                 owner = make_kuid(current_user_ns(), arg);
3192                 if (!uid_valid(owner)) {
3193                         ret = -EINVAL;
3194                         break;
3195                 }
3196                 tun->owner = owner;
3197                 do_notify = true;
3198                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3199                            from_kuid(&init_user_ns, tun->owner));
3200                 break;
3201
3202         case TUNSETGROUP:
3203                 /* Set group of the device */
3204                 group = make_kgid(current_user_ns(), arg);
3205                 if (!gid_valid(group)) {
3206                         ret = -EINVAL;
3207                         break;
3208                 }
3209                 tun->group = group;
3210                 do_notify = true;
3211                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3212                            from_kgid(&init_user_ns, tun->group));
3213                 break;
3214
3215         case TUNSETLINK:
3216                 /* Only allow setting the type when the interface is down */
3217                 if (tun->dev->flags & IFF_UP) {
3218                         netif_info(tun, drv, tun->dev,
3219                                    "Linktype set failed because interface is up\n");
3220                         ret = -EBUSY;
3221                 } else {
3222                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3223                                                        tun->dev);
3224                         ret = notifier_to_errno(ret);
3225                         if (ret) {
3226                                 netif_info(tun, drv, tun->dev,
3227                                            "Refused to change device type\n");
3228                                 break;
3229                         }
3230                         tun->dev->type = (int) arg;
3231                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3232                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3233                                    tun->dev->type);
3234                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3235                                                  tun->dev);
3236                 }
3237                 break;
3238
3239         case TUNSETDEBUG:
3240                 tun->msg_enable = (u32)arg;
3241                 break;
3242
3243         case TUNSETOFFLOAD:
3244                 ret = set_offload(tun, arg);
3245                 break;
3246
3247         case TUNSETTXFILTER:
3248                 /* Can be set only for TAPs */
3249                 ret = -EINVAL;
3250                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3251                         break;
3252                 ret = update_filter(&tun->txflt, (void __user *)arg);
3253                 break;
3254
3255         case SIOCGIFHWADDR:
3256                 /* Get hw address */
3257                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3258                 if (copy_to_user(argp, &ifr, ifreq_len))
3259                         ret = -EFAULT;
3260                 break;
3261
3262         case SIOCSIFHWADDR:
3263                 /* Set hw address */
3264                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3265                 break;
3266
3267         case TUNGETSNDBUF:
3268                 sndbuf = tfile->socket.sk->sk_sndbuf;
3269                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3270                         ret = -EFAULT;
3271                 break;
3272
3273         case TUNSETSNDBUF:
3274                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3275                         ret = -EFAULT;
3276                         break;
3277                 }
3278                 if (sndbuf <= 0) {
3279                         ret = -EINVAL;
3280                         break;
3281                 }
3282
3283                 tun->sndbuf = sndbuf;
3284                 tun_set_sndbuf(tun);
3285                 break;
3286
3287         case TUNGETVNETHDRSZ:
3288                 vnet_hdr_sz = tun->vnet_hdr_sz;
3289                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3290                         ret = -EFAULT;
3291                 break;
3292
3293         case TUNSETVNETHDRSZ:
3294                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3295                         ret = -EFAULT;
3296                         break;
3297                 }
3298                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3299                         ret = -EINVAL;
3300                         break;
3301                 }
3302
3303                 tun->vnet_hdr_sz = vnet_hdr_sz;
3304                 break;
3305
3306         case TUNGETVNETLE:
3307                 le = !!(tun->flags & TUN_VNET_LE);
3308                 if (put_user(le, (int __user *)argp))
3309                         ret = -EFAULT;
3310                 break;
3311
3312         case TUNSETVNETLE:
3313                 if (get_user(le, (int __user *)argp)) {
3314                         ret = -EFAULT;
3315                         break;
3316                 }
3317                 if (le)
3318                         tun->flags |= TUN_VNET_LE;
