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)
 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_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 272                                   NAPI_POLL_WEIGHT);
 273                 napi_enable(&tfile->napi);
 274         }
 275 }
 276
 277 static void tun_napi_enable(struct tun_file *tfile)
 278 {
 279         if (tfile->napi_enabled)
 280                 napi_enable(&tfile->napi);
 281 }
 282
 283 static void tun_napi_disable(struct tun_file *tfile)
 284 {
 285         if (tfile->napi_enabled)
 286                 napi_disable(&tfile->napi);
 287 }
 288
 289 static void tun_napi_del(struct tun_file *tfile)
 290 {
 291         if (tfile->napi_enabled)
 292                 netif_napi_del(&tfile->napi);
 293 }
 294
 295 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 296 {
 297         return tfile->napi_frags_enabled;
 298 }
 299
 300 #ifdef CONFIG_TUN_VNET_CROSS_LE
 301 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 302 {
 303         return tun->flags & TUN_VNET_BE ? false :
 304                 virtio_legacy_is_little_endian();
 305 }
 306
 307 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 308 {
 309         int be = !!(tun->flags & TUN_VNET_BE);
 310
 311         if (put_user(be, argp))
 312                 return -EFAULT;
 313
 314         return 0;
 315 }
 316
 317 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 318 {
 319         int be;
 320
 321         if (get_user(be, argp))
 322                 return -EFAULT;
 323
 324         if (be)
 325                 tun->flags |= TUN_VNET_BE;
 326         else
 327                 tun->flags &= ~TUN_VNET_BE;
 328
 329         return 0;
 330 }
 331 #else
 332 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 333 {
 334         return virtio_legacy_is_little_endian();
 335 }
 336
 337 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 338 {
 339         return -EINVAL;
 340 }
 341
 342 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 343 {
 344         return -EINVAL;
 345 }
 346 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 347
 348 static inline bool tun_is_little_endian(struct tun_struct *tun)
 349 {
 350         return tun->flags & TUN_VNET_LE ||
 351                 tun_legacy_is_little_endian(tun);
 352 }
 353
 354 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 355 {
 356         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 357 }
 358
 359 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 360 {
 361         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 362 }
 363
 364 static inline u32 tun_hashfn(u32 rxhash)
 365 {
 366         return rxhash & TUN_MASK_FLOW_ENTRIES;
 367 }
 368
 369 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 370 {
 371         struct tun_flow_entry *e;
 372
 373         hlist_for_each_entry_rcu(e, head, hash_link) {
 374                 if (e->rxhash == rxhash)
 375                         return e;
 376         }
 377         return NULL;
 378 }
 379
 380 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 381                                               struct hlist_head *head,
 382                                               u32 rxhash, u16 queue_index)
 383 {
 384         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 385
 386         if (e) {
 387                 netif_info(tun, tx_queued, tun->dev,
 388                            "create flow: hash %u index %u\n",
 389                            rxhash, queue_index);
 390                 e->updated = jiffies;
 391                 e->rxhash = rxhash;
 392                 e->rps_rxhash = 0;
 393                 e->queue_index = queue_index;
 394                 e->tun = tun;
 395                 hlist_add_head_rcu(&e->hash_link, head);
 396                 ++tun->flow_count;
 397         }
 398         return e;
 399 }
 400
 401 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 402 {
 403         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 404                    e->rxhash, e->queue_index);
 405         hlist_del_rcu(&e->hash_link);
 406         kfree_rcu(e, rcu);
 407         --tun->flow_count;
 408 }
 409
 410 static void tun_flow_flush(struct tun_struct *tun)
 411 {
 412         int i;
 413
 414         spin_lock_bh(&tun->lock);
 415         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 416                 struct tun_flow_entry *e;
 417                 struct hlist_node *n;
 418
 419                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 420                         tun_flow_delete(tun, e);
 421         }
 422         spin_unlock_bh(&tun->lock);
 423 }
 424
 425 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 426 {
 427         int i;
 428
 429         spin_lock_bh(&tun->lock);
 430         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 431                 struct tun_flow_entry *e;
 432                 struct hlist_node *n;
 433
 434                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 435                         if (e->queue_index == queue_index)
 436                                 tun_flow_delete(tun, e);
 437                 }
 438         }
 439         spin_unlock_bh(&tun->lock);
 440 }
 441
 442 static void tun_flow_cleanup(struct timer_list *t)
 443 {
 444         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 445         unsigned long delay = tun->ageing_time;
 446         unsigned long next_timer = jiffies + delay;
 447         unsigned long count = 0;
 448         int i;
 449
 450         spin_lock(&tun->lock);
 451         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 452                 struct tun_flow_entry *e;
 453                 struct hlist_node *n;
 454
 455                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 456                         unsigned long this_timer;
 457
 458                         this_timer = e->updated + delay;
 459                         if (time_before_eq(this_timer, jiffies)) {
 460                                 tun_flow_delete(tun, e);
 461                                 continue;
 462                         }
 463                         count++;
 464                         if (time_before(this_timer, next_timer))
 465                                 next_timer = this_timer;
 466                 }
 467         }
 468
 469         if (count)
 470                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 471         spin_unlock(&tun->lock);
 472 }
 473
 474 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 475                             struct tun_file *tfile)
 476 {
 477         struct hlist_head *head;
 478         struct tun_flow_entry *e;
 479         unsigned long delay = tun->ageing_time;
 480         u16 queue_index = tfile->queue_index;
 481
 482         head = &tun->flows[tun_hashfn(rxhash)];
 483
 484         rcu_read_lock();
 485
 486         e = tun_flow_find(head, rxhash);
 487         if (likely(e)) {
 488                 /* TODO: keep queueing to old queue until it's empty? */
 489                 if (READ_ONCE(e->queue_index) != queue_index)
 490                         WRITE_ONCE(e->queue_index, queue_index);
 491                 if (e->updated != jiffies)
 492                         e->updated = jiffies;
 493                 sock_rps_record_flow_hash(e->rps_rxhash);
 494         } else {
 495                 spin_lock_bh(&tun->lock);
 496                 if (!tun_flow_find(head, rxhash) &&
 497                     tun->flow_count < MAX_TAP_FLOWS)
 498                         tun_flow_create(tun, head, rxhash, queue_index);
 499
 500                 if (!timer_pending(&tun->flow_gc_timer))
 501                         mod_timer(&tun->flow_gc_timer,
 502                                   round_jiffies_up(jiffies + delay));
 503                 spin_unlock_bh(&tun->lock);
 504         }
 505
 506         rcu_read_unlock();
 507 }
 508
 509 /* Save the hash received in the stack receive path and update the
 510  * flow_hash table accordingly.
 511  */
 512 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 513 {
 514         if (unlikely(e->rps_rxhash != hash))
 515                 e->rps_rxhash = hash;
 516 }
 517
 518 /* We try to identify a flow through its rxhash. The reason that
 519  * we do not check rxq no. is because some cards(e.g 82599), chooses
 520  * the rxq based on the txq where the last packet of the flow comes. As
 521  * the userspace application move between processors, we may get a
 522  * different rxq no. here.
 523  */
 524 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 525 {
 526         struct tun_flow_entry *e;
 527         u32 txq = 0;
 528         u32 numqueues = 0;
 529
 530         numqueues = READ_ONCE(tun->numqueues);
 531
 532         txq = __skb_get_hash_symmetric(skb);
 533         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 534         if (e) {
 535                 tun_flow_save_rps_rxhash(e, txq);
 536                 txq = e->queue_index;
 537         } else {
 538                 /* use multiply and shift instead of expensive divide */
 539                 txq = ((u64)txq * numqueues) >> 32;
 540         }
 541
 542         return txq;
 543 }
 544
 545 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 546 {
 547         struct tun_prog *prog;
 548         u32 numqueues;
 549         u16 ret = 0;
 550
 551         numqueues = READ_ONCE(tun->numqueues);
 552         if (!numqueues)
 553                 return 0;
 554
 555         prog = rcu_dereference(tun->steering_prog);
 556         if (prog)
 557                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 558
 559         return ret % numqueues;
 560 }
 561
 562 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 563                             struct net_device *sb_dev)
 564 {
 565         struct tun_struct *tun = netdev_priv(dev);
 566         u16 ret;
 567
 568         rcu_read_lock();
 569         if (rcu_dereference(tun->steering_prog))
 570                 ret = tun_ebpf_select_queue(tun, skb);
 571         else
 572                 ret = tun_automq_select_queue(tun, skb);
 573         rcu_read_unlock();
 574
 575         return ret;
 576 }
 577
 578 static inline bool tun_not_capable(struct tun_struct *tun)
 579 {
 580         const struct cred *cred = current_cred();
 581         struct net *net = dev_net(tun->dev);
 582
 583         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 584                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 585                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 586 }
 587
 588 static void tun_set_real_num_queues(struct tun_struct *tun)
 589 {
 590         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 591         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 592 }
 593
 594 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 595 {
 596         tfile->detached = tun;
 597         list_add_tail(&tfile->next, &tun->disabled);
 598         ++tun->numdisabled;
 599 }
 600
 601 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 602 {
 603         struct tun_struct *tun = tfile->detached;
 604
 605         tfile->detached = NULL;
 606         list_del_init(&tfile->next);
 607         --tun->numdisabled;
 608         return tun;
 609 }
 610
 611 void tun_ptr_free(void *ptr)
 612 {
 613         if (!ptr)
 614                 return;
 615         if (tun_is_xdp_frame(ptr)) {
 616                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 617
 618                 xdp_return_frame(xdpf);
 619         } else {
 620                 __skb_array_destroy_skb(ptr);
 621         }
 622 }
 623 EXPORT_SYMBOL_GPL(tun_ptr_free);
 624
 625 static void tun_queue_purge(struct tun_file *tfile)
 626 {
 627         void *ptr;
 628
 629         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 630                 tun_ptr_free(ptr);
 631
 632         skb_queue_purge(&tfile->sk.sk_write_queue);
 633         skb_queue_purge(&tfile->sk.sk_error_queue);
 634 }
 635
 636 static void __tun_detach(struct tun_file *tfile, bool clean)
 637 {
 638         struct tun_file *ntfile;
 639         struct tun_struct *tun;
 640
 641         tun = rtnl_dereference(tfile->tun);
 642
 643         if (tun && clean) {
 644                 if (!tfile->detached)
 645                         tun_napi_disable(tfile);
 646                 tun_napi_del(tfile);
 647         }
 648
 649         if (tun && !tfile->detached) {
 650                 u16 index = tfile->queue_index;
 651                 BUG_ON(index >= tun->numqueues);
 652
 653                 rcu_assign_pointer(tun->tfiles[index],
 654                                    tun->tfiles[tun->numqueues - 1]);
 655                 ntfile = rtnl_dereference(tun->tfiles[index]);
 656                 ntfile->queue_index = index;
 657                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 658                                    NULL);
 659
 660                 --tun->numqueues;
 661                 if (clean) {
 662                         RCU_INIT_POINTER(tfile->tun, NULL);
 663                         sock_put(&tfile->sk);
 664                 } else {
 665                         tun_disable_queue(tun, tfile);
 666                         tun_napi_disable(tfile);
 667                 }
 668
 669                 synchronize_net();
 670                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 671                 /* Drop read queue */
 672                 tun_queue_purge(tfile);
 673                 tun_set_real_num_queues(tun);
 674         } else if (tfile->detached && clean) {
 675                 tun = tun_enable_queue(tfile);
 676                 sock_put(&tfile->sk);
 677         }
 678
 679         if (clean) {
 680                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 681                         netif_carrier_off(tun->dev);
 682
 683                         if (!(tun->flags & IFF_PERSIST) &&
 684                             tun->dev->reg_state == NETREG_REGISTERED)
 685                                 unregister_netdevice(tun->dev);
 686                 }
 687                 if (tun)
 688                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 690                 sock_put(&tfile->sk);
 691         }
 692 }
 693
 694 static void tun_detach(struct tun_file *tfile, bool clean)
 695 {
 696         struct tun_struct *tun;
 697         struct net_device *dev;
 698
 699         rtnl_lock();
 700         tun = rtnl_dereference(tfile->tun);
 701         dev = tun ? tun->dev : NULL;
 702         __tun_detach(tfile, clean);
 703         if (dev)
 704                 netdev_state_change(dev);
 705         rtnl_unlock();
 706 }
 707
 708 static void tun_detach_all(struct net_device *dev)
 709 {
 710         struct tun_struct *tun = netdev_priv(dev);
 711         struct tun_file *tfile, *tmp;
 712         int i, n = tun->numqueues;
 713
 714         for (i = 0; i < n; i++) {
 715                 tfile = rtnl_dereference(tun->tfiles[i]);
 716                 BUG_ON(!tfile);
 717                 tun_napi_disable(tfile);
 718                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 719                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 720                 RCU_INIT_POINTER(tfile->tun, NULL);
 721                 --tun->numqueues;
 722         }
 723         list_for_each_entry(tfile, &tun->disabled, next) {
 724                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 725                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 726                 RCU_INIT_POINTER(tfile->tun, NULL);
 727         }
 728         BUG_ON(tun->numqueues != 0);
 729
 730         synchronize_net();
 731         for (i = 0; i < n; i++) {
 732                 tfile = rtnl_dereference(tun->tfiles[i]);
 733                 tun_napi_del(tfile);
 734                 /* Drop read queue */
 735                 tun_queue_purge(tfile);
 736                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 737                 sock_put(&tfile->sk);
 738         }
 739         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 740                 tun_napi_del(tfile);
 741                 tun_enable_queue(tfile);
 742                 tun_queue_purge(tfile);
 743                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 744                 sock_put(&tfile->sk);
 745         }
 746         BUG_ON(tun->numdisabled != 0);
 747
 748         if (tun->flags & IFF_PERSIST)
 749                 module_put(THIS_MODULE);
 750 }
 751
 752 static int tun_attach(struct tun_struct *tun, struct file *file,
 753                       bool skip_filter, bool napi, bool napi_frags,
 754                       bool publish_tun)
 755 {
 756         struct tun_file *tfile = file->private_data;
 757         struct net_device *dev = tun->dev;
 758         int err;
 759
 760         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 761         if (err < 0)
 762                 goto out;
 763
 764         err = -EINVAL;
 765         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 766                 goto out;
 767
 768         err = -EBUSY;
 769         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 770                 goto out;
 771
 772         err = -E2BIG;
 773         if (!tfile->detached &&
 774             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 775                 goto out;
 776
 777         err = 0;
 778
 779         /* Re-attach the filter to persist device */
 780         if (!skip_filter && (tun->filter_attached == true)) {
 781                 lock_sock(tfile->socket.sk);
 782                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 783                 release_sock(tfile->socket.sk);
 784                 if (!err)
 785                         goto out;
 786         }
 787
 788         if (!tfile->detached &&
 789             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 790                             GFP_KERNEL, tun_ptr_free)) {
 791                 err = -ENOMEM;
 792                 goto out;
 793         }
 794
 795         tfile->queue_index = tun->numqueues;
 796         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 797
 798         if (tfile->detached) {
 799                 /* Re-attach detached tfile, updating XDP queue_index */
 800                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 801
 802                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 803                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 804         } else {
 805                 /* Setup XDP RX-queue info, for new tfile getting attached */
 806                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 807                                        tun->dev, tfile->queue_index, 0);
 808                 if (err < 0)
 809                         goto out;
 810                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 811                                                  MEM_TYPE_PAGE_SHARED, NULL);
 812                 if (err < 0) {
 813                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 814                         goto out;
 815                 }
 816                 err = 0;
 817         }
 818
 819         if (tfile->detached) {
 820                 tun_enable_queue(tfile);
 821                 tun_napi_enable(tfile);
 822         } else {
 823                 sock_hold(&tfile->sk);
 824                 tun_napi_init(tun, tfile, napi, napi_frags);
 825         }
 826
 827         if (rtnl_dereference(tun->xdp_prog))
 828                 sock_set_flag(&tfile->sk, SOCK_XDP);
 829
 830         /* device is allowed to go away first, so no need to hold extra
 831          * refcnt.
