drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <underley@underley.eu.org>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212 };
 213
 214 struct veth {
 215         __be16 h_vlan_proto;
 216         __be16 h_vlan_TCI;
 217 };
 218
 219 static int tun_napi_receive(struct napi_struct *napi, int budget)
 220 {
 221         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 222         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 223         struct sk_buff_head process_queue;
 224         struct sk_buff *skb;
 225         int received = 0;
 226
 227         __skb_queue_head_init(&process_queue);
 228
 229         spin_lock(&queue->lock);
 230         skb_queue_splice_tail_init(queue, &process_queue);
 231         spin_unlock(&queue->lock);
 232
 233         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 234                 napi_gro_receive(napi, skb);
 235                 ++received;
 236         }
 237
 238         if (!skb_queue_empty(&process_queue)) {
 239                 spin_lock(&queue->lock);
 240                 skb_queue_splice(&process_queue, queue);
 241                 spin_unlock(&queue->lock);
 242         }
 243
 244         return received;
 245 }
 246
 247 static int tun_napi_poll(struct napi_struct *napi, int budget)
 248 {
 249         unsigned int received;
 250
 251         received = tun_napi_receive(napi, budget);
 252
 253         if (received < budget)
 254                 napi_complete_done(napi, received);
 255
 256         return received;
 257 }
 258
 259 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 260                           bool napi_en, bool napi_frags)
 261 {
 262         tfile->napi_enabled = napi_en;
 263         tfile->napi_frags_enabled = napi_en && napi_frags;
 264         if (napi_en) {
 265                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 266                                   NAPI_POLL_WEIGHT);
 267                 napi_enable(&tfile->napi);
 268         }
 269 }
 270
 271 static void tun_napi_disable(struct tun_file *tfile)
 272 {
 273         if (tfile->napi_enabled)
 274                 napi_disable(&tfile->napi);
 275 }
 276
 277 static void tun_napi_del(struct tun_file *tfile)
 278 {
 279         if (tfile->napi_enabled)
 280                 netif_napi_del(&tfile->napi);
 281 }
 282
 283 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 284 {
 285         return tfile->napi_frags_enabled;
 286 }
 287
 288 #ifdef CONFIG_TUN_VNET_CROSS_LE
 289 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 290 {
 291         return tun->flags & TUN_VNET_BE ? false :
 292                 virtio_legacy_is_little_endian();
 293 }
 294
 295 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 296 {
 297         int be = !!(tun->flags & TUN_VNET_BE);
 298
 299         if (put_user(be, argp))
 300                 return -EFAULT;
 301
 302         return 0;
 303 }
 304
 305 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 306 {
 307         int be;
 308
 309         if (get_user(be, argp))
 310                 return -EFAULT;
 311
 312         if (be)
 313                 tun->flags |= TUN_VNET_BE;
 314         else
 315                 tun->flags &= ~TUN_VNET_BE;
 316
 317         return 0;
 318 }
 319 #else
 320 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 321 {
 322         return virtio_legacy_is_little_endian();
 323 }
 324
 325 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 326 {
 327         return -EINVAL;
 328 }
 329
 330 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 331 {
 332         return -EINVAL;
 333 }
 334 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 335
 336 static inline bool tun_is_little_endian(struct tun_struct *tun)
 337 {
 338         return tun->flags & TUN_VNET_LE ||
 339                 tun_legacy_is_little_endian(tun);
 340 }
 341
 342 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 343 {
 344         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 345 }
 346
 347 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 348 {
 349         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 350 }
 351
 352 static inline u32 tun_hashfn(u32 rxhash)
 353 {
 354         return rxhash & TUN_MASK_FLOW_ENTRIES;
 355 }
 356
 357 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 358 {
 359         struct tun_flow_entry *e;
 360
 361         hlist_for_each_entry_rcu(e, head, hash_link) {
 362                 if (e->rxhash == rxhash)
 363                         return e;
 364         }
 365         return NULL;
 366 }
 367
 368 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 369                                               struct hlist_head *head,
 370                                               u32 rxhash, u16 queue_index)
 371 {
 372         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 373
 374         if (e) {
 375                 netif_info(tun, tx_queued, tun->dev,
 376                            "create flow: hash %u index %u\n",
 377                            rxhash, queue_index);
 378                 e->updated = jiffies;
 379                 e->rxhash = rxhash;
 380                 e->rps_rxhash = 0;
 381                 e->queue_index = queue_index;
 382                 e->tun = tun;
 383                 hlist_add_head_rcu(&e->hash_link, head);
 384                 ++tun->flow_count;
 385         }
 386         return e;
 387 }
 388
 389 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 390 {
 391         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 392                    e->rxhash, e->queue_index);
 393         hlist_del_rcu(&e->hash_link);
 394         kfree_rcu(e, rcu);
 395         --tun->flow_count;
 396 }
 397
 398 static void tun_flow_flush(struct tun_struct *tun)
 399 {
 400         int i;
 401
 402         spin_lock_bh(&tun->lock);
 403         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 404                 struct tun_flow_entry *e;
 405                 struct hlist_node *n;
 406
 407                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 408                         tun_flow_delete(tun, e);
 409         }
 410         spin_unlock_bh(&tun->lock);
 411 }
 412
 413 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 414 {
 415         int i;
 416
 417         spin_lock_bh(&tun->lock);
 418         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 419                 struct tun_flow_entry *e;
 420                 struct hlist_node *n;
 421
 422                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 423                         if (e->queue_index == queue_index)
 424                                 tun_flow_delete(tun, e);
 425                 }
 426         }
 427         spin_unlock_bh(&tun->lock);
 428 }
 429
 430 static void tun_flow_cleanup(struct timer_list *t)
 431 {
 432         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 433         unsigned long delay = tun->ageing_time;
 434         unsigned long next_timer = jiffies + delay;
 435         unsigned long count = 0;
 436         int i;
 437
 438         spin_lock(&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                         unsigned long this_timer;
 445
 446                         this_timer = e->updated + delay;
 447                         if (time_before_eq(this_timer, jiffies)) {
 448                                 tun_flow_delete(tun, e);
 449                                 continue;
 450                         }
 451                         count++;
 452                         if (time_before(this_timer, next_timer))
 453                                 next_timer = this_timer;
 454                 }
 455         }
 456
 457         if (count)
 458                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 459         spin_unlock(&tun->lock);
 460 }
 461
 462 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 463                             struct tun_file *tfile)
 464 {
 465         struct hlist_head *head;
 466         struct tun_flow_entry *e;
 467         unsigned long delay = tun->ageing_time;
 468         u16 queue_index = tfile->queue_index;
 469
 470         head = &tun->flows[tun_hashfn(rxhash)];
 471
 472         rcu_read_lock();
 473
 474         e = tun_flow_find(head, rxhash);
 475         if (likely(e)) {
 476                 /* TODO: keep queueing to old queue until it's empty? */
 477                 if (READ_ONCE(e->queue_index) != queue_index)
 478                         WRITE_ONCE(e->queue_index, queue_index);
 479                 if (e->updated != jiffies)
 480                         e->updated = jiffies;
 481                 sock_rps_record_flow_hash(e->rps_rxhash);
 482         } else {
 483                 spin_lock_bh(&tun->lock);
 484                 if (!tun_flow_find(head, rxhash) &&
 485                     tun->flow_count < MAX_TAP_FLOWS)
 486                         tun_flow_create(tun, head, rxhash, queue_index);
 487
 488                 if (!timer_pending(&tun->flow_gc_timer))
 489                         mod_timer(&tun->flow_gc_timer,
 490                                   round_jiffies_up(jiffies + delay));
 491                 spin_unlock_bh(&tun->lock);
 492         }
 493
 494         rcu_read_unlock();
 495 }
 496
 497 /* Save the hash received in the stack receive path and update the
 498  * flow_hash table accordingly.
 499  */
 500 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 501 {
 502         if (unlikely(e->rps_rxhash != hash))
 503                 e->rps_rxhash = hash;
 504 }
 505
 506 /* We try to identify a flow through its rxhash. The reason that
 507  * we do not check rxq no. is because some cards(e.g 82599), chooses
 508  * the rxq based on the txq where the last packet of the flow comes. As
 509  * the userspace application move between processors, we may get a
 510  * different rxq no. here.
