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
  73 #include <linux/uaccess.h>
  74 #include <linux/proc_fs.h>
  75
  76 static void tun_default_link_ksettings(struct net_device *dev,
  77                                        struct ethtool_link_ksettings *cmd);
  78
  79 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  80
  81 /* TUN device flags */
  82
  83 /* IFF_ATTACH_QUEUE is never stored in device flags,
  84  * overload it to mean fasync when stored there.
  85  */
  86 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  87 /* High bits in flags field are unused. */
  88 #define TUN_VNET_LE     0x80000000
  89 #define TUN_VNET_BE     0x40000000
  90
  91 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
  92                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
  93
  94 #define GOODCOPY_LEN 128
  95
  96 #define FLT_EXACT_COUNT 8
  97 struct tap_filter {
  98         unsigned int    count;    /* Number of addrs. Zero means disabled */
  99         u32             mask[2];  /* Mask of the hashed addrs */
 100         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 101 };
 102
 103 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 104  * to max number of VCPUs in guest. */
 105 #define MAX_TAP_QUEUES 256
 106 #define MAX_TAP_FLOWS  4096
 107
 108 #define TUN_FLOW_EXPIRE (3 * HZ)
 109
 110 struct tun_pcpu_stats {
 111         u64_stats_t rx_packets;
 112         u64_stats_t rx_bytes;
 113         u64_stats_t tx_packets;
 114         u64_stats_t tx_bytes;
 115         struct u64_stats_sync syncp;
 116         u32 rx_dropped;
 117         u32 tx_dropped;
 118         u32 rx_frame_errors;
 119 };
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related structures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct tun_struct __rcu *tun;
 136         struct fasync_struct *fasync;
 137         /* only used for fasnyc */
 138         unsigned int flags;
 139         union {
 140                 u16 queue_index;
 141                 unsigned int ifindex;
 142         };
 143         struct napi_struct napi;
 144         bool napi_enabled;
 145         bool napi_frags_enabled;
 146         struct mutex napi_mutex;        /* Protects access to the above napi */
 147         struct list_head next;
 148         struct tun_struct *detached;
 149         struct ptr_ring tx_ring;
 150         struct xdp_rxq_info xdp_rxq;
 151 };
 152
 153 struct tun_page {
 154         struct page *page;
 155         int count;
 156 };
 157
 158 struct tun_flow_entry {
 159         struct hlist_node hash_link;
 160         struct rcu_head rcu;
 161         struct tun_struct *tun;
 162
 163         u32 rxhash;
 164         u32 rps_rxhash;
 165         int queue_index;
 166         unsigned long updated ____cacheline_aligned_in_smp;
 167 };
 168
 169 #define TUN_NUM_FLOW_ENTRIES 1024
 170 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 171
 172 struct tun_prog {
 173         struct rcu_head rcu;
 174         struct bpf_prog *prog;
 175 };
 176
 177 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 178  * device, socket filter, sndbuf and vnet header size were restore when the
 179  * file were attached to a persist device.
 180  */
 181 struct tun_struct {
 182         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 183         unsigned int            numqueues;
 184         unsigned int            flags;
 185         kuid_t                  owner;
 186         kgid_t                  group;
 187
 188         struct net_device       *dev;
 189         netdev_features_t       set_features;
 190 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 191                           NETIF_F_TSO6)
 192
 193         int                     align;
 194         int                     vnet_hdr_sz;
 195         int                     sndbuf;
 196         struct tap_filter       txflt;
 197         struct sock_fprog       fprog;
 198         /* protected by rtnl lock */
 199         bool                    filter_attached;
 200         u32                     msg_enable;
 201         spinlock_t lock;
 202         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 203         struct timer_list flow_gc_timer;
 204         unsigned long ageing_time;
 205         unsigned int numdisabled;
 206         struct list_head disabled;
 207         void *security;
 208         u32 flow_count;
 209         u32 rx_batched;
 210         struct tun_pcpu_stats __percpu *pcpu_stats;
 211         struct bpf_prog __rcu *xdp_prog;
 212         struct tun_prog __rcu *steering_prog;
 213         struct tun_prog __rcu *filter_prog;
 214         struct ethtool_link_ksettings link_ksettings;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 bool tun_is_xdp_frame(void *ptr)
 223 {
 224         return (unsigned long)ptr & TUN_XDP_FLAG;
 225 }
 226 EXPORT_SYMBOL(tun_is_xdp_frame);
 227
 228 void *tun_xdp_to_ptr(void *ptr)
 229 {
 230         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 231 }
 232 EXPORT_SYMBOL(tun_xdp_to_ptr);
 233
 234 void *tun_ptr_to_xdp(void *ptr)
 235 {
 236         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 237 }
 238 EXPORT_SYMBOL(tun_ptr_to_xdp);
 239
 240 static int tun_napi_receive(struct napi_struct *napi, int budget)
 241 {
 242         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 243         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 244         struct sk_buff_head process_queue;
 245         struct sk_buff *skb;
 246         int received = 0;
 247
 248         __skb_queue_head_init(&process_queue);
 249
 250         spin_lock(&queue->lock);
 251         skb_queue_splice_tail_init(queue, &process_queue);
 252         spin_unlock(&queue->lock);
 253
 254         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 255                 napi_gro_receive(napi, skb);
 256                 ++received;
 257         }
 258
 259         if (!skb_queue_empty(&process_queue)) {
 260                 spin_lock(&queue->lock);
 261                 skb_queue_splice(&process_queue, queue);
 262                 spin_unlock(&queue->lock);
 263         }
 264
 265         return received;
 266 }
 267
 268 static int tun_napi_poll(struct napi_struct *napi, int budget)
 269 {
 270         unsigned int received;
 271
 272         received = tun_napi_receive(napi, budget);
 273
 274         if (received < budget)
 275                 napi_complete_done(napi, received);
 276
 277         return received;
 278 }
 279
 280 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 281                           bool napi_en, bool napi_frags)
 282 {
 283         tfile->napi_enabled = napi_en;
 284         tfile->napi_frags_enabled = napi_en && napi_frags;
 285         if (napi_en) {
 286                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 287                                   NAPI_POLL_WEIGHT);
 288                 napi_enable(&tfile->napi);
 289         }
 290 }
 291
 292 static void tun_napi_disable(struct tun_file *tfile)
 293 {
 294         if (tfile->napi_enabled)
 295                 napi_disable(&tfile->napi);
 296 }
 297
 298 static void tun_napi_del(struct tun_file *tfile)
 299 {
 300         if (tfile->napi_enabled)
 301                 netif_napi_del(&tfile->napi);
 302 }
 303
 304 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 305 {
 306         return tfile->napi_frags_enabled;
 307 }
 308
 309 #ifdef CONFIG_TUN_VNET_CROSS_LE
 310 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 311 {
 312         return tun->flags & TUN_VNET_BE ? false :
 313                 virtio_legacy_is_little_endian();
 314 }
 315
 316 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 317 {
 318         int be = !!(tun->flags & TUN_VNET_BE);
 319
 320         if (put_user(be, argp))
 321                 return -EFAULT;
 322
 323         return 0;
 324 }
 325
 326 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 327 {
 328         int be;
 329
 330         if (get_user(be, argp))
 331                 return -EFAULT;
 332
 333         if (be)
 334                 tun->flags |= TUN_VNET_BE;
 335         else
 336                 tun->flags &= ~TUN_VNET_BE;
 337
 338         return 0;
 339 }
 340 #else
 341 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 342 {
 343         return virtio_legacy_is_little_endian();
 344 }
 345
 346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 347 {
 348         return -EINVAL;
 349 }
 350
 351 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 352 {
 353         return -EINVAL;
 354 }
 355 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 356
 357 static inline bool tun_is_little_endian(struct tun_struct *tun)
 358 {
 359         return tun->flags & TUN_VNET_LE ||
 360                 tun_legacy_is_little_endian(tun);
 361 }
 362
 363 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 364 {
 365         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 366 }
 367
 368 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 369 {
 370         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 371 }
 372
 373 static inline u32 tun_hashfn(u32 rxhash)
 374 {
 375         return rxhash & TUN_MASK_FLOW_ENTRIES;
 376 }
 377
 378 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 379 {
 380         struct tun_flow_entry *e;
 381
 382         hlist_for_each_entry_rcu(e, head, hash_link) {
 383                 if (e->rxhash == rxhash)
 384                         return e;
 385         }
 386         return NULL;
 387 }
 388
 389 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 390                                               struct hlist_head *head,
 391                                               u32 rxhash, u16 queue_index)
 392 {
 393         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 394
 395         if (e) {
 396                 netif_info(tun, tx_queued, tun->dev,
 397                            "create flow: hash %u index %u\n",
 398                            rxhash, queue_index);
 399                 e->updated = jiffies;
 400                 e->rxhash = rxhash;
 401                 e->rps_rxhash = 0;
 402                 e->queue_index = queue_index;
 403                 e->tun = tun;
 404                 hlist_add_head_rcu(&e->hash_link, head);
 405                 ++tun->flow_count;
 406         }
 407         return e;
 408 }
 409
 410 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 411 {
 412         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 413                    e->rxhash, e->queue_index);
 414         hlist_del_rcu(&e->hash_link);
 415         kfree_rcu(e, rcu);
 416         --tun->flow_count;
 417 }
 418
 419 static void tun_flow_flush(struct tun_struct *tun)
 420 {
 421         int i;
 422
 423         spin_lock_bh(&tun->lock);
 424         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 425                 struct tun_flow_entry *e;
 426                 struct hlist_node *n;
 427
 428                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 429                         tun_flow_delete(tun, e);
 430         }
 431         spin_unlock_bh(&tun->lock);
 432 }
 433
 434 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 435 {
 436         int i;
 437
 438         spin_lock_bh(&tun->lock);
 439         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 440                 struct tun_flow_entry *e;
 441                 struct hlist_node *n;
 442
 443                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 444                         if (e->queue_index == queue_index)
 445                                 tun_flow_delete(tun, e);
 446                 }
 447         }
 448         spin_unlock_bh(&tun->lock);
 449 }
 450
 451 static void tun_flow_cleanup(struct timer_list *t)
 452 {
 453         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 454         unsigned long delay = tun->ageing_time;
 455         unsigned long next_timer = jiffies + delay;
 456         unsigned long count = 0;
 457         int i;
 458
 459         spin_lock(&tun->lock);
 460         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 461                 struct tun_flow_entry *e;
 462                 struct hlist_node *n;
 463
 464                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 465                         unsigned long this_timer;
 466
 467                         this_timer = e->updated + delay;
 468                         if (time_before_eq(this_timer, jiffies)) {
 469                                 tun_flow_delete(tun, e);
 470                                 continue;
 471                         }
 472                         count++;
 473                         if (time_before(this_timer, next_timer))
 474                                 next_timer = this_timer;
 475                 }
 476         }
 477
 478         if (count)
 479                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 480         spin_unlock(&tun->lock);
 481 }
 482
 483 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 484                             struct tun_file *tfile)
 485 {
 486         struct hlist_head *head;
 487         struct tun_flow_entry *e;
 488         unsigned long delay = tun->ageing_time;
 489         u16 queue_index = tfile->queue_index;
 490
 491         head = &tun->flows[tun_hashfn(rxhash)];
 492
 493         rcu_read_lock();
 494
 495         e = tun_flow_find(head, rxhash);
 496         if (likely(e)) {
 497                 /* TODO: keep queueing to old queue until it's empty? */
 498                 if (READ_ONCE(e->queue_index) != queue_index)
 499                         WRITE_ONCE(e->queue_index, queue_index);
 500                 if (e->updated != jiffies)
 501                         e->updated = jiffies;
 502                 sock_rps_record_flow_hash(e->rps_rxhash);
 503         } else {
 504                 spin_lock_bh(&tun->lock);
 505                 if (!tun_flow_find(head, rxhash) &&
 506                     tun->flow_count < MAX_TAP_FLOWS)
 507                         tun_flow_create(tun, head, rxhash, queue_index);
 508
 509                 if (!timer_pending(&tun->flow_gc_timer))
 510                         mod_timer(&tun->flow_gc_timer,
 511                                   round_jiffies_up(jiffies + delay));
 512                 spin_unlock_bh(&tun->lock);
 513         }
 514
 515         rcu_read_unlock();
 516 }
 517
 518 /* Save the hash received in the stack receive path and update the
 519  * flow_hash table accordingly.
 520  */
 521 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 522 {
 523         if (unlikely(e->rps_rxhash != hash))
 524                 e->rps_rxhash = hash;
 525 }
 526
 527 /* We try to identify a flow through its rxhash. The reason that
 528  * we do not check rxq no. is because some cards(e.g 82599), chooses
 529  * the rxq based on the txq where the last packet of the flow comes. As
 530  * the userspace application move between processors, we may get a
 531  * different rxq no. here.
