drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <[email protected]> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <[email protected]>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <[email protected]>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <[email protected]>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <[email protected]>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212         /* init args */
 213         struct file *file;
 214         struct ifreq *ifr;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static void tun_flow_init(struct tun_struct *tun);
 223 static void tun_flow_uninit(struct tun_struct *tun);
 224
 225 static int tun_napi_receive(struct napi_struct *napi, int budget)
 226 {
 227         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 228         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 229         struct sk_buff_head process_queue;
 230         struct sk_buff *skb;
 231         int received = 0;
 232
 233         __skb_queue_head_init(&process_queue);
 234
 235         spin_lock(&queue->lock);
 236         skb_queue_splice_tail_init(queue, &process_queue);
 237         spin_unlock(&queue->lock);
 238
 239         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 240                 napi_gro_receive(napi, skb);
 241                 ++received;
 242         }
 243
 244         if (!skb_queue_empty(&process_queue)) {
 245                 spin_lock(&queue->lock);
 246                 skb_queue_splice(&process_queue, queue);
 247                 spin_unlock(&queue->lock);
 248         }
 249
 250         return received;
 251 }
 252
 253 static int tun_napi_poll(struct napi_struct *napi, int budget)
 254 {
 255         unsigned int received;
 256
 257         received = tun_napi_receive(napi, budget);
 258
 259         if (received < budget)
 260                 napi_complete_done(napi, received);
 261
 262         return received;
 263 }
 264
 265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 266                           bool napi_en, bool napi_frags)
 267 {
 268         tfile->napi_enabled = napi_en;
 269         tfile->napi_frags_enabled = napi_en && napi_frags;
 270         if (napi_en) {
 271                 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 272                 napi_enable(&tfile->napi);
 273         }
 274 }
 275
 276 static void tun_napi_disable(struct tun_file *tfile)
 277 {
 278         if (tfile->napi_enabled)
 279                 napi_disable(&tfile->napi);
 280 }
 281
 282 static void tun_napi_del(struct tun_file *tfile)
 283 {
 284         if (tfile->napi_enabled)
 285                 netif_napi_del(&tfile->napi);
 286 }
 287
 288 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 289 {
 290         return tfile->napi_frags_enabled;
 291 }
 292
 293 #ifdef CONFIG_TUN_VNET_CROSS_LE
 294 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 295 {
 296         return tun->flags & TUN_VNET_BE ? false :
 297                 virtio_legacy_is_little_endian();
 298 }
 299
 300 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 301 {
 302         int be = !!(tun->flags & TUN_VNET_BE);
 303
 304         if (put_user(be, argp))
 305                 return -EFAULT;
 306
 307         return 0;
 308 }
 309
 310 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 311 {
 312         int be;
 313
 314         if (get_user(be, argp))
 315                 return -EFAULT;
 316
 317         if (be)
 318                 tun->flags |= TUN_VNET_BE;
 319         else
 320                 tun->flags &= ~TUN_VNET_BE;
 321
 322         return 0;
 323 }
 324 #else
 325 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 326 {
 327         return virtio_legacy_is_little_endian();
 328 }
 329
 330 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 331 {
 332         return -EINVAL;
 333 }
 334
 335 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 336 {
 337         return -EINVAL;
 338 }
 339 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 340
 341 static inline bool tun_is_little_endian(struct tun_struct *tun)
 342 {
 343         return tun->flags & TUN_VNET_LE ||
 344                 tun_legacy_is_little_endian(tun);
 345 }
 346
 347 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 348 {
 349         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 350 }
 351
 352 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 353 {
 354         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 355 }
 356
 357 static inline u32 tun_hashfn(u32 rxhash)
 358 {
 359         return rxhash & TUN_MASK_FLOW_ENTRIES;
 360 }
 361
 362 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 363 {
 364         struct tun_flow_entry *e;
 365
 366         hlist_for_each_entry_rcu(e, head, hash_link) {
 367                 if (e->rxhash == rxhash)
 368                         return e;
 369         }
 370         return NULL;
 371 }
 372
 373 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 374                                               struct hlist_head *head,
 375                                               u32 rxhash, u16 queue_index)
 376 {
 377         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 378
 379         if (e) {
 380                 netif_info(tun, tx_queued, tun->dev,
 381                            "create flow: hash %u index %u\n",
 382                            rxhash, queue_index);
 383                 e->updated = jiffies;
 384                 e->rxhash = rxhash;
 385                 e->rps_rxhash = 0;
 386                 e->queue_index = queue_index;
 387                 e->tun = tun;
 388                 hlist_add_head_rcu(&e->hash_link, head);
 389                 ++tun->flow_count;
 390         }
 391         return e;
 392 }
 393
 394 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 395 {
 396         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 397                    e->rxhash, e->queue_index);
 398         hlist_del_rcu(&e->hash_link);
 399         kfree_rcu(e, rcu);
 400         --tun->flow_count;
 401 }
 402
 403 static void tun_flow_flush(struct tun_struct *tun)
 404 {
 405         int i;
 406
 407         spin_lock_bh(&tun->lock);
 408         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 409                 struct tun_flow_entry *e;
 410                 struct hlist_node *n;
 411
 412                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 413                         tun_flow_delete(tun, e);
 414         }
 415         spin_unlock_bh(&tun->lock);
 416 }
 417
 418 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 419 {
 420         int i;
 421
 422         spin_lock_bh(&tun->lock);
 423         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 424                 struct tun_flow_entry *e;
 425                 struct hlist_node *n;
 426
 427                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 428                         if (e->queue_index == queue_index)
 429                                 tun_flow_delete(tun, e);
 430                 }
 431         }
 432         spin_unlock_bh(&tun->lock);
 433 }
 434
 435 static void tun_flow_cleanup(struct timer_list *t)
 436 {
 437         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 438         unsigned long delay = tun->ageing_time;
 439         unsigned long next_timer = jiffies + delay;
 440         unsigned long count = 0;
 441         int i;
 442
 443         spin_lock(&tun->lock);
 444         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 445                 struct tun_flow_entry *e;
 446                 struct hlist_node *n;
 447
 448                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 449                         unsigned long this_timer;
 450
 451                         this_timer = e->updated + delay;
 452                         if (time_before_eq(this_timer, jiffies)) {
 453                                 tun_flow_delete(tun, e);
 454                                 continue;
 455                         }
 456                         count++;
 457                         if (time_before(this_timer, next_timer))
 458                                 next_timer = this_timer;
 459                 }
 460         }
 461
 462         if (count)
 463                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 464         spin_unlock(&tun->lock);
 465 }
 466
 467 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 468                             struct tun_file *tfile)
 469 {
 470         struct hlist_head *head;
 471         struct tun_flow_entry *e;
 472         unsigned long delay = tun->ageing_time;
 473         u16 queue_index = tfile->queue_index;
 474
 475         head = &tun->flows[tun_hashfn(rxhash)];
 476
 477         rcu_read_lock();
 478
 479         e = tun_flow_find(head, rxhash);
 480         if (likely(e)) {
 481                 /* TODO: keep queueing to old queue until it's empty? */
 482                 if (READ_ONCE(e->queue_index) != queue_index)
 483                         WRITE_ONCE(e->queue_index, queue_index);
 484                 if (e->updated != jiffies)
 485                         e->updated = jiffies;
 486                 sock_rps_record_flow_hash(e->rps_rxhash);
 487         } else {
 488                 spin_lock_bh(&tun->lock);
 489                 if (!tun_flow_find(head, rxhash) &&
 490                     tun->flow_count < MAX_TAP_FLOWS)
 491                         tun_flow_create(tun, head, rxhash, queue_index);
 492
 493                 if (!timer_pending(&tun->flow_gc_timer))
 494                         mod_timer(&tun->flow_gc_timer,
 495                                   round_jiffies_up(jiffies + delay));
 496                 spin_unlock_bh(&tun->lock);
 497         }
 498
 499         rcu_read_unlock();
 500 }
 501
 502 /* Save the hash received in the stack receive path and update the
 503  * flow_hash table accordingly.
 504  */
 505 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 506 {
 507         if (unlikely(e->rps_rxhash != hash))
 508                 e->rps_rxhash = hash;
 509 }
 510
 511 /* We try to identify a flow through its rxhash. The reason that
 512  * we do not check rxq no. is because some cards(e.g 82599), chooses
 513  * the rxq based on the txq where the last packet of the flow comes. As
 514  * the userspace application move between processors, we may get a
 515  * different rxq no. here.
 516  */
 517 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 518 {
 519         struct tun_flow_entry *e;
 520         u32 txq = 0;
 521         u32 numqueues = 0;
 522
 523         numqueues = READ_ONCE(tun->numqueues);
 524
 525         txq = __skb_get_hash_symmetric(skb);
 526         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 527         if (e) {
 528                 tun_flow_save_rps_rxhash(e, txq);
 529                 txq = e->queue_index;
 530         } else {
 531                 /* use multiply and shift instead of expensive divide */
 532                 txq = ((u64)txq * numqueues) >> 32;
 533         }
 534
 535         return txq;
 536 }
 537
 538 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 539 {
 540         struct tun_prog *prog;
 541         u32 numqueues;
 542         u16 ret = 0;
 543
 544         numqueues = READ_ONCE(tun->numqueues);
 545         if (!numqueues)
 546                 return 0;
 547
 548         prog = rcu_dereference(tun->steering_prog);
 549         if (prog)
 550                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 551
 552         return ret % numqueues;
 553 }
 554
 555 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 556                             struct net_device *sb_dev)
 557 {
 558         struct tun_struct *tun = netdev_priv(dev);
 559         u16 ret;
 560
 561         rcu_read_lock();
 562         if (rcu_dereference(tun->steering_prog))
 563                 ret = tun_ebpf_select_queue(tun, skb);
 564         else
 565                 ret = tun_automq_select_queue(tun, skb);
 566         rcu_read_unlock();
 567
 568         return ret;
 569 }
 570
 571 static inline bool tun_not_capable(struct tun_struct *tun)
 572 {
 573         const struct cred *cred = current_cred();
 574         struct net *net = dev_net(tun->dev);
 575
 576         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 577                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 578                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 579 }
 580
 581 static void tun_set_real_num_queues(struct tun_struct *tun)
 582 {
 583         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 584         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 585 }
 586
 587 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 588 {
 589         tfile->detached = tun;
 590         list_add_tail(&tfile->next, &tun->disabled);
 591         ++tun->numdisabled;
 592 }
 593
 594 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 595 {
 596         struct tun_struct *tun = tfile->detached;
 597
 598         tfile->detached = NULL;
 599         list_del_init(&tfile->next);
 600         --tun->numdisabled;
 601         return tun;
 602 }
 603
 604 void tun_ptr_free(void *ptr)
 605 {
 606         if (!ptr)
 607                 return;
 608         if (tun_is_xdp_frame(ptr)) {
 609                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 610
 611                 xdp_return_frame(xdpf);
 612         } else {
 613                 __skb_array_destroy_skb(ptr);
 614         }
 615 }
 616 EXPORT_SYMBOL_GPL(tun_ptr_free);
 617
 618 static void tun_queue_purge(struct tun_file *tfile)
 619 {
 620         void *ptr;
 621
 622         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 623                 tun_ptr_free(ptr);
 624
 625         skb_queue_purge(&tfile->sk.sk_write_queue);
 626         skb_queue_purge(&tfile->sk.sk_error_queue);
 627 }
 628
 629 static void __tun_detach(struct tun_file *tfile, bool clean)
 630 {
 631         struct tun_file *ntfile;
 632         struct tun_struct *tun;
 633
 634         tun = rtnl_dereference(tfile->tun);
 635
 636         if (tun && clean) {
 637                 tun_napi_disable(tfile);
 638                 tun_napi_del(tfile);
 639         }
 640
 641         if (tun && !tfile->detached) {
 642                 u16 index = tfile->queue_index;
 643                 BUG_ON(index >= tun->numqueues);
 644
 645                 rcu_assign_pointer(tun->tfiles[index],
 646                                    tun->tfiles[tun->numqueues - 1]);
 647                 ntfile = rtnl_dereference(tun->tfiles[index]);
 648                 ntfile->queue_index = index;
 649                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 650                                    NULL);
 651
 652                 --tun->numqueues;
 653                 if (clean) {
 654                         RCU_INIT_POINTER(tfile->tun, NULL);
 655                         sock_put(&tfile->sk);
 656                 } else
 657                         tun_disable_queue(tun, tfile);
 658
 659                 synchronize_net();
 660                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 661                 /* Drop read queue */
 662                 tun_queue_purge(tfile);
 663                 tun_set_real_num_queues(tun);
 664         } else if (tfile->detached && clean) {
 665                 tun = tun_enable_queue(tfile);
 666                 sock_put(&tfile->sk);
 667         }
 668
 669         if (clean) {
 670                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 671                         netif_carrier_off(tun->dev);
 672