 832          */
 833
 834         /* Publish tfile->tun and tun->tfiles only after we've fully
 835          * initialized tfile; otherwise we risk using half-initialized
 836          * object.
 837          */
 838         if (publish_tun)
 839                 rcu_assign_pointer(tfile->tun, tun);
 840         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 841         tun->numqueues++;
 842         tun_set_real_num_queues(tun);
 843 out:
 844         return err;
 845 }
 846
 847 static struct tun_struct *tun_get(struct tun_file *tfile)
 848 {
 849         struct tun_struct *tun;
 850
 851         rcu_read_lock();
 852         tun = rcu_dereference(tfile->tun);
 853         if (tun)
 854                 dev_hold(tun->dev);
 855         rcu_read_unlock();
 856
 857         return tun;
 858 }
 859
 860 static void tun_put(struct tun_struct *tun)
 861 {
 862         dev_put(tun->dev);
 863 }
 864
 865 /* TAP filtering */
 866 static void addr_hash_set(u32 *mask, const u8 *addr)
 867 {
 868         int n = ether_crc(ETH_ALEN, addr) >> 26;
 869         mask[n >> 5] |= (1 << (n & 31));
 870 }
 871
 872 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 873 {
 874         int n = ether_crc(ETH_ALEN, addr) >> 26;
 875         return mask[n >> 5] & (1 << (n & 31));
 876 }
 877
 878 static int update_filter(struct tap_filter *filter, void __user *arg)
 879 {
 880         struct { u8 u[ETH_ALEN]; } *addr;
 881         struct tun_filter uf;
 882         int err, alen, n, nexact;
 883
 884         if (copy_from_user(&uf, arg, sizeof(uf)))
 885                 return -EFAULT;
 886
 887         if (!uf.count) {
 888                 /* Disabled */
 889                 filter->count = 0;
 890                 return 0;
 891         }
 892
 893         alen = ETH_ALEN * uf.count;
 894         addr = memdup_user(arg + sizeof(uf), alen);
 895         if (IS_ERR(addr))
 896                 return PTR_ERR(addr);
 897
 898         /* The filter is updated without holding any locks. Which is
 899          * perfectly safe. We disable it first and in the worst
 900          * case we'll accept a few undesired packets. */
 901         filter->count = 0;
 902         wmb();
 903
 904         /* Use first set of addresses as an exact filter */
 905         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 906                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 907
 908         nexact = n;
 909
 910         /* Remaining multicast addresses are hashed,
 911          * unicast will leave the filter disabled. */
 912         memset(filter->mask, 0, sizeof(filter->mask));
 913         for (; n < uf.count; n++) {
 914                 if (!is_multicast_ether_addr(addr[n].u)) {
 915                         err = 0; /* no filter */
 916                         goto free_addr;
 917                 }
 918                 addr_hash_set(filter->mask, addr[n].u);
 919         }
 920
 921         /* For ALLMULTI just set the mask to all ones.
 922          * This overrides the mask populated above. */
 923         if ((uf.flags & TUN_FLT_ALLMULTI))
 924                 memset(filter->mask, ~0, sizeof(filter->mask));
 925
 926         /* Now enable the filter */
 927         wmb();
 928         filter->count = nexact;
 929
 930         /* Return the number of exact filters */
 931         err = nexact;
 932 free_addr:
 933         kfree(addr);
 934         return err;
 935 }
 936
 937 /* Returns: 0 - drop, !=0 - accept */
 938 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 939 {
 940         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 941          * at this point. */
 942         struct ethhdr *eh = (struct ethhdr *) skb->data;
 943         int i;
 944
 945         /* Exact match */
 946         for (i = 0; i < filter->count; i++)
 947                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 948                         return 1;
 949
 950         /* Inexact match (multicast only) */
 951         if (is_multicast_ether_addr(eh->h_dest))
 952                 return addr_hash_test(filter->mask, eh->h_dest);
 953
 954         return 0;
 955 }
 956
 957 /*
 958  * Checks whether the packet is accepted or not.
 959  * Returns: 0 - drop, !=0 - accept
 960  */
 961 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 962 {
 963         if (!filter->count)
 964                 return 1;
 965
 966         return run_filter(filter, skb);
 967 }
 968
 969 /* Network device part of the driver */
 970
 971 static const struct ethtool_ops tun_ethtool_ops;
 972
 973 static int tun_net_init(struct net_device *dev)
 974 {
 975         struct tun_struct *tun = netdev_priv(dev);
 976         struct ifreq *ifr = tun->ifr;
 977         int err;
 978
 979         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 980         if (!dev->tstats)
 981                 return -ENOMEM;
 982
 983         spin_lock_init(&tun->lock);
 984
 985         err = security_tun_dev_alloc_security(&tun->security);
 986         if (err < 0) {
 987                 free_percpu(dev->tstats);
 988                 return err;
 989         }
 990
 991         tun_flow_init(tun);
 992
 993         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 994                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 995                            NETIF_F_HW_VLAN_STAG_TX;
 996         dev->features = dev->hw_features | NETIF_F_LLTX;
 997         dev->vlan_features = dev->features &
 998                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 999                                NETIF_F_HW_VLAN_STAG_TX);
1000
1001         tun->flags = (tun->flags & ~TUN_FEATURES) |
1002                       (ifr->ifr_flags & TUN_FEATURES);
1003
1004         INIT_LIST_HEAD(&tun->disabled);
1005         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1006                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1007         if (err < 0) {
1008                 tun_flow_uninit(tun);
1009                 security_tun_dev_free_security(tun->security);
1010                 free_percpu(dev->tstats);
1011                 return err;
1012         }
1013         return 0;
1014 }
1015
1016 /* Net device detach from fd. */
1017 static void tun_net_uninit(struct net_device *dev)
1018 {
1019         tun_detach_all(dev);
1020 }
1021
1022 /* Net device open. */
1023 static int tun_net_open(struct net_device *dev)
1024 {
1025         netif_tx_start_all_queues(dev);
1026
1027         return 0;
1028 }
1029
1030 /* Net device close. */
1031 static int tun_net_close(struct net_device *dev)
1032 {
1033         netif_tx_stop_all_queues(dev);
1034         return 0;
1035 }
1036
1037 /* Net device start xmit */
1038 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1039 {
1040 #ifdef CONFIG_RPS
1041         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1042                 /* Select queue was not called for the skbuff, so we extract the
1043                  * RPS hash and save it into the flow_table here.
1044                  */
1045                 struct tun_flow_entry *e;
1046                 __u32 rxhash;
1047
1048                 rxhash = __skb_get_hash_symmetric(skb);
1049                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1050                 if (e)
1051                         tun_flow_save_rps_rxhash(e, rxhash);
1052         }
1053 #endif
1054 }
1055
1056 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1057                                     struct sk_buff *skb,
1058                                     int len)
1059 {
1060         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1061
1062         if (prog)
1063                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1064
1065         return len;
1066 }
1067
1068 /* Net device start xmit */
1069 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1070 {
1071         struct tun_struct *tun = netdev_priv(dev);
1072         int txq = skb->queue_mapping;
1073         struct netdev_queue *queue;
1074         struct tun_file *tfile;
1075         int len = skb->len;
1076
1077         rcu_read_lock();
1078         tfile = rcu_dereference(tun->tfiles[txq]);
1079
1080         /* Drop packet if interface is not attached */
1081         if (!tfile)
1082                 goto drop;
1083
1084         if (!rcu_dereference(tun->steering_prog))
1085                 tun_automq_xmit(tun, skb);
1086
1087         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1088
1089         /* Drop if the filter does not like it.
1090          * This is a noop if the filter is disabled.
1091          * Filter can be enabled only for the TAP devices. */
1092         if (!check_filter(&tun->txflt, skb))
1093                 goto drop;
1094
1095         if (tfile->socket.sk->sk_filter &&
1096             sk_filter(tfile->socket.sk, skb))
1097                 goto drop;
1098
1099         len = run_ebpf_filter(tun, skb, len);
1100         if (len == 0 || pskb_trim(skb, len))
1101                 goto drop;
1102
1103         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1104                 goto drop;
1105
1106         skb_tx_timestamp(skb);
1107
1108         /* Orphan the skb - required as we might hang on to it
1109          * for indefinite time.
1110          */
1111         skb_orphan(skb);
1112
1113         nf_reset_ct(skb);
1114
1115         if (ptr_ring_produce(&tfile->tx_ring, skb))
1116                 goto drop;
1117
1118         /* NETIF_F_LLTX requires to do our own update of trans_start */
1119         queue = netdev_get_tx_queue(dev, txq);
1120         queue->trans_start = jiffies;
1121
1122         /* Notify and wake up reader process */
1123         if (tfile->flags & TUN_FASYNC)
1124                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1125         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1126
1127         rcu_read_unlock();
1128         return NETDEV_TX_OK;
1129
1130 drop:
1131         atomic_long_inc(&dev->tx_dropped);
1132         skb_tx_error(skb);
1133         kfree_skb(skb);
1134         rcu_read_unlock();
1135         return NET_XMIT_DROP;
1136 }
1137
1138 static void tun_net_mclist(struct net_device *dev)
1139 {
1140         /*
1141          * This callback is supposed to deal with mc filter in
1142          * _rx_ path and has nothing to do with the _tx_ path.
1143          * In rx path we always accept everything userspace gives us.