 532  */
 533 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 534 {
 535         struct tun_flow_entry *e;
 536         u32 txq = 0;
 537         u32 numqueues = 0;
 538
 539         numqueues = READ_ONCE(tun->numqueues);
 540
 541         txq = __skb_get_hash_symmetric(skb);
 542         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 543         if (e) {
 544                 tun_flow_save_rps_rxhash(e, txq);
 545                 txq = e->queue_index;
 546         } else {
 547                 /* use multiply and shift instead of expensive divide */
 548                 txq = ((u64)txq * numqueues) >> 32;
 549         }
 550
 551         return txq;
 552 }
 553
 554 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 555 {
 556         struct tun_prog *prog;
 557         u32 numqueues;
 558         u16 ret = 0;
 559
 560         numqueues = READ_ONCE(tun->numqueues);
 561         if (!numqueues)
 562                 return 0;
 563
 564         prog = rcu_dereference(tun->steering_prog);
 565         if (prog)
 566                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 567
 568         return ret % numqueues;
 569 }
 570
 571 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 572                             struct net_device *sb_dev)
 573 {
 574         struct tun_struct *tun = netdev_priv(dev);
 575         u16 ret;
 576
 577         rcu_read_lock();
 578         if (rcu_dereference(tun->steering_prog))
 579                 ret = tun_ebpf_select_queue(tun, skb);
 580         else
 581                 ret = tun_automq_select_queue(tun, skb);
 582         rcu_read_unlock();
 583
 584         return ret;
 585 }
 586
 587 static inline bool tun_not_capable(struct tun_struct *tun)
 588 {
 589         const struct cred *cred = current_cred();
 590         struct net *net = dev_net(tun->dev);
 591
 592         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 593                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 594                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 595 }
 596
 597 static void tun_set_real_num_queues(struct tun_struct *tun)
 598 {
 599         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 600         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 601 }
 602
 603 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 604 {
 605         tfile->detached = tun;
 606         list_add_tail(&tfile->next, &tun->disabled);
 607         ++tun->numdisabled;
 608 }
 609
 610 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 611 {
 612         struct tun_struct *tun = tfile->detached;
 613
 614         tfile->detached = NULL;
 615         list_del_init(&tfile->next);
 616         --tun->numdisabled;
 617         return tun;
 618 }
 619
 620 void tun_ptr_free(void *ptr)
 621 {
 622         if (!ptr)
 623                 return;
 624         if (tun_is_xdp_frame(ptr)) {
 625                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 626
 627                 xdp_return_frame(xdpf);
 628         } else {
 629                 __skb_array_destroy_skb(ptr);
 630         }
 631 }
 632 EXPORT_SYMBOL_GPL(tun_ptr_free);
 633
 634 static void tun_queue_purge(struct tun_file *tfile)
 635 {
 636         void *ptr;
 637
 638         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 639                 tun_ptr_free(ptr);
 640
 641         skb_queue_purge(&tfile->sk.sk_write_queue);
 642         skb_queue_purge(&tfile->sk.sk_error_queue);
 643 }
 644
 645 static void __tun_detach(struct tun_file *tfile, bool clean)
 646 {
 647         struct tun_file *ntfile;
 648         struct tun_struct *tun;
 649
 650         tun = rtnl_dereference(tfile->tun);
 651
 652         if (tun && clean) {
 653                 tun_napi_disable(tfile);
 654                 tun_napi_del(tfile);
 655         }
 656
 657         if (tun && !tfile->detached) {
 658                 u16 index = tfile->queue_index;
 659                 BUG_ON(index >= tun->numqueues);
 660
 661                 rcu_assign_pointer(tun->tfiles[index],
 662                                    tun->tfiles[tun->numqueues - 1]);
 663                 ntfile = rtnl_dereference(tun->tfiles[index]);
 664                 ntfile->queue_index = index;
 665                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 666                                    NULL);
 667
 668                 --tun->numqueues;
 669                 if (clean) {
 670                         RCU_INIT_POINTER(tfile->tun, NULL);
 671                         sock_put(&tfile->sk);
 672                 } else
 673                         tun_disable_queue(tun, tfile);
 674
 675                 synchronize_net();
 676                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 677                 /* Drop read queue */
 678                 tun_queue_purge(tfile);
 679                 tun_set_real_num_queues(tun);
 680         } else if (tfile->detached && clean) {
 681                 tun = tun_enable_queue(tfile);
 682                 sock_put(&tfile->sk);
 683         }
 684
 685         if (clean) {
 686                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 687                         netif_carrier_off(tun->dev);
 688
 689                         if (!(tun->flags & IFF_PERSIST) &&
 690                             tun->dev->reg_state == NETREG_REGISTERED)
 691                                 unregister_netdevice(tun->dev);
 692                 }
 693                 if (tun)
 694                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 695                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 696                 sock_put(&tfile->sk);
 697         }
 698 }
 699
 700 static void tun_detach(struct tun_file *tfile, bool clean)
 701 {
 702         struct tun_struct *tun;
 703         struct net_device *dev;
 704
 705         rtnl_lock();
 706         tun = rtnl_dereference(tfile->tun);
 707         dev = tun ? tun->dev : NULL;
 708         __tun_detach(tfile, clean);
 709         if (dev)
 710                 netdev_state_change(dev);
 711         rtnl_unlock();
 712 }
 713
 714 static void tun_detach_all(struct net_device *dev)
 715 {
 716         struct tun_struct *tun = netdev_priv(dev);
 717         struct tun_file *tfile, *tmp;
 718         int i, n = tun->numqueues;
 719
 720         for (i = 0; i < n; i++) {
 721                 tfile = rtnl_dereference(tun->tfiles[i]);
 722                 BUG_ON(!tfile);
 723                 tun_napi_disable(tfile);
 724                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 725                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 726                 RCU_INIT_POINTER(tfile->tun, NULL);
 727                 --tun->numqueues;
 728         }
 729         list_for_each_entry(tfile, &tun->disabled, next) {
 730                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 731                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 732                 RCU_INIT_POINTER(tfile->tun, NULL);
 733         }
 734         BUG_ON(tun->numqueues != 0);
 735
 736         synchronize_net();
 737         for (i = 0; i < n; i++) {
 738                 tfile = rtnl_dereference(tun->tfiles[i]);
 739                 tun_napi_del(tfile);
 740                 /* Drop read queue */
 741                 tun_queue_purge(tfile);
 742                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 743                 sock_put(&tfile->sk);
 744         }
 745         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 746                 tun_enable_queue(tfile);
 747                 tun_queue_purge(tfile);
 748                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 749                 sock_put(&tfile->sk);
 750         }
 751         BUG_ON(tun->numdisabled != 0);
 752
 753         if (tun->flags & IFF_PERSIST)
 754                 module_put(THIS_MODULE);
 755 }
 756
 757 static int tun_attach(struct tun_struct *tun, struct file *file,
 758                       bool skip_filter, bool napi, bool napi_frags,
 759                       bool publish_tun)
 760 {
 761         struct tun_file *tfile = file->private_data;
 762         struct net_device *dev = tun->dev;
 763         int err;
 764
 765         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 766         if (err < 0)
 767                 goto out;
 768
 769         err = -EINVAL;
 770         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 771                 goto out;
 772
 773         err = -EBUSY;
 774         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 775                 goto out;
 776
 777         err = -E2BIG;
 778         if (!tfile->detached &&
 779             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 780                 goto out;
 781
 782         err = 0;
 783
 784         /* Re-attach the filter to persist device */
 785         if (!skip_filter && (tun->filter_attached == true)) {
 786                 lock_sock(tfile->socket.sk);
 787                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 788                 release_sock(tfile->socket.sk);
 789                 if (!err)
 790                         goto out;
 791         }
 792
 793         if (!tfile->detached &&
 794             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 795                             GFP_KERNEL, tun_ptr_free)) {
 796                 err = -ENOMEM;
 797                 goto out;
 798         }
 799
 800         tfile->queue_index = tun->numqueues;
 801         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 802
 803         if (tfile->detached) {
 804                 /* Re-attach detached tfile, updating XDP queue_index */
 805                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 806
 807                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 808                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 809         } else {
 810                 /* Setup XDP RX-queue info, for new tfile getting attached */
 811                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 812                                        tun->dev, tfile->queue_index);
 813                 if (err < 0)
 814                         goto out;
 815                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 816                                                  MEM_TYPE_PAGE_SHARED, NULL);
 817                 if (err < 0) {
 818                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 819                         goto out;
 820                 }
 821                 err = 0;
 822         }
 823
 824         if (tfile->detached) {
 825                 tun_enable_queue(tfile);
 826         } else {
 827                 sock_hold(&tfile->sk);
 828                 tun_napi_init(tun, tfile, napi, napi_frags);
 829         }
 830
 831         if (rtnl_dereference(tun->xdp_prog))
 832                 sock_set_flag(&tfile->sk, SOCK_XDP);
 833
 834         /* device is allowed to go away first, so no need to hold extra
 835          * refcnt.
 836          */
 837
 838         /* Publish tfile->tun and tun->tfiles only after we've fully
 839          * initialized tfile; otherwise we risk using half-initialized
 840          * object.
 841          */
 842         if (publish_tun)
 843                 rcu_assign_pointer(tfile->tun, tun);
 844         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 845         tun->numqueues++;
 846         tun_set_real_num_queues(tun);
 847 out:
 848         return err;
 849 }
 850
 851 static struct tun_struct *tun_get(struct tun_file *tfile)
 852 {
 853         struct tun_struct *tun;
 854
 855         rcu_read_lock();
 856         tun = rcu_dereference(tfile->tun);
 857         if (tun)
 858                 dev_hold(tun->dev);
 859         rcu_read_unlock();
 860
 861         return tun;
 862 }
 863
 864 static void tun_put(struct tun_struct *tun)
 865 {
 866         dev_put(tun->dev);
 867 }
 868
 869 /* TAP filtering */
 870 static void addr_hash_set(u32 *mask, const u8 *addr)
 871 {
 872         int n = ether_crc(ETH_ALEN, addr) >> 26;
 873         mask[n >> 5] |= (1 << (n & 31));
 874 }
 875
 876 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 877 {
 878         int n = ether_crc(ETH_ALEN, addr) >> 26;
 879         return mask[n >> 5] & (1 << (n & 31));
 880 }
 881
 882 static int update_filter(struct tap_filter *filter, void __user *arg)
 883 {
 884         struct { u8 u[ETH_ALEN]; } *addr;
 885         struct tun_filter uf;
 886         int err, alen, n, nexact;
 887
 888         if (copy_from_user(&uf, arg, sizeof(uf)))
 889                 return -EFAULT;
 890
 891         if (!uf.count) {
 892                 /* Disabled */
 893                 filter->count = 0;
 894                 return 0;
 895         }
 896
 897         alen = ETH_ALEN * uf.count;
 898         addr = memdup_user(arg + sizeof(uf), alen);
 899         if (IS_ERR(addr))
 900                 return PTR_ERR(addr);
 901
 902         /* The filter is updated without holding any locks. Which is
 903          * perfectly safe. We disable it first and in the worst
 904          * case we'll accept a few undesired packets. */
 905         filter->count = 0;
 906         wmb();
 907
 908         /* Use first set of addresses as an exact filter */
 909         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 910                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 911
 912         nexact = n;
 913
 914         /* Remaining multicast addresses are hashed,
 915          * unicast will leave the filter disabled. */
 916         memset(filter->mask, 0, sizeof(filter->mask));
 917         for (; n < uf.count; n++) {
 918                 if (!is_multicast_ether_addr(addr[n].u)) {
 919                         err = 0; /* no filter */
 920                         goto free_addr;
 921                 }
 922                 addr_hash_set(filter->mask, addr[n].u);
 923         }
 924
 925         /* For ALLMULTI just set the mask to all ones.
 926          * This overrides the mask populated above. */
 927         if ((uf.flags & TUN_FLT_ALLMULTI))
 928                 memset(filter->mask, ~0, sizeof(filter->mask));
 929
 930         /* Now enable the filter */
 931         wmb();
 932         filter->count = nexact;
 933
 934         /* Return the number of exact filters */
 935         err = nexact;
 936 free_addr:
 937         kfree(addr);
 938         return err;
 939 }
 940
 941 /* Returns: 0 - drop, !=0 - accept */
 942 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 943 {
 944         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 945          * at this point. */
 946         struct ethhdr *eh = (struct ethhdr *) skb->data;
 947         int i;
 948
 949         /* Exact match */
 950         for (i = 0; i < filter->count; i++)
 951                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 952                         return 1;
 953
 954         /* Inexact match (multicast only) */
 955         if (is_multicast_ether_addr(eh->h_dest))
 956                 return addr_hash_test(filter->mask, eh->h_dest);
 957
 958         return 0;
 959 }
 960
 961 /*
 962  * Checks whether the packet is accepted or not.
 963  * Returns: 0 - drop, !=0 - accept
 964  */
 965 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 966 {
 967         if (!filter->count)
 968                 return 1;
 969
 970         return run_filter(filter, skb);
 971 }
 972
 973 /* Network device part of the driver */
 974
 975 static const struct ethtool_ops tun_ethtool_ops;
 976
 977 /* Net device detach from fd. */
 978 static void tun_net_uninit(struct net_device *dev)
 979 {
 980         tun_detach_all(dev);
 981 }
 982
 983 /* Net device open. */
 984 static int tun_net_open(struct net_device *dev)
 985 {
 986         netif_tx_start_all_queues(dev);
 987
 988         return 0;
 989 }
 990
 991 /* Net device close. */
 992 static int tun_net_close(struct net_device *dev)
 993 {
 994         netif_tx_stop_all_queues(dev);
 995         return 0;
 996 }
 997
 998 /* Net device start xmit */
 999 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1000 {
1001 #ifdef CONFIG_RPS
1002         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1003                 /* Select queue was not called for the skbuff, so we extract the
1004                  * RPS hash and save it into the flow_table here.