 673                         if (!(tun->flags & IFF_PERSIST) &&
 674                             tun->dev->reg_state == NETREG_REGISTERED)
 675                                 unregister_netdevice(tun->dev);
 676                 }
 677                 if (tun)
 678                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 679                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 680                 sock_put(&tfile->sk);
 681         }
 682 }
 683
 684 static void tun_detach(struct tun_file *tfile, bool clean)
 685 {
 686         struct tun_struct *tun;
 687         struct net_device *dev;
 688
 689         rtnl_lock();
 690         tun = rtnl_dereference(tfile->tun);
 691         dev = tun ? tun->dev : NULL;
 692         __tun_detach(tfile, clean);
 693         if (dev)
 694                 netdev_state_change(dev);
 695         rtnl_unlock();
 696 }
 697
 698 static void tun_detach_all(struct net_device *dev)
 699 {
 700         struct tun_struct *tun = netdev_priv(dev);
 701         struct tun_file *tfile, *tmp;
 702         int i, n = tun->numqueues;
 703
 704         for (i = 0; i < n; i++) {
 705                 tfile = rtnl_dereference(tun->tfiles[i]);
 706                 BUG_ON(!tfile);
 707                 tun_napi_disable(tfile);
 708                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 709                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 710                 RCU_INIT_POINTER(tfile->tun, NULL);
 711                 --tun->numqueues;
 712         }
 713         list_for_each_entry(tfile, &tun->disabled, next) {
 714                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 715                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 716                 RCU_INIT_POINTER(tfile->tun, NULL);
 717         }
 718         BUG_ON(tun->numqueues != 0);
 719
 720         synchronize_net();
 721         for (i = 0; i < n; i++) {
 722                 tfile = rtnl_dereference(tun->tfiles[i]);
 723                 tun_napi_del(tfile);
 724                 /* Drop read queue */
 725                 tun_queue_purge(tfile);
 726                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 727                 sock_put(&tfile->sk);
 728         }
 729         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 730                 tun_enable_queue(tfile);
 731                 tun_queue_purge(tfile);
 732                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 733                 sock_put(&tfile->sk);
 734         }
 735         BUG_ON(tun->numdisabled != 0);
 736
 737         if (tun->flags & IFF_PERSIST)
 738                 module_put(THIS_MODULE);
 739 }
 740
 741 static int tun_attach(struct tun_struct *tun, struct file *file,
 742                       bool skip_filter, bool napi, bool napi_frags,
 743                       bool publish_tun)
 744 {
 745         struct tun_file *tfile = file->private_data;
 746         struct net_device *dev = tun->dev;
 747         int err;
 748
 749         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 750         if (err < 0)
 751                 goto out;
 752
 753         err = -EINVAL;
 754         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 755                 goto out;
 756
 757         err = -EBUSY;
 758         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 759                 goto out;
 760
 761         err = -E2BIG;
 762         if (!tfile->detached &&
 763             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 764                 goto out;
 765
 766         err = 0;
 767
 768         /* Re-attach the filter to persist device */
 769         if (!skip_filter && (tun->filter_attached == true)) {
 770                 lock_sock(tfile->socket.sk);
 771                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 772                 release_sock(tfile->socket.sk);
 773                 if (!err)
 774                         goto out;
 775         }
 776
 777         if (!tfile->detached &&
 778             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 779                             GFP_KERNEL, tun_ptr_free)) {
 780                 err = -ENOMEM;
 781                 goto out;
 782         }
 783
 784         tfile->queue_index = tun->numqueues;
 785         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 786
 787         if (tfile->detached) {
 788                 /* Re-attach detached tfile, updating XDP queue_index */
 789                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 790
 791                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 792                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 793         } else {
 794                 /* Setup XDP RX-queue info, for new tfile getting attached */
 795                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 796                                        tun->dev, tfile->queue_index, 0);
 797                 if (err < 0)
 798                         goto out;
 799                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 800                                                  MEM_TYPE_PAGE_SHARED, NULL);
 801                 if (err < 0) {
 802                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 803                         goto out;
 804                 }
 805                 err = 0;
 806         }
 807
 808         if (tfile->detached) {
 809                 tun_enable_queue(tfile);
 810         } else {
 811                 sock_hold(&tfile->sk);
 812                 tun_napi_init(tun, tfile, napi, napi_frags);
 813         }
 814
 815         if (rtnl_dereference(tun->xdp_prog))
 816                 sock_set_flag(&tfile->sk, SOCK_XDP);
 817
 818         /* device is allowed to go away first, so no need to hold extra
 819          * refcnt.
 820          */
 821
 822         /* Publish tfile->tun and tun->tfiles only after we've fully
 823          * initialized tfile; otherwise we risk using half-initialized
 824          * object.
 825          */
 826         if (publish_tun)
 827                 rcu_assign_pointer(tfile->tun, tun);
 828         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 829         tun->numqueues++;
 830         tun_set_real_num_queues(tun);
 831 out:
 832         return err;
 833 }
 834
 835 static struct tun_struct *tun_get(struct tun_file *tfile)
 836 {
 837         struct tun_struct *tun;
 838
 839         rcu_read_lock();
 840         tun = rcu_dereference(tfile->tun);
 841         if (tun)
 842                 dev_hold(tun->dev);
 843         rcu_read_unlock();
 844
 845         return tun;
 846 }
 847
 848 static void tun_put(struct tun_struct *tun)
 849 {
 850         dev_put(tun->dev);
 851 }
 852
 853 /* TAP filtering */
 854 static void addr_hash_set(u32 *mask, const u8 *addr)
 855 {
 856         int n = ether_crc(ETH_ALEN, addr) >> 26;
 857         mask[n >> 5] |= (1 << (n & 31));
 858 }
 859
 860 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 861 {
 862         int n = ether_crc(ETH_ALEN, addr) >> 26;
 863         return mask[n >> 5] & (1 << (n & 31));
 864 }
 865
 866 static int update_filter(struct tap_filter *filter, void __user *arg)
 867 {
 868         struct { u8 u[ETH_ALEN]; } *addr;
 869         struct tun_filter uf;
 870         int err, alen, n, nexact;
 871
 872         if (copy_from_user(&uf, arg, sizeof(uf)))
 873                 return -EFAULT;
 874
 875         if (!uf.count) {
 876                 /* Disabled */
 877                 filter->count = 0;
 878                 return 0;
 879         }
 880
 881         alen = ETH_ALEN * uf.count;
 882         addr = memdup_user(arg + sizeof(uf), alen);
 883         if (IS_ERR(addr))
 884                 return PTR_ERR(addr);
 885
 886         /* The filter is updated without holding any locks. Which is
 887          * perfectly safe. We disable it first and in the worst
 888          * case we'll accept a few undesired packets. */
 889         filter->count = 0;
 890         wmb();
 891
 892         /* Use first set of addresses as an exact filter */
 893         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 894                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 895
 896         nexact = n;
 897
 898         /* Remaining multicast addresses are hashed,
 899          * unicast will leave the filter disabled. */
 900         memset(filter->mask, 0, sizeof(filter->mask));
 901         for (; n < uf.count; n++) {
 902                 if (!is_multicast_ether_addr(addr[n].u)) {
 903                         err = 0; /* no filter */
 904                         goto free_addr;
 905                 }
 906                 addr_hash_set(filter->mask, addr[n].u);
 907         }
 908
 909         /* For ALLMULTI just set the mask to all ones.
 910          * This overrides the mask populated above. */
 911         if ((uf.flags & TUN_FLT_ALLMULTI))
 912                 memset(filter->mask, ~0, sizeof(filter->mask));
 913
 914         /* Now enable the filter */
 915         wmb();
 916         filter->count = nexact;
 917
 918         /* Return the number of exact filters */
 919         err = nexact;
 920 free_addr:
 921         kfree(addr);
 922         return err;
 923 }
 924
 925 /* Returns: 0 - drop, !=0 - accept */
 926 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 927 {
 928         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 929          * at this point. */
 930         struct ethhdr *eh = (struct ethhdr *) skb->data;
 931         int i;
 932
 933         /* Exact match */
 934         for (i = 0; i < filter->count; i++)
 935                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 936                         return 1;
 937
 938         /* Inexact match (multicast only) */
 939         if (is_multicast_ether_addr(eh->h_dest))
 940                 return addr_hash_test(filter->mask, eh->h_dest);
 941
 942         return 0;
 943 }
 944
 945 /*
 946  * Checks whether the packet is accepted or not.
 947  * Returns: 0 - drop, !=0 - accept
 948  */
 949 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 950 {
 951         if (!filter->count)
 952                 return 1;
 953
 954         return run_filter(filter, skb);
 955 }
 956
 957 /* Network device part of the driver */
 958
 959 static const struct ethtool_ops tun_ethtool_ops;
 960
 961 static int tun_net_init(struct net_device *dev)
 962 {
 963         struct tun_struct *tun = netdev_priv(dev);
 964         struct ifreq *ifr = tun->ifr;
 965         int err;
 966
 967         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 968         if (!dev->tstats)
 969                 return -ENOMEM;
 970
 971         spin_lock_init(&tun->lock);
 972
 973         err = security_tun_dev_alloc_security(&tun->security);
 974         if (err < 0) {
 975                 free_percpu(dev->tstats);
 976                 return err;
 977         }
 978
 979         tun_flow_init(tun);
 980
 981         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 982                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 983                            NETIF_F_HW_VLAN_STAG_TX;
 984         dev->features = dev->hw_features | NETIF_F_LLTX;
 985         dev->vlan_features = dev->features &
 986                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 987                                NETIF_F_HW_VLAN_STAG_TX);
 988
 989         tun->flags = (tun->flags & ~TUN_FEATURES) |
 990                       (ifr->ifr_flags & TUN_FEATURES);
 991
 992         INIT_LIST_HEAD(&tun->disabled);
 993         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
 994                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
 995         if (err < 0) {
 996                 tun_flow_uninit(tun);
 997                 security_tun_dev_free_security(tun->security);
 998                 free_percpu(dev->tstats);
 999                 return err;
1000         }
1001         return 0;
1002 }
1003
1004 /* Net device detach from fd. */
1005 static void tun_net_uninit(struct net_device *dev)
1006 {
1007         tun_detach_all(dev);
1008 }
1009
1010 /* Net device open. */
1011 static int tun_net_open(struct net_device *dev)
1012 {
1013         netif_tx_start_all_queues(dev);
1014
1015         return 0;
1016 }
1017
1018 /* Net device close. */
1019 static int tun_net_close(struct net_device *dev)
1020 {
1021         netif_tx_stop_all_queues(dev);
1022         return 0;
1023 }
1024
1025 /* Net device start xmit */
1026 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1027 {
1028 #ifdef CONFIG_RPS
1029         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1030                 /* Select queue was not called for the skbuff, so we extract the
1031                  * RPS hash and save it into the flow_table here.
1032                  */
1033                 struct tun_flow_entry *e;
1034                 __u32 rxhash;
1035
1036                 rxhash = __skb_get_hash_symmetric(skb);
1037                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1038                 if (e)
1039                         tun_flow_save_rps_rxhash(e, rxhash);
1040         }
1041 #endif
1042 }
1043
1044 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1045                                     struct sk_buff *skb,
1046                                     int len)
1047 {
1048         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1049
1050         if (prog)
1051                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1052
1053         return len;
1054 }
1055
1056 /* Net device start xmit */
1057 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1058 {
1059         struct tun_struct *tun = netdev_priv(dev);
1060         enum skb_drop_reason drop_reason;
1061         int txq = skb->queue_mapping;
1062         struct netdev_queue *queue;
1063         struct tun_file *tfile;
1064         int len = skb->len;
1065
1066         rcu_read_lock();
1067         tfile = rcu_dereference(tun->tfiles[txq]);
1068
1069         /* Drop packet if interface is not attached */
1070         if (!tfile) {
1071                 drop_reason = SKB_DROP_REASON_DEV_READY;
1072                 goto drop;
1073         }
1074
1075         if (!rcu_dereference(tun->steering_prog))
1076                 tun_automq_xmit(tun, skb);
1077
1078         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1079
1080         /* Drop if the filter does not like it.
1081          * This is a noop if the filter is disabled.
1082          * Filter can be enabled only for the TAP devices. */
1083         if (!check_filter(&tun->txflt, skb)) {
1084                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1085                 goto drop;
1086         }
1087
1088         if (tfile->socket.sk->sk_filter &&
1089             sk_filter(tfile->socket.sk, skb)) {
1090                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1091                 goto drop;
1092         }
1093
1094         len = run_ebpf_filter(tun, skb, len);
1095         if (len == 0) {
1096                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1097                 goto drop;
1098         }
1099
1100         if (pskb_trim(skb, len)) {
1101                 drop_reason = SKB_DROP_REASON_NOMEM;
1102                 goto drop;
1103         }
1104
1105         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1106                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1107                 goto drop;
1108         }
1109
1110         skb_tx_timestamp(skb);
1111
1112         /* Orphan the skb - required as we might hang on to it
1113          * for indefinite time.