 511  */
 512 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 513 {
 514         struct tun_flow_entry *e;
 515         u32 txq = 0;
 516         u32 numqueues = 0;
 517
 518         numqueues = READ_ONCE(tun->numqueues);
 519
 520         txq = __skb_get_hash_symmetric(skb);
 521         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 522         if (e) {
 523                 tun_flow_save_rps_rxhash(e, txq);
 524                 txq = e->queue_index;
 525         } else {
 526                 /* use multiply and shift instead of expensive divide */
 527                 txq = ((u64)txq * numqueues) >> 32;
 528         }
 529
 530         return txq;
 531 }
 532
 533 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 534 {
 535         struct tun_prog *prog;
 536         u32 numqueues;
 537         u16 ret = 0;
 538
 539         numqueues = READ_ONCE(tun->numqueues);
 540         if (!numqueues)
 541                 return 0;
 542
 543         prog = rcu_dereference(tun->steering_prog);
 544         if (prog)
 545                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 546
 547         return ret % numqueues;
 548 }
 549
 550 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 551                             struct net_device *sb_dev)
 552 {
 553         struct tun_struct *tun = netdev_priv(dev);
 554         u16 ret;
 555
 556         rcu_read_lock();
 557         if (rcu_dereference(tun->steering_prog))
 558                 ret = tun_ebpf_select_queue(tun, skb);
 559         else
 560                 ret = tun_automq_select_queue(tun, skb);
 561         rcu_read_unlock();
 562
 563         return ret;
 564 }
 565
 566 static inline bool tun_not_capable(struct tun_struct *tun)
 567 {
 568         const struct cred *cred = current_cred();
 569         struct net *net = dev_net(tun->dev);
 570
 571         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 572                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 573                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 574 }
 575
 576 static void tun_set_real_num_queues(struct tun_struct *tun)
 577 {
 578         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 579         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 580 }
 581
 582 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 583 {
 584         tfile->detached = tun;
 585         list_add_tail(&tfile->next, &tun->disabled);
 586         ++tun->numdisabled;
 587 }
 588
 589 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 590 {
 591         struct tun_struct *tun = tfile->detached;
 592
 593         tfile->detached = NULL;
 594         list_del_init(&tfile->next);
 595         --tun->numdisabled;
 596         return tun;
 597 }
 598
 599 void tun_ptr_free(void *ptr)
 600 {
 601         if (!ptr)
 602                 return;
 603         if (tun_is_xdp_frame(ptr)) {
 604                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 605
 606                 xdp_return_frame(xdpf);
 607         } else {
 608                 __skb_array_destroy_skb(ptr);
 609         }
 610 }
 611 EXPORT_SYMBOL_GPL(tun_ptr_free);
 612
 613 static void tun_queue_purge(struct tun_file *tfile)
 614 {
 615         void *ptr;
 616
 617         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 618                 tun_ptr_free(ptr);
 619
 620         skb_queue_purge(&tfile->sk.sk_write_queue);
 621         skb_queue_purge(&tfile->sk.sk_error_queue);
 622 }
 623
 624 static void __tun_detach(struct tun_file *tfile, bool clean)
 625 {
 626         struct tun_file *ntfile;
 627         struct tun_struct *tun;
 628
 629         tun = rtnl_dereference(tfile->tun);
 630
 631         if (tun && clean) {
 632                 tun_napi_disable(tfile);
 633                 tun_napi_del(tfile);
 634         }
 635
 636         if (tun && !tfile->detached) {
 637                 u16 index = tfile->queue_index;
 638                 BUG_ON(index >= tun->numqueues);
 639
 640                 rcu_assign_pointer(tun->tfiles[index],
 641                                    tun->tfiles[tun->numqueues - 1]);
 642                 ntfile = rtnl_dereference(tun->tfiles[index]);
 643                 ntfile->queue_index = index;
 644                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 645                                    NULL);
 646
 647                 --tun->numqueues;
 648                 if (clean) {
 649                         RCU_INIT_POINTER(tfile->tun, NULL);
 650                         sock_put(&tfile->sk);
 651                 } else
 652                         tun_disable_queue(tun, tfile);
 653
 654                 synchronize_net();
 655                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 656                 /* Drop read queue */
 657                 tun_queue_purge(tfile);
 658                 tun_set_real_num_queues(tun);
 659         } else if (tfile->detached && clean) {
 660                 tun = tun_enable_queue(tfile);
 661                 sock_put(&tfile->sk);
 662         }
 663
 664         if (clean) {
 665                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 666                         netif_carrier_off(tun->dev);
 667
 668                         if (!(tun->flags & IFF_PERSIST) &&
 669                             tun->dev->reg_state == NETREG_REGISTERED)
 670                                 unregister_netdevice(tun->dev);
 671                 }
 672                 if (tun)
 673                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 674                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 675                 sock_put(&tfile->sk);
 676         }
 677 }
 678
 679 static void tun_detach(struct tun_file *tfile, bool clean)
 680 {
 681         struct tun_struct *tun;
 682         struct net_device *dev;
 683
 684         rtnl_lock();
 685         tun = rtnl_dereference(tfile->tun);
 686         dev = tun ? tun->dev : NULL;
 687         __tun_detach(tfile, clean);
 688         if (dev)
 689                 netdev_state_change(dev);
 690         rtnl_unlock();
 691 }
 692
 693 static void tun_detach_all(struct net_device *dev)
 694 {
 695         struct tun_struct *tun = netdev_priv(dev);
 696         struct tun_file *tfile, *tmp;
 697         int i, n = tun->numqueues;
 698
 699         for (i = 0; i < n; i++) {
 700                 tfile = rtnl_dereference(tun->tfiles[i]);
 701                 BUG_ON(!tfile);
 702                 tun_napi_disable(tfile);
 703                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 704                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 705                 RCU_INIT_POINTER(tfile->tun, NULL);
 706                 --tun->numqueues;
 707         }
 708         list_for_each_entry(tfile, &tun->disabled, next) {
 709                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 710                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 711                 RCU_INIT_POINTER(tfile->tun, NULL);
 712         }
 713         BUG_ON(tun->numqueues != 0);
 714
 715         synchronize_net();
 716         for (i = 0; i < n; i++) {
 717                 tfile = rtnl_dereference(tun->tfiles[i]);
 718                 tun_napi_del(tfile);
 719                 /* Drop read queue */
 720                 tun_queue_purge(tfile);
 721                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 722                 sock_put(&tfile->sk);
 723         }
 724         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 725                 tun_enable_queue(tfile);
 726                 tun_queue_purge(tfile);
 727                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 728                 sock_put(&tfile->sk);
 729         }
 730         BUG_ON(tun->numdisabled != 0);
 731
 732         if (tun->flags & IFF_PERSIST)
 733                 module_put(THIS_MODULE);
 734 }
 735
 736 static int tun_attach(struct tun_struct *tun, struct file *file,
 737                       bool skip_filter, bool napi, bool napi_frags,
 738                       bool publish_tun)
 739 {
 740         struct tun_file *tfile = file->private_data;
 741         struct net_device *dev = tun->dev;
 742         int err;
 743
 744         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 745         if (err < 0)
 746                 goto out;
 747
 748         err = -EINVAL;
 749         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 750                 goto out;
 751
 752         err = -EBUSY;
 753         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 754                 goto out;
 755
 756         err = -E2BIG;
 757         if (!tfile->detached &&
 758             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 759                 goto out;
 760
 761         err = 0;
 762
 763         /* Re-attach the filter to persist device */
 764         if (!skip_filter && (tun->filter_attached == true)) {
 765                 lock_sock(tfile->socket.sk);
 766                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 767                 release_sock(tfile->socket.sk);
 768                 if (!err)
 769                         goto out;
 770         }
 771
 772         if (!tfile->detached &&
 773             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 774                             GFP_KERNEL, tun_ptr_free)) {
 775                 err = -ENOMEM;
 776                 goto out;
 777         }
 778
 779         tfile->queue_index = tun->numqueues;
 780         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 781
 782         if (tfile->detached) {
 783                 /* Re-attach detached tfile, updating XDP queue_index */
 784                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 785
 786                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 787                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 788         } else {
 789                 /* Setup XDP RX-queue info, for new tfile getting attached */
 790                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 791                                        tun->dev, tfile->queue_index, 0);
 792                 if (err < 0)
 793                         goto out;
 794                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 795                                                  MEM_TYPE_PAGE_SHARED, NULL);
 796                 if (err < 0) {
 797                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 798                         goto out;
 799                 }
 800                 err = 0;
 801         }
 802
 803         if (tfile->detached) {
 804                 tun_enable_queue(tfile);
 805         } else {
 806                 sock_hold(&tfile->sk);
 807                 tun_napi_init(tun, tfile, napi, napi_frags);
 808         }
 809
 810         if (rtnl_dereference(tun->xdp_prog))
 811                 sock_set_flag(&tfile->sk, SOCK_XDP);
 812
 813         /* device is allowed to go away first, so no need to hold extra
 814          * refcnt.
 815          */
 816
 817         /* Publish tfile->tun and tun->tfiles only after we've fully
 818          * initialized tfile; otherwise we risk using half-initialized
 819          * object.
 820          */
 821         if (publish_tun)
 822                 rcu_assign_pointer(tfile->tun, tun);
 823         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 824         tun->numqueues++;
 825         tun_set_real_num_queues(tun);
 826 out:
 827         return err;
 828 }
 829
 830 static struct tun_struct *tun_get(struct tun_file *tfile)
 831 {
 832         struct tun_struct *tun;
 833
 834         rcu_read_lock();
 835         tun = rcu_dereference(tfile->tun);
 836         if (tun)
 837                 dev_hold(tun->dev);
 838         rcu_read_unlock();
 839
 840         return tun;
 841 }
 842
 843 static void tun_put(struct tun_struct *tun)
 844 {
 845         dev_put(tun->dev);
 846 }
 847
 848 /* TAP filtering */
 849 static void addr_hash_set(u32 *mask, const u8 *addr)
 850 {
 851         int n = ether_crc(ETH_ALEN, addr) >> 26;
 852         mask[n >> 5] |= (1 << (n & 31));
 853 }
 854
 855 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 856 {
 857         int n = ether_crc(ETH_ALEN, addr) >> 26;
 858         return mask[n >> 5] & (1 << (n & 31));
 859 }
 860
 861 static int update_filter(struct tap_filter *filter, void __user *arg)
 862 {
 863         struct { u8 u[ETH_ALEN]; } *addr;
 864         struct tun_filter uf;
 865         int err, alen, n, nexact;
 866
 867         if (copy_from_user(&uf, arg, sizeof(uf)))
 868                 return -EFAULT;
 869
 870         if (!uf.count) {
 871                 /* Disabled */
 872                 filter->count = 0;
 873                 return 0;
 874         }
 875
 876         alen = ETH_ALEN * uf.count;
 877         addr = memdup_user(arg + sizeof(uf), alen);
 878         if (IS_ERR(addr))
 879                 return PTR_ERR(addr);
 880
 881         /* The filter is updated without holding any locks. Which is
 882          * perfectly safe. We disable it first and in the worst
 883          * case we'll accept a few undesired packets. */
 884         filter->count = 0;
 885         wmb();
 886
 887         /* Use first set of addresses as an exact filter */
 888         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 889                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 890
 891         nexact = n;
 892
 893         /* Remaining multicast addresses are hashed,
 894          * unicast will leave the filter disabled. */
 895         memset(filter->mask, 0, sizeof(filter->mask));
 896         for (; n < uf.count; n++) {
 897                 if (!is_multicast_ether_addr(addr[n].u)) {
 898                         err = 0; /* no filter */
 899                         goto free_addr;
 900                 }
 901                 addr_hash_set(filter->mask, addr[n].u);
 902         }
 903
 904         /* For ALLMULTI just set the mask to all ones.
 905          * This overrides the mask populated above. */
 906         if ((uf.flags & TUN_FLT_ALLMULTI))
 907                 memset(filter->mask, ~0, sizeof(filter->mask));
 908
 909         /* Now enable the filter */
 910         wmb();
 911         filter->count = nexact;
 912
 913         /* Return the number of exact filters */
 914         err = nexact;
 915 free_addr:
 916         kfree(addr);
 917         return err;
 918 }
 919
 920 /* Returns: 0 - drop, !=0 - accept */
 921 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 922 {
 923         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 924          * at this point. */
 925         struct ethhdr *eh = (struct ethhdr *) skb->data;
 926         int i;
 927
 928         /* Exact match */
 929         for (i = 0; i < filter->count; i++)
 930                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 931                         return 1;
 932
 933         /* Inexact match (multicast only) */
 934         if (is_multicast_ether_addr(eh->h_dest))
 935                 return addr_hash_test(filter->mask, eh->h_dest);
 936
 937         return 0;
 938 }
 939
 940 /*
 941  * Checks whether the packet is accepted or not.
 942  * Returns: 0 - drop, !=0 - accept
 943  */
 944 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 945 {
 946         if (!filter->count)
 947                 return 1;
 948
 949         return run_filter(filter, skb);
 950 }
 951
 952 /* Network device part of the driver */
 953
 954 static const struct ethtool_ops tun_ethtool_ops;
 955
 956 /* Net device detach from fd. */
 957 static void tun_net_uninit(struct net_device *dev)
 958 {
 959         tun_detach_all(dev);
 960 }
 961
 962 /* Net device open. */
 963 static int tun_net_open(struct net_device *dev)
 964 {
 965         netif_tx_start_all_queues(dev);
 966
 967         return 0;
 968 }
 969
 970 /* Net device close. */
 971 static int tun_net_close(struct net_device *dev)
 972 {
 973         netif_tx_stop_all_queues(dev);
 974         return 0;
 975 }
 976
 977 /* Net device start xmit */
 978 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
 979 {
 980 #ifdef CONFIG_RPS
 981         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
 982                 /* Select queue was not called for the skbuff, so we extract the
 983                  * RPS hash and save it into the flow_table here.
 984                  */
 985                 struct tun_flow_entry *e;
 986                 __u32 rxhash;
 987
 988                 rxhash = __skb_get_hash_symmetric(skb);
 989                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
 990                 if (e)
 991                         tun_flow_save_rps_rxhash(e, rxhash);
 992         }
 993 #endif
 994 }
 995
 996 static unsigned int run_ebpf_filter(struct tun_struct *tun,
 997                                     struct sk_buff *skb,
 998                                     int len)
 999 {
1000         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1001
1002         if (prog)
1003                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1004
1005         return len;
1006 }
1007
1008 /* Net device start xmit */
1009 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1010 {
1011         struct tun_struct *tun = netdev_priv(dev);
1012         int txq = skb->queue_mapping;
1013         struct tun_file *tfile;
1014         int len = skb->len;
1015
1016         rcu_read_lock();
1017         tfile = rcu_dereference(tun->tfiles[txq]);
1018
1019         /* Drop packet if interface is not attached */
1020         if (!tfile)
1021                 goto drop;
1022
1023         if (!rcu_dereference(tun->steering_prog))
1024                 tun_automq_xmit(tun, skb);
1025
1026         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1027
1028         /* Drop if the filter does not like it.