1005                  */
1006                 struct tun_flow_entry *e;
1007                 __u32 rxhash;
1008
1009                 rxhash = __skb_get_hash_symmetric(skb);
1010                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1011                 if (e)
1012                         tun_flow_save_rps_rxhash(e, rxhash);
1013         }
1014 #endif
1015 }
1016
1017 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1018                                     struct sk_buff *skb,
1019                                     int len)
1020 {
1021         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1022
1023         if (prog)
1024                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1025
1026         return len;
1027 }
1028
1029 /* Net device start xmit */
1030 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1031 {
1032         struct tun_struct *tun = netdev_priv(dev);
1033         int txq = skb->queue_mapping;
1034         struct tun_file *tfile;
1035         int len = skb->len;
1036
1037         rcu_read_lock();
1038         tfile = rcu_dereference(tun->tfiles[txq]);
1039
1040         /* Drop packet if interface is not attached */
1041         if (!tfile)
1042                 goto drop;
1043
1044         if (!rcu_dereference(tun->steering_prog))
1045                 tun_automq_xmit(tun, skb);
1046
1047         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1048
1049         /* Drop if the filter does not like it.
1050          * This is a noop if the filter is disabled.
1051          * Filter can be enabled only for the TAP devices. */
1052         if (!check_filter(&tun->txflt, skb))
1053                 goto drop;
1054
1055         if (tfile->socket.sk->sk_filter &&
1056             sk_filter(tfile->socket.sk, skb))
1057                 goto drop;
1058
1059         len = run_ebpf_filter(tun, skb, len);
1060         if (len == 0 || pskb_trim(skb, len))
1061                 goto drop;
1062
1063         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1064                 goto drop;
1065
1066         skb_tx_timestamp(skb);
1067
1068         /* Orphan the skb - required as we might hang on to it
1069          * for indefinite time.
1070          */
1071         skb_orphan(skb);
1072
1073         nf_reset_ct(skb);
1074
1075         if (ptr_ring_produce(&tfile->tx_ring, skb))
1076                 goto drop;
1077
1078         /* Notify and wake up reader process */
1079         if (tfile->flags & TUN_FASYNC)
1080                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1081         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1082
1083         rcu_read_unlock();
1084         return NETDEV_TX_OK;
1085
1086 drop:
1087         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1088         skb_tx_error(skb);
1089         kfree_skb(skb);
1090         rcu_read_unlock();
1091         return NET_XMIT_DROP;
1092 }
1093
1094 static void tun_net_mclist(struct net_device *dev)
1095 {
1096         /*
1097          * This callback is supposed to deal with mc filter in
1098          * _rx_ path and has nothing to do with the _tx_ path.
1099          * In rx path we always accept everything userspace gives us.
1100          */
1101 }
1102
1103 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1104         netdev_features_t features)
1105 {
1106         struct tun_struct *tun = netdev_priv(dev);
1107
1108         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1109 }
1110
1111 static void tun_set_headroom(struct net_device *dev, int new_hr)
1112 {
1113         struct tun_struct *tun = netdev_priv(dev);
1114
1115         if (new_hr < NET_SKB_PAD)
1116                 new_hr = NET_SKB_PAD;
1117
1118         tun->align = new_hr;
1119 }
1120
1121 static void
1122 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1123 {
1124         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1125         struct tun_struct *tun = netdev_priv(dev);
1126         struct tun_pcpu_stats *p;
1127         int i;
1128
1129         for_each_possible_cpu(i) {
1130                 u64 rxpackets, rxbytes, txpackets, txbytes;
1131                 unsigned int start;
1132
1133                 p = per_cpu_ptr(tun->pcpu_stats, i);
1134                 do {
1135                         start = u64_stats_fetch_begin(&p->syncp);
1136                         rxpackets       = u64_stats_read(&p->rx_packets);
1137                         rxbytes         = u64_stats_read(&p->rx_bytes);
1138                         txpackets       = u64_stats_read(&p->tx_packets);
1139                         txbytes         = u64_stats_read(&p->tx_bytes);
1140                 } while (u64_stats_fetch_retry(&p->syncp, start));
1141
1142                 stats->rx_packets       += rxpackets;
1143                 stats->rx_bytes         += rxbytes;
1144                 stats->tx_packets       += txpackets;
1145                 stats->tx_bytes         += txbytes;
1146
1147                 /* u32 counters */
1148                 rx_dropped      += p->rx_dropped;
1149                 rx_frame_errors += p->rx_frame_errors;
1150                 tx_dropped      += p->tx_dropped;
1151         }
1152         stats->rx_dropped  = rx_dropped;
1153         stats->rx_frame_errors = rx_frame_errors;
1154         stats->tx_dropped = tx_dropped;
1155 }
1156
1157 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1158                        struct netlink_ext_ack *extack)
1159 {
1160         struct tun_struct *tun = netdev_priv(dev);
1161         struct tun_file *tfile;
1162         struct bpf_prog *old_prog;
1163         int i;
1164
1165         old_prog = rtnl_dereference(tun->xdp_prog);
1166         rcu_assign_pointer(tun->xdp_prog, prog);
1167         if (old_prog)
1168                 bpf_prog_put(old_prog);
1169
1170         for (i = 0; i < tun->numqueues; i++) {
1171                 tfile = rtnl_dereference(tun->tfiles[i]);
1172                 if (prog)
1173                         sock_set_flag(&tfile->sk, SOCK_XDP);
1174                 else
1175                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1176         }
1177         list_for_each_entry(tfile, &tun->disabled, next) {
1178                 if (prog)
1179                         sock_set_flag(&tfile->sk, SOCK_XDP);
1180                 else
1181                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1182         }
1183
1184         return 0;
1185 }
1186
1187 static u32 tun_xdp_query(struct net_device *dev)
1188 {
1189         struct tun_struct *tun = netdev_priv(dev);
1190         const struct bpf_prog *xdp_prog;
1191
1192         xdp_prog = rtnl_dereference(tun->xdp_prog);
1193         if (xdp_prog)
1194                 return xdp_prog->aux->id;
1195
1196         return 0;
1197 }
1198
1199 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1200 {
1201         switch (xdp->command) {
1202         case XDP_SETUP_PROG:
1203                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1204         case XDP_QUERY_PROG:
1205                 xdp->prog_id = tun_xdp_query(dev);
1206                 return 0;
1207         default:
1208                 return -EINVAL;
1209         }
1210 }
1211
1212 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1213 {
1214         if (new_carrier) {
1215                 struct tun_struct *tun = netdev_priv(dev);
1216
1217                 if (!tun->numqueues)
1218                         return -EPERM;
1219
1220                 netif_carrier_on(dev);
1221         } else {
1222                 netif_carrier_off(dev);
1223         }
1224         return 0;
1225 }
1226
1227 static const struct net_device_ops tun_netdev_ops = {
1228         .ndo_uninit             = tun_net_uninit,
1229         .ndo_open               = tun_net_open,
1230         .ndo_stop               = tun_net_close,
1231         .ndo_start_xmit         = tun_net_xmit,
1232         .ndo_fix_features       = tun_net_fix_features,
1233         .ndo_select_queue       = tun_select_queue,
1234         .ndo_set_rx_headroom    = tun_set_headroom,
1235         .ndo_get_stats64        = tun_net_get_stats64,
1236         .ndo_change_carrier     = tun_net_change_carrier,
1237 };
1238
1239 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1240 {
1241         /* Notify and wake up reader process */
1242         if (tfile->flags & TUN_FASYNC)
1243                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1244         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1245 }
1246
1247 static int tun_xdp_xmit(struct net_device *dev, int n,
1248                         struct xdp_frame **frames, u32 flags)
1249 {
1250         struct tun_struct *tun = netdev_priv(dev);
1251         struct tun_file *tfile;
1252         u32 numqueues;
1253         int drops = 0;
1254         int cnt = n;
1255         int i;
1256
1257         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1258                 return -EINVAL;
1259
1260         rcu_read_lock();
1261
1262 resample:
1263         numqueues = READ_ONCE(tun->numqueues);
1264         if (!numqueues) {
1265                 rcu_read_unlock();
1266                 return -ENXIO; /* Caller will free/return all frames */
1267         }
1268
1269         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1270                                             numqueues]);
1271         if (unlikely(!tfile))
1272                 goto resample;
1273
1274         spin_lock(&tfile->tx_ring.producer_lock);
1275         for (i = 0; i < n; i++) {
1276                 struct xdp_frame *xdp = frames[i];
1277                 /* Encode the XDP flag into lowest bit for consumer to differ
1278                  * XDP buffer from sk_buff.
1279                  */
1280                 void *frame = tun_xdp_to_ptr(xdp);
1281
1282                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1283                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1284                         xdp_return_frame_rx_napi(xdp);
1285                         drops++;
1286                 }
1287         }
1288         spin_unlock(&tfile->tx_ring.producer_lock);
1289
1290         if (flags & XDP_XMIT_FLUSH)
1291                 __tun_xdp_flush_tfile(tfile);
1292
1293         rcu_read_unlock();
1294         return cnt - drops;
1295 }
1296
1297 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1298 {
1299         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1300
1301         if (unlikely(!frame))
1302                 return -EOVERFLOW;
1303
1304         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1305 }
1306
1307 static const struct net_device_ops tap_netdev_ops = {
1308         .ndo_uninit             = tun_net_uninit,
1309         .ndo_open               = tun_net_open,
1310         .ndo_stop               = tun_net_close,
1311         .ndo_start_xmit         = tun_net_xmit,
1312         .ndo_fix_features       = tun_net_fix_features,
1313         .ndo_set_rx_mode        = tun_net_mclist,
1314         .ndo_set_mac_address    = eth_mac_addr,
1315         .ndo_validate_addr      = eth_validate_addr,
1316         .ndo_select_queue       = tun_select_queue,
1317         .ndo_features_check     = passthru_features_check,
1318         .ndo_set_rx_headroom    = tun_set_headroom,
1319         .ndo_get_stats64        = tun_net_get_stats64,
1320         .ndo_bpf                = tun_xdp,
1321         .ndo_xdp_xmit           = tun_xdp_xmit,
1322         .ndo_change_carrier     = tun_net_change_carrier,
1323 };
1324
1325 static void tun_flow_init(struct tun_struct *tun)
1326 {
1327         int i;
1328
1329         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1330                 INIT_HLIST_HEAD(&tun->flows[i]);
1331
1332         tun->ageing_time = TUN_FLOW_EXPIRE;
1333         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1334         mod_timer(&tun->flow_gc_timer,
1335                   round_jiffies_up(jiffies + tun->ageing_time));
1336 }
1337
1338 static void tun_flow_uninit(struct tun_struct *tun)
1339 {
1340         del_timer_sync(&tun->flow_gc_timer);
1341         tun_flow_flush(tun);
1342 }
1343
1344 #define MIN_MTU 68
1345 #define MAX_MTU 65535
1346
1347 /* Initialize net device. */
1348 static void tun_net_init(struct net_device *dev)
1349 {
1350         struct tun_struct *tun = netdev_priv(dev);
1351
1352         switch (tun->flags & TUN_TYPE_MASK) {
1353         case IFF_TUN:
1354                 dev->netdev_ops = &tun_netdev_ops;
1355                 dev->header_ops = &ip_tunnel_header_ops;
1356
1357                 /* Point-to-Point TUN Device */
1358                 dev->hard_header_len = 0;
1359                 dev->addr_len = 0;
1360                 dev->mtu = 1500;
1361
1362                 /* Zero header length */
1363                 dev->type = ARPHRD_NONE;
1364                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1365                 break;
1366
1367         case IFF_TAP:
1368                 dev->netdev_ops = &tap_netdev_ops;
1369                 /* Ethernet TAP Device */
1370                 ether_setup(dev);
1371                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1372                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1373
1374                 eth_hw_addr_random(dev);
1375
1376                 break;
1377         }
1378
1379         dev->min_mtu = MIN_MTU;
1380         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1381 }
1382
1383 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1384 {
1385         struct sock *sk = tfile->socket.sk;
1386
1387         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1388 }
1389
1390 /* Character device part */
1391
1392 /* Poll */
1393 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1394 {
1395         struct tun_file *tfile = file->private_data;
1396         struct tun_struct *tun = tun_get(tfile);
1397         struct sock *sk;
1398         __poll_t mask = 0;
1399
1400         if (!tun)
1401                 return EPOLLERR;
1402
1403         sk = tfile->socket.sk;
1404
1405         poll_wait(file, sk_sleep(sk), wait);
1406
1407         if (!ptr_ring_empty(&tfile->tx_ring))
1408                 mask |= EPOLLIN | EPOLLRDNORM;
1409
1410         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1411          * guarantee EPOLLOUT to be raised by either here or
1412          * tun_sock_write_space(). Then process could get notification
1413          * after it writes to a down device and meets -EIO.