1114          */
1115         skb_orphan(skb);
1116
1117         nf_reset_ct(skb);
1118
1119         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1120                 drop_reason = SKB_DROP_REASON_FULL_RING;
1121                 goto drop;
1122         }
1123
1124         /* NETIF_F_LLTX requires to do our own update of trans_start */
1125         queue = netdev_get_tx_queue(dev, txq);
1126         txq_trans_cond_update(queue);
1127
1128         /* Notify and wake up reader process */
1129         if (tfile->flags & TUN_FASYNC)
1130                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1131         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1132
1133         rcu_read_unlock();
1134         return NETDEV_TX_OK;
1135
1136 drop:
1137         dev_core_stats_tx_dropped_inc(dev);
1138         skb_tx_error(skb);
1139         kfree_skb_reason(skb, drop_reason);
1140         rcu_read_unlock();
1141         return NET_XMIT_DROP;
1142 }
1143
1144 static void tun_net_mclist(struct net_device *dev)
1145 {
1146         /*
1147          * This callback is supposed to deal with mc filter in
1148          * _rx_ path and has nothing to do with the _tx_ path.
1149          * In rx path we always accept everything userspace gives us.
1150          */
1151 }
1152
1153 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1154         netdev_features_t features)
1155 {
1156         struct tun_struct *tun = netdev_priv(dev);
1157
1158         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1159 }
1160
1161 static void tun_set_headroom(struct net_device *dev, int new_hr)
1162 {
1163         struct tun_struct *tun = netdev_priv(dev);
1164
1165         if (new_hr < NET_SKB_PAD)
1166                 new_hr = NET_SKB_PAD;
1167
1168         tun->align = new_hr;
1169 }
1170
1171 static void
1172 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1173 {
1174         struct tun_struct *tun = netdev_priv(dev);
1175
1176         dev_get_tstats64(dev, stats);
1177
1178         stats->rx_frame_errors +=
1179                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1180 }
1181
1182 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1183                        struct netlink_ext_ack *extack)
1184 {
1185         struct tun_struct *tun = netdev_priv(dev);
1186         struct tun_file *tfile;
1187         struct bpf_prog *old_prog;
1188         int i;
1189
1190         old_prog = rtnl_dereference(tun->xdp_prog);
1191         rcu_assign_pointer(tun->xdp_prog, prog);
1192         if (old_prog)
1193                 bpf_prog_put(old_prog);
1194
1195         for (i = 0; i < tun->numqueues; i++) {
1196                 tfile = rtnl_dereference(tun->tfiles[i]);
1197                 if (prog)
1198                         sock_set_flag(&tfile->sk, SOCK_XDP);
1199                 else
1200                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1201         }
1202         list_for_each_entry(tfile, &tun->disabled, next) {
1203                 if (prog)
1204                         sock_set_flag(&tfile->sk, SOCK_XDP);
1205                 else
1206                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1207         }
1208
1209         return 0;
1210 }
1211
1212 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1213 {
1214         switch (xdp->command) {
1215         case XDP_SETUP_PROG:
1216                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1217         default:
1218                 return -EINVAL;
1219         }
1220 }
1221
1222 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1223 {
1224         if (new_carrier) {
1225                 struct tun_struct *tun = netdev_priv(dev);
1226
1227                 if (!tun->numqueues)
1228                         return -EPERM;
1229
1230                 netif_carrier_on(dev);
1231         } else {
1232                 netif_carrier_off(dev);
1233         }
1234         return 0;
1235 }
1236
1237 static const struct net_device_ops tun_netdev_ops = {
1238         .ndo_init               = tun_net_init,
1239         .ndo_uninit             = tun_net_uninit,
1240         .ndo_open               = tun_net_open,
1241         .ndo_stop               = tun_net_close,
1242         .ndo_start_xmit         = tun_net_xmit,
1243         .ndo_fix_features       = tun_net_fix_features,
1244         .ndo_select_queue       = tun_select_queue,
1245         .ndo_set_rx_headroom    = tun_set_headroom,
1246         .ndo_get_stats64        = tun_net_get_stats64,
1247         .ndo_change_carrier     = tun_net_change_carrier,
1248 };
1249
1250 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1251 {
1252         /* Notify and wake up reader process */
1253         if (tfile->flags & TUN_FASYNC)
1254                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1255         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1256 }
1257
1258 static int tun_xdp_xmit(struct net_device *dev, int n,
1259                         struct xdp_frame **frames, u32 flags)
1260 {
1261         struct tun_struct *tun = netdev_priv(dev);
1262         struct tun_file *tfile;
1263         u32 numqueues;
1264         int nxmit = 0;
1265         int i;
1266
1267         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1268                 return -EINVAL;
1269
1270         rcu_read_lock();
1271
1272 resample:
1273         numqueues = READ_ONCE(tun->numqueues);
1274         if (!numqueues) {
1275                 rcu_read_unlock();
1276                 return -ENXIO; /* Caller will free/return all frames */
1277         }
1278
1279         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1280                                             numqueues]);
1281         if (unlikely(!tfile))
1282                 goto resample;
1283
1284         spin_lock(&tfile->tx_ring.producer_lock);
1285         for (i = 0; i < n; i++) {
1286                 struct xdp_frame *xdp = frames[i];
1287                 /* Encode the XDP flag into lowest bit for consumer to differ
1288                  * XDP buffer from sk_buff.
1289                  */
1290                 void *frame = tun_xdp_to_ptr(xdp);
1291
1292                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1293                         dev_core_stats_tx_dropped_inc(dev);
1294                         break;
1295                 }
1296                 nxmit++;
1297         }
1298         spin_unlock(&tfile->tx_ring.producer_lock);
1299
1300         if (flags & XDP_XMIT_FLUSH)
1301                 __tun_xdp_flush_tfile(tfile);
1302
1303         rcu_read_unlock();
1304         return nxmit;
1305 }
1306
1307 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1308 {
1309         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1310         int nxmit;
1311
1312         if (unlikely(!frame))
1313                 return -EOVERFLOW;
1314
1315         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1316         if (!nxmit)
1317                 xdp_return_frame_rx_napi(frame);
1318         return nxmit;
1319 }
1320
1321 static const struct net_device_ops tap_netdev_ops = {
1322         .ndo_init               = tun_net_init,
1323         .ndo_uninit             = tun_net_uninit,
1324         .ndo_open               = tun_net_open,
1325         .ndo_stop               = tun_net_close,
1326         .ndo_start_xmit         = tun_net_xmit,
1327         .ndo_fix_features       = tun_net_fix_features,
1328         .ndo_set_rx_mode        = tun_net_mclist,
1329         .ndo_set_mac_address    = eth_mac_addr,
1330         .ndo_validate_addr      = eth_validate_addr,
1331         .ndo_select_queue       = tun_select_queue,
1332         .ndo_features_check     = passthru_features_check,
1333         .ndo_set_rx_headroom    = tun_set_headroom,
1334         .ndo_get_stats64        = dev_get_tstats64,
1335         .ndo_bpf                = tun_xdp,
1336         .ndo_xdp_xmit           = tun_xdp_xmit,
1337         .ndo_change_carrier     = tun_net_change_carrier,
1338 };
1339
1340 static void tun_flow_init(struct tun_struct *tun)
1341 {
1342         int i;
1343
1344         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1345                 INIT_HLIST_HEAD(&tun->flows[i]);
1346
1347         tun->ageing_time = TUN_FLOW_EXPIRE;
1348         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1349         mod_timer(&tun->flow_gc_timer,
1350                   round_jiffies_up(jiffies + tun->ageing_time));
1351 }
1352
1353 static void tun_flow_uninit(struct tun_struct *tun)
1354 {
1355         del_timer_sync(&tun->flow_gc_timer);
1356         tun_flow_flush(tun);
1357 }
1358
1359 #define MIN_MTU 68
1360 #define MAX_MTU 65535
1361
1362 /* Initialize net device. */
1363 static void tun_net_initialize(struct net_device *dev)
1364 {
1365         struct tun_struct *tun = netdev_priv(dev);
1366
1367         switch (tun->flags & TUN_TYPE_MASK) {
1368         case IFF_TUN:
1369                 dev->netdev_ops = &tun_netdev_ops;
1370                 dev->header_ops = &ip_tunnel_header_ops;
1371
1372                 /* Point-to-Point TUN Device */
1373                 dev->hard_header_len = 0;
1374                 dev->addr_len = 0;
1375                 dev->mtu = 1500;
1376
1377                 /* Zero header length */
1378                 dev->type = ARPHRD_NONE;
1379                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1380                 break;
1381
1382         case IFF_TAP:
1383                 dev->netdev_ops = &tap_netdev_ops;
1384                 /* Ethernet TAP Device */
1385                 ether_setup(dev);
1386                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1387                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1388
1389                 eth_hw_addr_random(dev);
1390
1391                 break;
1392         }
1393
1394         dev->min_mtu = MIN_MTU;
1395         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1396 }
1397
1398 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1399 {
1400         struct sock *sk = tfile->socket.sk;
1401
1402         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1403 }
1404
1405 /* Character device part */
1406
1407 /* Poll */
1408 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1409 {
1410         struct tun_file *tfile = file->private_data;
1411         struct tun_struct *tun = tun_get(tfile);
1412         struct sock *sk;
1413         __poll_t mask = 0;
1414
1415         if (!tun)
1416                 return EPOLLERR;
1417
1418         sk = tfile->socket.sk;
1419
1420         poll_wait(file, sk_sleep(sk), wait);
1421
1422         if (!ptr_ring_empty(&tfile->tx_ring))
1423                 mask |= EPOLLIN | EPOLLRDNORM;
1424
1425         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1426          * guarantee EPOLLOUT to be raised by either here or
1427          * tun_sock_write_space(). Then process could get notification
1428          * after it writes to a down device and meets -EIO.
1429          */
1430         if (tun_sock_writeable(tun, tfile) ||
1431             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1432              tun_sock_writeable(tun, tfile)))
1433                 mask |= EPOLLOUT | EPOLLWRNORM;
1434
1435         if (tun->dev->reg_state != NETREG_REGISTERED)
1436                 mask = EPOLLERR;
1437
1438         tun_put(tun);
1439         return mask;
1440 }
1441
1442 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1443                                             size_t len,
1444                                             const struct iov_iter *it)
1445 {
1446         struct sk_buff *skb;
1447         size_t linear;
1448         int err;
1449         int i;
1450
1451         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1452                 return ERR_PTR(-EMSGSIZE);
1453
1454         local_bh_disable();
1455         skb = napi_get_frags(&tfile->napi);
1456         local_bh_enable();
1457         if (!skb)
1458                 return ERR_PTR(-ENOMEM);
1459
1460         linear = iov_iter_single_seg_count(it);
1461         err = __skb_grow(skb, linear);
1462         if (err)
1463                 goto free;
1464
1465         skb->len = len;
1466         skb->data_len = len - linear;
1467         skb->truesize += skb->data_len;
1468
1469         for (i = 1; i < it->nr_segs; i++) {
1470                 size_t fragsz = it->iov[i].iov_len;
1471                 struct page *page;
1472                 void *frag;
1473
1474                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1475                         err = -EINVAL;
1476                         goto free;
1477                 }
1478                 frag = netdev_alloc_frag(fragsz);
1479                 if (!frag) {
1480                         err = -ENOMEM;
1481                         goto free;
1482                 }
1483                 page = virt_to_head_page(frag);
1484                 skb_fill_page_desc(skb, i - 1, page,
1485                                    frag - page_address(page), fragsz);
1486         }
1487
1488         return skb;
1489 free:
1490         /* frees skb and all frags allocated with napi_alloc_frag() */
1491         napi_free_frags(&tfile->napi);
1492         return ERR_PTR(err);
1493 }
1494
1495 /* prepad is the amount to reserve at front.  len is length after that.