1144          */
1145 }
1146
1147 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1148         netdev_features_t features)
1149 {
1150         struct tun_struct *tun = netdev_priv(dev);
1151
1152         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1153 }
1154
1155 static void tun_set_headroom(struct net_device *dev, int new_hr)
1156 {
1157         struct tun_struct *tun = netdev_priv(dev);
1158
1159         if (new_hr < NET_SKB_PAD)
1160                 new_hr = NET_SKB_PAD;
1161
1162         tun->align = new_hr;
1163 }
1164
1165 static void
1166 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1167 {
1168         struct tun_struct *tun = netdev_priv(dev);
1169
1170         dev_get_tstats64(dev, stats);
1171
1172         stats->rx_frame_errors +=
1173                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1174 }
1175
1176 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1177                        struct netlink_ext_ack *extack)
1178 {
1179         struct tun_struct *tun = netdev_priv(dev);
1180         struct tun_file *tfile;
1181         struct bpf_prog *old_prog;
1182         int i;
1183
1184         old_prog = rtnl_dereference(tun->xdp_prog);
1185         rcu_assign_pointer(tun->xdp_prog, prog);
1186         if (old_prog)
1187                 bpf_prog_put(old_prog);
1188
1189         for (i = 0; i < tun->numqueues; i++) {
1190                 tfile = rtnl_dereference(tun->tfiles[i]);
1191                 if (prog)
1192                         sock_set_flag(&tfile->sk, SOCK_XDP);
1193                 else
1194                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1195         }
1196         list_for_each_entry(tfile, &tun->disabled, next) {
1197                 if (prog)
1198                         sock_set_flag(&tfile->sk, SOCK_XDP);
1199                 else
1200                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1201         }
1202
1203         return 0;
1204 }
1205
1206 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1207 {
1208         switch (xdp->command) {
1209         case XDP_SETUP_PROG:
1210                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1211         default:
1212                 return -EINVAL;
1213         }
1214 }
1215
1216 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1217 {
1218         if (new_carrier) {
1219                 struct tun_struct *tun = netdev_priv(dev);
1220
1221                 if (!tun->numqueues)
1222                         return -EPERM;
1223
1224                 netif_carrier_on(dev);
1225         } else {
1226                 netif_carrier_off(dev);
1227         }
1228         return 0;
1229 }
1230
1231 static const struct net_device_ops tun_netdev_ops = {
1232         .ndo_init               = tun_net_init,
1233         .ndo_uninit             = tun_net_uninit,
1234         .ndo_open               = tun_net_open,
1235         .ndo_stop               = tun_net_close,
1236         .ndo_start_xmit         = tun_net_xmit,
1237         .ndo_fix_features       = tun_net_fix_features,
1238         .ndo_select_queue       = tun_select_queue,
1239         .ndo_set_rx_headroom    = tun_set_headroom,
1240         .ndo_get_stats64        = tun_net_get_stats64,
1241         .ndo_change_carrier     = tun_net_change_carrier,
1242 };
1243
1244 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1245 {
1246         /* Notify and wake up reader process */
1247         if (tfile->flags & TUN_FASYNC)
1248                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1249         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1250 }
1251
1252 static int tun_xdp_xmit(struct net_device *dev, int n,
1253                         struct xdp_frame **frames, u32 flags)
1254 {
1255         struct tun_struct *tun = netdev_priv(dev);
1256         struct tun_file *tfile;
1257         u32 numqueues;
1258         int nxmit = 0;
1259         int i;
1260
1261         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1262                 return -EINVAL;
1263
1264         rcu_read_lock();
1265
1266 resample:
1267         numqueues = READ_ONCE(tun->numqueues);
1268         if (!numqueues) {
1269                 rcu_read_unlock();
1270                 return -ENXIO; /* Caller will free/return all frames */
1271         }
1272
1273         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1274                                             numqueues]);
1275         if (unlikely(!tfile))
1276                 goto resample;
1277
1278         spin_lock(&tfile->tx_ring.producer_lock);
1279         for (i = 0; i < n; i++) {
1280                 struct xdp_frame *xdp = frames[i];
1281                 /* Encode the XDP flag into lowest bit for consumer to differ
1282                  * XDP buffer from sk_buff.
1283                  */
1284                 void *frame = tun_xdp_to_ptr(xdp);
1285
1286                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1287                         atomic_long_inc(&dev->tx_dropped);
1288                         break;
1289                 }
1290                 nxmit++;
1291         }
1292         spin_unlock(&tfile->tx_ring.producer_lock);
1293
1294         if (flags & XDP_XMIT_FLUSH)
1295                 __tun_xdp_flush_tfile(tfile);
1296
1297         rcu_read_unlock();
1298         return nxmit;
1299 }
1300
1301 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1302 {
1303         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1304         int nxmit;
1305
1306         if (unlikely(!frame))
1307                 return -EOVERFLOW;
1308
1309         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1310         if (!nxmit)
1311                 xdp_return_frame_rx_napi(frame);
1312         return nxmit;
1313 }
1314
1315 static const struct net_device_ops tap_netdev_ops = {
1316         .ndo_init               = tun_net_init,
1317         .ndo_uninit             = tun_net_uninit,
1318         .ndo_open               = tun_net_open,
1319         .ndo_stop               = tun_net_close,
1320         .ndo_start_xmit         = tun_net_xmit,
1321         .ndo_fix_features       = tun_net_fix_features,
1322         .ndo_set_rx_mode        = tun_net_mclist,
1323         .ndo_set_mac_address    = eth_mac_addr,
1324         .ndo_validate_addr      = eth_validate_addr,
1325         .ndo_select_queue       = tun_select_queue,
1326         .ndo_features_check     = passthru_features_check,
1327         .ndo_set_rx_headroom    = tun_set_headroom,
1328         .ndo_get_stats64        = dev_get_tstats64,
1329         .ndo_bpf                = tun_xdp,
1330         .ndo_xdp_xmit           = tun_xdp_xmit,
1331         .ndo_change_carrier     = tun_net_change_carrier,
1332 };
1333
1334 static void tun_flow_init(struct tun_struct *tun)
1335 {
1336         int i;
1337
1338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1339                 INIT_HLIST_HEAD(&tun->flows[i]);
1340
1341         tun->ageing_time = TUN_FLOW_EXPIRE;
1342         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1343         mod_timer(&tun->flow_gc_timer,
1344                   round_jiffies_up(jiffies + tun->ageing_time));
1345 }
1346
1347 static void tun_flow_uninit(struct tun_struct *tun)
1348 {
1349         del_timer_sync(&tun->flow_gc_timer);
1350         tun_flow_flush(tun);
1351 }
1352
1353 #define MIN_MTU 68
1354 #define MAX_MTU 65535
1355
1356 /* Initialize net device. */
1357 static void tun_net_initialize(struct net_device *dev)
1358 {
1359         struct tun_struct *tun = netdev_priv(dev);
1360
1361         switch (tun->flags & TUN_TYPE_MASK) {
1362         case IFF_TUN:
1363                 dev->netdev_ops = &tun_netdev_ops;
1364                 dev->header_ops = &ip_tunnel_header_ops;
1365
1366                 /* Point-to-Point TUN Device */
1367                 dev->hard_header_len = 0;
1368                 dev->addr_len = 0;
1369                 dev->mtu = 1500;
1370
1371                 /* Zero header length */
1372                 dev->type = ARPHRD_NONE;
1373                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1374                 break;
1375
1376         case IFF_TAP:
1377                 dev->netdev_ops = &tap_netdev_ops;
1378                 /* Ethernet TAP Device */
1379                 ether_setup(dev);
1380                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1381                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1382
1383                 eth_hw_addr_random(dev);
1384
1385                 break;
1386         }
1387
1388         dev->min_mtu = MIN_MTU;
1389         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1390 }
1391
1392 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1393 {
1394         struct sock *sk = tfile->socket.sk;
1395
1396         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1397 }
1398
1399 /* Character device part */
1400
1401 /* Poll */
1402 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1403 {
1404         struct tun_file *tfile = file->private_data;
1405         struct tun_struct *tun = tun_get(tfile);
1406         struct sock *sk;
1407         __poll_t mask = 0;
1408
1409         if (!tun)
1410                 return EPOLLERR;
1411
1412         sk = tfile->socket.sk;
1413
1414         poll_wait(file, sk_sleep(sk), wait);
1415
1416         if (!ptr_ring_empty(&tfile->tx_ring))
1417                 mask |= EPOLLIN | EPOLLRDNORM;
1418
1419         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1420          * guarantee EPOLLOUT to be raised by either here or
1421          * tun_sock_write_space(). Then process could get notification
1422          * after it writes to a down device and meets -EIO.
1423          */
1424         if (tun_sock_writeable(tun, tfile) ||
1425             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1426              tun_sock_writeable(tun, tfile)))
1427                 mask |= EPOLLOUT | EPOLLWRNORM;
1428
1429         if (tun->dev->reg_state != NETREG_REGISTERED)
1430                 mask = EPOLLERR;
1431
1432         tun_put(tun);
1433         return mask;
1434 }
1435
1436 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1437                                             size_t len,
1438                                             const struct iov_iter *it)
1439 {
1440         struct sk_buff *skb;
1441         size_t linear;
1442         int err;
1443         int i;
1444
1445         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1446                 return ERR_PTR(-EMSGSIZE);
1447
1448         local_bh_disable();
1449         skb = napi_get_frags(&tfile->napi);
1450         local_bh_enable();
1451         if (!skb)
1452                 return ERR_PTR(-ENOMEM);
1453
1454         linear = iov_iter_single_seg_count(it);
1455         err = __skb_grow(skb, linear);
1456         if (err)
1457                 goto free;
1458
1459         skb->len = len;
1460         skb->data_len = len - linear;
1461         skb->truesize += skb->data_len;
1462
1463         for (i = 1; i < it->nr_segs; i++) {
1464                 size_t fragsz = it->iov[i].iov_len;
1465                 struct page *page;
1466                 void *frag;
1467
1468                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1469                         err = -EINVAL;
1470                         goto free;
1471                 }
1472                 frag = netdev_alloc_frag(fragsz);
1473                 if (!frag) {
1474                         err = -ENOMEM;
1475                         goto free;
1476                 }
1477                 page = virt_to_head_page(frag);
1478                 skb_fill_page_desc(skb, i - 1, page,
1479                                    frag - page_address(page), fragsz);
1480         }
1481
1482         return skb;
1483 free:
1484         /* frees skb and all frags allocated with napi_alloc_frag() */
1485         napi_free_frags(&tfile->napi);
1486         return ERR_PTR(err);
1487 }
1488
1489 /* prepad is the amount to reserve at front.  len is length after that.