1029          * This is a noop if the filter is disabled.
1030          * Filter can be enabled only for the TAP devices. */
1031         if (!check_filter(&tun->txflt, skb))
1032                 goto drop;
1033
1034         if (tfile->socket.sk->sk_filter &&
1035             sk_filter(tfile->socket.sk, skb))
1036                 goto drop;
1037
1038         len = run_ebpf_filter(tun, skb, len);
1039         if (len == 0 || pskb_trim(skb, len))
1040                 goto drop;
1041
1042         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1043                 goto drop;
1044
1045         skb_tx_timestamp(skb);
1046
1047         /* Orphan the skb - required as we might hang on to it
1048          * for indefinite time.
1049          */
1050         skb_orphan(skb);
1051
1052         nf_reset_ct(skb);
1053
1054         if (ptr_ring_produce(&tfile->tx_ring, skb))
1055                 goto drop;
1056
1057         /* Notify and wake up reader process */
1058         if (tfile->flags & TUN_FASYNC)
1059                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1060         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1061
1062         rcu_read_unlock();
1063         return NETDEV_TX_OK;
1064
1065 drop:
1066         atomic_long_inc(&dev->tx_dropped);
1067         skb_tx_error(skb);
1068         kfree_skb(skb);
1069         rcu_read_unlock();
1070         return NET_XMIT_DROP;
1071 }
1072
1073 static void tun_net_mclist(struct net_device *dev)
1074 {
1075         /*
1076          * This callback is supposed to deal with mc filter in
1077          * _rx_ path and has nothing to do with the _tx_ path.
1078          * In rx path we always accept everything userspace gives us.
1079          */
1080 }
1081
1082 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1083         netdev_features_t features)
1084 {
1085         struct tun_struct *tun = netdev_priv(dev);
1086
1087         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1088 }
1089
1090 static void tun_set_headroom(struct net_device *dev, int new_hr)
1091 {
1092         struct tun_struct *tun = netdev_priv(dev);
1093
1094         if (new_hr < NET_SKB_PAD)
1095                 new_hr = NET_SKB_PAD;
1096
1097         tun->align = new_hr;
1098 }
1099
1100 static void
1101 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1102 {
1103         struct tun_struct *tun = netdev_priv(dev);
1104
1105         dev_get_tstats64(dev, stats);
1106
1107         stats->rx_frame_errors +=
1108                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1109 }
1110
1111 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1112                        struct netlink_ext_ack *extack)
1113 {
1114         struct tun_struct *tun = netdev_priv(dev);
1115         struct tun_file *tfile;
1116         struct bpf_prog *old_prog;
1117         int i;
1118
1119         old_prog = rtnl_dereference(tun->xdp_prog);
1120         rcu_assign_pointer(tun->xdp_prog, prog);
1121         if (old_prog)
1122                 bpf_prog_put(old_prog);
1123
1124         for (i = 0; i < tun->numqueues; i++) {
1125                 tfile = rtnl_dereference(tun->tfiles[i]);
1126                 if (prog)
1127                         sock_set_flag(&tfile->sk, SOCK_XDP);
1128                 else
1129                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1130         }
1131         list_for_each_entry(tfile, &tun->disabled, next) {
1132                 if (prog)
1133                         sock_set_flag(&tfile->sk, SOCK_XDP);
1134                 else
1135                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1136         }
1137
1138         return 0;
1139 }
1140
1141 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1142 {
1143         switch (xdp->command) {
1144         case XDP_SETUP_PROG:
1145                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1146         default:
1147                 return -EINVAL;
1148         }
1149 }
1150
1151 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1152 {
1153         if (new_carrier) {
1154                 struct tun_struct *tun = netdev_priv(dev);
1155
1156                 if (!tun->numqueues)
1157                         return -EPERM;
1158
1159                 netif_carrier_on(dev);
1160         } else {
1161                 netif_carrier_off(dev);
1162         }
1163         return 0;
1164 }
1165
1166 static const struct net_device_ops tun_netdev_ops = {
1167         .ndo_uninit             = tun_net_uninit,
1168         .ndo_open               = tun_net_open,
1169         .ndo_stop               = tun_net_close,
1170         .ndo_start_xmit         = tun_net_xmit,
1171         .ndo_fix_features       = tun_net_fix_features,
1172         .ndo_select_queue       = tun_select_queue,
1173         .ndo_set_rx_headroom    = tun_set_headroom,
1174         .ndo_get_stats64        = tun_net_get_stats64,
1175         .ndo_change_carrier     = tun_net_change_carrier,
1176 };
1177
1178 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1179 {
1180         /* Notify and wake up reader process */
1181         if (tfile->flags & TUN_FASYNC)
1182                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1183         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1184 }
1185
1186 static int tun_xdp_xmit(struct net_device *dev, int n,
1187                         struct xdp_frame **frames, u32 flags)
1188 {
1189         struct tun_struct *tun = netdev_priv(dev);
1190         struct tun_file *tfile;
1191         u32 numqueues;
1192         int drops = 0;
1193         int cnt = n;
1194         int i;
1195
1196         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1197                 return -EINVAL;
1198
1199         rcu_read_lock();
1200
1201 resample:
1202         numqueues = READ_ONCE(tun->numqueues);
1203         if (!numqueues) {
1204                 rcu_read_unlock();
1205                 return -ENXIO; /* Caller will free/return all frames */
1206         }
1207
1208         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1209                                             numqueues]);
1210         if (unlikely(!tfile))
1211                 goto resample;
1212
1213         spin_lock(&tfile->tx_ring.producer_lock);
1214         for (i = 0; i < n; i++) {
1215                 struct xdp_frame *xdp = frames[i];
1216                 /* Encode the XDP flag into lowest bit for consumer to differ
1217                  * XDP buffer from sk_buff.
1218                  */
1219                 void *frame = tun_xdp_to_ptr(xdp);
1220
1221                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1222                         atomic_long_inc(&dev->tx_dropped);
1223                         xdp_return_frame_rx_napi(xdp);
1224                         drops++;
1225                 }
1226         }
1227         spin_unlock(&tfile->tx_ring.producer_lock);
1228
1229         if (flags & XDP_XMIT_FLUSH)
1230                 __tun_xdp_flush_tfile(tfile);
1231
1232         rcu_read_unlock();
1233         return cnt - drops;
1234 }
1235
1236 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1237 {
1238         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1239
1240         if (unlikely(!frame))
1241                 return -EOVERFLOW;
1242
1243         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1244 }
1245
1246 static const struct net_device_ops tap_netdev_ops = {
1247         .ndo_uninit             = tun_net_uninit,
1248         .ndo_open               = tun_net_open,
1249         .ndo_stop               = tun_net_close,
1250         .ndo_start_xmit         = tun_net_xmit,
1251         .ndo_fix_features       = tun_net_fix_features,
1252         .ndo_set_rx_mode        = tun_net_mclist,
1253         .ndo_set_mac_address    = eth_mac_addr,
1254         .ndo_validate_addr      = eth_validate_addr,
1255         .ndo_select_queue       = tun_select_queue,
1256         .ndo_features_check     = passthru_features_check,
1257         .ndo_set_rx_headroom    = tun_set_headroom,
1258         .ndo_get_stats64        = dev_get_tstats64,
1259         .ndo_bpf                = tun_xdp,
1260         .ndo_xdp_xmit           = tun_xdp_xmit,
1261         .ndo_change_carrier     = tun_net_change_carrier,
1262 };
1263
1264 static void tun_flow_init(struct tun_struct *tun)
1265 {
1266         int i;
1267
1268         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1269                 INIT_HLIST_HEAD(&tun->flows[i]);
1270
1271         tun->ageing_time = TUN_FLOW_EXPIRE;
1272         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1273         mod_timer(&tun->flow_gc_timer,
1274                   round_jiffies_up(jiffies + tun->ageing_time));
1275 }
1276
1277 static void tun_flow_uninit(struct tun_struct *tun)
1278 {
1279         del_timer_sync(&tun->flow_gc_timer);
1280         tun_flow_flush(tun);
1281 }
1282
1283 #define MIN_MTU 68
1284 #define MAX_MTU 65535
1285
1286 /* Initialize net device. */
1287 static void tun_net_init(struct net_device *dev)
1288 {
1289         struct tun_struct *tun = netdev_priv(dev);
1290
1291         switch (tun->flags & TUN_TYPE_MASK) {
1292         case IFF_TUN:
1293                 dev->netdev_ops = &tun_netdev_ops;
1294                 dev->header_ops = &ip_tunnel_header_ops;
1295
1296                 /* Point-to-Point TUN Device */
1297                 dev->hard_header_len = 0;
1298                 dev->addr_len = 0;
1299                 dev->mtu = 1500;
1300
1301                 /* Zero header length */
1302                 dev->type = ARPHRD_NONE;
1303                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1304                 break;
1305
1306         case IFF_TAP:
1307                 dev->netdev_ops = &tap_netdev_ops;
1308                 /* Ethernet TAP Device */
1309                 ether_setup(dev);
1310                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1311                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1312
1313                 eth_hw_addr_random(dev);
1314
1315                 break;
1316         }
1317
1318         dev->min_mtu = MIN_MTU;
1319         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1320 }
1321
1322 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1323 {
1324         struct sock *sk = tfile->socket.sk;
1325
1326         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1327 }
1328
1329 /* Character device part */
1330
1331 /* Poll */
1332 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1333 {
1334         struct tun_file *tfile = file->private_data;
1335         struct tun_struct *tun = tun_get(tfile);
1336         struct sock *sk;
1337         __poll_t mask = 0;
1338
1339         if (!tun)
1340                 return EPOLLERR;
1341
1342         sk = tfile->socket.sk;
1343
1344         poll_wait(file, sk_sleep(sk), wait);
1345
1346         if (!ptr_ring_empty(&tfile->tx_ring))
1347                 mask |= EPOLLIN | EPOLLRDNORM;
1348
1349         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1350          * guarantee EPOLLOUT to be raised by either here or
1351          * tun_sock_write_space(). Then process could get notification
1352          * after it writes to a down device and meets -EIO.
1353          */
1354         if (tun_sock_writeable(tun, tfile) ||
1355             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1356              tun_sock_writeable(tun, tfile)))
1357                 mask |= EPOLLOUT | EPOLLWRNORM;
1358
1359         if (tun->dev->reg_state != NETREG_REGISTERED)
1360                 mask = EPOLLERR;
1361
1362         tun_put(tun);
1363         return mask;
1364 }
1365
1366 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1367                                             size_t len,
1368                                             const struct iov_iter *it)
1369 {
1370         struct sk_buff *skb;
1371         size_t linear;
1372         int err;
1373         int i;
1374
1375         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1376                 return ERR_PTR(-EMSGSIZE);
1377
1378         local_bh_disable();
1379         skb = napi_get_frags(&tfile->napi);
1380         local_bh_enable();
1381         if (!skb)
1382                 return ERR_PTR(-ENOMEM);
1383
1384         linear = iov_iter_single_seg_count(it);
1385         err = __skb_grow(skb, linear);
1386         if (err)
1387                 goto free;
1388
1389         skb->len = len;
1390         skb->data_len = len - linear;
1391         skb->truesize += skb->data_len;
1392
1393         for (i = 1; i < it->nr_segs; i++) {
1394                 size_t fragsz = it->iov[i].iov_len;
1395                 struct page *page;
1396                 void *frag;
1397
1398                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1399                         err = -EINVAL;
1400                         goto free;
1401                 }
1402                 frag = netdev_alloc_frag(fragsz);
1403                 if (!frag) {
1404                         err = -ENOMEM;
1405                         goto free;
1406                 }
1407                 page = virt_to_head_page(frag);
1408                 skb_fill_page_desc(skb, i - 1, page,
1409                                    frag - page_address(page), fragsz);
1410         }
1411
1412         return skb;
1413 free:
1414         /* frees skb and all frags allocated with napi_alloc_frag() */
1415         napi_free_frags(&tfile->napi);
1416         return ERR_PTR(err);
1417 }
1418
1419 /* prepad is the amount to reserve at front.  len is length after that.