1496  * linear is a hint as to how much to copy (usually headers). */
1497 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1498                                      size_t prepad, size_t len,
1499                                      size_t linear, int noblock)
1500 {
1501         struct sock *sk = tfile->socket.sk;
1502         struct sk_buff *skb;
1503         int err;
1504
1505         /* Under a page?  Don't bother with paged skb. */
1506         if (prepad + len < PAGE_SIZE || !linear)
1507                 linear = len;
1508
1509         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1510                                    &err, 0);
1511         if (!skb)
1512                 return ERR_PTR(err);
1513
1514         skb_reserve(skb, prepad);
1515         skb_put(skb, linear);
1516         skb->data_len = len - linear;
1517         skb->len += len - linear;
1518
1519         return skb;
1520 }
1521
1522 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1523                            struct sk_buff *skb, int more)
1524 {
1525         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1526         struct sk_buff_head process_queue;
1527         u32 rx_batched = tun->rx_batched;
1528         bool rcv = false;
1529
1530         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1531                 local_bh_disable();
1532                 skb_record_rx_queue(skb, tfile->queue_index);
1533                 netif_receive_skb(skb);
1534                 local_bh_enable();
1535                 return;
1536         }
1537
1538         spin_lock(&queue->lock);
1539         if (!more || skb_queue_len(queue) == rx_batched) {
1540                 __skb_queue_head_init(&process_queue);
1541                 skb_queue_splice_tail_init(queue, &process_queue);
1542                 rcv = true;
1543         } else {
1544                 __skb_queue_tail(queue, skb);
1545         }
1546         spin_unlock(&queue->lock);
1547
1548         if (rcv) {
1549                 struct sk_buff *nskb;
1550
1551                 local_bh_disable();
1552                 while ((nskb = __skb_dequeue(&process_queue))) {
1553                         skb_record_rx_queue(nskb, tfile->queue_index);
1554                         netif_receive_skb(nskb);
1555                 }
1556                 skb_record_rx_queue(skb, tfile->queue_index);
1557                 netif_receive_skb(skb);
1558                 local_bh_enable();
1559         }
1560 }
1561
1562 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1563                               int len, int noblock, bool zerocopy)
1564 {
1565         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1566                 return false;
1567
1568         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1569                 return false;
1570
1571         if (!noblock)
1572                 return false;
1573
1574         if (zerocopy)
1575                 return false;
1576
1577         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1578             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1579                 return false;
1580
1581         return true;
1582 }
1583
1584 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1585                                        struct page_frag *alloc_frag, char *buf,
1586                                        int buflen, int len, int pad)
1587 {
1588         struct sk_buff *skb = build_skb(buf, buflen);
1589
1590         if (!skb)
1591                 return ERR_PTR(-ENOMEM);
1592
1593         skb_reserve(skb, pad);
1594         skb_put(skb, len);
1595         skb_set_owner_w(skb, tfile->socket.sk);
1596
1597         get_page(alloc_frag->page);
1598         alloc_frag->offset += buflen;
1599
1600         return skb;
1601 }
1602
1603 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1604                        struct xdp_buff *xdp, u32 act)
1605 {
1606         int err;
1607
1608         switch (act) {
1609         case XDP_REDIRECT:
1610                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1611                 if (err)
1612                         return err;
1613                 break;
1614         case XDP_TX:
1615                 err = tun_xdp_tx(tun->dev, xdp);
1616                 if (err < 0)
1617                         return err;
1618                 break;
1619         case XDP_PASS:
1620                 break;
1621         default:
1622                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1623                 fallthrough;
1624         case XDP_ABORTED:
1625                 trace_xdp_exception(tun->dev, xdp_prog, act);
1626                 fallthrough;
1627         case XDP_DROP:
1628                 dev_core_stats_rx_dropped_inc(tun->dev);
1629                 break;
1630         }
1631
1632         return act;
1633 }
1634
1635 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1636                                      struct tun_file *tfile,
1637                                      struct iov_iter *from,
1638                                      struct virtio_net_hdr *hdr,
1639                                      int len, int *skb_xdp)
1640 {
1641         struct page_frag *alloc_frag = &current->task_frag;
1642         struct bpf_prog *xdp_prog;
1643         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1644         char *buf;
1645         size_t copied;
1646         int pad = TUN_RX_PAD;
1647         int err = 0;
1648
1649         rcu_read_lock();
1650         xdp_prog = rcu_dereference(tun->xdp_prog);
1651         if (xdp_prog)
1652                 pad += XDP_PACKET_HEADROOM;
1653         buflen += SKB_DATA_ALIGN(len + pad);
1654         rcu_read_unlock();
1655
1656         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1657         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1658                 return ERR_PTR(-ENOMEM);
1659
1660         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1661         copied = copy_page_from_iter(alloc_frag->page,
1662                                      alloc_frag->offset + pad,
1663                                      len, from);
1664         if (copied != len)
1665                 return ERR_PTR(-EFAULT);
1666
1667         /* There's a small window that XDP may be set after the check
1668          * of xdp_prog above, this should be rare and for simplicity
1669          * we do XDP on skb in case the headroom is not enough.
1670          */
1671         if (hdr->gso_type || !xdp_prog) {
1672                 *skb_xdp = 1;
1673                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1674                                        pad);
1675         }
1676
1677         *skb_xdp = 0;
1678
1679         local_bh_disable();
1680         rcu_read_lock();
1681         xdp_prog = rcu_dereference(tun->xdp_prog);
1682         if (xdp_prog) {
1683                 struct xdp_buff xdp;
1684                 u32 act;
1685
1686                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1687                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1688
1689                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1690                 if (act == XDP_REDIRECT || act == XDP_TX) {
1691                         get_page(alloc_frag->page);
1692                         alloc_frag->offset += buflen;
1693                 }
1694                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1695                 if (err < 0) {
1696                         if (act == XDP_REDIRECT || act == XDP_TX)
1697                                 put_page(alloc_frag->page);
1698                         goto out;
1699                 }
1700
1701                 if (err == XDP_REDIRECT)
1702                         xdp_do_flush();
1703                 if (err != XDP_PASS)
1704                         goto out;
1705
1706                 pad = xdp.data - xdp.data_hard_start;
1707                 len = xdp.data_end - xdp.data;
1708         }
1709         rcu_read_unlock();
1710         local_bh_enable();
1711
1712         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1713
1714 out:
1715         rcu_read_unlock();
1716         local_bh_enable();
1717         return NULL;
1718 }
1719
1720 /* Get packet from user space buffer */
1721 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1722                             void *msg_control, struct iov_iter *from,
1723                             int noblock, bool more)
1724 {
1725         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1726         struct sk_buff *skb;
1727         size_t total_len = iov_iter_count(from);
1728         size_t len = total_len, align = tun->align, linear;
1729         struct virtio_net_hdr gso = { 0 };
1730         int good_linear;
1731         int copylen;
1732         bool zerocopy = false;
1733         int err;
1734         u32 rxhash = 0;
1735         int skb_xdp = 1;
1736         bool frags = tun_napi_frags_enabled(tfile);
1737         enum skb_drop_reason drop_reason;
1738
1739         if (!(tun->flags & IFF_NO_PI)) {
1740                 if (len < sizeof(pi))
1741                         return -EINVAL;
1742                 len -= sizeof(pi);
1743
1744                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1745                         return -EFAULT;
1746         }
1747
1748         if (tun->flags & IFF_VNET_HDR) {
1749                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1750
1751                 if (len < vnet_hdr_sz)
1752                         return -EINVAL;
1753                 len -= vnet_hdr_sz;
1754
1755                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1756                         return -EFAULT;
1757
1758                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1759                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1760                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1761
1762                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1763                         return -EINVAL;
1764                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1765         }
1766
1767         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1768                 align += NET_IP_ALIGN;
1769                 if (unlikely(len < ETH_HLEN ||
1770                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1771                         return -EINVAL;
1772         }
1773
1774         good_linear = SKB_MAX_HEAD(align);
1775
1776         if (msg_control) {
1777                 struct iov_iter i = *from;
1778
1779                 /* There are 256 bytes to be copied in skb, so there is
1780                  * enough room for skb expand head in case it is used.
1781                  * The rest of the buffer is mapped from userspace.
1782                  */
1783                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1784                 if (copylen > good_linear)
1785                         copylen = good_linear;
1786                 linear = copylen;
1787                 iov_iter_advance(&i, copylen);
1788                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1789                         zerocopy = true;
1790         }
1791
1792         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1793                 /* For the packet that is not easy to be processed
1794                  * (e.g gso or jumbo packet), we will do it at after
1795                  * skb was created with generic XDP routine.
1796                  */
1797                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1798                 if (IS_ERR(skb)) {
1799                         dev_core_stats_rx_dropped_inc(tun->dev);
1800                         return PTR_ERR(skb);
1801                 }
1802                 if (!skb)
1803                         return total_len;
1804         } else {
1805                 if (!zerocopy) {
1806                         copylen = len;
1807                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1808                                 linear = good_linear;
1809                         else
1810                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1811                 }
1812
1813                 if (frags) {
1814                         mutex_lock(&tfile->napi_mutex);
1815                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1816                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1817                          * If zerocopy is enabled, then this layout will be
1818                          * overwritten by zerocopy_sg_from_iter().
1819                          */
1820                         zerocopy = false;
1821                 } else {
1822                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1823                                             noblock);
1824                 }
1825
1826                 if (IS_ERR(skb)) {
1827                         if (PTR_ERR(skb) != -EAGAIN)
1828                                 dev_core_stats_rx_dropped_inc(tun->dev);
1829                         if (frags)
1830                                 mutex_unlock(&tfile->napi_mutex);
1831                         return PTR_ERR(skb);
1832                 }
1833
1834                 if (zerocopy)
1835                         err = zerocopy_sg_from_iter(skb, from);
1836                 else
1837                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1838
1839                 if (err) {
1840                         err = -EFAULT;
1841                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1842 drop:
1843                         dev_core_stats_rx_dropped_inc(tun->dev);
1844                         kfree_skb_reason(skb, drop_reason);
1845                         if (frags) {
1846                                 tfile->napi.skb = NULL;
1847                                 mutex_unlock(&tfile->napi_mutex);
1848                         }
1849
1850                         return err;
1851                 }
1852         }
1853
1854         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1855                 atomic_long_inc(&tun->rx_frame_errors);
1856                 kfree_skb(skb);
1857                 if (frags) {
1858                         tfile->napi.skb = NULL;
1859                         mutex_unlock(&tfile->napi_mutex);
1860                 }
1861
1862                 return -EINVAL;
1863         }
1864
1865         switch (tun->flags & TUN_TYPE_MASK) {
1866         case IFF_TUN:
1867                 if (tun->flags & IFF_NO_PI) {
1868                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1869
1870                         switch (ip_version) {
1871                         case 4:
1872                                 pi.proto = htons(ETH_P_IP);
1873                                 break;
1874                         case 6:
1875                                 pi.proto = htons(ETH_P_IPV6);
1876                                 break;
1877                         default:
1878                                 dev_core_stats_rx_dropped_inc(tun->dev);
1879                                 kfree_skb(skb);
1880                                 return -EINVAL;
1881                         }
1882                 }
1883
1884                 skb_reset_mac_header(skb);
1885                 skb->protocol = pi.proto;
1886                 skb->dev = tun->dev;
1887                 break;
1888         case IFF_TAP:
1889                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1890                         err = -ENOMEM;
1891                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1892                         goto drop;
1893                 }
1894                 skb->protocol = eth_type_trans(skb, tun->dev);
1895                 break;
1896         }
1897
1898         /* copy skb_ubuf_info for callback when skb has no error */
1899         if (zerocopy) {
1900                 skb_zcopy_init(skb, msg_control);
1901         } else if (msg_control) {
1902                 struct ubuf_info *uarg = msg_control;
1903                 uarg->callback(NULL, uarg, false);
1904         }
1905
1906         skb_reset_network_header(skb);
1907         skb_probe_transport_header(skb);
1908         skb_record_rx_queue(skb, tfile->queue_index);
1909
1910         if (skb_xdp) {
1911                 struct bpf_prog *xdp_prog;
1912                 int ret;
1913
1914                 local_bh_disable();
1915                 rcu_read_lock();
1916                 xdp_prog = rcu_dereference(tun->xdp_prog);
1917                 if (xdp_prog) {
1918                         ret = do_xdp_generic(xdp_prog, skb);
1919                         if (ret != XDP_PASS) {
1920                                 rcu_read_unlock();
1921                                 local_bh_enable();
1922                                 if (frags) {
1923                                         tfile->napi.skb = NULL;
1924                                         mutex_unlock(&tfile->napi_mutex);
1925                                 }
1926                                 return total_len;
1927                         }
1928                 }
1929                 rcu_read_unlock();
1930                 local_bh_enable();
1931         }
1932
1933         /* Compute the costly rx hash only if needed for flow updates.
1934          * We may get a very small possibility of OOO during switching, not
1935          * worth to optimize.