1490  * linear is a hint as to how much to copy (usually headers). */
1491 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1492                                      size_t prepad, size_t len,
1493                                      size_t linear, int noblock)
1494 {
1495         struct sock *sk = tfile->socket.sk;
1496         struct sk_buff *skb;
1497         int err;
1498
1499         /* Under a page?  Don't bother with paged skb. */
1500         if (prepad + len < PAGE_SIZE || !linear)
1501                 linear = len;
1502
1503         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1504                                    &err, 0);
1505         if (!skb)
1506                 return ERR_PTR(err);
1507
1508         skb_reserve(skb, prepad);
1509         skb_put(skb, linear);
1510         skb->data_len = len - linear;
1511         skb->len += len - linear;
1512
1513         return skb;
1514 }
1515
1516 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1517                            struct sk_buff *skb, int more)
1518 {
1519         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1520         struct sk_buff_head process_queue;
1521         u32 rx_batched = tun->rx_batched;
1522         bool rcv = false;
1523
1524         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1525                 local_bh_disable();
1526                 skb_record_rx_queue(skb, tfile->queue_index);
1527                 netif_receive_skb(skb);
1528                 local_bh_enable();
1529                 return;
1530         }
1531
1532         spin_lock(&queue->lock);
1533         if (!more || skb_queue_len(queue) == rx_batched) {
1534                 __skb_queue_head_init(&process_queue);
1535                 skb_queue_splice_tail_init(queue, &process_queue);
1536                 rcv = true;
1537         } else {
1538                 __skb_queue_tail(queue, skb);
1539         }
1540         spin_unlock(&queue->lock);
1541
1542         if (rcv) {
1543                 struct sk_buff *nskb;
1544
1545                 local_bh_disable();
1546                 while ((nskb = __skb_dequeue(&process_queue))) {
1547                         skb_record_rx_queue(nskb, tfile->queue_index);
1548                         netif_receive_skb(nskb);
1549                 }
1550                 skb_record_rx_queue(skb, tfile->queue_index);
1551                 netif_receive_skb(skb);
1552                 local_bh_enable();
1553         }
1554 }
1555
1556 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1557                               int len, int noblock, bool zerocopy)
1558 {
1559         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1560                 return false;
1561
1562         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1563                 return false;
1564
1565         if (!noblock)
1566                 return false;
1567
1568         if (zerocopy)
1569                 return false;
1570
1571         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1572             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1573                 return false;
1574
1575         return true;
1576 }
1577
1578 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1579                                        struct page_frag *alloc_frag, char *buf,
1580                                        int buflen, int len, int pad)
1581 {
1582         struct sk_buff *skb = build_skb(buf, buflen);
1583
1584         if (!skb)
1585                 return ERR_PTR(-ENOMEM);
1586
1587         skb_reserve(skb, pad);
1588         skb_put(skb, len);
1589         skb_set_owner_w(skb, tfile->socket.sk);
1590
1591         get_page(alloc_frag->page);
1592         alloc_frag->offset += buflen;
1593
1594         return skb;
1595 }
1596
1597 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1598                        struct xdp_buff *xdp, u32 act)
1599 {
1600         int err;
1601
1602         switch (act) {
1603         case XDP_REDIRECT:
1604                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1605                 if (err)
1606                         return err;
1607                 break;
1608         case XDP_TX:
1609                 err = tun_xdp_tx(tun->dev, xdp);
1610                 if (err < 0)
1611                         return err;
1612                 break;
1613         case XDP_PASS:
1614                 break;
1615         default:
1616                 bpf_warn_invalid_xdp_action(act);
1617                 fallthrough;
1618         case XDP_ABORTED:
1619                 trace_xdp_exception(tun->dev, xdp_prog, act);
1620                 fallthrough;
1621         case XDP_DROP:
1622                 atomic_long_inc(&tun->dev->rx_dropped);
1623                 break;
1624         }
1625
1626         return act;
1627 }
1628
1629 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1630                                      struct tun_file *tfile,
1631                                      struct iov_iter *from,
1632                                      struct virtio_net_hdr *hdr,
1633                                      int len, int *skb_xdp)
1634 {
1635         struct page_frag *alloc_frag = &current->task_frag;
1636         struct bpf_prog *xdp_prog;
1637         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1638         char *buf;
1639         size_t copied;
1640         int pad = TUN_RX_PAD;
1641         int err = 0;
1642
1643         rcu_read_lock();
1644         xdp_prog = rcu_dereference(tun->xdp_prog);
1645         if (xdp_prog)
1646                 pad += XDP_PACKET_HEADROOM;
1647         buflen += SKB_DATA_ALIGN(len + pad);
1648         rcu_read_unlock();
1649
1650         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1651         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1652                 return ERR_PTR(-ENOMEM);
1653
1654         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1655         copied = copy_page_from_iter(alloc_frag->page,
1656                                      alloc_frag->offset + pad,
1657                                      len, from);
1658         if (copied != len)
1659                 return ERR_PTR(-EFAULT);
1660
1661         /* There's a small window that XDP may be set after the check
1662          * of xdp_prog above, this should be rare and for simplicity
1663          * we do XDP on skb in case the headroom is not enough.
1664          */
1665         if (hdr->gso_type || !xdp_prog) {
1666                 *skb_xdp = 1;
1667                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1668                                        pad);
1669         }
1670
1671         *skb_xdp = 0;
1672
1673         local_bh_disable();
1674         rcu_read_lock();
1675         xdp_prog = rcu_dereference(tun->xdp_prog);
1676         if (xdp_prog) {
1677                 struct xdp_buff xdp;
1678                 u32 act;
1679
1680                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1681                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1682
1683                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1684                 if (act == XDP_REDIRECT || act == XDP_TX) {
1685                         get_page(alloc_frag->page);
1686                         alloc_frag->offset += buflen;
1687                 }
1688                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1689                 if (err < 0) {
1690                         if (act == XDP_REDIRECT || act == XDP_TX)
1691                                 put_page(alloc_frag->page);
1692                         goto out;
1693                 }
1694
1695                 if (err == XDP_REDIRECT)
1696                         xdp_do_flush();
1697                 if (err != XDP_PASS)
1698                         goto out;
1699
1700                 pad = xdp.data - xdp.data_hard_start;
1701                 len = xdp.data_end - xdp.data;
1702         }
1703         rcu_read_unlock();
1704         local_bh_enable();
1705
1706         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1707
1708 out:
1709         rcu_read_unlock();
1710         local_bh_enable();
1711         return NULL;
1712 }
1713
1714 /* Get packet from user space buffer */
1715 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1716                             void *msg_control, struct iov_iter *from,
1717                             int noblock, bool more)
1718 {
1719         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1720         struct sk_buff *skb;
1721         size_t total_len = iov_iter_count(from);
1722         size_t len = total_len, align = tun->align, linear;
1723         struct virtio_net_hdr gso = { 0 };
1724         int good_linear;
1725         int copylen;
1726         bool zerocopy = false;
1727         int err;
1728         u32 rxhash = 0;
1729         int skb_xdp = 1;
1730         bool frags = tun_napi_frags_enabled(tfile);
1731
1732         if (!(tun->flags & IFF_NO_PI)) {
1733                 if (len < sizeof(pi))
1734                         return -EINVAL;
1735                 len -= sizeof(pi);
1736
1737                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1738                         return -EFAULT;
1739         }
1740
1741         if (tun->flags & IFF_VNET_HDR) {
1742                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1743
1744                 if (len < vnet_hdr_sz)
1745                         return -EINVAL;
1746                 len -= vnet_hdr_sz;
1747
1748                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1749                         return -EFAULT;
1750
1751                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1752                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1753                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1754
1755                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1756                         return -EINVAL;
1757                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1758         }
1759
1760         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1761                 align += NET_IP_ALIGN;
1762                 if (unlikely(len < ETH_HLEN ||
1763                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1764                         return -EINVAL;
1765         }
1766
1767         good_linear = SKB_MAX_HEAD(align);
1768
1769         if (msg_control) {
1770                 struct iov_iter i = *from;
1771
1772                 /* There are 256 bytes to be copied in skb, so there is
1773                  * enough room for skb expand head in case it is used.
1774                  * The rest of the buffer is mapped from userspace.
1775                  */
1776                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1777                 if (copylen > good_linear)
1778                         copylen = good_linear;
1779                 linear = copylen;
1780                 iov_iter_advance(&i, copylen);
1781                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1782                         zerocopy = true;
1783         }
1784
1785         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1786                 /* For the packet that is not easy to be processed
1787                  * (e.g gso or jumbo packet), we will do it at after
1788                  * skb was created with generic XDP routine.
1789                  */
1790                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1791                 if (IS_ERR(skb)) {
1792                         atomic_long_inc(&tun->dev->rx_dropped);
1793                         return PTR_ERR(skb);
1794                 }
1795                 if (!skb)
1796                         return total_len;
1797         } else {
1798                 if (!zerocopy) {
1799                         copylen = len;
1800                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1801                                 linear = good_linear;
1802                         else
1803                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1804                 }
1805
1806                 if (frags) {
1807                         mutex_lock(&tfile->napi_mutex);
1808                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1809                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1810                          * If zerocopy is enabled, then this layout will be
1811                          * overwritten by zerocopy_sg_from_iter().
1812                          */
1813                         zerocopy = false;
1814                 } else {
1815                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1816                                             noblock);
1817                 }
1818
1819                 if (IS_ERR(skb)) {
1820                         if (PTR_ERR(skb) != -EAGAIN)
1821                                 atomic_long_inc(&tun->dev->rx_dropped);
1822                         if (frags)
1823                                 mutex_unlock(&tfile->napi_mutex);
1824                         return PTR_ERR(skb);
1825                 }
1826
1827                 if (zerocopy)
1828                         err = zerocopy_sg_from_iter(skb, from);
1829                 else
1830                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1831
1832                 if (err) {
1833                         err = -EFAULT;
1834 drop:
1835                         atomic_long_inc(&tun->dev->rx_dropped);
1836                         kfree_skb(skb);
1837                         if (frags) {
1838                                 tfile->napi.skb = NULL;
1839                                 mutex_unlock(&tfile->napi_mutex);
1840                         }
1841
1842                         return err;
1843                 }
1844         }
1845
1846         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1847                 atomic_long_inc(&tun->rx_frame_errors);
1848                 kfree_skb(skb);
1849                 if (frags) {
1850                         tfile->napi.skb = NULL;
1851                         mutex_unlock(&tfile->napi_mutex);
1852                 }
1853
1854                 return -EINVAL;
1855         }
1856
1857         switch (tun->flags & TUN_TYPE_MASK) {
1858         case IFF_TUN:
1859                 if (tun->flags & IFF_NO_PI) {
1860                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1861
1862                         switch (ip_version) {
1863                         case 4:
1864                                 pi.proto = htons(ETH_P_IP);
1865                                 break;
1866                         case 6:
1867                                 pi.proto = htons(ETH_P_IPV6);
1868                                 break;
1869                         default:
1870                                 atomic_long_inc(&tun->dev->rx_dropped);
1871                                 kfree_skb(skb);
1872                                 return -EINVAL;
1873                         }
1874                 }
1875
1876                 skb_reset_mac_header(skb);
1877                 skb->protocol = pi.proto;
1878                 skb->dev = tun->dev;
1879                 break;
1880         case IFF_TAP:
1881                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1882                         err = -ENOMEM;
1883                         goto drop;
1884                 }
1885                 skb->protocol = eth_type_trans(skb, tun->dev);
1886                 break;
1887         }
1888
1889         /* copy skb_ubuf_info for callback when skb has no error */
1890         if (zerocopy) {
1891                 skb_zcopy_init(skb, msg_control);
1892         } else if (msg_control) {
1893                 struct ubuf_info *uarg = msg_control;
1894                 uarg->callback(NULL, uarg, false);
1895         }
1896
1897         skb_reset_network_header(skb);
1898         skb_probe_transport_header(skb);
1899         skb_record_rx_queue(skb, tfile->queue_index);
1900
1901         if (skb_xdp) {
1902                 struct bpf_prog *xdp_prog;
1903                 int ret;
1904
1905                 local_bh_disable();
1906                 rcu_read_lock();
1907                 xdp_prog = rcu_dereference(tun->xdp_prog);
1908                 if (xdp_prog) {
1909                         ret = do_xdp_generic(xdp_prog, skb);
1910                         if (ret != XDP_PASS) {
1911                                 rcu_read_unlock();
1912                                 local_bh_enable();
1913                                 if (frags) {
1914                                         tfile->napi.skb = NULL;
1915                                         mutex_unlock(&tfile->napi_mutex);
1916                                 }
1917                                 return total_len;
1918                         }
1919                 }
1920                 rcu_read_unlock();
1921                 local_bh_enable();
1922         }
1923
1924         /* Compute the costly rx hash only if needed for flow updates.
1925          * We may get a very small possibility of OOO during switching, not
1926          * worth to optimize.