1414          */
1415         if (tun_sock_writeable(tun, tfile) ||
1416             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1417              tun_sock_writeable(tun, tfile)))
1418                 mask |= EPOLLOUT | EPOLLWRNORM;
1419
1420         if (tun->dev->reg_state != NETREG_REGISTERED)
1421                 mask = EPOLLERR;
1422
1423         tun_put(tun);
1424         return mask;
1425 }
1426
1427 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1428                                             size_t len,
1429                                             const struct iov_iter *it)
1430 {
1431         struct sk_buff *skb;
1432         size_t linear;
1433         int err;
1434         int i;
1435
1436         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1437                 return ERR_PTR(-ENOMEM);
1438
1439         local_bh_disable();
1440         skb = napi_get_frags(&tfile->napi);
1441         local_bh_enable();
1442         if (!skb)
1443                 return ERR_PTR(-ENOMEM);
1444
1445         linear = iov_iter_single_seg_count(it);
1446         err = __skb_grow(skb, linear);
1447         if (err)
1448                 goto free;
1449
1450         skb->len = len;
1451         skb->data_len = len - linear;
1452         skb->truesize += skb->data_len;
1453
1454         for (i = 1; i < it->nr_segs; i++) {
1455                 size_t fragsz = it->iov[i].iov_len;
1456                 struct page *page;
1457                 void *frag;
1458
1459                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1460                         err = -EINVAL;
1461                         goto free;
1462                 }
1463                 frag = netdev_alloc_frag(fragsz);
1464                 if (!frag) {
1465                         err = -ENOMEM;
1466                         goto free;
1467                 }
1468                 page = virt_to_head_page(frag);
1469                 skb_fill_page_desc(skb, i - 1, page,
1470                                    frag - page_address(page), fragsz);
1471         }
1472
1473         return skb;
1474 free:
1475         /* frees skb and all frags allocated with napi_alloc_frag() */
1476         napi_free_frags(&tfile->napi);
1477         return ERR_PTR(err);
1478 }
1479
1480 /* prepad is the amount to reserve at front.  len is length after that.
1481  * linear is a hint as to how much to copy (usually headers). */
1482 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1483                                      size_t prepad, size_t len,
1484                                      size_t linear, int noblock)
1485 {
1486         struct sock *sk = tfile->socket.sk;
1487         struct sk_buff *skb;
1488         int err;
1489
1490         /* Under a page?  Don't bother with paged skb. */
1491         if (prepad + len < PAGE_SIZE || !linear)
1492                 linear = len;
1493
1494         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1495                                    &err, 0);
1496         if (!skb)
1497                 return ERR_PTR(err);
1498
1499         skb_reserve(skb, prepad);
1500         skb_put(skb, linear);
1501         skb->data_len = len - linear;
1502         skb->len += len - linear;
1503
1504         return skb;
1505 }
1506
1507 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1508                            struct sk_buff *skb, int more)
1509 {
1510         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1511         struct sk_buff_head process_queue;
1512         u32 rx_batched = tun->rx_batched;
1513         bool rcv = false;
1514
1515         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1516                 local_bh_disable();
1517                 skb_record_rx_queue(skb, tfile->queue_index);
1518                 netif_receive_skb(skb);
1519                 local_bh_enable();
1520                 return;
1521         }
1522
1523         spin_lock(&queue->lock);
1524         if (!more || skb_queue_len(queue) == rx_batched) {
1525                 __skb_queue_head_init(&process_queue);
1526                 skb_queue_splice_tail_init(queue, &process_queue);
1527                 rcv = true;
1528         } else {
1529                 __skb_queue_tail(queue, skb);
1530         }
1531         spin_unlock(&queue->lock);
1532
1533         if (rcv) {
1534                 struct sk_buff *nskb;
1535
1536                 local_bh_disable();
1537                 while ((nskb = __skb_dequeue(&process_queue))) {
1538                         skb_record_rx_queue(nskb, tfile->queue_index);
1539                         netif_receive_skb(nskb);
1540                 }
1541                 skb_record_rx_queue(skb, tfile->queue_index);
1542                 netif_receive_skb(skb);
1543                 local_bh_enable();
1544         }
1545 }
1546
1547 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1548                               int len, int noblock, bool zerocopy)
1549 {
1550         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1551                 return false;
1552
1553         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1554                 return false;
1555
1556         if (!noblock)
1557                 return false;
1558
1559         if (zerocopy)
1560                 return false;
1561
1562         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1563             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1564                 return false;
1565
1566         return true;
1567 }
1568
1569 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1570                                        struct page_frag *alloc_frag, char *buf,
1571                                        int buflen, int len, int pad)
1572 {
1573         struct sk_buff *skb = build_skb(buf, buflen);
1574
1575         if (!skb)
1576                 return ERR_PTR(-ENOMEM);
1577
1578         skb_reserve(skb, pad);
1579         skb_put(skb, len);
1580         skb_set_owner_w(skb, tfile->socket.sk);
1581
1582         get_page(alloc_frag->page);
1583         alloc_frag->offset += buflen;
1584
1585         return skb;
1586 }
1587
1588 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1589                        struct xdp_buff *xdp, u32 act)
1590 {
1591         int err;
1592
1593         switch (act) {
1594         case XDP_REDIRECT:
1595                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1596                 if (err)
1597                         return err;
1598                 break;
1599         case XDP_TX:
1600                 err = tun_xdp_tx(tun->dev, xdp);
1601                 if (err < 0)
1602                         return err;
1603                 break;
1604         case XDP_PASS:
1605                 break;
1606         default:
1607                 bpf_warn_invalid_xdp_action(act);
1608                 /* fall through */
1609         case XDP_ABORTED:
1610                 trace_xdp_exception(tun->dev, xdp_prog, act);
1611                 /* fall through */
1612         case XDP_DROP:
1613                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1614                 break;
1615         }
1616
1617         return act;
1618 }
1619
1620 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1621                                      struct tun_file *tfile,
1622                                      struct iov_iter *from,
1623                                      struct virtio_net_hdr *hdr,
1624                                      int len, int *skb_xdp)
1625 {
1626         struct page_frag *alloc_frag = &current->task_frag;
1627         struct bpf_prog *xdp_prog;
1628         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1629         char *buf;
1630         size_t copied;
1631         int pad = TUN_RX_PAD;
1632         int err = 0;
1633
1634         rcu_read_lock();
1635         xdp_prog = rcu_dereference(tun->xdp_prog);
1636         if (xdp_prog)
1637                 pad += XDP_PACKET_HEADROOM;
1638         buflen += SKB_DATA_ALIGN(len + pad);
1639         rcu_read_unlock();
1640
1641         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1642         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1643                 return ERR_PTR(-ENOMEM);
1644
1645         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1646         copied = copy_page_from_iter(alloc_frag->page,
1647                                      alloc_frag->offset + pad,
1648                                      len, from);
1649         if (copied != len)
1650                 return ERR_PTR(-EFAULT);
1651
1652         /* There's a small window that XDP may be set after the check
1653          * of xdp_prog above, this should be rare and for simplicity
1654          * we do XDP on skb in case the headroom is not enough.
1655          */
1656         if (hdr->gso_type || !xdp_prog) {
1657                 *skb_xdp = 1;
1658                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1659                                        pad);
1660         }
1661
1662         *skb_xdp = 0;
1663
1664         local_bh_disable();
1665         rcu_read_lock();
1666         xdp_prog = rcu_dereference(tun->xdp_prog);
1667         if (xdp_prog) {
1668                 struct xdp_buff xdp;
1669                 u32 act;
1670
1671                 xdp.data_hard_start = buf;
1672                 xdp.data = buf + pad;
1673                 xdp_set_data_meta_invalid(&xdp);
1674                 xdp.data_end = xdp.data + len;
1675                 xdp.rxq = &tfile->xdp_rxq;
1676                 xdp.frame_sz = buflen;
1677
1678                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1679                 if (act == XDP_REDIRECT || act == XDP_TX) {
1680                         get_page(alloc_frag->page);
1681                         alloc_frag->offset += buflen;
1682                 }
1683                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1684                 if (err < 0) {
1685                         if (act == XDP_REDIRECT || act == XDP_TX)
1686                                 put_page(alloc_frag->page);
1687                         goto out;
1688                 }
1689
1690                 if (err == XDP_REDIRECT)
1691                         xdp_do_flush();
1692                 if (err != XDP_PASS)
1693                         goto out;
1694
1695                 pad = xdp.data - xdp.data_hard_start;
1696                 len = xdp.data_end - xdp.data;
1697         }
1698         rcu_read_unlock();
1699         local_bh_enable();
1700
1701         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1702
1703 out:
1704         rcu_read_unlock();
1705         local_bh_enable();
1706         return NULL;
1707 }
1708
1709 /* Get packet from user space buffer */
1710 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1711                             void *msg_control, struct iov_iter *from,
1712                             int noblock, bool more)
1713 {
1714         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1715         struct sk_buff *skb;
1716         size_t total_len = iov_iter_count(from);
1717         size_t len = total_len, align = tun->align, linear;
1718         struct virtio_net_hdr gso = { 0 };
1719         struct tun_pcpu_stats *stats;
1720         int good_linear;
1721         int copylen;
1722         bool zerocopy = false;
1723         int err;
1724         u32 rxhash = 0;
1725         int skb_xdp = 1;
1726         bool frags = tun_napi_frags_enabled(tfile);
1727
1728         if (!(tun->flags & IFF_NO_PI)) {
1729                 if (len < sizeof(pi))
1730                         return -EINVAL;
1731                 len -= sizeof(pi);
1732
1733                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1734                         return -EFAULT;
1735         }
1736
1737         if (tun->flags & IFF_VNET_HDR) {
1738                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1739
1740                 if (len < vnet_hdr_sz)
1741                         return -EINVAL;
1742                 len -= vnet_hdr_sz;
1743
1744                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1745                         return -EFAULT;
1746
1747                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1748                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1749                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1750
1751                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1752                         return -EINVAL;
1753                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1754         }
1755
1756         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1757                 align += NET_IP_ALIGN;
1758                 if (unlikely(len < ETH_HLEN ||
1759                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1760                         return -EINVAL;
1761         }
1762
1763         good_linear = SKB_MAX_HEAD(align);
1764
1765         if (msg_control) {
1766                 struct iov_iter i = *from;
1767
1768                 /* There are 256 bytes to be copied in skb, so there is
1769                  * enough room for skb expand head in case it is used.
1770                  * The rest of the buffer is mapped from userspace.
1771                  */
1772                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1773                 if (copylen > good_linear)
1774                         copylen = good_linear;
1775                 linear = copylen;
1776                 iov_iter_advance(&i, copylen);
1777                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1778                         zerocopy = true;
1779         }
1780
1781         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1782                 /* For the packet that is not easy to be processed
1783                  * (e.g gso or jumbo packet), we will do it at after
1784                  * skb was created with generic XDP routine.
1785                  */
1786                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1787                 if (IS_ERR(skb)) {
1788                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1789                         return PTR_ERR(skb);
1790                 }
1791                 if (!skb)
1792                         return total_len;
1793         } else {
1794                 if (!zerocopy) {
1795                         copylen = len;
1796                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1797                                 linear = good_linear;
1798                         else
1799                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1800                 }
1801
1802                 if (frags) {
1803                         mutex_lock(&tfile->napi_mutex);
1804                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1805                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1806                          * If zerocopy is enabled, then this layout will be
1807                          * overwritten by zerocopy_sg_from_iter().
1808                          */
1809                         zerocopy = false;
1810                 } else {
1811                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1812                                             noblock);
1813                 }
1814
1815                 if (IS_ERR(skb)) {
1816                         if (PTR_ERR(skb) != -EAGAIN)
1817                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1818                         if (frags)
1819                                 mutex_unlock(&tfile->napi_mutex);
1820                         return PTR_ERR(skb);
1821                 }
1822
1823                 if (zerocopy)
1824                         err = zerocopy_sg_from_iter(skb, from);
1825                 else
1826                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1827
1828                 if (err) {
1829                         err = -EFAULT;
1830 drop:
1831                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1832                         kfree_skb(skb);
1833                         if (frags) {
1834                                 tfile->napi.skb = NULL;
1835                                 mutex_unlock(&tfile->napi_mutex);
1836                         }
1837
1838                         return err;
1839                 }
1840         }
1841
1842         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1843                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1844                 kfree_skb(skb);
1845                 if (frags) {
1846                         tfile->napi.skb = NULL;
1847                         mutex_unlock(&tfile->napi_mutex);
1848                 }
1849
1850                 return -EINVAL;
1851         }
1852
1853         switch (tun->flags & TUN_TYPE_MASK) {
1854         case IFF_TUN:
1855                 if (tun->flags & IFF_NO_PI) {
1856                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1857
1858                         switch (ip_version) {
1859                         case 4:
1860                                 pi.proto = htons(ETH_P_IP);
1861                                 break;
1862                         case 6:
1863                                 pi.proto = htons(ETH_P_IPV6);
1864                                 break;
1865                         default:
1866                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1867                                 kfree_skb(skb);
1868                                 return -EINVAL;
1869                         }
1870                 }
1871
1872                 skb_reset_mac_header(skb);
1873                 skb->protocol = pi.proto;
1874                 skb->dev = tun->dev;
1875                 break;
1876         case IFF_TAP:
1877                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1878                         err = -ENOMEM;
1879                         goto drop;
1880                 }
1881                 skb->protocol = eth_type_trans(skb, tun->dev);
1882                 break;
1883         }
1884
1885         /* copy skb_ubuf_info for callback when skb has no error */
1886         if (zerocopy) {
1887                 skb_shinfo(skb)->destructor_arg = msg_control;
1888                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1889                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1890         } else if (msg_control) {
1891                 struct ubuf_info *uarg = msg_control;
1892                 uarg->callback(uarg, false);
1893         }
1894
1895         skb_reset_network_header(skb);
1896         skb_probe_transport_header(skb);
1897         skb_record_rx_queue(skb, tfile->queue_index);
1898
1899         if (skb_xdp) {
1900                 struct bpf_prog *xdp_prog;
1901                 int ret;
1902
1903                 local_bh_disable();
1904                 rcu_read_lock();
1905                 xdp_prog = rcu_dereference(tun->xdp_prog);
1906                 if (xdp_prog) {
1907                         ret = do_xdp_generic(xdp_prog, skb);
1908                         if (ret != XDP_PASS) {
1909                                 rcu_read_unlock();
1910                                 local_bh_enable();
1911                                 if (frags) {
1912                                         tfile->napi.skb = NULL;
1913                                         mutex_unlock(&tfile->napi_mutex);
1914                                 }
1915                                 return total_len;
1916                         }
1917                 }
1918                 rcu_read_unlock();
1919                 local_bh_enable();
1920         }
1921
1922         /* Compute the costly rx hash only if needed for flow updates.