1936          */
1937         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1938             !tfile->detached)
1939                 rxhash = __skb_get_hash_symmetric(skb);
1940
1941         rcu_read_lock();
1942         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1943                 err = -EIO;
1944                 rcu_read_unlock();
1945                 drop_reason = SKB_DROP_REASON_DEV_READY;
1946                 goto drop;
1947         }
1948
1949         if (frags) {
1950                 u32 headlen;
1951
1952                 /* Exercise flow dissector code path. */
1953                 skb_push(skb, ETH_HLEN);
1954                 headlen = eth_get_headlen(tun->dev, skb->data,
1955                                           skb_headlen(skb));
1956
1957                 if (unlikely(headlen > skb_headlen(skb))) {
1958                         dev_core_stats_rx_dropped_inc(tun->dev);
1959                         napi_free_frags(&tfile->napi);
1960                         rcu_read_unlock();
1961                         mutex_unlock(&tfile->napi_mutex);
1962                         WARN_ON(1);
1963                         return -ENOMEM;
1964                 }
1965
1966                 local_bh_disable();
1967                 napi_gro_frags(&tfile->napi);
1968                 local_bh_enable();
1969                 mutex_unlock(&tfile->napi_mutex);
1970         } else if (tfile->napi_enabled) {
1971                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1972                 int queue_len;
1973
1974                 spin_lock_bh(&queue->lock);
1975                 __skb_queue_tail(queue, skb);
1976                 queue_len = skb_queue_len(queue);
1977                 spin_unlock(&queue->lock);
1978
1979                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1980                         napi_schedule(&tfile->napi);
1981
1982                 local_bh_enable();
1983         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1984                 tun_rx_batched(tun, tfile, skb, more);
1985         } else {
1986                 netif_rx(skb);
1987         }
1988         rcu_read_unlock();
1989
1990         preempt_disable();
1991         dev_sw_netstats_rx_add(tun->dev, len);
1992         preempt_enable();
1993
1994         if (rxhash)
1995                 tun_flow_update(tun, rxhash, tfile);
1996
1997         return total_len;
1998 }
1999
2000 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2001 {
2002         struct file *file = iocb->ki_filp;
2003         struct tun_file *tfile = file->private_data;
2004         struct tun_struct *tun = tun_get(tfile);
2005         ssize_t result;
2006         int noblock = 0;
2007
2008         if (!tun)
2009                 return -EBADFD;
2010
2011         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2012                 noblock = 1;
2013
2014         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2015
2016         tun_put(tun);
2017         return result;
2018 }
2019
2020 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2021                                 struct tun_file *tfile,
2022                                 struct xdp_frame *xdp_frame,
2023                                 struct iov_iter *iter)
2024 {
2025         int vnet_hdr_sz = 0;
2026         size_t size = xdp_frame->len;
2027         size_t ret;
2028
2029         if (tun->flags & IFF_VNET_HDR) {
2030                 struct virtio_net_hdr gso = { 0 };
2031
2032                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2033                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2034                         return -EINVAL;
2035                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2036                              sizeof(gso)))
2037                         return -EFAULT;
2038                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2039         }
2040
2041         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2042
2043         preempt_disable();
2044         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2045         preempt_enable();
2046
2047         return ret;
2048 }
2049
2050 /* Put packet to the user space buffer */
2051 static ssize_t tun_put_user(struct tun_struct *tun,
2052                             struct tun_file *tfile,
2053                             struct sk_buff *skb,
2054                             struct iov_iter *iter)
2055 {
2056         struct tun_pi pi = { 0, skb->protocol };
2057         ssize_t total;
2058         int vlan_offset = 0;
2059         int vlan_hlen = 0;
2060         int vnet_hdr_sz = 0;
2061
2062         if (skb_vlan_tag_present(skb))
2063                 vlan_hlen = VLAN_HLEN;
2064
2065         if (tun->flags & IFF_VNET_HDR)
2066                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2067
2068         total = skb->len + vlan_hlen + vnet_hdr_sz;
2069
2070         if (!(tun->flags & IFF_NO_PI)) {
2071                 if (iov_iter_count(iter) < sizeof(pi))
2072                         return -EINVAL;
2073
2074                 total += sizeof(pi);
2075                 if (iov_iter_count(iter) < total) {
2076                         /* Packet will be striped */
2077                         pi.flags |= TUN_PKT_STRIP;
2078                 }
2079
2080                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2081                         return -EFAULT;
2082         }
2083
2084         if (vnet_hdr_sz) {
2085                 struct virtio_net_hdr gso;
2086
2087                 if (iov_iter_count(iter) < vnet_hdr_sz)
2088                         return -EINVAL;
2089
2090                 if (virtio_net_hdr_from_skb(skb, &gso,
2091                                             tun_is_little_endian(tun), true,
2092                                             vlan_hlen)) {
2093                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2094                         pr_err("unexpected GSO type: "
2095                                "0x%x, gso_size %d, hdr_len %d\n",
2096                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2097                                tun16_to_cpu(tun, gso.hdr_len));
2098                         print_hex_dump(KERN_ERR, "tun: ",
2099                                        DUMP_PREFIX_NONE,
2100                                        16, 1, skb->head,
2101                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2102                         WARN_ON_ONCE(1);
2103                         return -EINVAL;
2104                 }
2105
2106                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2107                         return -EFAULT;
2108
2109                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2110         }
2111
2112         if (vlan_hlen) {
2113                 int ret;
2114                 struct veth veth;
2115
2116                 veth.h_vlan_proto = skb->vlan_proto;
2117                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2118
2119                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2120
2121                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2122                 if (ret || !iov_iter_count(iter))
2123                         goto done;
2124
2125                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2126                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2127                         goto done;
2128         }
2129
2130         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2131
2132 done:
2133         /* caller is in process context, */
2134         preempt_disable();
2135         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2136         preempt_enable();
2137
2138         return total;
2139 }
2140
2141 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2142 {
2143         DECLARE_WAITQUEUE(wait, current);
2144         void *ptr = NULL;
2145         int error = 0;
2146
2147         ptr = ptr_ring_consume(&tfile->tx_ring);
2148         if (ptr)
2149                 goto out;
2150         if (noblock) {
2151                 error = -EAGAIN;
2152                 goto out;
2153         }
2154
2155         add_wait_queue(&tfile->socket.wq.wait, &wait);
2156
2157         while (1) {
2158                 set_current_state(TASK_INTERRUPTIBLE);
2159                 ptr = ptr_ring_consume(&tfile->tx_ring);
2160                 if (ptr)
2161                         break;
2162                 if (signal_pending(current)) {
2163                         error = -ERESTARTSYS;
2164                         break;
2165                 }
2166                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2167                         error = -EFAULT;
2168                         break;
2169                 }
2170
2171                 schedule();
2172         }
2173
2174         __set_current_state(TASK_RUNNING);
2175         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2176
2177 out:
2178         *err = error;
2179         return ptr;
2180 }
2181
2182 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2183                            struct iov_iter *to,
2184                            int noblock, void *ptr)
2185 {
2186         ssize_t ret;
2187         int err;
2188
2189         if (!iov_iter_count(to)) {
2190                 tun_ptr_free(ptr);
2191                 return 0;
2192         }
2193
2194         if (!ptr) {
2195                 /* Read frames from ring */
2196                 ptr = tun_ring_recv(tfile, noblock, &err);
2197                 if (!ptr)
2198                         return err;
2199         }
2200
2201         if (tun_is_xdp_frame(ptr)) {
2202                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2203
2204                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2205                 xdp_return_frame(xdpf);
2206         } else {
2207                 struct sk_buff *skb = ptr;
2208
2209                 ret = tun_put_user(tun, tfile, skb, to);
2210                 if (unlikely(ret < 0))
2211                         kfree_skb(skb);
2212                 else
2213                         consume_skb(skb);
2214         }
2215
2216         return ret;
2217 }
2218
2219 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2220 {
2221         struct file *file = iocb->ki_filp;
2222         struct tun_file *tfile = file->private_data;
2223         struct tun_struct *tun = tun_get(tfile);
2224         ssize_t len = iov_iter_count(to), ret;
2225         int noblock = 0;
2226
2227         if (!tun)
2228                 return -EBADFD;
2229
2230         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2231                 noblock = 1;
2232
2233         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2234         ret = min_t(ssize_t, ret, len);
2235         if (ret > 0)
2236                 iocb->ki_pos = ret;
2237         tun_put(tun);
2238         return ret;
2239 }
2240
2241 static void tun_prog_free(struct rcu_head *rcu)
2242 {
2243         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2244
2245         bpf_prog_destroy(prog->prog);
2246         kfree(prog);
2247 }
2248
2249 static int __tun_set_ebpf(struct tun_struct *tun,
2250                           struct tun_prog __rcu **prog_p,
2251                           struct bpf_prog *prog)
2252 {
2253         struct tun_prog *old, *new = NULL;
2254
2255         if (prog) {
2256                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2257                 if (!new)
2258                         return -ENOMEM;
2259                 new->prog = prog;
2260         }
2261
2262         spin_lock_bh(&tun->lock);
2263         old = rcu_dereference_protected(*prog_p,
2264                                         lockdep_is_held(&tun->lock));
2265         rcu_assign_pointer(*prog_p, new);
2266         spin_unlock_bh(&tun->lock);
2267
2268         if (old)
2269                 call_rcu(&old->rcu, tun_prog_free);
2270
2271         return 0;
2272 }
2273
2274 static void tun_free_netdev(struct net_device *dev)
2275 {
2276         struct tun_struct *tun = netdev_priv(dev);
2277
2278         BUG_ON(!(list_empty(&tun->disabled)));
2279
2280         free_percpu(dev->tstats);
2281         tun_flow_uninit(tun);
2282         security_tun_dev_free_security(tun->security);
2283         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2284         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2285 }
2286
2287 static void tun_setup(struct net_device *dev)
2288 {
2289         struct tun_struct *tun = netdev_priv(dev);
2290
2291         tun->owner = INVALID_UID;
2292         tun->group = INVALID_GID;
2293         tun_default_link_ksettings(dev, &tun->link_ksettings);
2294
2295         dev->ethtool_ops = &tun_ethtool_ops;
2296         dev->needs_free_netdev = true;
2297         dev->priv_destructor = tun_free_netdev;
2298         /* We prefer our own queue length */
2299         dev->tx_queue_len = TUN_READQ_SIZE;
2300 }
2301
2302 /* Trivial set of netlink ops to allow deleting tun or tap
2303  * device with netlink.