1927          */
1928         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1929             !tfile->detached)
1930                 rxhash = __skb_get_hash_symmetric(skb);
1931
1932         rcu_read_lock();
1933         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1934                 err = -EIO;
1935                 rcu_read_unlock();
1936                 goto drop;
1937         }
1938
1939         if (frags) {
1940                 u32 headlen;
1941
1942                 /* Exercise flow dissector code path. */
1943                 skb_push(skb, ETH_HLEN);
1944                 headlen = eth_get_headlen(tun->dev, skb->data,
1945                                           skb_headlen(skb));
1946
1947                 if (unlikely(headlen > skb_headlen(skb))) {
1948                         atomic_long_inc(&tun->dev->rx_dropped);
1949                         napi_free_frags(&tfile->napi);
1950                         rcu_read_unlock();
1951                         mutex_unlock(&tfile->napi_mutex);
1952                         WARN_ON(1);
1953                         return -ENOMEM;
1954                 }
1955
1956                 local_bh_disable();
1957                 napi_gro_frags(&tfile->napi);
1958                 local_bh_enable();
1959                 mutex_unlock(&tfile->napi_mutex);
1960         } else if (tfile->napi_enabled) {
1961                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1962                 int queue_len;
1963
1964                 spin_lock_bh(&queue->lock);
1965                 __skb_queue_tail(queue, skb);
1966                 queue_len = skb_queue_len(queue);
1967                 spin_unlock(&queue->lock);
1968
1969                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1970                         napi_schedule(&tfile->napi);
1971
1972                 local_bh_enable();
1973         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1974                 tun_rx_batched(tun, tfile, skb, more);
1975         } else {
1976                 netif_rx_ni(skb);
1977         }
1978         rcu_read_unlock();
1979
1980         preempt_disable();
1981         dev_sw_netstats_rx_add(tun->dev, len);
1982         preempt_enable();
1983
1984         if (rxhash)
1985                 tun_flow_update(tun, rxhash, tfile);
1986
1987         return total_len;
1988 }
1989
1990 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1991 {
1992         struct file *file = iocb->ki_filp;
1993         struct tun_file *tfile = file->private_data;
1994         struct tun_struct *tun = tun_get(tfile);
1995         ssize_t result;
1996         int noblock = 0;
1997
1998         if (!tun)
1999                 return -EBADFD;
2000
2001         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2002                 noblock = 1;
2003
2004         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2005
2006         tun_put(tun);
2007         return result;
2008 }
2009
2010 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2011                                 struct tun_file *tfile,
2012                                 struct xdp_frame *xdp_frame,
2013                                 struct iov_iter *iter)
2014 {
2015         int vnet_hdr_sz = 0;
2016         size_t size = xdp_frame->len;
2017         size_t ret;
2018
2019         if (tun->flags & IFF_VNET_HDR) {
2020                 struct virtio_net_hdr gso = { 0 };
2021
2022                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2023                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2024                         return -EINVAL;
2025                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2026                              sizeof(gso)))
2027                         return -EFAULT;
2028                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2029         }
2030
2031         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2032
2033         preempt_disable();
2034         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2035         preempt_enable();
2036
2037         return ret;
2038 }
2039
2040 /* Put packet to the user space buffer */
2041 static ssize_t tun_put_user(struct tun_struct *tun,
2042                             struct tun_file *tfile,
2043                             struct sk_buff *skb,
2044                             struct iov_iter *iter)
2045 {
2046         struct tun_pi pi = { 0, skb->protocol };
2047         ssize_t total;
2048         int vlan_offset = 0;
2049         int vlan_hlen = 0;
2050         int vnet_hdr_sz = 0;
2051
2052         if (skb_vlan_tag_present(skb))
2053                 vlan_hlen = VLAN_HLEN;
2054
2055         if (tun->flags & IFF_VNET_HDR)
2056                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2057
2058         total = skb->len + vlan_hlen + vnet_hdr_sz;
2059
2060         if (!(tun->flags & IFF_NO_PI)) {
2061                 if (iov_iter_count(iter) < sizeof(pi))
2062                         return -EINVAL;
2063
2064                 total += sizeof(pi);
2065                 if (iov_iter_count(iter) < total) {
2066                         /* Packet will be striped */
2067                         pi.flags |= TUN_PKT_STRIP;
2068                 }
2069
2070                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2071                         return -EFAULT;
2072         }
2073
2074         if (vnet_hdr_sz) {
2075                 struct virtio_net_hdr gso;
2076
2077                 if (iov_iter_count(iter) < vnet_hdr_sz)
2078                         return -EINVAL;
2079
2080                 if (virtio_net_hdr_from_skb(skb, &gso,
2081                                             tun_is_little_endian(tun), true,
2082                                             vlan_hlen)) {
2083                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2084                         pr_err("unexpected GSO type: "
2085                                "0x%x, gso_size %d, hdr_len %d\n",
2086                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2087                                tun16_to_cpu(tun, gso.hdr_len));
2088                         print_hex_dump(KERN_ERR, "tun: ",
2089                                        DUMP_PREFIX_NONE,
2090                                        16, 1, skb->head,
2091                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2092                         WARN_ON_ONCE(1);
2093                         return -EINVAL;
2094                 }
2095
2096                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2097                         return -EFAULT;
2098
2099                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2100         }
2101
2102         if (vlan_hlen) {
2103                 int ret;
2104                 struct veth veth;
2105
2106                 veth.h_vlan_proto = skb->vlan_proto;
2107                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2108
2109                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2110
2111                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2112                 if (ret || !iov_iter_count(iter))
2113                         goto done;
2114
2115                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2116                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2117                         goto done;
2118         }
2119
2120         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2121
2122 done:
2123         /* caller is in process context, */
2124         preempt_disable();
2125         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2126         preempt_enable();
2127
2128         return total;
2129 }
2130
2131 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2132 {
2133         DECLARE_WAITQUEUE(wait, current);
2134         void *ptr = NULL;
2135         int error = 0;
2136
2137         ptr = ptr_ring_consume(&tfile->tx_ring);
2138         if (ptr)
2139                 goto out;
2140         if (noblock) {
2141                 error = -EAGAIN;
2142                 goto out;
2143         }
2144
2145         add_wait_queue(&tfile->socket.wq.wait, &wait);
2146
2147         while (1) {
2148                 set_current_state(TASK_INTERRUPTIBLE);
2149                 ptr = ptr_ring_consume(&tfile->tx_ring);
2150                 if (ptr)
2151                         break;
2152                 if (signal_pending(current)) {
2153                         error = -ERESTARTSYS;
2154                         break;
2155                 }
2156                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2157                         error = -EFAULT;
2158                         break;
2159                 }
2160
2161                 schedule();
2162         }
2163
2164         __set_current_state(TASK_RUNNING);
2165         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2166
2167 out:
2168         *err = error;
2169         return ptr;
2170 }
2171
2172 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2173                            struct iov_iter *to,
2174                            int noblock, void *ptr)
2175 {
2176         ssize_t ret;
2177         int err;
2178
2179         if (!iov_iter_count(to)) {
2180                 tun_ptr_free(ptr);
2181                 return 0;
2182         }
2183
2184         if (!ptr) {
2185                 /* Read frames from ring */
2186                 ptr = tun_ring_recv(tfile, noblock, &err);
2187                 if (!ptr)
2188                         return err;
2189         }
2190
2191         if (tun_is_xdp_frame(ptr)) {
2192                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2193
2194                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2195                 xdp_return_frame(xdpf);
2196         } else {
2197                 struct sk_buff *skb = ptr;
2198
2199                 ret = tun_put_user(tun, tfile, skb, to);
2200                 if (unlikely(ret < 0))
2201                         kfree_skb(skb);
2202                 else
2203                         consume_skb(skb);
2204         }
2205
2206         return ret;
2207 }
2208
2209 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2210 {
2211         struct file *file = iocb->ki_filp;
2212         struct tun_file *tfile = file->private_data;
2213         struct tun_struct *tun = tun_get(tfile);
2214         ssize_t len = iov_iter_count(to), ret;
2215         int noblock = 0;
2216
2217         if (!tun)
2218                 return -EBADFD;
2219
2220         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2221                 noblock = 1;
2222
2223         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2224         ret = min_t(ssize_t, ret, len);
2225         if (ret > 0)
2226                 iocb->ki_pos = ret;
2227         tun_put(tun);
2228         return ret;
2229 }
2230
2231 static void tun_prog_free(struct rcu_head *rcu)
2232 {
2233         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2234
2235         bpf_prog_destroy(prog->prog);
2236         kfree(prog);
2237 }
2238
2239 static int __tun_set_ebpf(struct tun_struct *tun,
2240                           struct tun_prog __rcu **prog_p,
2241                           struct bpf_prog *prog)
2242 {
2243         struct tun_prog *old, *new = NULL;
2244
2245         if (prog) {
2246                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2247                 if (!new)
2248                         return -ENOMEM;
2249                 new->prog = prog;
2250         }
2251
2252         spin_lock_bh(&tun->lock);
2253         old = rcu_dereference_protected(*prog_p,
2254                                         lockdep_is_held(&tun->lock));
2255         rcu_assign_pointer(*prog_p, new);
2256         spin_unlock_bh(&tun->lock);
2257
2258         if (old)
2259                 call_rcu(&old->rcu, tun_prog_free);
2260
2261         return 0;
2262 }
2263
2264 static void tun_free_netdev(struct net_device *dev)
2265 {
2266         struct tun_struct *tun = netdev_priv(dev);
2267
2268         BUG_ON(!(list_empty(&tun->disabled)));
2269
2270         free_percpu(dev->tstats);
2271         tun_flow_uninit(tun);
2272         security_tun_dev_free_security(tun->security);
2273         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2274         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2275 }
2276
2277 static void tun_setup(struct net_device *dev)
2278 {
2279         struct tun_struct *tun = netdev_priv(dev);
2280
2281         tun->owner = INVALID_UID;
2282         tun->group = INVALID_GID;
2283         tun_default_link_ksettings(dev, &tun->link_ksettings);
2284
2285         dev->ethtool_ops = &tun_ethtool_ops;
2286         dev->needs_free_netdev = true;
2287         dev->priv_destructor = tun_free_netdev;
2288         /* We prefer our own queue length */
2289         dev->tx_queue_len = TUN_READQ_SIZE;
2290 }
2291
2292 /* Trivial set of netlink ops to allow deleting tun or tap
2293  * device with netlink.
2294  */
2295 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2296                         struct netlink_ext_ack *extack)
2297 {
2298         NL_SET_ERR_MSG(extack,
2299                        "tun/tap creation via rtnetlink is not supported.");
2300         return -EOPNOTSUPP;
2301 }
2302
2303 static size_t tun_get_size(const struct net_device *dev)
2304 {
2305         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2306         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2307
2308         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2309                nla_total_size(sizeof(gid_t)) + /* GROUP */
2310                nla_total_size(sizeof(u8)) + /* TYPE */
2311                nla_total_size(sizeof(u8)) + /* PI */
2312                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2313                nla_total_size(sizeof(u8)) + /* PERSIST */
2314                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2315                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2316                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2317                0;
2318 }
2319
2320 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2321 {
2322         struct tun_struct *tun = netdev_priv(dev);
2323
2324         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2325                 goto nla_put_failure;
2326         if (uid_valid(tun->owner) &&
2327             nla_put_u32(skb, IFLA_TUN_OWNER,
2328                         from_kuid_munged(current_user_ns(), tun->owner)))
2329                 goto nla_put_failure;
2330         if (gid_valid(tun->group) &&
2331             nla_put_u32(skb, IFLA_TUN_GROUP,
2332                         from_kgid_munged(current_user_ns(), tun->group)))
2333                 goto nla_put_failure;
2334         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2335                 goto nla_put_failure;
2336         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2337                 goto nla_put_failure;
2338         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2339                 goto nla_put_failure;
2340         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2341                        !!(tun->flags & IFF_MULTI_QUEUE)))
2342                 goto nla_put_failure;
2343         if (tun->flags & IFF_MULTI_QUEUE) {
2344                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2345                         goto nla_put_failure;
2346                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2347                                 tun->numdisabled))
2348                         goto nla_put_failure;
2349         }
2350
2351         return 0;
2352
2353 nla_put_failure:
2354         return -EMSGSIZE;
2355 }
2356
2357 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2358         .kind           = DRV_NAME,