3319                 else
3320                         tun->flags &= ~TUN_VNET_LE;
3321                 break;
3322
3323         case TUNGETVNETBE:
3324                 ret = tun_get_vnet_be(tun, argp);
3325                 break;
3326
3327         case TUNSETVNETBE:
3328                 ret = tun_set_vnet_be(tun, argp);
3329                 break;
3330
3331         case TUNATTACHFILTER:
3332                 /* Can be set only for TAPs */
3333                 ret = -EINVAL;
3334                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3335                         break;
3336                 ret = -EFAULT;
3337                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3338                         break;
3339
3340                 ret = tun_attach_filter(tun);
3341                 break;
3342
3343         case TUNDETACHFILTER:
3344                 /* Can be set only for TAPs */
3345                 ret = -EINVAL;
3346                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3347                         break;
3348                 ret = 0;
3349                 tun_detach_filter(tun, tun->numqueues);
3350                 break;
3351
3352         case TUNGETFILTER:
3353                 ret = -EINVAL;
3354                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3355                         break;
3356                 ret = -EFAULT;
3357                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3358                         break;
3359                 ret = 0;
3360                 break;
3361
3362         case TUNSETSTEERINGEBPF:
3363                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3364                 break;
3365
3366         case TUNSETFILTEREBPF:
3367                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3368                 break;
3369
3370         case TUNSETCARRIER:
3371                 ret = -EFAULT;
3372                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3373                         goto unlock;
3374
3375                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3376                 break;
3377
3378         case TUNGETDEVNETNS:
3379                 ret = -EPERM;
3380                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3381                         goto unlock;
3382                 ret = open_related_ns(&net->ns, get_net_ns);
3383                 break;
3384
3385         default:
3386                 ret = -EINVAL;
3387                 break;
3388         }
3389
3390         if (do_notify)
3391                 netdev_state_change(tun->dev);
3392
3393 unlock:
3394         rtnl_unlock();
3395         if (tun)
3396                 tun_put(tun);
3397         return ret;
3398 }
3399
3400 static long tun_chr_ioctl(struct file *file,
3401                           unsigned int cmd, unsigned long arg)
3402 {
3403         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3404 }
3405
3406 #ifdef CONFIG_COMPAT
3407 static long tun_chr_compat_ioctl(struct file *file,
3408                          unsigned int cmd, unsigned long arg)
3409 {
3410         switch (cmd) {
3411         case TUNSETIFF:
3412         case TUNGETIFF:
3413         case TUNSETTXFILTER:
3414         case TUNGETSNDBUF:
3415         case TUNSETSNDBUF:
3416         case SIOCGIFHWADDR:
3417         case SIOCSIFHWADDR:
3418                 arg = (unsigned long)compat_ptr(arg);
3419                 break;
3420         default:
3421                 arg = (compat_ulong_t)arg;
3422                 break;
3423         }
3424
3425         /*
3426          * compat_ifreq is shorter than ifreq, so we must not access beyond
3427          * the end of that structure. All fields that are used in this
3428          * driver are compatible though, we don't need to convert the
3429          * contents.
3430          */
3431         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3432 }
3433 #endif /* CONFIG_COMPAT */
3434
3435 static int tun_chr_fasync(int fd, struct file *file, int on)
3436 {
3437         struct tun_file *tfile = file->private_data;
3438         int ret;
3439
3440         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3441                 goto out;
3442
3443         if (on) {
3444                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3445                 tfile->flags |= TUN_FASYNC;
3446         } else
3447                 tfile->flags &= ~TUN_FASYNC;
3448         ret = 0;
3449 out:
3450         return ret;
3451 }
3452
3453 static int tun_chr_open(struct inode *inode, struct file * file)
3454 {
3455         struct net *net = current->nsproxy->net_ns;
3456         struct tun_file *tfile;
3457
3458         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3459                                             &tun_proto, 0);
3460         if (!tfile)
3461                 return -ENOMEM;
3462         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3463                 sk_free(&tfile->sk);
3464                 return -ENOMEM;
3465         }
3466
3467         mutex_init(&tfile->napi_mutex);
3468         RCU_INIT_POINTER(tfile->tun, NULL);
3469         tfile->flags = 0;
3470         tfile->ifindex = 0;
3471
3472         init_waitqueue_head(&tfile->socket.wq.wait);
3473
3474         tfile->socket.file = file;