2304  */
2305 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2306                         struct netlink_ext_ack *extack)
2307 {
2308         NL_SET_ERR_MSG(extack,
2309                        "tun/tap creation via rtnetlink is not supported.");
2310         return -EOPNOTSUPP;
2311 }
2312
2313 static size_t tun_get_size(const struct net_device *dev)
2314 {
2315         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2316         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2317
2318         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2319                nla_total_size(sizeof(gid_t)) + /* GROUP */
2320                nla_total_size(sizeof(u8)) + /* TYPE */
2321                nla_total_size(sizeof(u8)) + /* PI */
2322                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2323                nla_total_size(sizeof(u8)) + /* PERSIST */
2324                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2325                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2326                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2327                0;
2328 }
2329
2330 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2331 {
2332         struct tun_struct *tun = netdev_priv(dev);
2333
2334         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2335                 goto nla_put_failure;
2336         if (uid_valid(tun->owner) &&
2337             nla_put_u32(skb, IFLA_TUN_OWNER,
2338                         from_kuid_munged(current_user_ns(), tun->owner)))
2339                 goto nla_put_failure;
2340         if (gid_valid(tun->group) &&
2341             nla_put_u32(skb, IFLA_TUN_GROUP,
2342                         from_kgid_munged(current_user_ns(), tun->group)))
2343                 goto nla_put_failure;
2344         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2345                 goto nla_put_failure;
2346         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2347                 goto nla_put_failure;
2348         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2349                 goto nla_put_failure;
2350         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2351                        !!(tun->flags & IFF_MULTI_QUEUE)))
2352                 goto nla_put_failure;
2353         if (tun->flags & IFF_MULTI_QUEUE) {
2354                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2355                         goto nla_put_failure;
2356                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2357                                 tun->numdisabled))
2358                         goto nla_put_failure;
2359         }
2360
2361         return 0;
2362
2363 nla_put_failure:
2364         return -EMSGSIZE;
2365 }
2366
2367 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2368         .kind           = DRV_NAME,
2369         .priv_size      = sizeof(struct tun_struct),
2370         .setup          = tun_setup,
2371         .validate       = tun_validate,
2372         .get_size       = tun_get_size,
2373         .fill_info      = tun_fill_info,
2374 };
2375
2376 static void tun_sock_write_space(struct sock *sk)
2377 {
2378         struct tun_file *tfile;
2379         wait_queue_head_t *wqueue;
2380
2381         if (!sock_writeable(sk))
2382                 return;
2383
2384         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2385                 return;
2386
2387         wqueue = sk_sleep(sk);
2388         if (wqueue && waitqueue_active(wqueue))
2389                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2390                                                 EPOLLWRNORM | EPOLLWRBAND);
2391
2392         tfile = container_of(sk, struct tun_file, sk);
2393         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2394 }
2395
2396 static void tun_put_page(struct tun_page *tpage)
2397 {
2398         if (tpage->page)
2399                 __page_frag_cache_drain(tpage->page, tpage->count);
2400 }
2401
2402 static int tun_xdp_one(struct tun_struct *tun,
2403                        struct tun_file *tfile,
2404                        struct xdp_buff *xdp, int *flush,
2405                        struct tun_page *tpage)
2406 {
2407         unsigned int datasize = xdp->data_end - xdp->data;
2408         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2409         struct virtio_net_hdr *gso = &hdr->gso;
2410         struct bpf_prog *xdp_prog;
2411         struct sk_buff *skb = NULL;
2412         struct sk_buff_head *queue;
2413         u32 rxhash = 0, act;
2414         int buflen = hdr->buflen;
2415         int ret = 0;
2416         bool skb_xdp = false;
2417         struct page *page;
2418
2419         xdp_prog = rcu_dereference(tun->xdp_prog);
2420         if (xdp_prog) {
2421                 if (gso->gso_type) {
2422                         skb_xdp = true;
2423                         goto build;
2424                 }
2425
2426                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2427                 xdp_set_data_meta_invalid(xdp);
2428
2429                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2430                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2431                 if (ret < 0) {
2432                         put_page(virt_to_head_page(xdp->data));
2433                         return ret;
2434                 }
2435
2436                 switch (ret) {
2437                 case XDP_REDIRECT:
2438                         *flush = true;
2439                         fallthrough;
2440                 case XDP_TX:
2441                         return 0;
2442                 case XDP_PASS:
2443                         break;
2444                 default:
2445                         page = virt_to_head_page(xdp->data);
2446                         if (tpage->page == page) {
2447                                 ++tpage->count;
2448                         } else {
2449                                 tun_put_page(tpage);
2450                                 tpage->page = page;
2451                                 tpage->count = 1;
2452                         }
2453                         return 0;
2454                 }
2455         }
2456
2457 build:
2458         skb = build_skb(xdp->data_hard_start, buflen);
2459         if (!skb) {
2460                 ret = -ENOMEM;
2461                 goto out;
2462         }
2463
2464         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2465         skb_put(skb, xdp->data_end - xdp->data);
2466
2467         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2468                 atomic_long_inc(&tun->rx_frame_errors);
2469                 kfree_skb(skb);
2470                 ret = -EINVAL;
2471                 goto out;
2472         }
2473
2474         skb->protocol = eth_type_trans(skb, tun->dev);
2475         skb_reset_network_header(skb);
2476         skb_probe_transport_header(skb);
2477         skb_record_rx_queue(skb, tfile->queue_index);
2478
2479         if (skb_xdp) {
2480                 ret = do_xdp_generic(xdp_prog, skb);
2481                 if (ret != XDP_PASS) {
2482                         ret = 0;
2483                         goto out;
2484                 }
2485         }
2486
2487         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2488             !tfile->detached)
2489                 rxhash = __skb_get_hash_symmetric(skb);
2490
2491         if (tfile->napi_enabled) {
2492                 queue = &tfile->sk.sk_write_queue;
2493                 spin_lock(&queue->lock);
2494                 __skb_queue_tail(queue, skb);
2495                 spin_unlock(&queue->lock);
2496                 ret = 1;
2497         } else {
2498                 netif_receive_skb(skb);
2499                 ret = 0;
2500         }
2501
2502         /* No need to disable preemption here since this function is
2503          * always called with bh disabled
2504          */
2505         dev_sw_netstats_rx_add(tun->dev, datasize);
2506
2507         if (rxhash)
2508                 tun_flow_update(tun, rxhash, tfile);
2509
2510 out:
2511         return ret;
2512 }
2513
2514 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2515 {
2516         int ret, i;
2517         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2518         struct tun_struct *tun = tun_get(tfile);
2519         struct tun_msg_ctl *ctl = m->msg_control;
2520         struct xdp_buff *xdp;
2521
2522         if (!tun)
2523                 return -EBADFD;
2524
2525         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2526             ctl && ctl->type == TUN_MSG_PTR) {
2527                 struct tun_page tpage;
2528                 int n = ctl->num;
2529                 int flush = 0, queued = 0;
2530
2531                 memset(&tpage, 0, sizeof(tpage));
2532
2533                 local_bh_disable();
2534                 rcu_read_lock();
2535
2536                 for (i = 0; i < n; i++) {
2537                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2538                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2539                         if (ret > 0)
2540                                 queued += ret;
2541                 }
2542
2543                 if (flush)
2544                         xdp_do_flush();
2545
2546                 if (tfile->napi_enabled && queued > 0)
2547                         napi_schedule(&tfile->napi);
2548
2549                 rcu_read_unlock();
2550                 local_bh_enable();
2551
2552                 tun_put_page(&tpage);
2553
2554                 ret = total_len;
2555                 goto out;
2556         }
2557
2558         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2559                            m->msg_flags & MSG_DONTWAIT,
2560                            m->msg_flags & MSG_MORE);
2561 out:
2562         tun_put(tun);
2563         return ret;
2564 }
2565
2566 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2567                        int flags)
2568 {
2569         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2570         struct tun_struct *tun = tun_get(tfile);
2571         void *ptr = m->msg_control;
2572         int ret;
2573
2574         if (!tun) {
2575                 ret = -EBADFD;
2576                 goto out_free;
2577         }
2578
2579         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2580                 ret = -EINVAL;
2581                 goto out_put_tun;
2582         }
2583         if (flags & MSG_ERRQUEUE) {
2584                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2585                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2586                 goto out;
2587         }
2588         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2589         if (ret > (ssize_t)total_len) {
2590                 m->msg_flags |= MSG_TRUNC;
2591                 ret = flags & MSG_TRUNC ? ret : total_len;
2592         }
2593 out:
2594         tun_put(tun);
2595         return ret;
2596
2597 out_put_tun:
2598         tun_put(tun);
2599 out_free:
2600         tun_ptr_free(ptr);
2601         return ret;
2602 }
2603
2604 static int tun_ptr_peek_len(void *ptr)
2605 {
2606         if (likely(ptr)) {
2607                 if (tun_is_xdp_frame(ptr)) {
2608                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2609
2610                         return xdpf->len;
2611                 }
2612                 return __skb_array_len_with_tag(ptr);
2613         } else {
2614                 return 0;
2615         }
2616 }
2617
2618 static int tun_peek_len(struct socket *sock)
2619 {
2620         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2621         struct tun_struct *tun;
2622         int ret = 0;
2623
2624         tun = tun_get(tfile);
2625         if (!tun)
2626                 return 0;
2627
2628         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2629         tun_put(tun);
2630
2631         return ret;
2632 }
2633
2634 /* Ops structure to mimic raw sockets with tun */
2635 static const struct proto_ops tun_socket_ops = {
2636         .peek_len = tun_peek_len,
2637         .sendmsg = tun_sendmsg,
2638         .recvmsg = tun_recvmsg,
2639 };
2640
2641 static struct proto tun_proto = {
2642         .name           = "tun",
2643         .owner          = THIS_MODULE,
2644         .obj_size       = sizeof(struct tun_file),
2645 };
2646
2647 static int tun_flags(struct tun_struct *tun)
2648 {
2649         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2650 }
2651
2652 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2653                               char *buf)
2654 {
2655         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2656         return sprintf(buf, "0x%x\n", tun_flags(tun));
2657 }
2658
2659 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2660                           char *buf)
2661 {
2662         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2663         return uid_valid(tun->owner)?
2664                 sprintf(buf, "%u\n",
2665                         from_kuid_munged(current_user_ns(), tun->owner)):
2666                 sprintf(buf, "-1\n");
2667 }
2668
2669 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2670                           char *buf)
2671 {
2672         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2673         return gid_valid(tun->group) ?
2674                 sprintf(buf, "%u\n",
2675                         from_kgid_munged(current_user_ns(), tun->group)):
2676                 sprintf(buf, "-1\n");
2677 }
2678
2679 static DEVICE_ATTR_RO(tun_flags);
2680 static DEVICE_ATTR_RO(owner);
2681 static DEVICE_ATTR_RO(group);
2682
2683 static struct attribute *tun_dev_attrs[] = {
2684         &dev_attr_tun_flags.attr,
2685         &dev_attr_owner.attr,
2686         &dev_attr_group.attr,
2687         NULL
2688 };
2689
2690 static const struct attribute_group tun_attr_group = {
2691         .attrs = tun_dev_attrs
2692 };
2693
2694 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2695 {
2696         struct tun_struct *tun;
2697         struct tun_file *tfile = file->private_data;
2698         struct net_device *dev;
2699         int err;
2700
2701         if (tfile->detached)
2702                 return -EINVAL;
2703
2704         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2705                 if (!capable(CAP_NET_ADMIN))
2706                         return -EPERM;
2707
2708                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2709                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2710                         return -EINVAL;
2711         }
2712
2713         dev = __dev_get_by_name(net, ifr->ifr_name);
2714         if (dev) {
2715                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2716                         return -EBUSY;
2717                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2718                         tun = netdev_priv(dev);
2719                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2720                         tun = netdev_priv(dev);
2721                 else
2722                         return -EINVAL;
2723
2724                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2725                     !!(tun->flags & IFF_MULTI_QUEUE))
2726                         return -EINVAL;
2727
2728                 if (tun_not_capable(tun))
2729                         return -EPERM;
2730                 err = security_tun_dev_open(tun->security);
2731                 if (err < 0)
2732                         return err;
2733
2734                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2735                                  ifr->ifr_flags & IFF_NAPI,
2736                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2737                 if (err < 0)
2738                         return err;
2739
2740                 if (tun->flags & IFF_MULTI_QUEUE &&
2741                     (tun->numqueues + tun->numdisabled > 1)) {
2742                         /* One or more queue has already been attached, no need
2743                          * to initialize the device again.
2744                          */
2745                         netdev_state_change(dev);
2746                         return 0;
2747                 }
2748
2749                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2750                               (ifr->ifr_flags & TUN_FEATURES);
2751
2752                 netdev_state_change(dev);
2753         } else {
2754                 char *name;
2755                 unsigned long flags = 0;
2756                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2757                              MAX_TAP_QUEUES : 1;
2758
2759                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2760                         return -EPERM;
2761                 err = security_tun_dev_create();
2762                 if (err < 0)
2763                         return err;
2764
2765                 /* Set dev type */
2766                 if (ifr->ifr_flags & IFF_TUN) {
2767                         /* TUN device */
2768                         flags |= IFF_TUN;
2769                         name = "tun%d";
2770                 } else if (ifr->ifr_flags & IFF_TAP) {
2771                         /* TAP device */
2772                         flags |= IFF_TAP;
2773                         name = "tap%d";
2774                 } else
2775                         return -EINVAL;
2776
2777                 if (*ifr->ifr_name)
2778                         name = ifr->ifr_name;
2779
2780                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2781                                        NET_NAME_UNKNOWN, tun_setup, queues,
2782                                        queues);
2783
2784                 if (!dev)
2785                         return -ENOMEM;
2786
2787                 dev_net_set(dev, net);
2788                 dev->rtnl_link_ops = &tun_link_ops;
2789                 dev->ifindex = tfile->ifindex;
2790                 dev->sysfs_groups[0] = &tun_attr_group;
2791
2792                 tun = netdev_priv(dev);
2793                 tun->dev = dev;
2794                 tun->flags = flags;
2795                 tun->txflt.count = 0;
2796                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2797
2798                 tun->align = NET_SKB_PAD;
2799                 tun->filter_attached = false;
2800                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2801                 tun->rx_batched = 0;
2802                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2803
2804                 tun->ifr = ifr;
2805                 tun->file = file;
2806
2807                 tun_net_initialize(dev);
2808
2809                 err = register_netdevice(tun->dev);
2810                 if (err < 0) {
2811                         free_netdev(dev);
2812                         return err;
2813                 }
2814                 /* free_netdev() won't check refcnt, to avoid race
2815                  * with dev_put() we need publish tun after registration.
2816                  */
2817                 rcu_assign_pointer(tfile->tun, tun);
2818         }
2819
2820         netif_carrier_on(tun->dev);
2821
2822         /* Make sure persistent devices do not get stuck in
2823          * xoff state.