2359         .priv_size      = sizeof(struct tun_struct),
2360         .setup          = tun_setup,
2361         .validate       = tun_validate,
2362         .get_size       = tun_get_size,
2363         .fill_info      = tun_fill_info,
2364 };
2365
2366 static void tun_sock_write_space(struct sock *sk)
2367 {
2368         struct tun_file *tfile;
2369         wait_queue_head_t *wqueue;
2370
2371         if (!sock_writeable(sk))
2372                 return;
2373
2374         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2375                 return;
2376
2377         wqueue = sk_sleep(sk);
2378         if (wqueue && waitqueue_active(wqueue))
2379                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2380                                                 EPOLLWRNORM | EPOLLWRBAND);
2381
2382         tfile = container_of(sk, struct tun_file, sk);
2383         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2384 }
2385
2386 static void tun_put_page(struct tun_page *tpage)
2387 {
2388         if (tpage->page)
2389                 __page_frag_cache_drain(tpage->page, tpage->count);
2390 }
2391
2392 static int tun_xdp_one(struct tun_struct *tun,
2393                        struct tun_file *tfile,
2394                        struct xdp_buff *xdp, int *flush,
2395                        struct tun_page *tpage)
2396 {
2397         unsigned int datasize = xdp->data_end - xdp->data;
2398         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2399         struct virtio_net_hdr *gso = &hdr->gso;
2400         struct bpf_prog *xdp_prog;
2401         struct sk_buff *skb = NULL;
2402         u32 rxhash = 0, act;
2403         int buflen = hdr->buflen;
2404         int err = 0;
2405         bool skb_xdp = false;
2406         struct page *page;
2407
2408         xdp_prog = rcu_dereference(tun->xdp_prog);
2409         if (xdp_prog) {
2410                 if (gso->gso_type) {
2411                         skb_xdp = true;
2412                         goto build;
2413                 }
2414
2415                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2416                 xdp_set_data_meta_invalid(xdp);
2417
2418                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2419                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2420                 if (err < 0) {
2421                         put_page(virt_to_head_page(xdp->data));
2422                         return err;
2423                 }
2424
2425                 switch (err) {
2426                 case XDP_REDIRECT:
2427                         *flush = true;
2428                         fallthrough;
2429                 case XDP_TX:
2430                         return 0;
2431                 case XDP_PASS:
2432                         break;
2433                 default:
2434                         page = virt_to_head_page(xdp->data);
2435                         if (tpage->page == page) {
2436                                 ++tpage->count;
2437                         } else {
2438                                 tun_put_page(tpage);
2439                                 tpage->page = page;
2440                                 tpage->count = 1;
2441                         }
2442                         return 0;
2443                 }
2444         }
2445
2446 build:
2447         skb = build_skb(xdp->data_hard_start, buflen);
2448         if (!skb) {
2449                 err = -ENOMEM;
2450                 goto out;
2451         }
2452
2453         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2454         skb_put(skb, xdp->data_end - xdp->data);
2455
2456         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2457                 atomic_long_inc(&tun->rx_frame_errors);
2458                 kfree_skb(skb);
2459                 err = -EINVAL;
2460                 goto out;
2461         }
2462
2463         skb->protocol = eth_type_trans(skb, tun->dev);
2464         skb_reset_network_header(skb);
2465         skb_probe_transport_header(skb);
2466         skb_record_rx_queue(skb, tfile->queue_index);
2467
2468         if (skb_xdp) {
2469                 err = do_xdp_generic(xdp_prog, skb);
2470                 if (err != XDP_PASS)
2471                         goto out;
2472         }
2473
2474         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2475             !tfile->detached)
2476                 rxhash = __skb_get_hash_symmetric(skb);
2477
2478         netif_receive_skb(skb);
2479
2480         /* No need to disable preemption here since this function is
2481          * always called with bh disabled
2482          */
2483         dev_sw_netstats_rx_add(tun->dev, datasize);
2484
2485         if (rxhash)
2486                 tun_flow_update(tun, rxhash, tfile);
2487
2488 out:
2489         return err;
2490 }
2491
2492 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2493 {
2494         int ret, i;
2495         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2496         struct tun_struct *tun = tun_get(tfile);
2497         struct tun_msg_ctl *ctl = m->msg_control;
2498         struct xdp_buff *xdp;
2499
2500         if (!tun)
2501                 return -EBADFD;
2502
2503         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2504             ctl && ctl->type == TUN_MSG_PTR) {
2505                 struct tun_page tpage;
2506                 int n = ctl->num;
2507                 int flush = 0;
2508
2509                 memset(&tpage, 0, sizeof(tpage));
2510
2511                 local_bh_disable();
2512                 rcu_read_lock();
2513
2514                 for (i = 0; i < n; i++) {
2515                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2516                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2517                 }
2518
2519                 if (flush)
2520                         xdp_do_flush();
2521
2522                 rcu_read_unlock();
2523                 local_bh_enable();
2524
2525                 tun_put_page(&tpage);
2526
2527                 ret = total_len;
2528                 goto out;
2529         }
2530
2531         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2532                            m->msg_flags & MSG_DONTWAIT,
2533                            m->msg_flags & MSG_MORE);
2534 out:
2535         tun_put(tun);
2536         return ret;
2537 }
2538
2539 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2540                        int flags)
2541 {
2542         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2543         struct tun_struct *tun = tun_get(tfile);
2544         void *ptr = m->msg_control;
2545         int ret;
2546
2547         if (!tun) {
2548                 ret = -EBADFD;
2549                 goto out_free;
2550         }
2551
2552         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2553                 ret = -EINVAL;
2554                 goto out_put_tun;
2555         }
2556         if (flags & MSG_ERRQUEUE) {
2557                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2558                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2559                 goto out;
2560         }
2561         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2562         if (ret > (ssize_t)total_len) {
2563                 m->msg_flags |= MSG_TRUNC;
2564                 ret = flags & MSG_TRUNC ? ret : total_len;
2565         }
2566 out:
2567         tun_put(tun);
2568         return ret;
2569
2570 out_put_tun:
2571         tun_put(tun);
2572 out_free:
2573         tun_ptr_free(ptr);
2574         return ret;
2575 }
2576
2577 static int tun_ptr_peek_len(void *ptr)
2578 {
2579         if (likely(ptr)) {
2580                 if (tun_is_xdp_frame(ptr)) {
2581                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2582
2583                         return xdpf->len;
2584                 }
2585                 return __skb_array_len_with_tag(ptr);
2586         } else {
2587                 return 0;
2588         }
2589 }
2590
2591 static int tun_peek_len(struct socket *sock)
2592 {
2593         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2594         struct tun_struct *tun;
2595         int ret = 0;
2596
2597         tun = tun_get(tfile);
2598         if (!tun)
2599                 return 0;
2600
2601         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2602         tun_put(tun);
2603
2604         return ret;
2605 }
2606
2607 /* Ops structure to mimic raw sockets with tun */
2608 static const struct proto_ops tun_socket_ops = {
2609         .peek_len = tun_peek_len,
2610         .sendmsg = tun_sendmsg,
2611         .recvmsg = tun_recvmsg,
2612 };
2613
2614 static struct proto tun_proto = {
2615         .name           = "tun",
2616         .owner          = THIS_MODULE,
2617         .obj_size       = sizeof(struct tun_file),
2618 };
2619
2620 static int tun_flags(struct tun_struct *tun)
2621 {
2622         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2623 }
2624
2625 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2626                               char *buf)
2627 {
2628         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2629         return sprintf(buf, "0x%x\n", tun_flags(tun));
2630 }
2631
2632 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2633                           char *buf)
2634 {
2635         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2636         return uid_valid(tun->owner)?
2637                 sprintf(buf, "%u\n",
2638                         from_kuid_munged(current_user_ns(), tun->owner)):
2639                 sprintf(buf, "-1\n");
2640 }
2641
2642 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2643                           char *buf)
2644 {
2645         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2646         return gid_valid(tun->group) ?
2647                 sprintf(buf, "%u\n",
2648                         from_kgid_munged(current_user_ns(), tun->group)):
2649                 sprintf(buf, "-1\n");
2650 }
2651
2652 static DEVICE_ATTR_RO(tun_flags);
2653 static DEVICE_ATTR_RO(owner);
2654 static DEVICE_ATTR_RO(group);
2655
2656 static struct attribute *tun_dev_attrs[] = {
2657         &dev_attr_tun_flags.attr,
2658         &dev_attr_owner.attr,
2659         &dev_attr_group.attr,
2660         NULL
2661 };
2662
2663 static const struct attribute_group tun_attr_group = {
2664         .attrs = tun_dev_attrs
2665 };
2666
2667 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2668 {
2669         struct tun_struct *tun;
2670         struct tun_file *tfile = file->private_data;
2671         struct net_device *dev;
2672         int err;
2673
2674         if (tfile->detached)
2675                 return -EINVAL;
2676
2677         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2678                 if (!capable(CAP_NET_ADMIN))
2679                         return -EPERM;
2680
2681                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2682                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2683                         return -EINVAL;
2684         }
2685
2686         dev = __dev_get_by_name(net, ifr->ifr_name);
2687         if (dev) {
2688                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2689                         return -EBUSY;
2690                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2691                         tun = netdev_priv(dev);
2692                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2693                         tun = netdev_priv(dev);
2694                 else
2695                         return -EINVAL;
2696
2697                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2698                     !!(tun->flags & IFF_MULTI_QUEUE))
2699                         return -EINVAL;
2700
2701                 if (tun_not_capable(tun))
2702                         return -EPERM;
2703                 err = security_tun_dev_open(tun->security);
2704                 if (err < 0)
2705                         return err;
2706
2707                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2708                                  ifr->ifr_flags & IFF_NAPI,
2709                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2710                 if (err < 0)
2711                         return err;
2712
2713                 if (tun->flags & IFF_MULTI_QUEUE &&
2714                     (tun->numqueues + tun->numdisabled > 1)) {
2715                         /* One or more queue has already been attached, no need
2716                          * to initialize the device again.
2717                          */
2718                         netdev_state_change(dev);
2719                         return 0;
2720                 }
2721
2722                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2723                               (ifr->ifr_flags & TUN_FEATURES);
2724
2725                 netdev_state_change(dev);
2726         } else {
2727                 char *name;
2728                 unsigned long flags = 0;
2729                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2730                              MAX_TAP_QUEUES : 1;
2731
2732                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2733                         return -EPERM;
2734                 err = security_tun_dev_create();
2735                 if (err < 0)
2736                         return err;
2737
2738                 /* Set dev type */
2739                 if (ifr->ifr_flags & IFF_TUN) {
2740                         /* TUN device */
2741                         flags |= IFF_TUN;
2742                         name = "tun%d";
2743                 } else if (ifr->ifr_flags & IFF_TAP) {
2744                         /* TAP device */
2745                         flags |= IFF_TAP;
2746                         name = "tap%d";
2747                 } else
2748                         return -EINVAL;
2749
2750                 if (*ifr->ifr_name)
2751                         name = ifr->ifr_name;
2752
2753                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2754                                        NET_NAME_UNKNOWN, tun_setup, queues,
2755                                        queues);
2756
2757                 if (!dev)
2758                         return -ENOMEM;
2759
2760                 dev_net_set(dev, net);
2761                 dev->rtnl_link_ops = &tun_link_ops;
2762                 dev->ifindex = tfile->ifindex;
2763                 dev->sysfs_groups[0] = &tun_attr_group;
2764
2765                 tun = netdev_priv(dev);
2766                 tun->dev = dev;
2767                 tun->flags = flags;
2768                 tun->txflt.count = 0;
2769                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2770
2771                 tun->align = NET_SKB_PAD;
2772                 tun->filter_attached = false;
2773                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2774                 tun->rx_batched = 0;
2775                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2776
2777                 tun->ifr = ifr;
2778                 tun->file = file;
2779
2780                 tun_net_initialize(dev);
2781
2782                 err = register_netdevice(tun->dev);
2783                 if (err < 0) {
2784                         free_netdev(dev);
2785                         return err;
2786                 }
2787                 /* free_netdev() won't check refcnt, to avoid race
2788                  * with dev_put() we need publish tun after registration.
2789                  */
2790                 rcu_assign_pointer(tfile->tun, tun);
2791         }
2792
2793         netif_carrier_on(tun->dev);
2794
2795         /* Make sure persistent devices do not get stuck in
2796          * xoff state.