1420  * linear is a hint as to how much to copy (usually headers). */
1421 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1422                                      size_t prepad, size_t len,
1423                                      size_t linear, int noblock)
1424 {
1425         struct sock *sk = tfile->socket.sk;
1426         struct sk_buff *skb;
1427         int err;
1428
1429         /* Under a page?  Don't bother with paged skb. */
1430         if (prepad + len < PAGE_SIZE || !linear)
1431                 linear = len;
1432
1433         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1434                                    &err, 0);
1435         if (!skb)
1436                 return ERR_PTR(err);
1437
1438         skb_reserve(skb, prepad);
1439         skb_put(skb, linear);
1440         skb->data_len = len - linear;
1441         skb->len += len - linear;
1442
1443         return skb;
1444 }
1445
1446 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1447                            struct sk_buff *skb, int more)
1448 {
1449         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1450         struct sk_buff_head process_queue;
1451         u32 rx_batched = tun->rx_batched;
1452         bool rcv = false;
1453
1454         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1455                 local_bh_disable();
1456                 skb_record_rx_queue(skb, tfile->queue_index);
1457                 netif_receive_skb(skb);
1458                 local_bh_enable();
1459                 return;
1460         }
1461
1462         spin_lock(&queue->lock);
1463         if (!more || skb_queue_len(queue) == rx_batched) {
1464                 __skb_queue_head_init(&process_queue);
1465                 skb_queue_splice_tail_init(queue, &process_queue);
1466                 rcv = true;
1467         } else {
1468                 __skb_queue_tail(queue, skb);
1469         }
1470         spin_unlock(&queue->lock);
1471
1472         if (rcv) {
1473                 struct sk_buff *nskb;
1474
1475                 local_bh_disable();
1476                 while ((nskb = __skb_dequeue(&process_queue))) {
1477                         skb_record_rx_queue(nskb, tfile->queue_index);
1478                         netif_receive_skb(nskb);
1479                 }
1480                 skb_record_rx_queue(skb, tfile->queue_index);
1481                 netif_receive_skb(skb);
1482                 local_bh_enable();
1483         }
1484 }
1485
1486 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1487                               int len, int noblock, bool zerocopy)
1488 {
1489         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1490                 return false;
1491
1492         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1493                 return false;
1494
1495         if (!noblock)
1496                 return false;
1497
1498         if (zerocopy)
1499                 return false;
1500
1501         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1502             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1503                 return false;
1504
1505         return true;
1506 }
1507
1508 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1509                                        struct page_frag *alloc_frag, char *buf,
1510                                        int buflen, int len, int pad)
1511 {
1512         struct sk_buff *skb = build_skb(buf, buflen);
1513
1514         if (!skb)
1515                 return ERR_PTR(-ENOMEM);
1516
1517         skb_reserve(skb, pad);
1518         skb_put(skb, len);
1519         skb_set_owner_w(skb, tfile->socket.sk);
1520
1521         get_page(alloc_frag->page);
1522         alloc_frag->offset += buflen;
1523
1524         return skb;
1525 }
1526
1527 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1528                        struct xdp_buff *xdp, u32 act)
1529 {
1530         int err;
1531
1532         switch (act) {
1533         case XDP_REDIRECT:
1534                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1535                 if (err)
1536                         return err;
1537                 break;
1538         case XDP_TX:
1539                 err = tun_xdp_tx(tun->dev, xdp);
1540                 if (err < 0)
1541                         return err;
1542                 break;
1543         case XDP_PASS:
1544                 break;
1545         default:
1546                 bpf_warn_invalid_xdp_action(act);
1547                 fallthrough;
1548         case XDP_ABORTED:
1549                 trace_xdp_exception(tun->dev, xdp_prog, act);
1550                 fallthrough;
1551         case XDP_DROP:
1552                 atomic_long_inc(&tun->dev->rx_dropped);
1553                 break;
1554         }
1555
1556         return act;
1557 }
1558
1559 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1560                                      struct tun_file *tfile,
1561                                      struct iov_iter *from,
1562                                      struct virtio_net_hdr *hdr,
1563                                      int len, int *skb_xdp)
1564 {
1565         struct page_frag *alloc_frag = &current->task_frag;
1566         struct bpf_prog *xdp_prog;
1567         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1568         char *buf;
1569         size_t copied;
1570         int pad = TUN_RX_PAD;
1571         int err = 0;
1572
1573         rcu_read_lock();
1574         xdp_prog = rcu_dereference(tun->xdp_prog);
1575         if (xdp_prog)
1576                 pad += XDP_PACKET_HEADROOM;
1577         buflen += SKB_DATA_ALIGN(len + pad);
1578         rcu_read_unlock();
1579
1580         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1581         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1582                 return ERR_PTR(-ENOMEM);
1583
1584         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1585         copied = copy_page_from_iter(alloc_frag->page,
1586                                      alloc_frag->offset + pad,
1587                                      len, from);
1588         if (copied != len)
1589                 return ERR_PTR(-EFAULT);
1590
1591         /* There's a small window that XDP may be set after the check
1592          * of xdp_prog above, this should be rare and for simplicity
1593          * we do XDP on skb in case the headroom is not enough.
1594          */
1595         if (hdr->gso_type || !xdp_prog) {
1596                 *skb_xdp = 1;
1597                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1598                                        pad);
1599         }
1600
1601         *skb_xdp = 0;
1602
1603         local_bh_disable();
1604         rcu_read_lock();
1605         xdp_prog = rcu_dereference(tun->xdp_prog);
1606         if (xdp_prog) {
1607                 struct xdp_buff xdp;
1608                 u32 act;
1609
1610                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1611                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1612
1613                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1614                 if (act == XDP_REDIRECT || act == XDP_TX) {
1615                         get_page(alloc_frag->page);
1616                         alloc_frag->offset += buflen;
1617                 }
1618                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1619                 if (err < 0) {
1620                         if (act == XDP_REDIRECT || act == XDP_TX)
1621                                 put_page(alloc_frag->page);
1622                         goto out;
1623                 }
1624
1625                 if (err == XDP_REDIRECT)
1626                         xdp_do_flush();
1627                 if (err != XDP_PASS)
1628                         goto out;
1629
1630                 pad = xdp.data - xdp.data_hard_start;
1631                 len = xdp.data_end - xdp.data;
1632         }
1633         rcu_read_unlock();
1634         local_bh_enable();
1635
1636         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1637
1638 out:
1639         rcu_read_unlock();
1640         local_bh_enable();
1641         return NULL;
1642 }
1643
1644 /* Get packet from user space buffer */
1645 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1646                             void *msg_control, struct iov_iter *from,
1647                             int noblock, bool more)
1648 {
1649         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1650         struct sk_buff *skb;
1651         size_t total_len = iov_iter_count(from);
1652         size_t len = total_len, align = tun->align, linear;
1653         struct virtio_net_hdr gso = { 0 };
1654         int good_linear;
1655         int copylen;
1656         bool zerocopy = false;
1657         int err;
1658         u32 rxhash = 0;
1659         int skb_xdp = 1;
1660         bool frags = tun_napi_frags_enabled(tfile);
1661
1662         if (!(tun->flags & IFF_NO_PI)) {
1663                 if (len < sizeof(pi))
1664                         return -EINVAL;
1665                 len -= sizeof(pi);
1666
1667                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1668                         return -EFAULT;
1669         }
1670
1671         if (tun->flags & IFF_VNET_HDR) {
1672                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1673
1674                 if (len < vnet_hdr_sz)
1675                         return -EINVAL;
1676                 len -= vnet_hdr_sz;
1677
1678                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1679                         return -EFAULT;
1680
1681                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1682                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1683                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1684
1685                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1686                         return -EINVAL;
1687                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1688         }
1689
1690         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1691                 align += NET_IP_ALIGN;
1692                 if (unlikely(len < ETH_HLEN ||
1693                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1694                         return -EINVAL;
1695         }
1696
1697         good_linear = SKB_MAX_HEAD(align);
1698
1699         if (msg_control) {
1700                 struct iov_iter i = *from;
1701
1702                 /* There are 256 bytes to be copied in skb, so there is
1703                  * enough room for skb expand head in case it is used.
1704                  * The rest of the buffer is mapped from userspace.
1705                  */
1706                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1707                 if (copylen > good_linear)
1708                         copylen = good_linear;
1709                 linear = copylen;
1710                 iov_iter_advance(&i, copylen);
1711                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1712                         zerocopy = true;
1713         }
1714
1715         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1716                 /* For the packet that is not easy to be processed
1717                  * (e.g gso or jumbo packet), we will do it at after
1718                  * skb was created with generic XDP routine.
1719                  */
1720                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1721                 if (IS_ERR(skb)) {
1722                         atomic_long_inc(&tun->dev->rx_dropped);
1723                         return PTR_ERR(skb);
1724                 }
1725                 if (!skb)
1726                         return total_len;
1727         } else {
1728                 if (!zerocopy) {
1729                         copylen = len;
1730                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1731                                 linear = good_linear;
1732                         else
1733                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1734                 }
1735
1736                 if (frags) {
1737                         mutex_lock(&tfile->napi_mutex);
1738                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1739                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1740                          * If zerocopy is enabled, then this layout will be
1741                          * overwritten by zerocopy_sg_from_iter().