1923          * We may get a very small possibility of OOO during switching, not
1924          * worth to optimize.
1925          */
1926         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1927             !tfile->detached)
1928                 rxhash = __skb_get_hash_symmetric(skb);
1929
1930         rcu_read_lock();
1931         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1932                 err = -EIO;
1933                 rcu_read_unlock();
1934                 goto drop;
1935         }
1936
1937         if (frags) {
1938                 u32 headlen;
1939
1940                 /* Exercise flow dissector code path. */
1941                 skb_push(skb, ETH_HLEN);
1942                 headlen = eth_get_headlen(tun->dev, skb->data,
1943                                           skb_headlen(skb));
1944
1945                 if (unlikely(headlen > skb_headlen(skb))) {
1946                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1947                         napi_free_frags(&tfile->napi);
1948                         rcu_read_unlock();
1949                         mutex_unlock(&tfile->napi_mutex);
1950                         WARN_ON(1);
1951                         return -ENOMEM;
1952                 }
1953
1954                 local_bh_disable();
1955                 napi_gro_frags(&tfile->napi);
1956                 local_bh_enable();
1957                 mutex_unlock(&tfile->napi_mutex);
1958         } else if (tfile->napi_enabled) {
1959                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1960                 int queue_len;
1961
1962                 spin_lock_bh(&queue->lock);
1963                 __skb_queue_tail(queue, skb);
1964                 queue_len = skb_queue_len(queue);
1965                 spin_unlock(&queue->lock);
1966
1967                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1968                         napi_schedule(&tfile->napi);
1969
1970                 local_bh_enable();
1971         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1972                 tun_rx_batched(tun, tfile, skb, more);
1973         } else {
1974                 netif_rx_ni(skb);
1975         }
1976         rcu_read_unlock();
1977
1978         stats = get_cpu_ptr(tun->pcpu_stats);
1979         u64_stats_update_begin(&stats->syncp);
1980         u64_stats_inc(&stats->rx_packets);
1981         u64_stats_add(&stats->rx_bytes, len);
1982         u64_stats_update_end(&stats->syncp);
1983         put_cpu_ptr(stats);
1984
1985         if (rxhash)
1986                 tun_flow_update(tun, rxhash, tfile);
1987
1988         return total_len;
1989 }
1990
1991 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1992 {
1993         struct file *file = iocb->ki_filp;
1994         struct tun_file *tfile = file->private_data;
1995         struct tun_struct *tun = tun_get(tfile);
1996         ssize_t result;
1997
1998         if (!tun)
1999                 return -EBADFD;
2000
2001         result = tun_get_user(tun, tfile, NULL, from,
2002                               file->f_flags & O_NONBLOCK, false);
2003
2004         tun_put(tun);
2005         return result;
2006 }
2007
2008 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2009                                 struct tun_file *tfile,
2010                                 struct xdp_frame *xdp_frame,
2011                                 struct iov_iter *iter)
2012 {
2013         int vnet_hdr_sz = 0;
2014         size_t size = xdp_frame->len;
2015         struct tun_pcpu_stats *stats;
2016         size_t ret;
2017
2018         if (tun->flags & IFF_VNET_HDR) {
2019                 struct virtio_net_hdr gso = { 0 };
2020
2021                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2022                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2023                         return -EINVAL;
2024                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2025                              sizeof(gso)))
2026                         return -EFAULT;
2027                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2028         }
2029
2030         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2031
2032         stats = get_cpu_ptr(tun->pcpu_stats);
2033         u64_stats_update_begin(&stats->syncp);
2034         u64_stats_inc(&stats->tx_packets);
2035         u64_stats_add(&stats->tx_bytes, ret);
2036         u64_stats_update_end(&stats->syncp);
2037         put_cpu_ptr(tun->pcpu_stats);
2038
2039         return ret;
2040 }
2041
2042 /* Put packet to the user space buffer */
2043 static ssize_t tun_put_user(struct tun_struct *tun,
2044                             struct tun_file *tfile,
2045                             struct sk_buff *skb,
2046                             struct iov_iter *iter)
2047 {
2048         struct tun_pi pi = { 0, skb->protocol };
2049         struct tun_pcpu_stats *stats;
2050         ssize_t total;
2051         int vlan_offset = 0;
2052         int vlan_hlen = 0;
2053         int vnet_hdr_sz = 0;
2054
2055         if (skb_vlan_tag_present(skb))
2056                 vlan_hlen = VLAN_HLEN;
2057
2058         if (tun->flags & IFF_VNET_HDR)
2059                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2060
2061         total = skb->len + vlan_hlen + vnet_hdr_sz;
2062
2063         if (!(tun->flags & IFF_NO_PI)) {
2064                 if (iov_iter_count(iter) < sizeof(pi))
2065                         return -EINVAL;
2066
2067                 total += sizeof(pi);
2068                 if (iov_iter_count(iter) < total) {
2069                         /* Packet will be striped */
2070                         pi.flags |= TUN_PKT_STRIP;
2071                 }
2072
2073                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2074                         return -EFAULT;
2075         }
2076
2077         if (vnet_hdr_sz) {
2078                 struct virtio_net_hdr gso;
2079
2080                 if (iov_iter_count(iter) < vnet_hdr_sz)
2081                         return -EINVAL;
2082
2083                 if (virtio_net_hdr_from_skb(skb, &gso,
2084                                             tun_is_little_endian(tun), true,
2085                                             vlan_hlen)) {
2086                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2087                         pr_err("unexpected GSO type: "
2088                                "0x%x, gso_size %d, hdr_len %d\n",
2089                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2090                                tun16_to_cpu(tun, gso.hdr_len));
2091                         print_hex_dump(KERN_ERR, "tun: ",
2092                                        DUMP_PREFIX_NONE,
2093                                        16, 1, skb->head,
2094                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2095                         WARN_ON_ONCE(1);
2096                         return -EINVAL;
2097                 }
2098
2099                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2100                         return -EFAULT;
2101
2102                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2103         }
2104
2105         if (vlan_hlen) {
2106                 int ret;
2107                 struct veth veth;
2108
2109                 veth.h_vlan_proto = skb->vlan_proto;
2110                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2111
2112                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2113
2114                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2115                 if (ret || !iov_iter_count(iter))
2116                         goto done;
2117
2118                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2119                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2120                         goto done;
2121         }
2122
2123         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2124
2125 done:
2126         /* caller is in process context, */
2127         stats = get_cpu_ptr(tun->pcpu_stats);
2128         u64_stats_update_begin(&stats->syncp);
2129         u64_stats_inc(&stats->tx_packets);
2130         u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2131         u64_stats_update_end(&stats->syncp);
2132         put_cpu_ptr(tun->pcpu_stats);
2133
2134         return total;
2135 }
2136
2137 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2138 {
2139         DECLARE_WAITQUEUE(wait, current);
2140         void *ptr = NULL;
2141         int error = 0;
2142
2143         ptr = ptr_ring_consume(&tfile->tx_ring);
2144         if (ptr)
2145                 goto out;
2146         if (noblock) {
2147                 error = -EAGAIN;
2148                 goto out;
2149         }
2150
2151         add_wait_queue(&tfile->socket.wq.wait, &wait);
2152
2153         while (1) {
2154                 set_current_state(TASK_INTERRUPTIBLE);
2155                 ptr = ptr_ring_consume(&tfile->tx_ring);
2156                 if (ptr)
2157                         break;
2158                 if (signal_pending(current)) {
2159                         error = -ERESTARTSYS;
2160                         break;
2161                 }
2162                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2163                         error = -EFAULT;
2164                         break;
2165                 }
2166
2167                 schedule();
2168         }
2169
2170         __set_current_state(TASK_RUNNING);
2171         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2172
2173 out:
2174         *err = error;
2175         return ptr;
2176 }
2177
2178 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2179                            struct iov_iter *to,
2180                            int noblock, void *ptr)
2181 {
2182         ssize_t ret;
2183         int err;
2184
2185         if (!iov_iter_count(to)) {
2186                 tun_ptr_free(ptr);
2187                 return 0;
2188         }
2189
2190         if (!ptr) {
2191                 /* Read frames from ring */
2192                 ptr = tun_ring_recv(tfile, noblock, &err);
2193                 if (!ptr)
2194                         return err;
2195         }
2196
2197         if (tun_is_xdp_frame(ptr)) {
2198                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2199
2200                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2201                 xdp_return_frame(xdpf);
2202         } else {
2203                 struct sk_buff *skb = ptr;
2204
2205                 ret = tun_put_user(tun, tfile, skb, to);
2206                 if (unlikely(ret < 0))
2207                         kfree_skb(skb);
2208                 else
2209                         consume_skb(skb);
2210         }
2211
2212         return ret;
2213 }
2214
2215 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2216 {
2217         struct file *file = iocb->ki_filp;
2218         struct tun_file *tfile = file->private_data;
2219         struct tun_struct *tun = tun_get(tfile);
2220         ssize_t len = iov_iter_count(to), ret;
2221
2222         if (!tun)
2223                 return -EBADFD;
2224         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2225         ret = min_t(ssize_t, ret, len);
2226         if (ret > 0)
2227                 iocb->ki_pos = ret;
2228         tun_put(tun);
2229         return ret;
2230 }
2231
2232 static void tun_prog_free(struct rcu_head *rcu)
2233 {
2234         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2235
2236         bpf_prog_destroy(prog->prog);
2237         kfree(prog);
2238 }
2239
2240 static int __tun_set_ebpf(struct tun_struct *tun,
2241                           struct tun_prog __rcu **prog_p,
2242                           struct bpf_prog *prog)
2243 {
2244         struct tun_prog *old, *new = NULL;
2245
2246         if (prog) {
2247                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2248                 if (!new)
2249                         return -ENOMEM;
2250                 new->prog = prog;
2251         }
2252
2253         spin_lock_bh(&tun->lock);
2254         old = rcu_dereference_protected(*prog_p,
2255                                         lockdep_is_held(&tun->lock));
2256         rcu_assign_pointer(*prog_p, new);
2257         spin_unlock_bh(&tun->lock);
2258
2259         if (old)
2260                 call_rcu(&old->rcu, tun_prog_free);
2261
2262         return 0;
2263 }
2264
2265 static void tun_free_netdev(struct net_device *dev)
2266 {
2267         struct tun_struct *tun = netdev_priv(dev);
2268
2269         BUG_ON(!(list_empty(&tun->disabled)));
2270
2271         free_percpu(tun->pcpu_stats);
2272         /* We clear pcpu_stats so that tun_set_iff() can tell if
2273          * tun_free_netdev() has been called from register_netdevice().
2274          */
2275         tun->pcpu_stats = NULL;
2276
2277         tun_flow_uninit(tun);
2278         security_tun_dev_free_security(tun->security);
2279         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2280         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2281 }
2282
2283 static void tun_setup(struct net_device *dev)
2284 {
2285         struct tun_struct *tun = netdev_priv(dev);
2286
2287         tun->owner = INVALID_UID;
2288         tun->group = INVALID_GID;
2289         tun_default_link_ksettings(dev, &tun->link_ksettings);
2290
2291         dev->ethtool_ops = &tun_ethtool_ops;
2292         dev->needs_free_netdev = true;
2293         dev->priv_destructor = tun_free_netdev;
2294         /* We prefer our own queue length */
2295         dev->tx_queue_len = TUN_READQ_SIZE;
2296 }
2297
2298 /* Trivial set of netlink ops to allow deleting tun or tap
2299  * device with netlink.