2797          */
2798         if (netif_running(tun->dev))
2799                 netif_tx_wake_all_queues(tun->dev);
2800
2801         strcpy(ifr->ifr_name, tun->dev->name);
2802         return 0;
2803 }
2804
2805 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2806 {
2807         strcpy(ifr->ifr_name, tun->dev->name);
2808
2809         ifr->ifr_flags = tun_flags(tun);
2810
2811 }
2812
2813 /* This is like a cut-down ethtool ops, except done via tun fd so no
2814  * privs required. */
2815 static int set_offload(struct tun_struct *tun, unsigned long arg)
2816 {
2817         netdev_features_t features = 0;
2818
2819         if (arg & TUN_F_CSUM) {
2820                 features |= NETIF_F_HW_CSUM;
2821                 arg &= ~TUN_F_CSUM;
2822
2823                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2824                         if (arg & TUN_F_TSO_ECN) {
2825                                 features |= NETIF_F_TSO_ECN;
2826                                 arg &= ~TUN_F_TSO_ECN;
2827                         }
2828                         if (arg & TUN_F_TSO4)
2829                                 features |= NETIF_F_TSO;
2830                         if (arg & TUN_F_TSO6)
2831                                 features |= NETIF_F_TSO6;
2832                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2833                 }
2834
2835                 arg &= ~TUN_F_UFO;
2836         }
2837
2838         /* This gives the user a way to test for new features in future by
2839          * trying to set them. */
2840         if (arg)
2841                 return -EINVAL;
2842
2843         tun->set_features = features;
2844         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2845         tun->dev->wanted_features |= features;
2846         netdev_update_features(tun->dev);
2847
2848         return 0;
2849 }
2850
2851 static void tun_detach_filter(struct tun_struct *tun, int n)
2852 {
2853         int i;
2854         struct tun_file *tfile;
2855
2856         for (i = 0; i < n; i++) {
2857                 tfile = rtnl_dereference(tun->tfiles[i]);
2858                 lock_sock(tfile->socket.sk);
2859                 sk_detach_filter(tfile->socket.sk);
2860                 release_sock(tfile->socket.sk);
2861         }
2862
2863         tun->filter_attached = false;
2864 }
2865
2866 static int tun_attach_filter(struct tun_struct *tun)
2867 {
2868         int i, ret = 0;
2869         struct tun_file *tfile;
2870
2871         for (i = 0; i < tun->numqueues; i++) {
2872                 tfile = rtnl_dereference(tun->tfiles[i]);
2873                 lock_sock(tfile->socket.sk);
2874                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2875                 release_sock(tfile->socket.sk);
2876                 if (ret) {
2877                         tun_detach_filter(tun, i);
2878                         return ret;
2879                 }
2880         }
2881
2882         tun->filter_attached = true;
2883         return ret;
2884 }
2885
2886 static void tun_set_sndbuf(struct tun_struct *tun)
2887 {
2888         struct tun_file *tfile;
2889         int i;
2890
2891         for (i = 0; i < tun->numqueues; i++) {
2892                 tfile = rtnl_dereference(tun->tfiles[i]);
2893                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2894         }
2895 }
2896
2897 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2898 {
2899         struct tun_file *tfile = file->private_data;
2900         struct tun_struct *tun;
2901         int ret = 0;
2902
2903         rtnl_lock();
2904
2905         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2906                 tun = tfile->detached;
2907                 if (!tun) {
2908                         ret = -EINVAL;
2909                         goto unlock;
2910                 }
2911                 ret = security_tun_dev_attach_queue(tun->security);
2912                 if (ret < 0)
2913                         goto unlock;
2914                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2915                                  tun->flags & IFF_NAPI_FRAGS, true);
2916         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2917                 tun = rtnl_dereference(tfile->tun);
2918                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2919                         ret = -EINVAL;
2920                 else
2921                         __tun_detach(tfile, false);
2922         } else
2923                 ret = -EINVAL;
2924
2925         if (ret >= 0)
2926                 netdev_state_change(tun->dev);
2927
2928 unlock:
2929         rtnl_unlock();
2930         return ret;
2931 }
2932
2933 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2934                         void __user *data)
2935 {
2936         struct bpf_prog *prog;
2937         int fd;
2938
2939         if (copy_from_user(&fd, data, sizeof(fd)))
2940                 return -EFAULT;
2941
2942         if (fd == -1) {
2943                 prog = NULL;
2944         } else {
2945                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2946                 if (IS_ERR(prog))
2947                         return PTR_ERR(prog);
2948         }
2949
2950         return __tun_set_ebpf(tun, prog_p, prog);
2951 }
2952
2953 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2954 static unsigned char tun_get_addr_len(unsigned short type)
2955 {
2956         switch (type) {
2957         case ARPHRD_IP6GRE:
2958         case ARPHRD_TUNNEL6:
2959                 return sizeof(struct in6_addr);
2960         case ARPHRD_IPGRE:
2961         case ARPHRD_TUNNEL:
2962         case ARPHRD_SIT:
2963                 return 4;
2964         case ARPHRD_ETHER:
2965                 return ETH_ALEN;
2966         case ARPHRD_IEEE802154:
2967         case ARPHRD_IEEE802154_MONITOR:
2968                 return IEEE802154_EXTENDED_ADDR_LEN;
2969         case ARPHRD_PHONET_PIPE:
2970         case ARPHRD_PPP:
2971         case ARPHRD_NONE:
2972                 return 0;
2973         case ARPHRD_6LOWPAN:
2974                 return EUI64_ADDR_LEN;
2975         case ARPHRD_FDDI:
2976                 return FDDI_K_ALEN;
2977         case ARPHRD_HIPPI:
2978                 return HIPPI_ALEN;
2979         case ARPHRD_IEEE802:
2980                 return FC_ALEN;
2981         case ARPHRD_ROSE:
2982                 return ROSE_ADDR_LEN;
2983         case ARPHRD_NETROM:
2984                 return AX25_ADDR_LEN;
2985         case ARPHRD_LOCALTLK:
2986                 return LTALK_ALEN;
2987         default:
2988                 return 0;
2989         }
2990 }
2991
2992 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2993                             unsigned long arg, int ifreq_len)
2994 {
2995         struct tun_file *tfile = file->private_data;
2996         struct net *net = sock_net(&tfile->sk);
2997         struct tun_struct *tun;
2998         void __user* argp = (void __user*)arg;
2999         unsigned int ifindex, carrier;
3000         struct ifreq ifr;
3001         kuid_t owner;
3002         kgid_t group;
3003         int sndbuf;
3004         int vnet_hdr_sz;
3005         int le;
3006         int ret;
3007         bool do_notify = false;
3008
3009         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3010             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3011                 if (copy_from_user(&ifr, argp, ifreq_len))
3012                         return -EFAULT;
3013         } else {
3014                 memset(&ifr, 0, sizeof(ifr));
3015         }
3016         if (cmd == TUNGETFEATURES) {
3017                 /* Currently this just means: "what IFF flags are valid?".
3018                  * This is needed because we never checked for invalid flags on
3019                  * TUNSETIFF.
3020                  */
3021                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3022                                 (unsigned int __user*)argp);
3023         } else if (cmd == TUNSETQUEUE) {
3024                 return tun_set_queue(file, &ifr);
3025         } else if (cmd == SIOCGSKNS) {
3026                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3027                         return -EPERM;
3028                 return open_related_ns(&net->ns, get_net_ns);
3029         }
3030
3031         rtnl_lock();
3032
3033         tun = tun_get(tfile);
3034         if (cmd == TUNSETIFF) {
3035                 ret = -EEXIST;
3036                 if (tun)
3037                         goto unlock;
3038
3039                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3040
3041                 ret = tun_set_iff(net, file, &ifr);
3042
3043                 if (ret)
3044                         goto unlock;
3045
3046                 if (copy_to_user(argp, &ifr, ifreq_len))
3047                         ret = -EFAULT;
3048                 goto unlock;
3049         }
3050         if (cmd == TUNSETIFINDEX) {
3051                 ret = -EPERM;
3052                 if (tun)
3053                         goto unlock;
3054
3055                 ret = -EFAULT;
3056                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3057                         goto unlock;
3058
3059                 ret = 0;
3060                 tfile->ifindex = ifindex;
3061                 goto unlock;
3062         }
3063
3064         ret = -EBADFD;
3065         if (!tun)
3066                 goto unlock;
3067
3068         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3069
3070         net = dev_net(tun->dev);
3071         ret = 0;
3072         switch (cmd) {
3073         case TUNGETIFF:
3074                 tun_get_iff(tun, &ifr);
3075
3076                 if (tfile->detached)
3077                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3078                 if (!tfile->socket.sk->sk_filter)
3079                         ifr.ifr_flags |= IFF_NOFILTER;
3080
3081                 if (copy_to_user(argp, &ifr, ifreq_len))
3082                         ret = -EFAULT;
3083                 break;
3084
3085         case TUNSETNOCSUM:
3086                 /* Disable/Enable checksum */
3087
3088                 /* [unimplemented] */
3089                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3090                            arg ? "disabled" : "enabled");
3091                 break;
3092
3093         case TUNSETPERSIST:
3094                 /* Disable/Enable persist mode. Keep an extra reference to the
3095                  * module to prevent the module being unprobed.
3096                  */
3097                 if (arg && !(tun->flags & IFF_PERSIST)) {
3098                         tun->flags |= IFF_PERSIST;
3099                         __module_get(THIS_MODULE);
3100                         do_notify = true;
3101                 }
3102                 if (!arg && (tun->flags & IFF_PERSIST)) {
3103                         tun->flags &= ~IFF_PERSIST;
3104                         module_put(THIS_MODULE);
3105                         do_notify = true;
3106                 }
3107
3108                 netif_info(tun, drv, tun->dev, "persist %s\n",
3109                            arg ? "enabled" : "disabled");
3110                 break;
3111
3112         case TUNSETOWNER:
3113                 /* Set owner of the device */
3114                 owner = make_kuid(current_user_ns(), arg);
3115                 if (!uid_valid(owner)) {
3116                         ret = -EINVAL;
3117                         break;
3118                 }
3119                 tun->owner = owner;
3120                 do_notify = true;
3121                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3122                            from_kuid(&init_user_ns, tun->owner));
3123                 break;
3124
3125         case TUNSETGROUP:
3126                 /* Set group of the device */
3127                 group = make_kgid(current_user_ns(), arg);
3128                 if (!gid_valid(group)) {
3129                         ret = -EINVAL;
3130                         break;
3131                 }
3132                 tun->group = group;
3133                 do_notify = true;
3134                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3135                            from_kgid(&init_user_ns, tun->group));
3136                 break;
3137
3138         case TUNSETLINK:
3139                 /* Only allow setting the type when the interface is down */
3140                 if (tun->dev->flags & IFF_UP) {
3141                         netif_info(tun, drv, tun->dev,
3142                                    "Linktype set failed because interface is up\n");
3143                         ret = -EBUSY;
3144                 } else {
3145                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3146                                                        tun->dev);
3147                         ret = notifier_to_errno(ret);
3148                         if (ret) {
3149                                 netif_info(tun, drv, tun->dev,
3150                                            "Refused to change device type\n");
3151                                 break;
3152                         }
3153                         tun->dev->type = (int) arg;
3154                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3155                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3156                                    tun->dev->type);
3157                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3158                                                  tun->dev);
3159                 }
3160                 break;
3161
3162         case TUNSETDEBUG:
3163                 tun->msg_enable = (u32)arg;
3164                 break;
3165
3166         case TUNSETOFFLOAD:
3167                 ret = set_offload(tun, arg);
3168                 break;
3169
3170         case TUNSETTXFILTER:
3171                 /* Can be set only for TAPs */
3172                 ret = -EINVAL;
3173                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3174                         break;
3175                 ret = update_filter(&tun->txflt, (void __user *)arg);
3176                 break;
3177
3178         case SIOCGIFHWADDR:
3179                 /* Get hw address */
3180                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3181                 if (copy_to_user(argp, &ifr, ifreq_len))
3182                         ret = -EFAULT;
3183                 break;
3184
3185         case SIOCSIFHWADDR:
3186                 /* Set hw address */
3187                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3188                 break;
3189
3190         case TUNGETSNDBUF:
3191                 sndbuf = tfile->socket.sk->sk_sndbuf;
3192                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3193                         ret = -EFAULT;
3194                 break;
3195
3196         case TUNSETSNDBUF:
3197                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3198                         ret = -EFAULT;
3199                         break;
3200                 }
3201                 if (sndbuf <= 0) {
3202                         ret = -EINVAL;
3203                         break;
3204                 }
3205
3206                 tun->sndbuf = sndbuf;
3207                 tun_set_sndbuf(tun);
3208                 break;
3209
3210         case TUNGETVNETHDRSZ:
3211                 vnet_hdr_sz = tun->vnet_hdr_sz;
3212                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3213                         ret = -EFAULT;
3214                 break;
3215
3216         case TUNSETVNETHDRSZ:
3217                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3218                         ret = -EFAULT;
3219                         break;
3220                 }
3221                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3222                         ret = -EINVAL;
3223                         break;
3224                 }
3225
3226                 tun->vnet_hdr_sz = vnet_hdr_sz;
3227                 break;
3228
3229         case TUNGETVNETLE:
3230                 le = !!(tun->flags & TUN_VNET_LE);
3231                 if (put_user(le, (int __user *)argp))
3232                         ret = -EFAULT;
3233                 break;
3234
3235         case TUNSETVNETLE:
3236                 if (get_user(le, (int __user *)argp)) {
3237                         ret = -EFAULT;
3238                         break;
3239                 }
3240                 if (le)
3241                         tun->flags |= TUN_VNET_LE;
3242                 else
3243                         tun->flags &= ~TUN_VNET_LE;
3244                 break;
3245
3246         case TUNGETVNETBE:
3247                 ret = tun_get_vnet_be(tun, argp);
3248                 break;
3249
3250         case TUNSETVNETBE:
3251                 ret = tun_set_vnet_be(tun, argp);
3252                 break;
3253
3254         case TUNATTACHFILTER:
3255                 /* Can be set only for TAPs */
3256                 ret = -EINVAL;
3257                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3258                         break;
3259                 ret = -EFAULT;
3260                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3261                         break;
3262
3263                 ret = tun_attach_filter(tun);
3264                 break;
3265
3266         case TUNDETACHFILTER:
3267                 /* Can be set only for TAPs */
3268                 ret = -EINVAL;
3269                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3270                         break;
3271                 ret = 0;
3272                 tun_detach_filter(tun, tun->numqueues);
3273                 break;
3274
3275         case TUNGETFILTER:
3276                 ret = -EINVAL;
3277                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3278                         break;
3279                 ret = -EFAULT;
3280                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3281                         break;
3282                 ret = 0;
3283                 break;
3284
3285         case TUNSETSTEERINGEBPF:
3286                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3287                 break;
3288
3289         case TUNSETFILTEREBPF:
3290                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3291                 break;
3292
3293         case TUNSETCARRIER:
3294                 ret = -EFAULT;
3295                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3296                         goto unlock;
3297
3298                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3299                 break;
3300
3301         case TUNGETDEVNETNS:
3302                 ret = -EPERM;
3303                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3304                         goto unlock;
3305                 ret = open_related_ns(&net->ns, get_net_ns);
3306                 break;
3307
3308         default:
3309                 ret = -EINVAL;
3310                 break;
3311         }
3312
3313         if (do_notify)
3314                 netdev_state_change(tun->dev);
3315
3316 unlock:
3317         rtnl_unlock();
3318         if (tun)
3319                 tun_put(tun);
3320         return ret;
3321 }
3322
3323 static long tun_chr_ioctl(struct file *file,
3324                           unsigned int cmd, unsigned long arg)
3325 {
3326         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3327 }
3328
3329 #ifdef CONFIG_COMPAT
3330 static long tun_chr_compat_ioctl(struct file *file,
3331                          unsigned int cmd, unsigned long arg)
3332 {
3333         switch (cmd) {
3334         case TUNSETIFF:
3335         case TUNGETIFF:
3336         case TUNSETTXFILTER:
3337         case TUNGETSNDBUF:
3338         case TUNSETSNDBUF:
3339         case SIOCGIFHWADDR:
3340         case SIOCSIFHWADDR:
3341                 arg = (unsigned long)compat_ptr(arg);
3342                 break;
3343         default:
3344                 arg = (compat_ulong_t)arg;
3345                 break;
3346         }
3347
3348         /*
3349          * compat_ifreq is shorter than ifreq, so we must not access beyond
3350          * the end of that structure. All fields that are used in this
3351          * driver are compatible though, we don't need to convert the
3352          * contents.