1742                          */
1743                         zerocopy = false;
1744                 } else {
1745                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1746                                             noblock);
1747                 }
1748
1749                 if (IS_ERR(skb)) {
1750                         if (PTR_ERR(skb) != -EAGAIN)
1751                                 atomic_long_inc(&tun->dev->rx_dropped);
1752                         if (frags)
1753                                 mutex_unlock(&tfile->napi_mutex);
1754                         return PTR_ERR(skb);
1755                 }
1756
1757                 if (zerocopy)
1758                         err = zerocopy_sg_from_iter(skb, from);
1759                 else
1760                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1761
1762                 if (err) {
1763                         err = -EFAULT;
1764 drop:
1765                         atomic_long_inc(&tun->dev->rx_dropped);
1766                         kfree_skb(skb);
1767                         if (frags) {
1768                                 tfile->napi.skb = NULL;
1769                                 mutex_unlock(&tfile->napi_mutex);
1770                         }
1771
1772                         return err;
1773                 }
1774         }
1775
1776         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1777                 atomic_long_inc(&tun->rx_frame_errors);
1778                 kfree_skb(skb);
1779                 if (frags) {
1780                         tfile->napi.skb = NULL;
1781                         mutex_unlock(&tfile->napi_mutex);
1782                 }
1783
1784                 return -EINVAL;
1785         }
1786
1787         switch (tun->flags & TUN_TYPE_MASK) {
1788         case IFF_TUN:
1789                 if (tun->flags & IFF_NO_PI) {
1790                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1791
1792                         switch (ip_version) {
1793                         case 4:
1794                                 pi.proto = htons(ETH_P_IP);
1795                                 break;
1796                         case 6:
1797                                 pi.proto = htons(ETH_P_IPV6);
1798                                 break;
1799                         default:
1800                                 atomic_long_inc(&tun->dev->rx_dropped);
1801                                 kfree_skb(skb);
1802                                 return -EINVAL;
1803                         }
1804                 }
1805
1806                 skb_reset_mac_header(skb);
1807                 skb->protocol = pi.proto;
1808                 skb->dev = tun->dev;
1809                 break;
1810         case IFF_TAP:
1811                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1812                         err = -ENOMEM;
1813                         goto drop;
1814                 }
1815                 skb->protocol = eth_type_trans(skb, tun->dev);
1816                 break;
1817         }
1818
1819         /* copy skb_ubuf_info for callback when skb has no error */
1820         if (zerocopy) {
1821                 skb_zcopy_init(skb, msg_control);
1822         } else if (msg_control) {
1823                 struct ubuf_info *uarg = msg_control;
1824                 uarg->callback(NULL, uarg, false);
1825         }
1826
1827         skb_reset_network_header(skb);
1828         skb_probe_transport_header(skb);
1829         skb_record_rx_queue(skb, tfile->queue_index);
1830
1831         if (skb_xdp) {
1832                 struct bpf_prog *xdp_prog;
1833                 int ret;
1834
1835                 local_bh_disable();
1836                 rcu_read_lock();
1837                 xdp_prog = rcu_dereference(tun->xdp_prog);
1838                 if (xdp_prog) {
1839                         ret = do_xdp_generic(xdp_prog, skb);
1840                         if (ret != XDP_PASS) {
1841                                 rcu_read_unlock();
1842                                 local_bh_enable();
1843                                 if (frags) {
1844                                         tfile->napi.skb = NULL;
1845                                         mutex_unlock(&tfile->napi_mutex);
1846                                 }
1847                                 return total_len;
1848                         }
1849                 }
1850                 rcu_read_unlock();
1851                 local_bh_enable();
1852         }
1853
1854         /* Compute the costly rx hash only if needed for flow updates.
1855          * We may get a very small possibility of OOO during switching, not
1856          * worth to optimize.
1857          */
1858         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1859             !tfile->detached)
1860                 rxhash = __skb_get_hash_symmetric(skb);
1861
1862         rcu_read_lock();
1863         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1864                 err = -EIO;
1865                 rcu_read_unlock();
1866                 goto drop;
1867         }
1868
1869         if (frags) {
1870                 u32 headlen;
1871
1872                 /* Exercise flow dissector code path. */
1873                 skb_push(skb, ETH_HLEN);
1874                 headlen = eth_get_headlen(tun->dev, skb->data,
1875                                           skb_headlen(skb));
1876
1877                 if (unlikely(headlen > skb_headlen(skb))) {
1878                         atomic_long_inc(&tun->dev->rx_dropped);
1879                         napi_free_frags(&tfile->napi);
1880                         rcu_read_unlock();
1881                         mutex_unlock(&tfile->napi_mutex);
1882                         WARN_ON(1);
1883                         return -ENOMEM;
1884                 }
1885
1886                 local_bh_disable();
1887                 napi_gro_frags(&tfile->napi);
1888                 local_bh_enable();
1889                 mutex_unlock(&tfile->napi_mutex);
1890         } else if (tfile->napi_enabled) {
1891                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1892                 int queue_len;
1893
1894                 spin_lock_bh(&queue->lock);
1895                 __skb_queue_tail(queue, skb);
1896                 queue_len = skb_queue_len(queue);
1897                 spin_unlock(&queue->lock);
1898
1899                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1900                         napi_schedule(&tfile->napi);
1901
1902                 local_bh_enable();
1903         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1904                 tun_rx_batched(tun, tfile, skb, more);
1905         } else {
1906                 netif_rx_ni(skb);
1907         }
1908         rcu_read_unlock();
1909
1910         preempt_disable();
1911         dev_sw_netstats_rx_add(tun->dev, len);
1912         preempt_enable();
1913
1914         if (rxhash)
1915                 tun_flow_update(tun, rxhash, tfile);
1916
1917         return total_len;
1918 }
1919
1920 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1921 {
1922         struct file *file = iocb->ki_filp;
1923         struct tun_file *tfile = file->private_data;
1924         struct tun_struct *tun = tun_get(tfile);
1925         ssize_t result;
1926         int noblock = 0;
1927
1928         if (!tun)
1929                 return -EBADFD;
1930
1931         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1932                 noblock = 1;
1933
1934         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1935
1936         tun_put(tun);
1937         return result;
1938 }
1939
1940 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1941                                 struct tun_file *tfile,
1942                                 struct xdp_frame *xdp_frame,
1943                                 struct iov_iter *iter)
1944 {
1945         int vnet_hdr_sz = 0;
1946         size_t size = xdp_frame->len;
1947         size_t ret;
1948
1949         if (tun->flags & IFF_VNET_HDR) {
1950                 struct virtio_net_hdr gso = { 0 };
1951
1952                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1953                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1954                         return -EINVAL;
1955                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1956                              sizeof(gso)))
1957                         return -EFAULT;
1958                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1959         }
1960
1961         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1962
1963         preempt_disable();
1964         dev_sw_netstats_tx_add(tun->dev, 1, ret);
1965         preempt_enable();
1966
1967         return ret;
1968 }
1969
1970 /* Put packet to the user space buffer */
1971 static ssize_t tun_put_user(struct tun_struct *tun,
1972                             struct tun_file *tfile,
1973                             struct sk_buff *skb,
1974                             struct iov_iter *iter)
1975 {
1976         struct tun_pi pi = { 0, skb->protocol };
1977         ssize_t total;
1978         int vlan_offset = 0;
1979         int vlan_hlen = 0;
1980         int vnet_hdr_sz = 0;
1981
1982         if (skb_vlan_tag_present(skb))
1983                 vlan_hlen = VLAN_HLEN;
1984
1985         if (tun->flags & IFF_VNET_HDR)
1986                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1987
1988         total = skb->len + vlan_hlen + vnet_hdr_sz;
1989
1990         if (!(tun->flags & IFF_NO_PI)) {
1991                 if (iov_iter_count(iter) < sizeof(pi))
1992                         return -EINVAL;
1993
1994                 total += sizeof(pi);
1995                 if (iov_iter_count(iter) < total) {
1996                         /* Packet will be striped */
1997                         pi.flags |= TUN_PKT_STRIP;
1998                 }
1999
2000                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2001                         return -EFAULT;
2002         }
2003
2004         if (vnet_hdr_sz) {
2005                 struct virtio_net_hdr gso;
2006
2007                 if (iov_iter_count(iter) < vnet_hdr_sz)
2008                         return -EINVAL;
2009
2010                 if (virtio_net_hdr_from_skb(skb, &gso,
2011                                             tun_is_little_endian(tun), true,
2012                                             vlan_hlen)) {
2013                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2014                         pr_err("unexpected GSO type: "
2015                                "0x%x, gso_size %d, hdr_len %d\n",
2016                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2017                                tun16_to_cpu(tun, gso.hdr_len));
2018                         print_hex_dump(KERN_ERR, "tun: ",
2019                                        DUMP_PREFIX_NONE,
2020                                        16, 1, skb->head,
2021                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2022                         WARN_ON_ONCE(1);
2023                         return -EINVAL;
2024                 }
2025
2026                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2027                         return -EFAULT;
2028
2029                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2030         }
2031
2032         if (vlan_hlen) {
2033                 int ret;
2034                 struct veth veth;
2035
2036                 veth.h_vlan_proto = skb->vlan_proto;
2037                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2038
2039                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2040
2041                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2042                 if (ret || !iov_iter_count(iter))
2043                         goto done;
2044
2045                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2046                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2047                         goto done;
2048         }
2049
2050         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2051
2052 done:
2053         /* caller is in process context, */
2054         preempt_disable();
2055         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2056         preempt_enable();
2057
2058         return total;
2059 }
2060
2061 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2062 {
2063         DECLARE_WAITQUEUE(wait, current);
2064         void *ptr = NULL;
2065         int error = 0;
2066
2067         ptr = ptr_ring_consume(&tfile->tx_ring);
2068         if (ptr)
2069                 goto out;
2070         if (noblock) {
2071                 error = -EAGAIN;
2072                 goto out;
2073         }
2074
2075         add_wait_queue(&tfile->socket.wq.wait, &wait);
2076
2077         while (1) {
2078                 set_current_state(TASK_INTERRUPTIBLE);
2079                 ptr = ptr_ring_consume(&tfile->tx_ring);
2080                 if (ptr)
2081                         break;
2082                 if (signal_pending(current)) {
2083                         error = -ERESTARTSYS;
2084                         break;
2085                 }
2086                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2087                         error = -EFAULT;
2088                         break;
2089                 }
2090
2091                 schedule();
2092         }
2093
2094         __set_current_state(TASK_RUNNING);
2095         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2096
2097 out:
2098         *err = error;
2099         return ptr;
2100 }
2101
2102 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2103                            struct iov_iter *to,
2104                            int noblock, void *ptr)
2105 {
2106         ssize_t ret;
2107         int err;
2108
2109         if (!iov_iter_count(to)) {
2110                 tun_ptr_free(ptr);
2111                 return 0;
2112         }
2113
2114         if (!ptr) {
2115                 /* Read frames from ring */
2116                 ptr = tun_ring_recv(tfile, noblock, &err);
2117                 if (!ptr)
2118                         return err;
2119         }
2120
2121         if (tun_is_xdp_frame(ptr)) {
2122                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2123
2124                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2125                 xdp_return_frame(xdpf);
2126         } else {
2127                 struct sk_buff *skb = ptr;
2128
2129                 ret = tun_put_user(tun, tfile, skb, to);
2130                 if (unlikely(ret < 0))
2131                         kfree_skb(skb);
2132                 else
2133                         consume_skb(skb);
2134         }
2135
2136         return ret;
2137 }
2138
2139 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2140 {
2141         struct file *file = iocb->ki_filp;
2142         struct tun_file *tfile = file->private_data;
2143         struct tun_struct *tun = tun_get(tfile);
2144         ssize_t len = iov_iter_count(to), ret;
2145         int noblock = 0;
2146
2147         if (!tun)
2148                 return -EBADFD;
2149
2150         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2151                 noblock = 1;
2152
2153         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2154         ret = min_t(ssize_t, ret, len);
2155         if (ret > 0)
2156                 iocb->ki_pos = ret;
2157         tun_put(tun);
2158         return ret;
2159 }
2160
2161 static void tun_prog_free(struct rcu_head *rcu)
2162 {
2163         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2164
2165         bpf_prog_destroy(prog->prog);
2166         kfree(prog);
2167 }
2168
2169 static int __tun_set_ebpf(struct tun_struct *tun,
2170                           struct tun_prog __rcu **prog_p,
2171                           struct bpf_prog *prog)
2172 {
2173         struct tun_prog *old, *new = NULL;
2174
2175         if (prog) {
2176                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2177                 if (!new)
2178                         return -ENOMEM;
2179                 new->prog = prog;
2180         }
2181
2182         spin_lock_bh(&tun->lock);
2183         old = rcu_dereference_protected(*prog_p,
2184                                         lockdep_is_held(&tun->lock));
2185         rcu_assign_pointer(*prog_p, new);
2186         spin_unlock_bh(&tun->lock);
2187
2188         if (old)
2189                 call_rcu(&old->rcu, tun_prog_free);
2190
2191         return 0;
2192 }
2193
2194 static void tun_free_netdev(struct net_device *dev)
2195 {
2196         struct tun_struct *tun = netdev_priv(dev);
2197
2198         BUG_ON(!(list_empty(&tun->disabled)));
2199
2200         free_percpu(dev->tstats);
2201         /* We clear tstats so that tun_set_iff() can tell if
2202          * tun_free_netdev() has been called from register_netdevice().