3475         tfile->socket.ops = &tun_socket_ops;
3476
3477         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3478
3479         tfile->sk.sk_write_space = tun_sock_write_space;
3480         tfile->sk.sk_sndbuf = INT_MAX;
3481
3482         file->private_data = tfile;
3483         INIT_LIST_HEAD(&tfile->next);
3484
3485         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3486
3487         /* tun groks IOCB_NOWAIT just fine, mark it as such */
3488         file->f_mode |= FMODE_NOWAIT;
3489         return 0;
3490 }
3491
3492 static int tun_chr_close(struct inode *inode, struct file *file)
3493 {
3494         struct tun_file *tfile = file->private_data;
3495
3496         tun_detach(tfile, true);
3497
3498         return 0;
3499 }
3500
3501 #ifdef CONFIG_PROC_FS
3502 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3503 {
3504         struct tun_file *tfile = file->private_data;
3505         struct tun_struct *tun;
3506         struct ifreq ifr;
3507
3508         memset(&ifr, 0, sizeof(ifr));
3509
3510         rtnl_lock();
3511         tun = tun_get(tfile);
3512         if (tun)
3513                 tun_get_iff(tun, &ifr);
3514         rtnl_unlock();
3515
3516         if (tun)
3517                 tun_put(tun);
3518
3519         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3520 }
3521 #endif
3522
3523 static const struct file_operations tun_fops = {
3524         .owner  = THIS_MODULE,
3525         .llseek = no_llseek,
3526         .read_iter  = tun_chr_read_iter,
3527         .write_iter = tun_chr_write_iter,
3528         .poll   = tun_chr_poll,
3529         .unlocked_ioctl = tun_chr_ioctl,
3530 #ifdef CONFIG_COMPAT
3531         .compat_ioctl = tun_chr_compat_ioctl,
3532 #endif
3533         .open   = tun_chr_open,
3534         .release = tun_chr_close,
3535         .fasync = tun_chr_fasync,
3536 #ifdef CONFIG_PROC_FS
3537         .show_fdinfo = tun_chr_show_fdinfo,
3538 #endif
3539 };
3540
3541 static struct miscdevice tun_miscdev = {
3542         .minor = TUN_MINOR,
3543         .name = "tun",
3544         .nodename = "net/tun",
3545         .fops = &tun_fops,
3546 };
3547
3548 /* ethtool interface */
3549
3550 static void tun_default_link_ksettings(struct net_device *dev,
3551                                        struct ethtool_link_ksettings *cmd)
3552 {
3553         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3554         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3555         cmd->base.speed         = SPEED_10000;
3556         cmd->base.duplex        = DUPLEX_FULL;
3557         cmd->base.port          = PORT_TP;
3558         cmd->base.phy_address   = 0;
3559         cmd->base.autoneg       = AUTONEG_DISABLE;
3560 }
3561
3562 static int tun_get_link_ksettings(struct net_device *dev,
3563                                   struct ethtool_link_ksettings *cmd)
3564 {
3565         struct tun_struct *tun = netdev_priv(dev);
3566
3567         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3568         return 0;
3569 }
3570
3571 static int tun_set_link_ksettings(struct net_device *dev,
3572                                   const struct ethtool_link_ksettings *cmd)
3573 {
3574         struct tun_struct *tun = netdev_priv(dev);
3575
3576         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3577         return 0;
3578 }
3579
3580 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3581 {
3582         struct tun_struct *tun = netdev_priv(dev);
3583
3584         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3585         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3586
3587         switch (tun->flags & TUN_TYPE_MASK) {
3588         case IFF_TUN:
3589                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3590                 break;
3591         case IFF_TAP:
3592                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3593                 break;
3594         }
3595 }
3596
3597 static u32 tun_get_msglevel(struct net_device *dev)
3598 {
3599         struct tun_struct *tun = netdev_priv(dev);
3600
3601         return tun->msg_enable;
3602 }
3603
3604 static void tun_set_msglevel(struct net_device *dev, u32 value)
3605 {
3606         struct tun_struct *tun = netdev_priv(dev);
3607
3608         tun->msg_enable = value;
3609 }
3610
3611 static int tun_get_coalesce(struct net_device *dev,
3612                             struct ethtool_coalesce *ec,
3613                             struct kernel_ethtool_coalesce *kernel_coal,
3614                             struct netlink_ext_ack *extack)
3615 {
3616         struct tun_struct *tun = netdev_priv(dev);
3617
3618         ec->rx_max_coalesced_frames = tun->rx_batched;
3619
3620         return 0;
3621 }
3622
3623 static int tun_set_coalesce(struct net_device *dev,
3624                             struct ethtool_coalesce *ec,
3625                             struct kernel_ethtool_coalesce *kernel_coal,
3626                             struct netlink_ext_ack *extack)
3627 {
3628         struct tun_struct *tun = netdev_priv(dev);
3629
3630         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3631                 tun->rx_batched = NAPI_POLL_WEIGHT;
3632         else
3633                 tun->rx_batched = ec->rx_max_coalesced_frames;
3634
3635         return 0;
3636 }
3637