3353          */
3354         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3355 }
3356 #endif /* CONFIG_COMPAT */
3357
3358 static int tun_chr_fasync(int fd, struct file *file, int on)
3359 {
3360         struct tun_file *tfile = file->private_data;
3361         int ret;
3362
3363         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3364                 goto out;
3365
3366         if (on) {
3367                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3368                 tfile->flags |= TUN_FASYNC;
3369         } else
3370                 tfile->flags &= ~TUN_FASYNC;
3371         ret = 0;
3372 out:
3373         return ret;
3374 }
3375
3376 static int tun_chr_open(struct inode *inode, struct file * file)
3377 {
3378         struct net *net = current->nsproxy->net_ns;
3379         struct tun_file *tfile;
3380
3381         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3382                                             &tun_proto, 0);
3383         if (!tfile)
3384                 return -ENOMEM;
3385         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3386                 sk_free(&tfile->sk);
3387                 return -ENOMEM;
3388         }
3389
3390         mutex_init(&tfile->napi_mutex);
3391         RCU_INIT_POINTER(tfile->tun, NULL);
3392         tfile->flags = 0;
3393         tfile->ifindex = 0;
3394
3395         init_waitqueue_head(&tfile->socket.wq.wait);
3396
3397         tfile->socket.file = file;
3398         tfile->socket.ops = &tun_socket_ops;
3399
3400         sock_init_data(&tfile->socket, &tfile->sk);
3401
3402         tfile->sk.sk_write_space = tun_sock_write_space;
3403         tfile->sk.sk_sndbuf = INT_MAX;
3404
3405         file->private_data = tfile;
3406         INIT_LIST_HEAD(&tfile->next);
3407
3408         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3409
3410         return 0;
3411 }
3412
3413 static int tun_chr_close(struct inode *inode, struct file *file)
3414 {
3415         struct tun_file *tfile = file->private_data;
3416
3417         tun_detach(tfile, true);
3418
3419         return 0;
3420 }
3421
3422 #ifdef CONFIG_PROC_FS
3423 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3424 {
3425         struct tun_file *tfile = file->private_data;
3426         struct tun_struct *tun;
3427         struct ifreq ifr;
3428
3429         memset(&ifr, 0, sizeof(ifr));
3430
3431         rtnl_lock();
3432         tun = tun_get(tfile);
3433         if (tun)
3434                 tun_get_iff(tun, &ifr);
3435         rtnl_unlock();
3436
3437         if (tun)
3438                 tun_put(tun);
3439
3440         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3441 }
3442 #endif
3443
3444 static const struct file_operations tun_fops = {
3445         .owner  = THIS_MODULE,
3446         .llseek = no_llseek,
3447         .read_iter  = tun_chr_read_iter,
3448         .write_iter = tun_chr_write_iter,
3449         .poll   = tun_chr_poll,
3450         .unlocked_ioctl = tun_chr_ioctl,
3451 #ifdef CONFIG_COMPAT
3452         .compat_ioctl = tun_chr_compat_ioctl,
3453 #endif
3454         .open   = tun_chr_open,
3455         .release = tun_chr_close,
3456         .fasync = tun_chr_fasync,
3457 #ifdef CONFIG_PROC_FS
3458         .show_fdinfo = tun_chr_show_fdinfo,
3459 #endif
3460 };
3461
3462 static struct miscdevice tun_miscdev = {
3463         .minor = TUN_MINOR,
3464         .name = "tun",
3465         .nodename = "net/tun",
3466         .fops = &tun_fops,
3467 };
3468
3469 /* ethtool interface */
3470
3471 static void tun_default_link_ksettings(struct net_device *dev,
3472                                        struct ethtool_link_ksettings *cmd)
3473 {
3474         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3475         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3476         cmd->base.speed         = SPEED_10;
3477         cmd->base.duplex        = DUPLEX_FULL;
3478         cmd->base.port          = PORT_TP;
3479         cmd->base.phy_address   = 0;
3480         cmd->base.autoneg       = AUTONEG_DISABLE;
3481 }
3482
3483 static int tun_get_link_ksettings(struct net_device *dev,
3484                                   struct ethtool_link_ksettings *cmd)
3485 {
3486         struct tun_struct *tun = netdev_priv(dev);
3487
3488         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3489         return 0;
3490 }
3491
3492 static int tun_set_link_ksettings(struct net_device *dev,
3493                                   const struct ethtool_link_ksettings *cmd)
3494 {
3495         struct tun_struct *tun = netdev_priv(dev);
3496
3497         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3498         return 0;
3499 }
3500
3501 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3502 {
3503         struct tun_struct *tun = netdev_priv(dev);
3504
3505         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3506         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3507
3508         switch (tun->flags & TUN_TYPE_MASK) {
3509         case IFF_TUN:
3510                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3511                 break;
3512         case IFF_TAP:
3513                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3514                 break;
3515         }
3516 }
3517
3518 static u32 tun_get_msglevel(struct net_device *dev)
3519 {
3520         struct tun_struct *tun = netdev_priv(dev);
3521
3522         return tun->msg_enable;
3523 }
3524
3525 static void tun_set_msglevel(struct net_device *dev, u32 value)
3526 {
3527         struct tun_struct *tun = netdev_priv(dev);
3528
3529         tun->msg_enable = value;
3530 }
3531
3532 static int tun_get_coalesce(struct net_device *dev,
3533                             struct ethtool_coalesce *ec,
3534                             struct kernel_ethtool_coalesce *kernel_coal,
3535                             struct netlink_ext_ack *extack)
3536 {
3537         struct tun_struct *tun = netdev_priv(dev);
3538
3539         ec->rx_max_coalesced_frames = tun->rx_batched;
3540
3541         return 0;
3542 }
3543
3544 static int tun_set_coalesce(struct net_device *dev,
3545                             struct ethtool_coalesce *ec,
3546                             struct kernel_ethtool_coalesce *kernel_coal,
3547                             struct netlink_ext_ack *extack)
3548 {
3549         struct tun_struct *tun = netdev_priv(dev);
3550
3551         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3552                 tun->rx_batched = NAPI_POLL_WEIGHT;
3553         else
3554                 tun->rx_batched = ec->rx_max_coalesced_frames;
3555
3556         return 0;
3557 }
3558
3559 static const struct ethtool_ops tun_ethtool_ops = {
3560         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3561         .get_drvinfo    = tun_get_drvinfo,
3562         .get_msglevel   = tun_get_msglevel,
3563         .set_msglevel   = tun_set_msglevel,
3564         .get_link       = ethtool_op_get_link,
3565         .get_ts_info    = ethtool_op_get_ts_info,
3566         .get_coalesce   = tun_get_coalesce,
3567         .set_coalesce   = tun_set_coalesce,
3568         .get_link_ksettings = tun_get_link_ksettings,
3569         .set_link_ksettings = tun_set_link_ksettings,
3570 };
3571
3572 static int tun_queue_resize(struct tun_struct *tun)
3573 {
3574         struct net_device *dev = tun->dev;
3575         struct tun_file *tfile;
3576         struct ptr_ring **rings;
3577         int n = tun->numqueues + tun->numdisabled;
3578         int ret, i;
3579
3580         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3581         if (!rings)
3582                 return -ENOMEM;
3583
3584         for (i = 0; i < tun->numqueues; i++) {
3585                 tfile = rtnl_dereference(tun->tfiles[i]);
3586                 rings[i] = &tfile->tx_ring;
3587         }
3588         list_for_each_entry(tfile, &tun->disabled, next)
3589                 rings[i++] = &tfile->tx_ring;
3590
3591         ret = ptr_ring_resize_multiple(rings, n,
3592                                        dev->tx_queue_len, GFP_KERNEL,
3593                                        tun_ptr_free);
3594
3595         kfree(rings);
3596         return ret;
3597 }
3598
3599 static int tun_device_event(struct notifier_block *unused,
3600                             unsigned long event, void *ptr)
3601 {
3602         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3603         struct tun_struct *tun = netdev_priv(dev);
3604         int i;
3605
3606         if (dev->rtnl_link_ops != &tun_link_ops)
3607                 return NOTIFY_DONE;
3608
3609         switch (event) {
3610         case NETDEV_CHANGE_TX_QUEUE_LEN:
3611                 if (tun_queue_resize(tun))
3612                         return NOTIFY_BAD;
3613                 break;
3614         case NETDEV_UP:
3615                 for (i = 0; i < tun->numqueues; i++) {
3616                         struct tun_file *tfile;
3617
3618                         tfile = rtnl_dereference(tun->tfiles[i]);
3619                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3620                 }
3621                 break;
3622         default:
3623                 break;
3624         }
3625
3626         return NOTIFY_DONE;
3627 }
3628
3629 static struct notifier_block tun_notifier_block __read_mostly = {
3630         .notifier_call  = tun_device_event,
3631 };
3632
3633 static int __init tun_init(void)
3634 {
3635         int ret = 0;
3636
3637         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3638
3639         ret = rtnl_link_register(&tun_link_ops);
3640         if (ret) {
3641                 pr_err("Can't register link_ops\n");
3642                 goto err_linkops;
3643         }
3644
3645         ret = misc_register(&tun_miscdev);
3646         if (ret) {
3647                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3648                 goto err_misc;
3649         }
3650
3651         ret = register_netdevice_notifier(&tun_notifier_block);
3652         if (ret) {
3653                 pr_err("Can't register netdevice notifier\n");
3654                 goto err_notifier;
3655         }
3656
3657         return  0;
3658
3659 err_notifier:
3660         misc_deregister(&tun_miscdev);
3661 err_misc:
3662         rtnl_link_unregister(&tun_link_ops);
3663 err_linkops:
3664         return ret;
3665 }
3666
3667 static void tun_cleanup(void)
3668 {
3669         misc_deregister(&tun_miscdev);
3670         rtnl_link_unregister(&tun_link_ops);
3671         unregister_netdevice_notifier(&tun_notifier_block);
3672 }
3673
3674 /* Get an underlying socket object from tun file.  Returns error unless file is
3675  * attached to a device.  The returned object works like a packet socket, it
3676  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3677  * holding a reference to the file for as long as the socket is in use. */
3678 struct socket *tun_get_socket(struct file *file)
3679 {
3680         struct tun_file *tfile;
3681         if (file->f_op != &tun_fops)
3682                 return ERR_PTR(-EINVAL);
3683         tfile = file->private_data;
3684         if (!tfile)
3685                 return ERR_PTR(-EBADFD);
3686         return &tfile->socket;
3687 }
3688 EXPORT_SYMBOL_GPL(tun_get_socket);
3689
3690 struct ptr_ring *tun_get_tx_ring(struct file *file)
3691 {
3692         struct tun_file *tfile;
3693
3694         if (file->f_op != &tun_fops)
3695                 return ERR_PTR(-EINVAL);
3696         tfile = file->private_data;
3697         if (!tfile)
3698                 return ERR_PTR(-EBADFD);
3699         return &tfile->tx_ring;
3700 }
3701 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3702
3703 module_init(tun_init);
3704 module_exit(tun_cleanup);
3705 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3706 MODULE_AUTHOR(DRV_COPYRIGHT);
3707 MODULE_LICENSE("GPL");
3708 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3709 MODULE_ALIAS("devname:net/tun");