2203          */
2204         dev->tstats = NULL;
2205
2206         tun_flow_uninit(tun);
2207         security_tun_dev_free_security(tun->security);
2208         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2209         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2210 }
2211
2212 static void tun_setup(struct net_device *dev)
2213 {
2214         struct tun_struct *tun = netdev_priv(dev);
2215
2216         tun->owner = INVALID_UID;
2217         tun->group = INVALID_GID;
2218         tun_default_link_ksettings(dev, &tun->link_ksettings);
2219
2220         dev->ethtool_ops = &tun_ethtool_ops;
2221         dev->needs_free_netdev = true;
2222         dev->priv_destructor = tun_free_netdev;
2223         /* We prefer our own queue length */
2224         dev->tx_queue_len = TUN_READQ_SIZE;
2225 }
2226
2227 /* Trivial set of netlink ops to allow deleting tun or tap
2228  * device with netlink.
2229  */
2230 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2231                         struct netlink_ext_ack *extack)
2232 {
2233         NL_SET_ERR_MSG(extack,
2234                        "tun/tap creation via rtnetlink is not supported.");
2235         return -EOPNOTSUPP;
2236 }
2237
2238 static size_t tun_get_size(const struct net_device *dev)
2239 {
2240         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2241         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2242
2243         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2244                nla_total_size(sizeof(gid_t)) + /* GROUP */
2245                nla_total_size(sizeof(u8)) + /* TYPE */
2246                nla_total_size(sizeof(u8)) + /* PI */
2247                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2248                nla_total_size(sizeof(u8)) + /* PERSIST */
2249                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2250                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2251                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2252                0;
2253 }
2254
2255 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2256 {
2257         struct tun_struct *tun = netdev_priv(dev);
2258
2259         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2260                 goto nla_put_failure;
2261         if (uid_valid(tun->owner) &&
2262             nla_put_u32(skb, IFLA_TUN_OWNER,
2263                         from_kuid_munged(current_user_ns(), tun->owner)))
2264                 goto nla_put_failure;
2265         if (gid_valid(tun->group) &&
2266             nla_put_u32(skb, IFLA_TUN_GROUP,
2267                         from_kgid_munged(current_user_ns(), tun->group)))
2268                 goto nla_put_failure;
2269         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2270                 goto nla_put_failure;
2271         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2272                 goto nla_put_failure;
2273         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2274                 goto nla_put_failure;
2275         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2276                        !!(tun->flags & IFF_MULTI_QUEUE)))
2277                 goto nla_put_failure;
2278         if (tun->flags & IFF_MULTI_QUEUE) {
2279                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2280                         goto nla_put_failure;
2281                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2282                                 tun->numdisabled))
2283                         goto nla_put_failure;
2284         }
2285
2286         return 0;
2287
2288 nla_put_failure:
2289         return -EMSGSIZE;
2290 }
2291
2292 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2293         .kind           = DRV_NAME,
2294         .priv_size      = sizeof(struct tun_struct),
2295         .setup          = tun_setup,
2296         .validate       = tun_validate,
2297         .get_size       = tun_get_size,
2298         .fill_info      = tun_fill_info,
2299 };
2300
2301 static void tun_sock_write_space(struct sock *sk)
2302 {
2303         struct tun_file *tfile;
2304         wait_queue_head_t *wqueue;
2305
2306         if (!sock_writeable(sk))
2307                 return;
2308
2309         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2310                 return;
2311
2312         wqueue = sk_sleep(sk);
2313         if (wqueue && waitqueue_active(wqueue))
2314                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2315                                                 EPOLLWRNORM | EPOLLWRBAND);
2316
2317         tfile = container_of(sk, struct tun_file, sk);
2318         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2319 }
2320
2321 static void tun_put_page(struct tun_page *tpage)
2322 {
2323         if (tpage->page)
2324                 __page_frag_cache_drain(tpage->page, tpage->count);
2325 }
2326
2327 static int tun_xdp_one(struct tun_struct *tun,
2328                        struct tun_file *tfile,
2329                        struct xdp_buff *xdp, int *flush,
2330                        struct tun_page *tpage)
2331 {
2332         unsigned int datasize = xdp->data_end - xdp->data;
2333         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2334         struct virtio_net_hdr *gso = &hdr->gso;
2335         struct bpf_prog *xdp_prog;
2336         struct sk_buff *skb = NULL;
2337         u32 rxhash = 0, act;
2338         int buflen = hdr->buflen;
2339         int err = 0;
2340         bool skb_xdp = false;
2341         struct page *page;
2342
2343         xdp_prog = rcu_dereference(tun->xdp_prog);
2344         if (xdp_prog) {
2345                 if (gso->gso_type) {
2346                         skb_xdp = true;
2347                         goto build;
2348                 }
2349
2350                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2351                 xdp_set_data_meta_invalid(xdp);
2352
2353                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2354                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2355                 if (err < 0) {
2356                         put_page(virt_to_head_page(xdp->data));
2357                         return err;
2358                 }
2359
2360                 switch (err) {
2361                 case XDP_REDIRECT:
2362                         *flush = true;
2363                         fallthrough;
2364                 case XDP_TX:
2365                         return 0;
2366                 case XDP_PASS:
2367                         break;
2368                 default:
2369                         page = virt_to_head_page(xdp->data);
2370                         if (tpage->page == page) {
2371                                 ++tpage->count;
2372                         } else {
2373                                 tun_put_page(tpage);
2374                                 tpage->page = page;
2375                                 tpage->count = 1;
2376                         }
2377                         return 0;
2378                 }
2379         }
2380
2381 build:
2382         skb = build_skb(xdp->data_hard_start, buflen);
2383         if (!skb) {
2384                 err = -ENOMEM;
2385                 goto out;
2386         }
2387
2388         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2389         skb_put(skb, xdp->data_end - xdp->data);
2390
2391         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2392                 atomic_long_inc(&tun->rx_frame_errors);
2393                 kfree_skb(skb);
2394                 err = -EINVAL;
2395                 goto out;
2396         }
2397
2398         skb->protocol = eth_type_trans(skb, tun->dev);
2399         skb_reset_network_header(skb);
2400         skb_probe_transport_header(skb);
2401         skb_record_rx_queue(skb, tfile->queue_index);
2402
2403         if (skb_xdp) {
2404                 err = do_xdp_generic(xdp_prog, skb);
2405                 if (err != XDP_PASS)
2406                         goto out;
2407         }
2408
2409         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2410             !tfile->detached)
2411                 rxhash = __skb_get_hash_symmetric(skb);
2412
2413         netif_receive_skb(skb);
2414
2415         /* No need to disable preemption here since this function is
2416          * always called with bh disabled
2417          */
2418         dev_sw_netstats_rx_add(tun->dev, datasize);
2419
2420         if (rxhash)
2421                 tun_flow_update(tun, rxhash, tfile);
2422
2423 out:
2424         return err;
2425 }
2426
2427 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2428 {
2429         int ret, i;
2430         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2431         struct tun_struct *tun = tun_get(tfile);
2432         struct tun_msg_ctl *ctl = m->msg_control;
2433         struct xdp_buff *xdp;
2434
2435         if (!tun)
2436                 return -EBADFD;
2437
2438         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2439                 struct tun_page tpage;
2440                 int n = ctl->num;
2441                 int flush = 0;
2442
2443                 memset(&tpage, 0, sizeof(tpage));
2444
2445                 local_bh_disable();
2446                 rcu_read_lock();
2447
2448                 for (i = 0; i < n; i++) {
2449                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2450                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2451                 }
2452
2453                 if (flush)
2454                         xdp_do_flush();
2455
2456                 rcu_read_unlock();
2457                 local_bh_enable();
2458
2459                 tun_put_page(&tpage);
2460
2461                 ret = total_len;
2462                 goto out;
2463         }
2464
2465         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2466                            m->msg_flags & MSG_DONTWAIT,
2467                            m->msg_flags & MSG_MORE);
2468 out:
2469         tun_put(tun);
2470         return ret;
2471 }
2472
2473 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2474                        int flags)
2475 {
2476         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2477         struct tun_struct *tun = tun_get(tfile);
2478         void *ptr = m->msg_control;
2479         int ret;
2480
2481         if (!tun) {
2482                 ret = -EBADFD;
2483                 goto out_free;
2484         }
2485
2486         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2487                 ret = -EINVAL;
2488                 goto out_put_tun;
2489         }
2490         if (flags & MSG_ERRQUEUE) {
2491                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2492                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2493                 goto out;
2494         }
2495         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2496         if (ret > (ssize_t)total_len) {
2497                 m->msg_flags |= MSG_TRUNC;
2498                 ret = flags & MSG_TRUNC ? ret : total_len;
2499         }
2500 out:
2501         tun_put(tun);
2502         return ret;
2503
2504 out_put_tun:
2505         tun_put(tun);
2506 out_free:
2507         tun_ptr_free(ptr);
2508         return ret;
2509 }
2510
2511 static int tun_ptr_peek_len(void *ptr)
2512 {
2513         if (likely(ptr)) {
2514                 if (tun_is_xdp_frame(ptr)) {
2515                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2516
2517                         return xdpf->len;
2518                 }
2519                 return __skb_array_len_with_tag(ptr);
2520         } else {
2521                 return 0;
2522         }
2523 }
2524
2525 static int tun_peek_len(struct socket *sock)
2526 {
2527         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2528         struct tun_struct *tun;
2529         int ret = 0;
2530
2531         tun = tun_get(tfile);
2532         if (!tun)
2533                 return 0;
2534
2535         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2536         tun_put(tun);
2537
2538         return ret;
2539 }
2540
2541 /* Ops structure to mimic raw sockets with tun */
2542 static const struct proto_ops tun_socket_ops = {
2543         .peek_len = tun_peek_len,
2544         .sendmsg = tun_sendmsg,
2545         .recvmsg = tun_recvmsg,
2546 };
2547
2548 static struct proto tun_proto = {
2549         .name           = "tun",
2550         .owner          = THIS_MODULE,
2551         .obj_size       = sizeof(struct tun_file),
2552 };
2553
2554 static int tun_flags(struct tun_struct *tun)
2555 {
2556         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2557 }
2558
2559 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2560                               char *buf)
2561 {
2562         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2563         return sprintf(buf, "0x%x\n", tun_flags(tun));
2564 }
2565
2566 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2567                               char *buf)
2568 {
2569         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2570         return uid_valid(tun->owner)?
2571                 sprintf(buf, "%u\n",
2572                         from_kuid_munged(current_user_ns(), tun->owner)):
2573                 sprintf(buf, "-1\n");
2574 }
2575
2576 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2577                               char *buf)
2578 {
2579         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2580         return gid_valid(tun->group) ?