2824          */
2825         if (netif_running(tun->dev))
2826                 netif_tx_wake_all_queues(tun->dev);
2827
2828         strcpy(ifr->ifr_name, tun->dev->name);
2829         return 0;
2830 }
2831
2832 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2833 {
2834         strcpy(ifr->ifr_name, tun->dev->name);
2835
2836         ifr->ifr_flags = tun_flags(tun);
2837
2838 }
2839
2840 /* This is like a cut-down ethtool ops, except done via tun fd so no
2841  * privs required. */
2842 static int set_offload(struct tun_struct *tun, unsigned long arg)
2843 {
2844         netdev_features_t features = 0;
2845
2846         if (arg & TUN_F_CSUM) {
2847                 features |= NETIF_F_HW_CSUM;
2848                 arg &= ~TUN_F_CSUM;
2849
2850                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2851                         if (arg & TUN_F_TSO_ECN) {
2852                                 features |= NETIF_F_TSO_ECN;
2853                                 arg &= ~TUN_F_TSO_ECN;
2854                         }
2855                         if (arg & TUN_F_TSO4)
2856                                 features |= NETIF_F_TSO;
2857                         if (arg & TUN_F_TSO6)
2858                                 features |= NETIF_F_TSO6;
2859                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2860                 }
2861
2862                 arg &= ~TUN_F_UFO;
2863         }
2864
2865         /* This gives the user a way to test for new features in future by
2866          * trying to set them. */
2867         if (arg)
2868                 return -EINVAL;
2869
2870         tun->set_features = features;
2871         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2872         tun->dev->wanted_features |= features;
2873         netdev_update_features(tun->dev);
2874
2875         return 0;
2876 }
2877
2878 static void tun_detach_filter(struct tun_struct *tun, int n)
2879 {
2880         int i;
2881         struct tun_file *tfile;
2882
2883         for (i = 0; i < n; i++) {
2884                 tfile = rtnl_dereference(tun->tfiles[i]);
2885                 lock_sock(tfile->socket.sk);
2886                 sk_detach_filter(tfile->socket.sk);
2887                 release_sock(tfile->socket.sk);
2888         }
2889
2890         tun->filter_attached = false;
2891 }
2892
2893 static int tun_attach_filter(struct tun_struct *tun)
2894 {
2895         int i, ret = 0;
2896         struct tun_file *tfile;
2897
2898         for (i = 0; i < tun->numqueues; i++) {
2899                 tfile = rtnl_dereference(tun->tfiles[i]);
2900                 lock_sock(tfile->socket.sk);
2901                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2902                 release_sock(tfile->socket.sk);
2903                 if (ret) {
2904                         tun_detach_filter(tun, i);
2905                         return ret;
2906                 }
2907         }
2908
2909         tun->filter_attached = true;
2910         return ret;
2911 }
2912
2913 static void tun_set_sndbuf(struct tun_struct *tun)
2914 {
2915         struct tun_file *tfile;
2916         int i;
2917
2918         for (i = 0; i < tun->numqueues; i++) {
2919                 tfile = rtnl_dereference(tun->tfiles[i]);
2920                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2921         }
2922 }
2923
2924 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2925 {
2926         struct tun_file *tfile = file->private_data;
2927         struct tun_struct *tun;
2928         int ret = 0;
2929
2930         rtnl_lock();
2931
2932         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2933                 tun = tfile->detached;
2934                 if (!tun) {
2935                         ret = -EINVAL;
2936                         goto unlock;
2937                 }
2938                 ret = security_tun_dev_attach_queue(tun->security);
2939                 if (ret < 0)
2940                         goto unlock;
2941                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2942                                  tun->flags & IFF_NAPI_FRAGS, true);
2943         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2944                 tun = rtnl_dereference(tfile->tun);
2945                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2946                         ret = -EINVAL;
2947                 else
2948                         __tun_detach(tfile, false);
2949         } else
2950                 ret = -EINVAL;
2951
2952         if (ret >= 0)
2953                 netdev_state_change(tun->dev);
2954
2955 unlock:
2956         rtnl_unlock();
2957         return ret;
2958 }
2959
2960 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2961                         void __user *data)
2962 {
2963         struct bpf_prog *prog;
2964         int fd;
2965
2966         if (copy_from_user(&fd, data, sizeof(fd)))
2967                 return -EFAULT;
2968
2969         if (fd == -1) {
2970                 prog = NULL;
2971         } else {
2972                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2973                 if (IS_ERR(prog))
2974                         return PTR_ERR(prog);
2975         }
2976
2977         return __tun_set_ebpf(tun, prog_p, prog);
2978 }
2979
2980 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2981 static unsigned char tun_get_addr_len(unsigned short type)
2982 {
2983         switch (type) {
2984         case ARPHRD_IP6GRE:
2985         case ARPHRD_TUNNEL6:
2986                 return sizeof(struct in6_addr);
2987         case ARPHRD_IPGRE:
2988         case ARPHRD_TUNNEL:
2989         case ARPHRD_SIT:
2990                 return 4;
2991         case ARPHRD_ETHER:
2992                 return ETH_ALEN;
2993         case ARPHRD_IEEE802154:
2994         case ARPHRD_IEEE802154_MONITOR:
2995                 return IEEE802154_EXTENDED_ADDR_LEN;
2996         case ARPHRD_PHONET_PIPE:
2997         case ARPHRD_PPP:
2998         case ARPHRD_NONE:
2999                 return 0;
3000         case ARPHRD_6LOWPAN:
3001                 return EUI64_ADDR_LEN;
3002         case ARPHRD_FDDI:
3003                 return FDDI_K_ALEN;
3004         case ARPHRD_HIPPI:
3005                 return HIPPI_ALEN;
3006         case ARPHRD_IEEE802:
3007                 return FC_ALEN;
3008         case ARPHRD_ROSE:
3009                 return ROSE_ADDR_LEN;
3010         case ARPHRD_NETROM:
3011                 return AX25_ADDR_LEN;
3012         case ARPHRD_LOCALTLK:
3013                 return LTALK_ALEN;
3014         default:
3015                 return 0;
3016         }
3017 }
3018
3019 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3020                             unsigned long arg, int ifreq_len)
3021 {
3022         struct tun_file *tfile = file->private_data;
3023         struct net *net = sock_net(&tfile->sk);
3024         struct tun_struct *tun;
3025         void __user* argp = (void __user*)arg;
3026         unsigned int ifindex, carrier;
3027         struct ifreq ifr;
3028         kuid_t owner;
3029         kgid_t group;
3030         int sndbuf;
3031         int vnet_hdr_sz;
3032         int le;
3033         int ret;
3034         bool do_notify = false;
3035
3036         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3037             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3038                 if (copy_from_user(&ifr, argp, ifreq_len))
3039                         return -EFAULT;
3040         } else {
3041                 memset(&ifr, 0, sizeof(ifr));
3042         }
3043         if (cmd == TUNGETFEATURES) {
3044                 /* Currently this just means: "what IFF flags are valid?".
3045                  * This is needed because we never checked for invalid flags on
3046                  * TUNSETIFF.
3047                  */
3048                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3049                                 (unsigned int __user*)argp);
3050         } else if (cmd == TUNSETQUEUE) {
3051                 return tun_set_queue(file, &ifr);
3052         } else if (cmd == SIOCGSKNS) {
3053                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3054                         return -EPERM;
3055                 return open_related_ns(&net->ns, get_net_ns);
3056         }
3057
3058         rtnl_lock();
3059
3060         tun = tun_get(tfile);
3061         if (cmd == TUNSETIFF) {
3062                 ret = -EEXIST;
3063                 if (tun)
3064                         goto unlock;
3065
3066                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3067
3068                 ret = tun_set_iff(net, file, &ifr);
3069
3070                 if (ret)
3071                         goto unlock;
3072
3073                 if (copy_to_user(argp, &ifr, ifreq_len))
3074                         ret = -EFAULT;
3075                 goto unlock;
3076         }
3077         if (cmd == TUNSETIFINDEX) {
3078                 ret = -EPERM;
3079                 if (tun)
3080                         goto unlock;
3081
3082                 ret = -EFAULT;
3083                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3084                         goto unlock;
3085
3086                 ret = 0;
3087                 tfile->ifindex = ifindex;
3088                 goto unlock;
3089         }
3090
3091         ret = -EBADFD;
3092         if (!tun)
3093                 goto unlock;
3094
3095         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3096
3097         net = dev_net(tun->dev);
3098         ret = 0;
3099         switch (cmd) {
3100         case TUNGETIFF:
3101                 tun_get_iff(tun, &ifr);
3102
3103                 if (tfile->detached)
3104                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3105                 if (!tfile->socket.sk->sk_filter)
3106                         ifr.ifr_flags |= IFF_NOFILTER;
3107
3108                 if (copy_to_user(argp, &ifr, ifreq_len))
3109                         ret = -EFAULT;
3110                 break;
3111
3112         case TUNSETNOCSUM:
3113                 /* Disable/Enable checksum */
3114
3115                 /* [unimplemented] */
3116                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3117                            arg ? "disabled" : "enabled");
3118                 break;
3119
3120         case TUNSETPERSIST:
3121                 /* Disable/Enable persist mode. Keep an extra reference to the
3122                  * module to prevent the module being unprobed.
3123                  */
3124                 if (arg && !(tun->flags & IFF_PERSIST)) {
3125                         tun->flags |= IFF_PERSIST;
3126                         __module_get(THIS_MODULE);
3127                         do_notify = true;
3128                 }
3129                 if (!arg && (tun->flags & IFF_PERSIST)) {
3130                         tun->flags &= ~IFF_PERSIST;
3131                         module_put(THIS_MODULE);
3132                         do_notify = true;
3133                 }
3134
3135                 netif_info(tun, drv, tun->dev, "persist %s\n",
3136                            arg ? "enabled" : "disabled");
3137                 break;
3138
3139         case TUNSETOWNER:
3140                 /* Set owner of the device */
3141                 owner = make_kuid(current_user_ns(), arg);
3142                 if (!uid_valid(owner)) {
3143                         ret = -EINVAL;
3144                         break;
3145                 }
3146                 tun->owner = owner;
3147                 do_notify = true;
3148                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3149                            from_kuid(&init_user_ns, tun->owner));
3150                 break;
3151
3152         case TUNSETGROUP:
3153                 /* Set group of the device */
3154                 group = make_kgid(current_user_ns(), arg);
3155                 if (!gid_valid(group)) {
3156                         ret = -EINVAL;
3157                         break;
3158                 }
3159                 tun->group = group;
3160                 do_notify = true;
3161                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3162                            from_kgid(&init_user_ns, tun->group));
3163                 break;
3164
3165         case TUNSETLINK:
3166                 /* Only allow setting the type when the interface is down */
3167                 if (tun->dev->flags & IFF_UP) {
3168                         netif_info(tun, drv, tun->dev,
3169                                    "Linktype set failed because interface is up\n");
3170                         ret = -EBUSY;
3171                 } else {
3172                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3173                                                        tun->dev);
3174                         ret = notifier_to_errno(ret);
3175                         if (ret) {
3176                                 netif_info(tun, drv, tun->dev,
3177                                            "Refused to change device type\n");
3178                                 break;
3179                         }
3180                         tun->dev->type = (int) arg;
3181                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3182                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3183                                    tun->dev->type);
3184                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3185                                                  tun->dev);
3186                 }
3187                 break;
3188
3189         case TUNSETDEBUG:
3190                 tun->msg_enable = (u32)arg;
3191                 break;
3192
3193         case TUNSETOFFLOAD:
3194                 ret = set_offload(tun, arg);
3195                 break;
3196
3197         case TUNSETTXFILTER:
3198                 /* Can be set only for TAPs */
3199                 ret = -EINVAL;
3200                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3201                         break;
3202                 ret = update_filter(&tun->txflt, (void __user *)arg);
3203                 break;
3204
3205         case SIOCGIFHWADDR:
3206                 /* Get hw address */
3207                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3208                 if (copy_to_user(argp, &ifr, ifreq_len))
3209                         ret = -EFAULT;
3210                 break;
3211
3212         case SIOCSIFHWADDR:
3213                 /* Set hw address */
3214                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3215                 break;
3216
3217         case TUNGETSNDBUF:
3218                 sndbuf = tfile->socket.sk->sk_sndbuf;
3219                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3220                         ret = -EFAULT;
3221                 break;
3222
3223         case TUNSETSNDBUF:
3224                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3225                         ret = -EFAULT;
3226                         break;
3227                 }
3228                 if (sndbuf <= 0) {
3229                         ret = -EINVAL;
3230                         break;
3231                 }
3232
3233                 tun->sndbuf = sndbuf;
3234                 tun_set_sndbuf(tun);
3235                 break;
3236
3237         case TUNGETVNETHDRSZ:
3238                 vnet_hdr_sz = tun->vnet_hdr_sz;
3239                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3240                         ret = -EFAULT;
3241                 break;
3242
3243         case TUNSETVNETHDRSZ:
3244                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3245                         ret = -EFAULT;
3246                         break;
3247                 }
3248                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3249                         ret = -EINVAL;
3250                         break;
3251                 }
3252
3253                 tun->vnet_hdr_sz = vnet_hdr_sz;
3254                 break;
3255
3256         case TUNGETVNETLE:
3257                 le = !!(tun->flags & TUN_VNET_LE);
3258                 if (put_user(le, (int __user *)argp))
3259                         ret = -EFAULT;
3260                 break;
3261
3262         case TUNSETVNETLE:
3263                 if (get_user(le, (int __user *)argp)) {
3264                         ret = -EFAULT;
3265                         break;
3266                 }
3267                 if (le)
3268                         tun->flags |= TUN_VNET_LE;
3269                 else
3270                         tun->flags &= ~TUN_VNET_LE;
3271                 break;
3272
3273         case TUNGETVNETBE:
3274                 ret = tun_get_vnet_be(tun, argp);
3275                 break;
3276
3277         case TUNSETVNETBE:
3278                 ret = tun_set_vnet_be(tun, argp);
3279                 break;
3280
3281         case TUNATTACHFILTER:
3282                 /* Can be set only for TAPs */
3283                 ret = -EINVAL;
3284                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3285                         break;
3286                 ret = -EFAULT;
3287                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3288                         break;
3289
3290                 ret = tun_attach_filter(tun);
3291                 break;
3292
3293         case TUNDETACHFILTER:
3294                 /* Can be set only for TAPs */
3295                 ret = -EINVAL;
3296                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3297                         break;
3298                 ret = 0;
3299                 tun_detach_filter(tun, tun->numqueues);
3300                 break;
3301
3302         case TUNGETFILTER:
3303                 ret = -EINVAL;
3304                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3305                         break;
3306                 ret = -EFAULT;
3307                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3308                         break;
3309                 ret = 0;
3310                 break;
3311
3312         case TUNSETSTEERINGEBPF:
3313                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3314                 break;
3315
3316         case TUNSETFILTEREBPF:
3317                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3318                 break;
3319
3320         case TUNSETCARRIER:
3321                 ret = -EFAULT;
3322                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3323                         goto unlock;
3324
3325                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3326                 break;
3327
3328         case TUNGETDEVNETNS:
3329                 ret = -EPERM;
3330                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3331                         goto unlock;
3332                 ret = open_related_ns(&net->ns, get_net_ns);
3333                 break;
3334
3335         default:
3336                 ret = -EINVAL;
3337                 break;
3338         }
3339
3340         if (do_notify)
3341                 netdev_state_change(tun->dev);
3342
3343 unlock:
3344         rtnl_unlock();
3345         if (tun)
3346                 tun_put(tun);
3347         return ret;
3348 }
3349
3350 static long tun_chr_ioctl(struct file *file,
3351                           unsigned int cmd, unsigned long arg)
3352 {
3353         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3354 }
3355
3356 #ifdef CONFIG_COMPAT
3357 static long tun_chr_compat_ioctl(struct file *file,
3358                          unsigned int cmd, unsigned long arg)
3359 {
3360         switch (cmd) {
3361         case TUNSETIFF:
3362         case TUNGETIFF:
3363         case TUNSETTXFILTER:
3364         case TUNGETSNDBUF:
3365         case TUNSETSNDBUF:
3366         case SIOCGIFHWADDR:
3367         case SIOCSIFHWADDR:
3368                 arg = (unsigned long)compat_ptr(arg);
3369                 break;
3370         default:
3371                 arg = (compat_ulong_t)arg;
3372                 break;
3373         }
3374
3375         /*
3376          * compat_ifreq is shorter than ifreq, so we must not access beyond
3377          * the end of that structure. All fields that are used in this
3378          * driver are compatible though, we don't need to convert the
3379          * contents.