2300  */
2301 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2302                         struct netlink_ext_ack *extack)
2303 {
2304         NL_SET_ERR_MSG(extack,
2305                        "tun/tap creation via rtnetlink is not supported.");
2306         return -EOPNOTSUPP;
2307 }
2308
2309 static size_t tun_get_size(const struct net_device *dev)
2310 {
2311         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2312         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2313
2314         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2315                nla_total_size(sizeof(gid_t)) + /* GROUP */
2316                nla_total_size(sizeof(u8)) + /* TYPE */
2317                nla_total_size(sizeof(u8)) + /* PI */
2318                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2319                nla_total_size(sizeof(u8)) + /* PERSIST */
2320                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2321                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2322                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2323                0;
2324 }
2325
2326 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2327 {
2328         struct tun_struct *tun = netdev_priv(dev);
2329
2330         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2331                 goto nla_put_failure;
2332         if (uid_valid(tun->owner) &&
2333             nla_put_u32(skb, IFLA_TUN_OWNER,
2334                         from_kuid_munged(current_user_ns(), tun->owner)))
2335                 goto nla_put_failure;
2336         if (gid_valid(tun->group) &&
2337             nla_put_u32(skb, IFLA_TUN_GROUP,
2338                         from_kgid_munged(current_user_ns(), tun->group)))
2339                 goto nla_put_failure;
2340         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2341                 goto nla_put_failure;
2342         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2343                 goto nla_put_failure;
2344         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2345                 goto nla_put_failure;
2346         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2347                        !!(tun->flags & IFF_MULTI_QUEUE)))
2348                 goto nla_put_failure;
2349         if (tun->flags & IFF_MULTI_QUEUE) {
2350                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2351                         goto nla_put_failure;
2352                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2353                                 tun->numdisabled))
2354                         goto nla_put_failure;
2355         }
2356
2357         return 0;
2358
2359 nla_put_failure:
2360         return -EMSGSIZE;
2361 }
2362
2363 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2364         .kind           = DRV_NAME,
2365         .priv_size      = sizeof(struct tun_struct),
2366         .setup          = tun_setup,
2367         .validate       = tun_validate,
2368         .get_size       = tun_get_size,
2369         .fill_info      = tun_fill_info,
2370 };
2371
2372 static void tun_sock_write_space(struct sock *sk)
2373 {
2374         struct tun_file *tfile;
2375         wait_queue_head_t *wqueue;
2376
2377         if (!sock_writeable(sk))
2378                 return;
2379
2380         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2381                 return;
2382
2383         wqueue = sk_sleep(sk);
2384         if (wqueue && waitqueue_active(wqueue))
2385                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2386                                                 EPOLLWRNORM | EPOLLWRBAND);
2387
2388         tfile = container_of(sk, struct tun_file, sk);
2389         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2390 }
2391
2392 static void tun_put_page(struct tun_page *tpage)
2393 {
2394         if (tpage->page)
2395                 __page_frag_cache_drain(tpage->page, tpage->count);
2396 }
2397
2398 static int tun_xdp_one(struct tun_struct *tun,
2399                        struct tun_file *tfile,
2400                        struct xdp_buff *xdp, int *flush,
2401                        struct tun_page *tpage)
2402 {
2403         unsigned int datasize = xdp->data_end - xdp->data;
2404         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2405         struct virtio_net_hdr *gso = &hdr->gso;
2406         struct tun_pcpu_stats *stats;
2407         struct bpf_prog *xdp_prog;
2408         struct sk_buff *skb = NULL;
2409         u32 rxhash = 0, act;
2410         int buflen = hdr->buflen;
2411         int err = 0;
2412         bool skb_xdp = false;
2413         struct page *page;
2414
2415         xdp_prog = rcu_dereference(tun->xdp_prog);
2416         if (xdp_prog) {
2417                 if (gso->gso_type) {
2418                         skb_xdp = true;
2419                         goto build;
2420                 }
2421                 xdp_set_data_meta_invalid(xdp);
2422                 xdp->rxq = &tfile->xdp_rxq;
2423                 xdp->frame_sz = buflen;
2424
2425                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2426                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2427                 if (err < 0) {
2428                         put_page(virt_to_head_page(xdp->data));
2429                         return err;
2430                 }
2431
2432                 switch (err) {
2433                 case XDP_REDIRECT:
2434                         *flush = true;
2435                         /* fall through */
2436                 case XDP_TX:
2437                         return 0;
2438                 case XDP_PASS:
2439                         break;
2440                 default:
2441                         page = virt_to_head_page(xdp->data);
2442                         if (tpage->page == page) {
2443                                 ++tpage->count;
2444                         } else {
2445                                 tun_put_page(tpage);
2446                                 tpage->page = page;
2447                                 tpage->count = 1;
2448                         }
2449                         return 0;
2450                 }
2451         }
2452
2453 build:
2454         skb = build_skb(xdp->data_hard_start, buflen);
2455         if (!skb) {
2456                 err = -ENOMEM;
2457                 goto out;
2458         }
2459
2460         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2461         skb_put(skb, xdp->data_end - xdp->data);
2462
2463         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2464                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2465                 kfree_skb(skb);
2466                 err = -EINVAL;
2467                 goto out;
2468         }
2469
2470         skb->protocol = eth_type_trans(skb, tun->dev);
2471         skb_reset_network_header(skb);
2472         skb_probe_transport_header(skb);
2473         skb_record_rx_queue(skb, tfile->queue_index);
2474
2475         if (skb_xdp) {
2476                 err = do_xdp_generic(xdp_prog, skb);
2477                 if (err != XDP_PASS)
2478                         goto out;
2479         }
2480
2481         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2482             !tfile->detached)
2483                 rxhash = __skb_get_hash_symmetric(skb);
2484
2485         netif_receive_skb(skb);
2486
2487         /* No need for get_cpu_ptr() here since this function is
2488          * always called with bh disabled
2489          */
2490         stats = this_cpu_ptr(tun->pcpu_stats);
2491         u64_stats_update_begin(&stats->syncp);
2492         u64_stats_inc(&stats->rx_packets);
2493         u64_stats_add(&stats->rx_bytes, datasize);
2494         u64_stats_update_end(&stats->syncp);
2495
2496         if (rxhash)
2497                 tun_flow_update(tun, rxhash, tfile);
2498
2499 out:
2500         return err;
2501 }
2502
2503 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2504 {
2505         int ret, i;
2506         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2507         struct tun_struct *tun = tun_get(tfile);
2508         struct tun_msg_ctl *ctl = m->msg_control;
2509         struct xdp_buff *xdp;
2510
2511         if (!tun)
2512                 return -EBADFD;
2513
2514         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2515                 struct tun_page tpage;
2516                 int n = ctl->num;
2517                 int flush = 0;
2518
2519                 memset(&tpage, 0, sizeof(tpage));
2520
2521                 local_bh_disable();
2522                 rcu_read_lock();
2523
2524                 for (i = 0; i < n; i++) {
2525                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2526                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2527                 }
2528
2529                 if (flush)
2530                         xdp_do_flush();
2531
2532                 rcu_read_unlock();
2533                 local_bh_enable();
2534
2535                 tun_put_page(&tpage);
2536
2537                 ret = total_len;
2538                 goto out;
2539         }
2540
2541         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2542                            m->msg_flags & MSG_DONTWAIT,
2543                            m->msg_flags & MSG_MORE);
2544 out:
2545         tun_put(tun);
2546         return ret;
2547 }
2548
2549 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2550                        int flags)
2551 {
2552         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2553         struct tun_struct *tun = tun_get(tfile);
2554         void *ptr = m->msg_control;
2555         int ret;
2556
2557         if (!tun) {
2558                 ret = -EBADFD;
2559                 goto out_free;
2560         }
2561
2562         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2563                 ret = -EINVAL;
2564                 goto out_put_tun;
2565         }
2566         if (flags & MSG_ERRQUEUE) {
2567                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2568                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2569                 goto out;
2570         }
2571         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2572         if (ret > (ssize_t)total_len) {
2573                 m->msg_flags |= MSG_TRUNC;
2574                 ret = flags & MSG_TRUNC ? ret : total_len;
2575         }
2576 out:
2577         tun_put(tun);
2578         return ret;
2579
2580 out_put_tun:
2581         tun_put(tun);
2582 out_free:
2583         tun_ptr_free(ptr);
2584         return ret;
2585 }
2586
2587 static int tun_ptr_peek_len(void *ptr)
2588 {
2589         if (likely(ptr)) {
2590                 if (tun_is_xdp_frame(ptr)) {
2591                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2592
2593                         return xdpf->len;
2594                 }
2595                 return __skb_array_len_with_tag(ptr);
2596         } else {
2597                 return 0;
2598         }
2599 }
2600
2601 static int tun_peek_len(struct socket *sock)
2602 {
2603         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2604         struct tun_struct *tun;
2605         int ret = 0;
2606
2607         tun = tun_get(tfile);
2608         if (!tun)
2609                 return 0;
2610
2611         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2612         tun_put(tun);
2613
2614         return ret;
2615 }
2616
2617 /* Ops structure to mimic raw sockets with tun */
2618 static const struct proto_ops tun_socket_ops = {
2619         .peek_len = tun_peek_len,
2620         .sendmsg = tun_sendmsg,
2621         .recvmsg = tun_recvmsg,
2622 };
2623
2624 static struct proto tun_proto = {
2625         .name           = "tun",
2626         .owner          = THIS_MODULE,
2627         .obj_size       = sizeof(struct tun_file),
2628 };
2629
2630 static int tun_flags(struct tun_struct *tun)
2631 {
2632         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2633 }
2634
2635 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2636                               char *buf)
2637 {
2638         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2639         return sprintf(buf, "0x%x\n", tun_flags(tun));
2640 }
2641
2642 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2643                               char *buf)
2644 {
2645         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2646         return uid_valid(tun->owner)?
2647                 sprintf(buf, "%u\n",
2648                         from_kuid_munged(current_user_ns(), tun->owner)):
2649                 sprintf(buf, "-1\n");
2650 }
2651
2652 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2653                               char *buf)
2654 {
2655         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2656         return gid_valid(tun->group) ?
2657                 sprintf(buf, "%u\n",
2658                         from_kgid_munged(current_user_ns(), tun->group)):
2659                 sprintf(buf, "-1\n");
2660 }
2661
2662 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2663 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2664 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2665
2666 static struct attribute *tun_dev_attrs[] = {
2667         &dev_attr_tun_flags.attr,
2668         &dev_attr_owner.attr,
2669         &dev_attr_group.attr,
2670         NULL
2671 };
2672
2673 static const struct attribute_group tun_attr_group = {
2674         .attrs = tun_dev_attrs
2675 };
2676
2677 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2678 {
2679         struct tun_struct *tun;
2680         struct tun_file *tfile = file->private_data;
2681         struct net_device *dev;
2682         int err;
2683
2684         if (tfile->detached)
2685                 return -EINVAL;
2686
2687         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2688                 if (!capable(CAP_NET_ADMIN))
2689                         return -EPERM;
2690
2691                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2692                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2693                         return -EINVAL;
2694         }
2695
2696         dev = __dev_get_by_name(net, ifr->ifr_name);
2697         if (dev) {
2698                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2699                         return -EBUSY;
2700                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2701                         tun = netdev_priv(dev);
2702                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2703                         tun = netdev_priv(dev);
2704                 else
2705                         return -EINVAL;
2706
2707                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2708                     !!(tun->flags & IFF_MULTI_QUEUE))
2709                         return -EINVAL;
2710
2711                 if (tun_not_capable(tun))
2712                         return -EPERM;
2713                 err = security_tun_dev_open(tun->security);
2714                 if (err < 0)
2715                         return err;
2716
2717                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2718                                  ifr->ifr_flags & IFF_NAPI,
2719                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2720                 if (err < 0)
2721                         return err;
2722
2723                 if (tun->flags & IFF_MULTI_QUEUE &&
2724                     (tun->numqueues + tun->numdisabled > 1)) {
2725                         /* One or more queue has already been attached, no need
2726                          * to initialize the device again.
2727                          */
2728                         netdev_state_change(dev);
2729                         return 0;
2730                 }
2731
2732                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2733                               (ifr->ifr_flags & TUN_FEATURES);
2734
2735                 netdev_state_change(dev);
2736         } else {
2737                 char *name;
2738                 unsigned long flags = 0;
2739                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2740                              MAX_TAP_QUEUES : 1;
2741
2742                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2743                         return -EPERM;
2744                 err = security_tun_dev_create();
2745                 if (err < 0)
2746                         return err;
2747
2748                 /* Set dev type */
2749                 if (ifr->ifr_flags & IFF_TUN) {
2750                         /* TUN device */
2751                         flags |= IFF_TUN;
2752                         name = "tun%d";
2753                 } else if (ifr->ifr_flags & IFF_TAP) {
2754                         /* TAP device */
2755                         flags |= IFF_TAP;
2756                         name = "tap%d";
2757                 } else
2758                         return -EINVAL;
2759
2760                 if (*ifr->ifr_name)
2761                         name = ifr->ifr_name;
2762
2763                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2764                                        NET_NAME_UNKNOWN, tun_setup, queues,
2765                                        queues);
2766
2767                 if (!dev)
2768                         return -ENOMEM;
2769
2770                 dev_net_set(dev, net);
2771                 dev->rtnl_link_ops = &tun_link_ops;
2772                 dev->ifindex = tfile->ifindex;
2773                 dev->sysfs_groups[0] = &tun_attr_group;
2774
2775                 tun = netdev_priv(dev);
2776                 tun->dev = dev;
2777                 tun->flags = flags;
2778                 tun->txflt.count = 0;
2779                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2780
2781                 tun->align = NET_SKB_PAD;
2782                 tun->filter_attached = false;
2783                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2784                 tun->rx_batched = 0;
2785                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2786
2787                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2788                 if (!tun->pcpu_stats) {
2789                         err = -ENOMEM;
2790                         goto err_free_dev;
2791                 }
2792
2793                 spin_lock_init(&tun->lock);
2794
2795                 err = security_tun_dev_alloc_security(&tun->security);
2796                 if (err < 0)
2797                         goto err_free_stat;
2798
2799                 tun_net_init(dev);
2800                 tun_flow_init(tun);
2801
2802                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2803                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2804                                    NETIF_F_HW_VLAN_STAG_TX;
2805                 dev->features = dev->hw_features | NETIF_F_LLTX;
2806                 dev->vlan_features = dev->features &
2807                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2808                                        NETIF_F_HW_VLAN_STAG_TX);
2809
2810                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2811                               (ifr->ifr_flags & TUN_FEATURES);
2812
2813                 INIT_LIST_HEAD(&tun->disabled);
2814                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2815                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2816                 if (err < 0)
2817                         goto err_free_flow;
2818
2819                 err = register_netdevice(tun->dev);
2820                 if (err < 0)
2821                         goto err_detach;
2822                 /* free_netdev() won't check refcnt, to aovid race
2823                  * with dev_put() we need publish tun after registration.