2581                 sprintf(buf, "%u\n",
2582                         from_kgid_munged(current_user_ns(), tun->group)):
2583                 sprintf(buf, "-1\n");
2584 }
2585
2586 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2587 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2588 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2589
2590 static struct attribute *tun_dev_attrs[] = {
2591         &dev_attr_tun_flags.attr,
2592         &dev_attr_owner.attr,
2593         &dev_attr_group.attr,
2594         NULL
2595 };
2596
2597 static const struct attribute_group tun_attr_group = {
2598         .attrs = tun_dev_attrs
2599 };
2600
2601 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2602 {
2603         struct tun_struct *tun;
2604         struct tun_file *tfile = file->private_data;
2605         struct net_device *dev;
2606         int err;
2607
2608         if (tfile->detached)
2609                 return -EINVAL;
2610
2611         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2612                 if (!capable(CAP_NET_ADMIN))
2613                         return -EPERM;
2614
2615                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2616                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2617                         return -EINVAL;
2618         }
2619
2620         dev = __dev_get_by_name(net, ifr->ifr_name);
2621         if (dev) {
2622                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2623                         return -EBUSY;
2624                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2625                         tun = netdev_priv(dev);
2626                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2627                         tun = netdev_priv(dev);
2628                 else
2629                         return -EINVAL;
2630
2631                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2632                     !!(tun->flags & IFF_MULTI_QUEUE))
2633                         return -EINVAL;
2634
2635                 if (tun_not_capable(tun))
2636                         return -EPERM;
2637                 err = security_tun_dev_open(tun->security);
2638                 if (err < 0)
2639                         return err;
2640
2641                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2642                                  ifr->ifr_flags & IFF_NAPI,
2643                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2644                 if (err < 0)
2645                         return err;
2646
2647                 if (tun->flags & IFF_MULTI_QUEUE &&
2648                     (tun->numqueues + tun->numdisabled > 1)) {
2649                         /* One or more queue has already been attached, no need
2650                          * to initialize the device again.
2651                          */
2652                         netdev_state_change(dev);
2653                         return 0;
2654                 }
2655
2656                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2657                               (ifr->ifr_flags & TUN_FEATURES);
2658
2659                 netdev_state_change(dev);
2660         } else {
2661                 char *name;
2662                 unsigned long flags = 0;
2663                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2664                              MAX_TAP_QUEUES : 1;
2665
2666                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2667                         return -EPERM;
2668                 err = security_tun_dev_create();
2669                 if (err < 0)
2670                         return err;
2671
2672                 /* Set dev type */
2673                 if (ifr->ifr_flags & IFF_TUN) {
2674                         /* TUN device */
2675                         flags |= IFF_TUN;
2676                         name = "tun%d";
2677                 } else if (ifr->ifr_flags & IFF_TAP) {
2678                         /* TAP device */
2679                         flags |= IFF_TAP;
2680                         name = "tap%d";
2681                 } else
2682                         return -EINVAL;
2683
2684                 if (*ifr->ifr_name)
2685                         name = ifr->ifr_name;
2686
2687                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2688                                        NET_NAME_UNKNOWN, tun_setup, queues,
2689                                        queues);
2690
2691                 if (!dev)
2692                         return -ENOMEM;
2693
2694                 dev_net_set(dev, net);
2695                 dev->rtnl_link_ops = &tun_link_ops;
2696                 dev->ifindex = tfile->ifindex;
2697                 dev->sysfs_groups[0] = &tun_attr_group;
2698
2699                 tun = netdev_priv(dev);
2700                 tun->dev = dev;
2701                 tun->flags = flags;
2702                 tun->txflt.count = 0;
2703                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2704
2705                 tun->align = NET_SKB_PAD;
2706                 tun->filter_attached = false;
2707                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2708                 tun->rx_batched = 0;
2709                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2710
2711                 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2712                 if (!dev->tstats) {
2713                         err = -ENOMEM;
2714                         goto err_free_dev;
2715                 }
2716
2717                 spin_lock_init(&tun->lock);
2718
2719                 err = security_tun_dev_alloc_security(&tun->security);
2720                 if (err < 0)
2721                         goto err_free_stat;
2722
2723                 tun_net_init(dev);
2724                 tun_flow_init(tun);
2725
2726                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2727                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2728                                    NETIF_F_HW_VLAN_STAG_TX;
2729                 dev->features = dev->hw_features | NETIF_F_LLTX;
2730                 dev->vlan_features = dev->features &
2731                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2732                                        NETIF_F_HW_VLAN_STAG_TX);
2733
2734                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2735                               (ifr->ifr_flags & TUN_FEATURES);
2736
2737                 INIT_LIST_HEAD(&tun->disabled);
2738                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2739                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2740                 if (err < 0)
2741                         goto err_free_flow;
2742
2743                 err = register_netdevice(tun->dev);
2744                 if (err < 0)
2745                         goto err_detach;
2746                 /* free_netdev() won't check refcnt, to avoid race
2747                  * with dev_put() we need publish tun after registration.
2748                  */
2749                 rcu_assign_pointer(tfile->tun, tun);
2750         }
2751
2752         netif_carrier_on(tun->dev);
2753
2754         /* Make sure persistent devices do not get stuck in
2755          * xoff state.
2756          */
2757         if (netif_running(tun->dev))
2758                 netif_tx_wake_all_queues(tun->dev);
2759
2760         strcpy(ifr->ifr_name, tun->dev->name);
2761         return 0;
2762
2763 err_detach:
2764         tun_detach_all(dev);
2765         /* We are here because register_netdevice() has failed.
2766          * If register_netdevice() already called tun_free_netdev()
2767          * while dealing with the error, dev->stats has been cleared.
2768          */
2769         if (!dev->tstats)
2770                 goto err_free_dev;
2771
2772 err_free_flow:
2773         tun_flow_uninit(tun);
2774         security_tun_dev_free_security(tun->security);
2775 err_free_stat:
2776         free_percpu(dev->tstats);
2777 err_free_dev:
2778         free_netdev(dev);
2779         return err;
2780 }
2781
2782 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2783 {
2784         strcpy(ifr->ifr_name, tun->dev->name);
2785
2786         ifr->ifr_flags = tun_flags(tun);
2787
2788 }
2789
2790 /* This is like a cut-down ethtool ops, except done via tun fd so no
2791  * privs required. */
2792 static int set_offload(struct tun_struct *tun, unsigned long arg)
2793 {
2794         netdev_features_t features = 0;
2795
2796         if (arg & TUN_F_CSUM) {
2797                 features |= NETIF_F_HW_CSUM;
2798                 arg &= ~TUN_F_CSUM;
2799
2800                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2801                         if (arg & TUN_F_TSO_ECN) {
2802                                 features |= NETIF_F_TSO_ECN;
2803                                 arg &= ~TUN_F_TSO_ECN;
2804                         }
2805                         if (arg & TUN_F_TSO4)
2806                                 features |= NETIF_F_TSO;
2807                         if (arg & TUN_F_TSO6)
2808                                 features |= NETIF_F_TSO6;
2809                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2810                 }
2811
2812                 arg &= ~TUN_F_UFO;
2813         }
2814
2815         /* This gives the user a way to test for new features in future by
2816          * trying to set them. */
2817         if (arg)
2818                 return -EINVAL;
2819
2820         tun->set_features = features;
2821         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2822         tun->dev->wanted_features |= features;
2823         netdev_update_features(tun->dev);
2824
2825         return 0;
2826 }
2827
2828 static void tun_detach_filter(struct tun_struct *tun, int n)
2829 {
2830         int i;
2831         struct tun_file *tfile;
2832
2833         for (i = 0; i < n; i++) {
2834                 tfile = rtnl_dereference(tun->tfiles[i]);
2835                 lock_sock(tfile->socket.sk);
2836                 sk_detach_filter(tfile->socket.sk);
2837                 release_sock(tfile->socket.sk);
2838         }
2839
2840         tun->filter_attached = false;
2841 }
2842
2843 static int tun_attach_filter(struct tun_struct *tun)
2844 {
2845         int i, ret = 0;
2846         struct tun_file *tfile;
2847
2848         for (i = 0; i < tun->numqueues; i++) {
2849                 tfile = rtnl_dereference(tun->tfiles[i]);
2850                 lock_sock(tfile->socket.sk);
2851                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2852                 release_sock(tfile->socket.sk);
2853                 if (ret) {
2854                         tun_detach_filter(tun, i);
2855                         return ret;
2856                 }
2857         }
2858
2859         tun->filter_attached = true;
2860         return ret;
2861 }
2862
2863 static void tun_set_sndbuf(struct tun_struct *tun)
2864 {
2865         struct tun_file *tfile;
2866         int i;
2867
2868         for (i = 0; i < tun->numqueues; i++) {
2869                 tfile = rtnl_dereference(tun->tfiles[i]);
2870                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2871         }
2872 }
2873
2874 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2875 {
2876         struct tun_file *tfile = file->private_data;
2877         struct tun_struct *tun;
2878         int ret = 0;
2879
2880         rtnl_lock();
2881
2882         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2883                 tun = tfile->detached;
2884                 if (!tun) {
2885                         ret = -EINVAL;
2886                         goto unlock;
2887                 }
2888                 ret = security_tun_dev_attach_queue(tun->security);
2889                 if (ret < 0)
2890                         goto unlock;
2891                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2892                                  tun->flags & IFF_NAPI_FRAGS, true);
2893         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2894                 tun = rtnl_dereference(tfile->tun);
2895                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2896                         ret = -EINVAL;
2897                 else
2898                         __tun_detach(tfile, false);
2899         } else
2900                 ret = -EINVAL;
2901
2902         if (ret >= 0)
2903                 netdev_state_change(tun->dev);
2904
2905 unlock:
2906         rtnl_unlock();
2907         return ret;
2908 }
2909
2910 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2911                         void __user *data)
2912 {
2913         struct bpf_prog *prog;
2914         int fd;
2915
2916         if (copy_from_user(&fd, data, sizeof(fd)))
2917                 return -EFAULT;
2918
2919         if (fd == -1) {
2920                 prog = NULL;
2921         } else {
2922                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2923                 if (IS_ERR(prog))
2924                         return PTR_ERR(prog);
2925         }
2926
2927         return __tun_set_ebpf(tun, prog_p, prog);
2928 }
2929
2930 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2931 static unsigned char tun_get_addr_len(unsigned short type)
2932 {
2933         switch (type) {
2934         case ARPHRD_IP6GRE:
2935         case ARPHRD_TUNNEL6:
2936                 return sizeof(struct in6_addr);
2937         case ARPHRD_IPGRE:
2938         case ARPHRD_TUNNEL:
2939         case ARPHRD_SIT:
2940                 return 4;
2941         case ARPHRD_ETHER:
2942                 return ETH_ALEN;
2943         case ARPHRD_IEEE802154:
2944         case ARPHRD_IEEE802154_MONITOR:
2945                 return IEEE802154_EXTENDED_ADDR_LEN;
2946         case ARPHRD_PHONET_PIPE:
2947         case ARPHRD_PPP:
2948         case ARPHRD_NONE:
2949                 return 0;
2950         case ARPHRD_6LOWPAN:
2951                 return EUI64_ADDR_LEN;
2952         case ARPHRD_FDDI:
2953                 return FDDI_K_ALEN;
2954         case ARPHRD_HIPPI:
2955                 return HIPPI_ALEN;
2956         case ARPHRD_IEEE802:
2957                 return FC_ALEN;
2958         case ARPHRD_ROSE:
2959                 return ROSE_ADDR_LEN;
2960         case ARPHRD_NETROM:
2961                 return AX25_ADDR_LEN;
2962         case ARPHRD_LOCALTLK:
2963                 return LTALK_ALEN;
2964         default:
2965                 return 0;
2966         }
2967 }
2968
2969 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2970                             unsigned long arg, int ifreq_len)
2971 {
2972         struct tun_file *tfile = file->private_data;
2973         struct net *net = sock_net(&tfile->sk);
2974         struct tun_struct *tun;
2975         void __user* argp = (void __user*)arg;
2976         unsigned int ifindex, carrier;
2977         struct ifreq ifr;
2978         kuid_t owner;
2979         kgid_t group;
2980         int sndbuf;
2981         int vnet_hdr_sz;
2982         int le;
2983         int ret;
2984         bool do_notify = false;
2985
2986         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2987             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2988                 if (copy_from_user(&ifr, argp, ifreq_len))
2989                         return -EFAULT;
2990         } else {
2991                 memset(&ifr, 0, sizeof(ifr));
2992         }
2993         if (cmd == TUNGETFEATURES) {
2994                 /* Currently this just means: "what IFF flags are valid?".