3638 static const struct ethtool_ops tun_ethtool_ops = {
3639         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3640         .get_drvinfo    = tun_get_drvinfo,
3641         .get_msglevel   = tun_get_msglevel,
3642         .set_msglevel   = tun_set_msglevel,
3643         .get_link       = ethtool_op_get_link,
3644         .get_ts_info    = ethtool_op_get_ts_info,
3645         .get_coalesce   = tun_get_coalesce,
3646         .set_coalesce   = tun_set_coalesce,
3647         .get_link_ksettings = tun_get_link_ksettings,
3648         .set_link_ksettings = tun_set_link_ksettings,
3649 };
3650
3651 static int tun_queue_resize(struct tun_struct *tun)
3652 {
3653         struct net_device *dev = tun->dev;
3654         struct tun_file *tfile;
3655         struct ptr_ring **rings;
3656         int n = tun->numqueues + tun->numdisabled;
3657         int ret, i;
3658
3659         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3660         if (!rings)
3661                 return -ENOMEM;
3662
3663         for (i = 0; i < tun->numqueues; i++) {
3664                 tfile = rtnl_dereference(tun->tfiles[i]);
3665                 rings[i] = &tfile->tx_ring;
3666         }
3667         list_for_each_entry(tfile, &tun->disabled, next)
3668                 rings[i++] = &tfile->tx_ring;
3669
3670         ret = ptr_ring_resize_multiple(rings, n,
3671                                        dev->tx_queue_len, GFP_KERNEL,
3672                                        tun_ptr_free);
3673
3674         kfree(rings);
3675         return ret;
3676 }
3677
3678 static int tun_device_event(struct notifier_block *unused,
3679                             unsigned long event, void *ptr)
3680 {
3681         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3682         struct tun_struct *tun = netdev_priv(dev);
3683         int i;
3684
3685         if (dev->rtnl_link_ops != &tun_link_ops)
3686                 return NOTIFY_DONE;
3687
3688         switch (event) {
3689         case NETDEV_CHANGE_TX_QUEUE_LEN:
3690                 if (tun_queue_resize(tun))
3691                         return NOTIFY_BAD;
3692                 break;
3693         case NETDEV_UP:
3694                 for (i = 0; i < tun->numqueues; i++) {
3695                         struct tun_file *tfile;
3696
3697                         tfile = rtnl_dereference(tun->tfiles[i]);
3698                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3699                 }
3700                 break;
3701         default:
3702                 break;
3703         }
3704
3705         return NOTIFY_DONE;
3706 }
3707
3708 static struct notifier_block tun_notifier_block __read_mostly = {
3709         .notifier_call  = tun_device_event,
3710 };
3711
3712 static int __init tun_init(void)
3713 {
3714         int ret = 0;
3715
3716         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3717
3718         ret = rtnl_link_register(&tun_link_ops);
3719         if (ret) {
3720                 pr_err("Can't register link_ops\n");
3721                 goto err_linkops;
3722         }
3723
3724         ret = misc_register(&tun_miscdev);
3725         if (ret) {
3726                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3727                 goto err_misc;
3728         }
3729
3730         ret = register_netdevice_notifier(&tun_notifier_block);
3731         if (ret) {
3732                 pr_err("Can't register netdevice notifier\n");
3733                 goto err_notifier;
3734         }
3735
3736         return  0;
3737
3738 err_notifier:
3739         misc_deregister(&tun_miscdev);
3740 err_misc:
3741         rtnl_link_unregister(&tun_link_ops);
3742 err_linkops:
3743         return ret;
3744 }
3745
3746 static void __exit tun_cleanup(void)
3747 {
3748         misc_deregister(&tun_miscdev);
3749         rtnl_link_unregister(&tun_link_ops);
3750         unregister_netdevice_notifier(&tun_notifier_block);
3751 }
3752
3753 /* Get an underlying socket object from tun file.  Returns error unless file is
3754  * attached to a device.  The returned object works like a packet socket, it
3755  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3756  * holding a reference to the file for as long as the socket is in use. */
3757 struct socket *tun_get_socket(struct file *file)
3758 {
3759         struct tun_file *tfile;
3760         if (file->f_op != &tun_fops)
3761                 return ERR_PTR(-EINVAL);
3762         tfile = file->private_data;
3763         if (!tfile)
3764                 return ERR_PTR(-EBADFD);
3765         return &tfile->socket;
3766 }
3767 EXPORT_SYMBOL_GPL(tun_get_socket);
3768
3769 struct ptr_ring *tun_get_tx_ring(struct file *file)
3770 {
3771         struct tun_file *tfile;
3772
3773         if (file->f_op != &tun_fops)
3774                 return ERR_PTR(-EINVAL);
3775         tfile = file->private_data;
3776         if (!tfile)
3777                 return ERR_PTR(-EBADFD);
3778         return &tfile->tx_ring;
3779 }
3780 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3781
3782 module_init(tun_init);
3783 module_exit(tun_cleanup);
3784 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3785 MODULE_AUTHOR(DRV_COPYRIGHT);
3786 MODULE_LICENSE("GPL");
3787 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3788 MODULE_ALIAS("devname:net/tun");