2824                  */
2825                 rcu_assign_pointer(tfile->tun, tun);
2826         }
2827
2828         netif_carrier_on(tun->dev);
2829
2830         /* Make sure persistent devices do not get stuck in
2831          * xoff state.
2832          */
2833         if (netif_running(tun->dev))
2834                 netif_tx_wake_all_queues(tun->dev);
2835
2836         strcpy(ifr->ifr_name, tun->dev->name);
2837         return 0;
2838
2839 err_detach:
2840         tun_detach_all(dev);
2841         /* We are here because register_netdevice() has failed.
2842          * If register_netdevice() already called tun_free_netdev()
2843          * while dealing with the error, tun->pcpu_stats has been cleared.
2844          */
2845         if (!tun->pcpu_stats)
2846                 goto err_free_dev;
2847
2848 err_free_flow:
2849         tun_flow_uninit(tun);
2850         security_tun_dev_free_security(tun->security);
2851 err_free_stat:
2852         free_percpu(tun->pcpu_stats);
2853 err_free_dev:
2854         free_netdev(dev);
2855         return err;
2856 }
2857
2858 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2859 {
2860         strcpy(ifr->ifr_name, tun->dev->name);
2861
2862         ifr->ifr_flags = tun_flags(tun);
2863
2864 }
2865
2866 /* This is like a cut-down ethtool ops, except done via tun fd so no
2867  * privs required. */
2868 static int set_offload(struct tun_struct *tun, unsigned long arg)
2869 {
2870         netdev_features_t features = 0;
2871
2872         if (arg & TUN_F_CSUM) {
2873                 features |= NETIF_F_HW_CSUM;
2874                 arg &= ~TUN_F_CSUM;
2875
2876                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2877                         if (arg & TUN_F_TSO_ECN) {
2878                                 features |= NETIF_F_TSO_ECN;
2879                                 arg &= ~TUN_F_TSO_ECN;
2880                         }
2881                         if (arg & TUN_F_TSO4)
2882                                 features |= NETIF_F_TSO;
2883                         if (arg & TUN_F_TSO6)
2884                                 features |= NETIF_F_TSO6;
2885                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2886                 }
2887
2888                 arg &= ~TUN_F_UFO;
2889         }
2890
2891         /* This gives the user a way to test for new features in future by
2892          * trying to set them. */
2893         if (arg)
2894                 return -EINVAL;
2895
2896         tun->set_features = features;
2897         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2898         tun->dev->wanted_features |= features;
2899         netdev_update_features(tun->dev);
2900
2901         return 0;
2902 }
2903
2904 static void tun_detach_filter(struct tun_struct *tun, int n)
2905 {
2906         int i;
2907         struct tun_file *tfile;
2908
2909         for (i = 0; i < n; i++) {
2910                 tfile = rtnl_dereference(tun->tfiles[i]);
2911                 lock_sock(tfile->socket.sk);
2912                 sk_detach_filter(tfile->socket.sk);
2913                 release_sock(tfile->socket.sk);
2914         }
2915
2916         tun->filter_attached = false;
2917 }
2918
2919 static int tun_attach_filter(struct tun_struct *tun)
2920 {
2921         int i, ret = 0;
2922         struct tun_file *tfile;
2923
2924         for (i = 0; i < tun->numqueues; i++) {
2925                 tfile = rtnl_dereference(tun->tfiles[i]);
2926                 lock_sock(tfile->socket.sk);
2927                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2928                 release_sock(tfile->socket.sk);
2929                 if (ret) {
2930                         tun_detach_filter(tun, i);
2931                         return ret;
2932                 }
2933         }
2934
2935         tun->filter_attached = true;
2936         return ret;
2937 }
2938
2939 static void tun_set_sndbuf(struct tun_struct *tun)
2940 {
2941         struct tun_file *tfile;
2942         int i;
2943
2944         for (i = 0; i < tun->numqueues; i++) {
2945                 tfile = rtnl_dereference(tun->tfiles[i]);
2946                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2947         }
2948 }
2949
2950 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2951 {
2952         struct tun_file *tfile = file->private_data;
2953         struct tun_struct *tun;
2954         int ret = 0;
2955
2956         rtnl_lock();
2957
2958         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2959                 tun = tfile->detached;
2960                 if (!tun) {
2961                         ret = -EINVAL;
2962                         goto unlock;
2963                 }
2964                 ret = security_tun_dev_attach_queue(tun->security);
2965                 if (ret < 0)
2966                         goto unlock;
2967                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2968                                  tun->flags & IFF_NAPI_FRAGS, true);
2969         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2970                 tun = rtnl_dereference(tfile->tun);
2971                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2972                         ret = -EINVAL;
2973                 else
2974                         __tun_detach(tfile, false);
2975         } else
2976                 ret = -EINVAL;
2977
2978         if (ret >= 0)
2979                 netdev_state_change(tun->dev);
2980
2981 unlock:
2982         rtnl_unlock();
2983         return ret;
2984 }
2985
2986 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2987                         void __user *data)
2988 {
2989         struct bpf_prog *prog;
2990         int fd;
2991
2992         if (copy_from_user(&fd, data, sizeof(fd)))
2993                 return -EFAULT;
2994
2995         if (fd == -1) {
2996                 prog = NULL;
2997         } else {
2998                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2999                 if (IS_ERR(prog))
3000                         return PTR_ERR(prog);
3001         }
3002
3003         return __tun_set_ebpf(tun, prog_p, prog);
3004 }
3005
3006 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3007                             unsigned long arg, int ifreq_len)
3008 {
3009         struct tun_file *tfile = file->private_data;
3010         struct net *net = sock_net(&tfile->sk);
3011         struct tun_struct *tun;
3012         void __user* argp = (void __user*)arg;
3013         unsigned int ifindex, carrier;
3014         struct ifreq ifr;
3015         kuid_t owner;
3016         kgid_t group;
3017         int sndbuf;
3018         int vnet_hdr_sz;
3019         int le;
3020         int ret;
3021         bool do_notify = false;
3022
3023         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3024             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3025                 if (copy_from_user(&ifr, argp, ifreq_len))
3026                         return -EFAULT;
3027         } else {
3028                 memset(&ifr, 0, sizeof(ifr));
3029         }
3030         if (cmd == TUNGETFEATURES) {
3031                 /* Currently this just means: "what IFF flags are valid?".
3032                  * This is needed because we never checked for invalid flags on
3033                  * TUNSETIFF.
3034                  */
3035                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3036                                 (unsigned int __user*)argp);
3037         } else if (cmd == TUNSETQUEUE) {
3038                 return tun_set_queue(file, &ifr);
3039         } else if (cmd == SIOCGSKNS) {
3040                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3041                         return -EPERM;
3042                 return open_related_ns(&net->ns, get_net_ns);
3043         }
3044
3045         ret = 0;
3046         rtnl_lock();
3047
3048         tun = tun_get(tfile);
3049         if (cmd == TUNSETIFF) {
3050                 ret = -EEXIST;
3051                 if (tun)
3052                         goto unlock;
3053
3054                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3055
3056                 ret = tun_set_iff(net, file, &ifr);
3057
3058                 if (ret)
3059                         goto unlock;
3060
3061                 if (copy_to_user(argp, &ifr, ifreq_len))
3062                         ret = -EFAULT;
3063                 goto unlock;
3064         }
3065         if (cmd == TUNSETIFINDEX) {
3066                 ret = -EPERM;
3067                 if (tun)
3068                         goto unlock;
3069
3070                 ret = -EFAULT;
3071                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3072                         goto unlock;
3073
3074                 ret = 0;
3075                 tfile->ifindex = ifindex;
3076                 goto unlock;
3077         }
3078
3079         ret = -EBADFD;
3080         if (!tun)
3081                 goto unlock;
3082
3083         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3084
3085         net = dev_net(tun->dev);
3086         ret = 0;
3087         switch (cmd) {
3088         case TUNGETIFF:
3089                 tun_get_iff(tun, &ifr);
3090
3091                 if (tfile->detached)
3092                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3093                 if (!tfile->socket.sk->sk_filter)
3094                         ifr.ifr_flags |= IFF_NOFILTER;
3095
3096                 if (copy_to_user(argp, &ifr, ifreq_len))
3097                         ret = -EFAULT;
3098                 break;
3099
3100         case TUNSETNOCSUM:
3101                 /* Disable/Enable checksum */
3102
3103                 /* [unimplemented] */
3104                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3105                            arg ? "disabled" : "enabled");
3106                 break;
3107
3108         case TUNSETPERSIST:
3109                 /* Disable/Enable persist mode. Keep an extra reference to the
3110                  * module to prevent the module being unprobed.
3111                  */
3112                 if (arg && !(tun->flags & IFF_PERSIST)) {
3113                         tun->flags |= IFF_PERSIST;
3114                         __module_get(THIS_MODULE);
3115                         do_notify = true;
3116                 }
3117                 if (!arg && (tun->flags & IFF_PERSIST)) {
3118                         tun->flags &= ~IFF_PERSIST;
3119                         module_put(THIS_MODULE);
3120                         do_notify = true;
3121                 }
3122
3123                 netif_info(tun, drv, tun->dev, "persist %s\n",
3124                            arg ? "enabled" : "disabled");
3125                 break;
3126
3127         case TUNSETOWNER:
3128                 /* Set owner of the device */
3129                 owner = make_kuid(current_user_ns(), arg);
3130                 if (!uid_valid(owner)) {
3131                         ret = -EINVAL;
3132                         break;
3133                 }
3134                 tun->owner = owner;
3135                 do_notify = true;
3136                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3137                            from_kuid(&init_user_ns, tun->owner));
3138                 break;
3139
3140         case TUNSETGROUP:
3141                 /* Set group of the device */
3142                 group = make_kgid(current_user_ns(), arg);
3143                 if (!gid_valid(group)) {
3144                         ret = -EINVAL;
3145                         break;
3146                 }
3147                 tun->group = group;
3148                 do_notify = true;
3149                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3150                            from_kgid(&init_user_ns, tun->group));
3151                 break;
3152
3153         case TUNSETLINK:
3154                 /* Only allow setting the type when the interface is down */
3155                 if (tun->dev->flags & IFF_UP) {
3156                         netif_info(tun, drv, tun->dev,
3157                                    "Linktype set failed because interface is up\n");
3158                         ret = -EBUSY;
3159                 } else {
3160                         tun->dev->type = (int) arg;
3161                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3162                                    tun->dev->type);
3163                         ret = 0;
3164                 }
3165                 break;
3166
3167         case TUNSETDEBUG:
3168                 tun->msg_enable = (u32)arg;
3169                 break;
3170
3171         case TUNSETOFFLOAD:
3172                 ret = set_offload(tun, arg);
3173                 break;
3174
3175         case TUNSETTXFILTER:
3176                 /* Can be set only for TAPs */
3177                 ret = -EINVAL;
3178                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3179                         break;
3180                 ret = update_filter(&tun->txflt, (void __user *)arg);
3181                 break;
3182
3183         case SIOCGIFHWADDR:
3184                 /* Get hw address */
3185                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3186                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3187                 if (copy_to_user(argp, &ifr, ifreq_len))
3188                         ret = -EFAULT;
3189                 break;
3190
3191         case SIOCSIFHWADDR:
3192                 /* Set hw address */
3193                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3194                 break;
3195
3196         case TUNGETSNDBUF:
3197                 sndbuf = tfile->socket.sk->sk_sndbuf;
3198                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3199                         ret = -EFAULT;
3200                 break;
3201
3202         case TUNSETSNDBUF:
3203                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3204                         ret = -EFAULT;
3205                         break;
3206                 }
3207                 if (sndbuf <= 0) {
3208                         ret = -EINVAL;
3209                         break;
3210                 }
3211
3212                 tun->sndbuf = sndbuf;
3213                 tun_set_sndbuf(tun);
3214                 break;
3215
3216         case TUNGETVNETHDRSZ:
3217                 vnet_hdr_sz = tun->vnet_hdr_sz;
3218                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3219                         ret = -EFAULT;
3220                 break;
3221
3222         case TUNSETVNETHDRSZ:
3223                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3224                         ret = -EFAULT;
3225                         break;
3226                 }
3227                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3228                         ret = -EINVAL;
3229                         break;
3230                 }
3231
3232                 tun->vnet_hdr_sz = vnet_hdr_sz;
3233                 break;
3234
3235         case TUNGETVNETLE:
3236                 le = !!(tun->flags & TUN_VNET_LE);
3237                 if (put_user(le, (int __user *)argp))
3238                         ret = -EFAULT;
3239                 break;
3240
3241         case TUNSETVNETLE:
3242                 if (get_user(le, (int __user *)argp)) {
3243                         ret = -EFAULT;
3244                         break;
3245                 }
3246                 if (le)
3247                         tun->flags |= TUN_VNET_LE;
3248                 else
3249                         tun->flags &= ~TUN_VNET_LE;
3250                 break;
3251
3252         case TUNGETVNETBE:
3253                 ret = tun_get_vnet_be(tun, argp);
3254                 break;
3255
3256         case TUNSETVNETBE:
3257                 ret = tun_set_vnet_be(tun, argp);
3258                 break;
3259
3260         case TUNATTACHFILTER:
3261                 /* Can be set only for TAPs */
3262                 ret = -EINVAL;
3263                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3264                         break;
3265                 ret = -EFAULT;
3266                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3267                         break;
3268
3269                 ret = tun_attach_filter(tun);
3270                 break;
3271
3272         case TUNDETACHFILTER:
3273                 /* Can be set only for TAPs */
3274                 ret = -EINVAL;
3275                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3276                         break;
3277                 ret = 0;
3278                 tun_detach_filter(tun, tun->numqueues);
3279                 break;
3280
3281         case TUNGETFILTER:
3282                 ret = -EINVAL;
3283                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3284                         break;
3285                 ret = -EFAULT;
3286                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3287                         break;
3288                 ret = 0;
3289                 break;
3290
3291         case TUNSETSTEERINGEBPF:
3292                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3293                 break;
3294
3295         case TUNSETFILTEREBPF:
3296                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3297                 break;
3298
3299         case TUNSETCARRIER:
3300                 ret = -EFAULT;
3301                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3302                         goto unlock;
3303
3304                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3305                 break;
3306
3307         case TUNGETDEVNETNS:
3308                 ret = -EPERM;
3309                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3310                         goto unlock;
3311                 ret = open_related_ns(&net->ns, get_net_ns);
3312                 break;
3313
3314         default:
3315                 ret = -EINVAL;
3316                 break;
3317         }
3318
3319         if (do_notify)
3320                 netdev_state_change(tun->dev);
3321
3322 unlock:
3323         rtnl_unlock();
3324         if (tun)
3325                 tun_put(tun);
3326         return ret;
3327 }
3328
3329 static long tun_chr_ioctl(struct file *file,
3330                           unsigned int cmd, unsigned long arg)
3331 {
3332         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3333 }
3334
3335 #ifdef CONFIG_COMPAT
3336 static long tun_chr_compat_ioctl(struct file *file,
3337                          unsigned int cmd, unsigned long arg)
3338 {
3339         switch (cmd) {
3340         case TUNSETIFF:
3341         case TUNGETIFF:
3342         case TUNSETTXFILTER:
3343         case TUNGETSNDBUF:
3344         case TUNSETSNDBUF:
3345         case SIOCGIFHWADDR:
3346         case SIOCSIFHWADDR:
3347                 arg = (unsigned long)compat_ptr(arg);
3348                 break;
3349         default:
3350                 arg = (compat_ulong_t)arg;
3351                 break;
3352         }
3353
3354         /*
3355          * compat_ifreq is shorter than ifreq, so we must not access beyond
3356          * the end of that structure. All fields that are used in this
3357          * driver are compatible though, we don't need to convert the
3358          * contents.