3380          */
3381         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3382 }
3383 #endif /* CONFIG_COMPAT */
3384
3385 static int tun_chr_fasync(int fd, struct file *file, int on)
3386 {
3387         struct tun_file *tfile = file->private_data;
3388         int ret;
3389
3390         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3391                 goto out;
3392
3393         if (on) {
3394                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3395                 tfile->flags |= TUN_FASYNC;
3396         } else
3397                 tfile->flags &= ~TUN_FASYNC;
3398         ret = 0;
3399 out:
3400         return ret;
3401 }
3402
3403 static int tun_chr_open(struct inode *inode, struct file * file)
3404 {
3405         struct net *net = current->nsproxy->net_ns;
3406         struct tun_file *tfile;
3407
3408         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3409                                             &tun_proto, 0);
3410         if (!tfile)
3411                 return -ENOMEM;
3412         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3413                 sk_free(&tfile->sk);
3414                 return -ENOMEM;
3415         }
3416
3417         mutex_init(&tfile->napi_mutex);
3418         RCU_INIT_POINTER(tfile->tun, NULL);
3419         tfile->flags = 0;
3420         tfile->ifindex = 0;
3421
3422         init_waitqueue_head(&tfile->socket.wq.wait);
3423
3424         tfile->socket.file = file;
3425         tfile->socket.ops = &tun_socket_ops;
3426
3427         sock_init_data(&tfile->socket, &tfile->sk);
3428
3429         tfile->sk.sk_write_space = tun_sock_write_space;
3430         tfile->sk.sk_sndbuf = INT_MAX;
3431
3432         file->private_data = tfile;
3433         INIT_LIST_HEAD(&tfile->next);
3434
3435         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3436
3437         return 0;
3438 }
3439
3440 static int tun_chr_close(struct inode *inode, struct file *file)
3441 {
3442         struct tun_file *tfile = file->private_data;
3443
3444         tun_detach(tfile, true);
3445
3446         return 0;
3447 }
3448
3449 #ifdef CONFIG_PROC_FS
3450 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3451 {
3452         struct tun_file *tfile = file->private_data;
3453         struct tun_struct *tun;
3454         struct ifreq ifr;
3455
3456         memset(&ifr, 0, sizeof(ifr));
3457
3458         rtnl_lock();
3459         tun = tun_get(tfile);
3460         if (tun)
3461                 tun_get_iff(tun, &ifr);
3462         rtnl_unlock();
3463
3464         if (tun)
3465                 tun_put(tun);
3466
3467         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3468 }
3469 #endif
3470
3471 static const struct file_operations tun_fops = {
3472         .owner  = THIS_MODULE,
3473         .llseek = no_llseek,
3474         .read_iter  = tun_chr_read_iter,
3475         .write_iter = tun_chr_write_iter,
3476         .poll   = tun_chr_poll,
3477         .unlocked_ioctl = tun_chr_ioctl,
3478 #ifdef CONFIG_COMPAT
3479         .compat_ioctl = tun_chr_compat_ioctl,
3480 #endif
3481         .open   = tun_chr_open,
3482         .release = tun_chr_close,
3483         .fasync = tun_chr_fasync,
3484 #ifdef CONFIG_PROC_FS
3485         .show_fdinfo = tun_chr_show_fdinfo,
3486 #endif
3487 };
3488
3489 static struct miscdevice tun_miscdev = {
3490         .minor = TUN_MINOR,
3491         .name = "tun",
3492         .nodename = "net/tun",
3493         .fops = &tun_fops,
3494 };
3495
3496 /* ethtool interface */
3497
3498 static void tun_default_link_ksettings(struct net_device *dev,
3499                                        struct ethtool_link_ksettings *cmd)
3500 {
3501         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3502         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3503         cmd->base.speed         = SPEED_10;
3504         cmd->base.duplex        = DUPLEX_FULL;
3505         cmd->base.port          = PORT_TP;
3506         cmd->base.phy_address   = 0;
3507         cmd->base.autoneg       = AUTONEG_DISABLE;
3508 }
3509
3510 static int tun_get_link_ksettings(struct net_device *dev,
3511                                   struct ethtool_link_ksettings *cmd)
3512 {
3513         struct tun_struct *tun = netdev_priv(dev);
3514
3515         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3516         return 0;
3517 }
3518
3519 static int tun_set_link_ksettings(struct net_device *dev,
3520                                   const struct ethtool_link_ksettings *cmd)
3521 {
3522         struct tun_struct *tun = netdev_priv(dev);
3523
3524         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3525         return 0;
3526 }
3527
3528 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3529 {
3530         struct tun_struct *tun = netdev_priv(dev);
3531
3532         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3533         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3534
3535         switch (tun->flags & TUN_TYPE_MASK) {
3536         case IFF_TUN:
3537                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3538                 break;
3539         case IFF_TAP:
3540                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3541                 break;
3542         }
3543 }
3544
3545 static u32 tun_get_msglevel(struct net_device *dev)
3546 {
3547         struct tun_struct *tun = netdev_priv(dev);
3548
3549         return tun->msg_enable;
3550 }
3551
3552 static void tun_set_msglevel(struct net_device *dev, u32 value)
3553 {
3554         struct tun_struct *tun = netdev_priv(dev);
3555
3556         tun->msg_enable = value;
3557 }
3558
3559 static int tun_get_coalesce(struct net_device *dev,
3560                             struct ethtool_coalesce *ec,
3561                             struct kernel_ethtool_coalesce *kernel_coal,
3562                             struct netlink_ext_ack *extack)
3563 {
3564         struct tun_struct *tun = netdev_priv(dev);
3565
3566         ec->rx_max_coalesced_frames = tun->rx_batched;
3567
3568         return 0;
3569 }
3570
3571 static int tun_set_coalesce(struct net_device *dev,
3572                             struct ethtool_coalesce *ec,
3573                             struct kernel_ethtool_coalesce *kernel_coal,
3574                             struct netlink_ext_ack *extack)
3575 {
3576         struct tun_struct *tun = netdev_priv(dev);
3577
3578         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3579                 tun->rx_batched = NAPI_POLL_WEIGHT;
3580         else
3581                 tun->rx_batched = ec->rx_max_coalesced_frames;
3582
3583         return 0;
3584 }
3585
3586 static const struct ethtool_ops tun_ethtool_ops = {
3587         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3588         .get_drvinfo    = tun_get_drvinfo,
3589         .get_msglevel   = tun_get_msglevel,
3590         .set_msglevel   = tun_set_msglevel,
3591         .get_link       = ethtool_op_get_link,
3592         .get_ts_info    = ethtool_op_get_ts_info,
3593         .get_coalesce   = tun_get_coalesce,
3594         .set_coalesce   = tun_set_coalesce,
3595         .get_link_ksettings = tun_get_link_ksettings,
3596         .set_link_ksettings = tun_set_link_ksettings,
3597 };
3598
3599 static int tun_queue_resize(struct tun_struct *tun)
3600 {
3601         struct net_device *dev = tun->dev;
3602         struct tun_file *tfile;
3603         struct ptr_ring **rings;
3604         int n = tun->numqueues + tun->numdisabled;
3605         int ret, i;
3606
3607         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3608         if (!rings)
3609                 return -ENOMEM;
3610
3611         for (i = 0; i < tun->numqueues; i++) {
3612                 tfile = rtnl_dereference(tun->tfiles[i]);
3613                 rings[i] = &tfile->tx_ring;
3614         }
3615         list_for_each_entry(tfile, &tun->disabled, next)
3616                 rings[i++] = &tfile->tx_ring;
3617
3618         ret = ptr_ring_resize_multiple(rings, n,
3619                                        dev->tx_queue_len, GFP_KERNEL,
3620                                        tun_ptr_free);
3621
3622         kfree(rings);
3623         return ret;
3624 }
3625
3626 static int tun_device_event(struct notifier_block *unused,
3627                             unsigned long event, void *ptr)
3628 {
3629         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3630         struct tun_struct *tun = netdev_priv(dev);
3631         int i;
3632
3633         if (dev->rtnl_link_ops != &tun_link_ops)
3634                 return NOTIFY_DONE;
3635
3636         switch (event) {
3637         case NETDEV_CHANGE_TX_QUEUE_LEN:
3638                 if (tun_queue_resize(tun))
3639                         return NOTIFY_BAD;
3640                 break;
3641         case NETDEV_UP:
3642                 for (i = 0; i < tun->numqueues; i++) {
3643                         struct tun_file *tfile;
3644
3645                         tfile = rtnl_dereference(tun->tfiles[i]);
3646                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3647                 }
3648                 break;
3649         default:
3650                 break;
3651         }
3652
3653         return NOTIFY_DONE;
3654 }
3655
3656 static struct notifier_block tun_notifier_block __read_mostly = {
3657         .notifier_call  = tun_device_event,
3658 };
3659
3660 static int __init tun_init(void)
3661 {
3662         int ret = 0;
3663
3664         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3665
3666         ret = rtnl_link_register(&tun_link_ops);
3667         if (ret) {
3668                 pr_err("Can't register link_ops\n");
3669                 goto err_linkops;
3670         }
3671
3672         ret = misc_register(&tun_miscdev);
3673         if (ret) {
3674                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3675                 goto err_misc;
3676         }
3677
3678         ret = register_netdevice_notifier(&tun_notifier_block);
3679         if (ret) {
3680                 pr_err("Can't register netdevice notifier\n");
3681                 goto err_notifier;
3682         }
3683
3684         return  0;
3685
3686 err_notifier:
3687         misc_deregister(&tun_miscdev);
3688 err_misc:
3689         rtnl_link_unregister(&tun_link_ops);
3690 err_linkops:
3691         return ret;
3692 }
3693
3694 static void tun_cleanup(void)
3695 {
3696         misc_deregister(&tun_miscdev);
3697         rtnl_link_unregister(&tun_link_ops);
3698         unregister_netdevice_notifier(&tun_notifier_block);
3699 }
3700
3701 /* Get an underlying socket object from tun file.  Returns error unless file is
3702  * attached to a device.  The returned object works like a packet socket, it
3703  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3704  * holding a reference to the file for as long as the socket is in use. */
3705 struct socket *tun_get_socket(struct file *file)
3706 {
3707         struct tun_file *tfile;
3708         if (file->f_op != &tun_fops)
3709                 return ERR_PTR(-EINVAL);
3710         tfile = file->private_data;
3711         if (!tfile)
3712                 return ERR_PTR(-EBADFD);
3713         return &tfile->socket;
3714 }
3715 EXPORT_SYMBOL_GPL(tun_get_socket);
3716
3717 struct ptr_ring *tun_get_tx_ring(struct file *file)
3718 {
3719         struct tun_file *tfile;
3720
3721         if (file->f_op != &tun_fops)
3722                 return ERR_PTR(-EINVAL);
3723         tfile = file->private_data;
3724         if (!tfile)
3725                 return ERR_PTR(-EBADFD);
3726         return &tfile->tx_ring;
3727 }
3728 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3729
3730 module_init(tun_init);
3731 module_exit(tun_cleanup);
3732 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3733 MODULE_AUTHOR(DRV_COPYRIGHT);
3734 MODULE_LICENSE("GPL");
3735 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3736 MODULE_ALIAS("devname:net/tun");