3359          */
3360         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3361 }
3362 #endif /* CONFIG_COMPAT */
3363
3364 static int tun_chr_fasync(int fd, struct file *file, int on)
3365 {
3366         struct tun_file *tfile = file->private_data;
3367         int ret;
3368
3369         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3370                 goto out;
3371
3372         if (on) {
3373                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3374                 tfile->flags |= TUN_FASYNC;
3375         } else
3376                 tfile->flags &= ~TUN_FASYNC;
3377         ret = 0;
3378 out:
3379         return ret;
3380 }
3381
3382 static int tun_chr_open(struct inode *inode, struct file * file)
3383 {
3384         struct net *net = current->nsproxy->net_ns;
3385         struct tun_file *tfile;
3386
3387         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3388                                             &tun_proto, 0);
3389         if (!tfile)
3390                 return -ENOMEM;
3391         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3392                 sk_free(&tfile->sk);
3393                 return -ENOMEM;
3394         }
3395
3396         mutex_init(&tfile->napi_mutex);
3397         RCU_INIT_POINTER(tfile->tun, NULL);
3398         tfile->flags = 0;
3399         tfile->ifindex = 0;
3400
3401         init_waitqueue_head(&tfile->socket.wq.wait);
3402
3403         tfile->socket.file = file;
3404         tfile->socket.ops = &tun_socket_ops;
3405
3406         sock_init_data(&tfile->socket, &tfile->sk);
3407
3408         tfile->sk.sk_write_space = tun_sock_write_space;
3409         tfile->sk.sk_sndbuf = INT_MAX;
3410
3411         file->private_data = tfile;
3412         INIT_LIST_HEAD(&tfile->next);
3413
3414         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3415
3416         return 0;
3417 }
3418
3419 static int tun_chr_close(struct inode *inode, struct file *file)
3420 {
3421         struct tun_file *tfile = file->private_data;
3422
3423         tun_detach(tfile, true);
3424
3425         return 0;
3426 }
3427
3428 #ifdef CONFIG_PROC_FS
3429 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3430 {
3431         struct tun_file *tfile = file->private_data;
3432         struct tun_struct *tun;
3433         struct ifreq ifr;
3434
3435         memset(&ifr, 0, sizeof(ifr));
3436
3437         rtnl_lock();
3438         tun = tun_get(tfile);
3439         if (tun)
3440                 tun_get_iff(tun, &ifr);
3441         rtnl_unlock();
3442
3443         if (tun)
3444                 tun_put(tun);
3445
3446         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3447 }
3448 #endif
3449
3450 static const struct file_operations tun_fops = {
3451         .owner  = THIS_MODULE,
3452         .llseek = no_llseek,
3453         .read_iter  = tun_chr_read_iter,
3454         .write_iter = tun_chr_write_iter,
3455         .poll   = tun_chr_poll,
3456         .unlocked_ioctl = tun_chr_ioctl,
3457 #ifdef CONFIG_COMPAT
3458         .compat_ioctl = tun_chr_compat_ioctl,
3459 #endif
3460         .open   = tun_chr_open,
3461         .release = tun_chr_close,
3462         .fasync = tun_chr_fasync,
3463 #ifdef CONFIG_PROC_FS
3464         .show_fdinfo = tun_chr_show_fdinfo,
3465 #endif
3466 };
3467
3468 static struct miscdevice tun_miscdev = {
3469         .minor = TUN_MINOR,
3470         .name = "tun",
3471         .nodename = "net/tun",
3472         .fops = &tun_fops,
3473 };
3474
3475 /* ethtool interface */
3476
3477 static void tun_default_link_ksettings(struct net_device *dev,
3478                                        struct ethtool_link_ksettings *cmd)
3479 {
3480         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3481         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3482         cmd->base.speed         = SPEED_10;
3483         cmd->base.duplex        = DUPLEX_FULL;
3484         cmd->base.port          = PORT_TP;
3485         cmd->base.phy_address   = 0;
3486         cmd->base.autoneg       = AUTONEG_DISABLE;
3487 }
3488
3489 static int tun_get_link_ksettings(struct net_device *dev,
3490                                   struct ethtool_link_ksettings *cmd)
3491 {
3492         struct tun_struct *tun = netdev_priv(dev);
3493
3494         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3495         return 0;
3496 }
3497
3498 static int tun_set_link_ksettings(struct net_device *dev,
3499                                   const struct ethtool_link_ksettings *cmd)
3500 {
3501         struct tun_struct *tun = netdev_priv(dev);
3502
3503         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3504         return 0;
3505 }
3506
3507 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3508 {
3509         struct tun_struct *tun = netdev_priv(dev);
3510
3511         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3512         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3513
3514         switch (tun->flags & TUN_TYPE_MASK) {
3515         case IFF_TUN:
3516                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3517                 break;
3518         case IFF_TAP:
3519                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3520                 break;
3521         }
3522 }
3523
3524 static u32 tun_get_msglevel(struct net_device *dev)
3525 {
3526         struct tun_struct *tun = netdev_priv(dev);
3527
3528         return tun->msg_enable;
3529 }
3530
3531 static void tun_set_msglevel(struct net_device *dev, u32 value)
3532 {
3533         struct tun_struct *tun = netdev_priv(dev);
3534
3535         tun->msg_enable = value;
3536 }
3537
3538 static int tun_get_coalesce(struct net_device *dev,
3539                             struct ethtool_coalesce *ec)
3540 {
3541         struct tun_struct *tun = netdev_priv(dev);
3542
3543         ec->rx_max_coalesced_frames = tun->rx_batched;
3544
3545         return 0;
3546 }
3547
3548 static int tun_set_coalesce(struct net_device *dev,
3549                             struct ethtool_coalesce *ec)
3550 {
3551         struct tun_struct *tun = netdev_priv(dev);
3552
3553         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3554                 tun->rx_batched = NAPI_POLL_WEIGHT;
3555         else
3556                 tun->rx_batched = ec->rx_max_coalesced_frames;
3557
3558         return 0;
3559 }
3560
3561 static const struct ethtool_ops tun_ethtool_ops = {
3562         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3563         .get_drvinfo    = tun_get_drvinfo,
3564         .get_msglevel   = tun_get_msglevel,
3565         .set_msglevel   = tun_set_msglevel,
3566         .get_link       = ethtool_op_get_link,
3567         .get_ts_info    = ethtool_op_get_ts_info,
3568         .get_coalesce   = tun_get_coalesce,
3569         .set_coalesce   = tun_set_coalesce,
3570         .get_link_ksettings = tun_get_link_ksettings,
3571         .set_link_ksettings = tun_set_link_ksettings,
3572 };
3573
3574 static int tun_queue_resize(struct tun_struct *tun)
3575 {
3576         struct net_device *dev = tun->dev;
3577         struct tun_file *tfile;
3578         struct ptr_ring **rings;
3579         int n = tun->numqueues + tun->numdisabled;
3580         int ret, i;
3581
3582         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3583         if (!rings)
3584                 return -ENOMEM;
3585
3586         for (i = 0; i < tun->numqueues; i++) {
3587                 tfile = rtnl_dereference(tun->tfiles[i]);
3588                 rings[i] = &tfile->tx_ring;
3589         }
3590         list_for_each_entry(tfile, &tun->disabled, next)
3591                 rings[i++] = &tfile->tx_ring;
3592
3593         ret = ptr_ring_resize_multiple(rings, n,
3594                                        dev->tx_queue_len, GFP_KERNEL,
3595                                        tun_ptr_free);
3596
3597         kfree(rings);
3598         return ret;
3599 }
3600
3601 static int tun_device_event(struct notifier_block *unused,
3602                             unsigned long event, void *ptr)
3603 {
3604         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3605         struct tun_struct *tun = netdev_priv(dev);
3606         int i;
3607
3608         if (dev->rtnl_link_ops != &tun_link_ops)
3609                 return NOTIFY_DONE;
3610
3611         switch (event) {
3612         case NETDEV_CHANGE_TX_QUEUE_LEN:
3613                 if (tun_queue_resize(tun))
3614                         return NOTIFY_BAD;
3615                 break;
3616         case NETDEV_UP:
3617                 for (i = 0; i < tun->numqueues; i++) {
3618                         struct tun_file *tfile;
3619
3620                         tfile = rtnl_dereference(tun->tfiles[i]);
3621                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3622                 }
3623                 break;
3624         default:
3625                 break;
3626         }
3627
3628         return NOTIFY_DONE;
3629 }
3630
3631 static struct notifier_block tun_notifier_block __read_mostly = {
3632         .notifier_call  = tun_device_event,
3633 };
3634
3635 static int __init tun_init(void)
3636 {
3637         int ret = 0;
3638
3639         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3640
3641         ret = rtnl_link_register(&tun_link_ops);
3642         if (ret) {
3643                 pr_err("Can't register link_ops\n");
3644                 goto err_linkops;
3645         }
3646
3647         ret = misc_register(&tun_miscdev);
3648         if (ret) {
3649                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3650                 goto err_misc;
3651         }
3652
3653         ret = register_netdevice_notifier(&tun_notifier_block);
3654         if (ret) {
3655                 pr_err("Can't register netdevice notifier\n");
3656                 goto err_notifier;
3657         }
3658
3659         return  0;
3660
3661 err_notifier:
3662         misc_deregister(&tun_miscdev);
3663 err_misc:
3664         rtnl_link_unregister(&tun_link_ops);
3665 err_linkops:
3666         return ret;
3667 }
3668
3669 static void tun_cleanup(void)
3670 {
3671         misc_deregister(&tun_miscdev);
3672         rtnl_link_unregister(&tun_link_ops);
3673         unregister_netdevice_notifier(&tun_notifier_block);
3674 }
3675
3676 /* Get an underlying socket object from tun file.  Returns error unless file is
3677  * attached to a device.  The returned object works like a packet socket, it
3678  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3679  * holding a reference to the file for as long as the socket is in use. */
3680 struct socket *tun_get_socket(struct file *file)
3681 {
3682         struct tun_file *tfile;
3683         if (file->f_op != &tun_fops)
3684                 return ERR_PTR(-EINVAL);
3685         tfile = file->private_data;
3686         if (!tfile)
3687                 return ERR_PTR(-EBADFD);
3688         return &tfile->socket;
3689 }
3690 EXPORT_SYMBOL_GPL(tun_get_socket);
3691
3692 struct ptr_ring *tun_get_tx_ring(struct file *file)
3693 {
3694         struct tun_file *tfile;
3695
3696         if (file->f_op != &tun_fops)
3697                 return ERR_PTR(-EINVAL);
3698         tfile = file->private_data;
3699         if (!tfile)
3700                 return ERR_PTR(-EBADFD);
3701         return &tfile->tx_ring;
3702 }
3703 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3704
3705 module_init(tun_init);
3706 module_exit(tun_cleanup);
3707 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3708 MODULE_AUTHOR(DRV_COPYRIGHT);
3709 MODULE_LICENSE("GPL");
3710 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3711 MODULE_ALIAS("devname:net/tun");