2995                  * This is needed because we never checked for invalid flags on
2996                  * TUNSETIFF.
2997                  */
2998                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2999                                 (unsigned int __user*)argp);
3000         } else if (cmd == TUNSETQUEUE) {
3001                 return tun_set_queue(file, &ifr);
3002         } else if (cmd == SIOCGSKNS) {
3003                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3004                         return -EPERM;
3005                 return open_related_ns(&net->ns, get_net_ns);
3006         }
3007
3008         ret = 0;
3009         rtnl_lock();
3010
3011         tun = tun_get(tfile);
3012         if (cmd == TUNSETIFF) {
3013                 ret = -EEXIST;
3014                 if (tun)
3015                         goto unlock;
3016
3017                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3018
3019                 ret = tun_set_iff(net, file, &ifr);
3020
3021                 if (ret)
3022                         goto unlock;
3023
3024                 if (copy_to_user(argp, &ifr, ifreq_len))
3025                         ret = -EFAULT;
3026                 goto unlock;
3027         }
3028         if (cmd == TUNSETIFINDEX) {
3029                 ret = -EPERM;
3030                 if (tun)
3031                         goto unlock;
3032
3033                 ret = -EFAULT;
3034                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3035                         goto unlock;
3036
3037                 ret = 0;
3038                 tfile->ifindex = ifindex;
3039                 goto unlock;
3040         }
3041
3042         ret = -EBADFD;
3043         if (!tun)
3044                 goto unlock;
3045
3046         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3047
3048         net = dev_net(tun->dev);
3049         ret = 0;
3050         switch (cmd) {
3051         case TUNGETIFF:
3052                 tun_get_iff(tun, &ifr);
3053
3054                 if (tfile->detached)
3055                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3056                 if (!tfile->socket.sk->sk_filter)
3057                         ifr.ifr_flags |= IFF_NOFILTER;
3058
3059                 if (copy_to_user(argp, &ifr, ifreq_len))
3060                         ret = -EFAULT;
3061                 break;
3062
3063         case TUNSETNOCSUM:
3064                 /* Disable/Enable checksum */
3065
3066                 /* [unimplemented] */
3067                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3068                            arg ? "disabled" : "enabled");
3069                 break;
3070
3071         case TUNSETPERSIST:
3072                 /* Disable/Enable persist mode. Keep an extra reference to the
3073                  * module to prevent the module being unprobed.
3074                  */
3075                 if (arg && !(tun->flags & IFF_PERSIST)) {
3076                         tun->flags |= IFF_PERSIST;
3077                         __module_get(THIS_MODULE);
3078                         do_notify = true;
3079                 }
3080                 if (!arg && (tun->flags & IFF_PERSIST)) {
3081                         tun->flags &= ~IFF_PERSIST;
3082                         module_put(THIS_MODULE);
3083                         do_notify = true;
3084                 }
3085
3086                 netif_info(tun, drv, tun->dev, "persist %s\n",
3087                            arg ? "enabled" : "disabled");
3088                 break;
3089
3090         case TUNSETOWNER:
3091                 /* Set owner of the device */
3092                 owner = make_kuid(current_user_ns(), arg);
3093                 if (!uid_valid(owner)) {
3094                         ret = -EINVAL;
3095                         break;
3096                 }
3097                 tun->owner = owner;
3098                 do_notify = true;
3099                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3100                            from_kuid(&init_user_ns, tun->owner));
3101                 break;
3102
3103         case TUNSETGROUP:
3104                 /* Set group of the device */
3105                 group = make_kgid(current_user_ns(), arg);
3106                 if (!gid_valid(group)) {
3107                         ret = -EINVAL;
3108                         break;
3109                 }
3110                 tun->group = group;
3111                 do_notify = true;
3112                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3113                            from_kgid(&init_user_ns, tun->group));
3114                 break;
3115
3116         case TUNSETLINK:
3117                 /* Only allow setting the type when the interface is down */
3118                 if (tun->dev->flags & IFF_UP) {
3119                         netif_info(tun, drv, tun->dev,
3120                                    "Linktype set failed because interface is up\n");
3121                         ret = -EBUSY;
3122                 } else {
3123                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3124                                                        tun->dev);
3125                         ret = notifier_to_errno(ret);
3126                         if (ret) {
3127                                 netif_info(tun, drv, tun->dev,
3128                                            "Refused to change device type\n");
3129                                 break;
3130                         }
3131                         tun->dev->type = (int) arg;
3132                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3133                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3134                                    tun->dev->type);
3135                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3136                                                  tun->dev);
3137                 }
3138                 break;
3139
3140         case TUNSETDEBUG:
3141                 tun->msg_enable = (u32)arg;
3142                 break;
3143
3144         case TUNSETOFFLOAD:
3145                 ret = set_offload(tun, arg);
3146                 break;
3147
3148         case TUNSETTXFILTER:
3149                 /* Can be set only for TAPs */
3150                 ret = -EINVAL;
3151                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3152                         break;
3153                 ret = update_filter(&tun->txflt, (void __user *)arg);
3154                 break;
3155
3156         case SIOCGIFHWADDR:
3157                 /* Get hw address */
3158                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3159                 if (copy_to_user(argp, &ifr, ifreq_len))
3160                         ret = -EFAULT;
3161                 break;
3162
3163         case SIOCSIFHWADDR:
3164                 /* Set hw address */
3165                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3166                 break;
3167
3168         case TUNGETSNDBUF:
3169                 sndbuf = tfile->socket.sk->sk_sndbuf;
3170                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3171                         ret = -EFAULT;
3172                 break;
3173
3174         case TUNSETSNDBUF:
3175                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3176                         ret = -EFAULT;
3177                         break;
3178                 }
3179                 if (sndbuf <= 0) {
3180                         ret = -EINVAL;
3181                         break;
3182                 }
3183
3184                 tun->sndbuf = sndbuf;
3185                 tun_set_sndbuf(tun);
3186                 break;
3187
3188         case TUNGETVNETHDRSZ:
3189                 vnet_hdr_sz = tun->vnet_hdr_sz;
3190                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3191                         ret = -EFAULT;
3192                 break;
3193
3194         case TUNSETVNETHDRSZ:
3195                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3196                         ret = -EFAULT;
3197                         break;
3198                 }
3199                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3200                         ret = -EINVAL;
3201                         break;
3202                 }
3203
3204                 tun->vnet_hdr_sz = vnet_hdr_sz;
3205                 break;
3206
3207         case TUNGETVNETLE:
3208                 le = !!(tun->flags & TUN_VNET_LE);
3209                 if (put_user(le, (int __user *)argp))
3210                         ret = -EFAULT;
3211                 break;
3212
3213         case TUNSETVNETLE:
3214                 if (get_user(le, (int __user *)argp)) {
3215                         ret = -EFAULT;
3216                         break;
3217                 }
3218                 if (le)
3219                         tun->flags |= TUN_VNET_LE;
3220                 else
3221                         tun->flags &= ~TUN_VNET_LE;
3222                 break;
3223
3224         case TUNGETVNETBE:
3225                 ret = tun_get_vnet_be(tun, argp);
3226                 break;
3227
3228         case TUNSETVNETBE:
3229                 ret = tun_set_vnet_be(tun, argp);
3230                 break;
3231
3232         case TUNATTACHFILTER:
3233                 /* Can be set only for TAPs */
3234                 ret = -EINVAL;
3235                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3236                         break;
3237                 ret = -EFAULT;
3238                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3239                         break;
3240
3241                 ret = tun_attach_filter(tun);
3242                 break;
3243
3244         case TUNDETACHFILTER:
3245                 /* Can be set only for TAPs */
3246                 ret = -EINVAL;
3247                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3248                         break;
3249                 ret = 0;
3250                 tun_detach_filter(tun, tun->numqueues);
3251                 break;
3252
3253         case TUNGETFILTER:
3254                 ret = -EINVAL;
3255                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3256                         break;
3257                 ret = -EFAULT;
3258                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3259                         break;
3260                 ret = 0;
3261                 break;
3262
3263         case TUNSETSTEERINGEBPF:
3264                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3265                 break;
3266
3267         case TUNSETFILTEREBPF:
3268                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3269                 break;
3270
3271         case TUNSETCARRIER:
3272                 ret = -EFAULT;
3273                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3274                         goto unlock;
3275
3276                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3277                 break;
3278
3279         case TUNGETDEVNETNS:
3280                 ret = -EPERM;
3281                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3282                         goto unlock;
3283                 ret = open_related_ns(&net->ns, get_net_ns);
3284                 break;
3285
3286         default:
3287                 ret = -EINVAL;
3288                 break;
3289         }
3290
3291         if (do_notify)
3292                 netdev_state_change(tun->dev);
3293
3294 unlock:
3295         rtnl_unlock();
3296         if (tun)
3297                 tun_put(tun);
3298         return ret;
3299 }
3300
3301 static long tun_chr_ioctl(struct file *file,
3302                           unsigned int cmd, unsigned long arg)
3303 {
3304         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3305 }
3306
3307 #ifdef CONFIG_COMPAT
3308 static long tun_chr_compat_ioctl(struct file *file,
3309                          unsigned int cmd, unsigned long arg)
3310 {
3311         switch (cmd) {
3312         case TUNSETIFF:
3313         case TUNGETIFF:
3314         case TUNSETTXFILTER:
3315         case TUNGETSNDBUF:
3316         case TUNSETSNDBUF:
3317         case SIOCGIFHWADDR:
3318         case SIOCSIFHWADDR:
3319                 arg = (unsigned long)compat_ptr(arg);
3320                 break;
3321         default:
3322                 arg = (compat_ulong_t)arg;
3323                 break;
3324         }
3325
3326         /*
3327          * compat_ifreq is shorter than ifreq, so we must not access beyond
3328          * the end of that structure. All fields that are used in this
3329          * driver are compatible though, we don't need to convert the
3330          * contents.
3331          */
3332         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3333 }
3334 #endif /* CONFIG_COMPAT */
3335
3336 static int tun_chr_fasync(int fd, struct file *file, int on)
3337 {
3338         struct tun_file *tfile = file->private_data;
3339         int ret;
3340
3341         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3342                 goto out;
3343
3344         if (on) {
3345                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3346                 tfile->flags |= TUN_FASYNC;
3347         } else
3348                 tfile->flags &= ~TUN_FASYNC;
3349         ret = 0;
3350 out:
3351         return ret;
3352 }
3353
3354 static int tun_chr_open(struct inode *inode, struct file * file)
3355 {
3356         struct net *net = current->nsproxy->net_ns;
3357         struct tun_file *tfile;
3358
3359         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3360                                             &tun_proto, 0);
3361         if (!tfile)
3362                 return -ENOMEM;
3363         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3364                 sk_free(&tfile->sk);
3365                 return -ENOMEM;
3366         }
3367
3368         mutex_init(&tfile->napi_mutex);
3369         RCU_INIT_POINTER(tfile->tun, NULL);
3370         tfile->flags = 0;
3371         tfile->ifindex = 0;
3372
3373         init_waitqueue_head(&tfile->socket.wq.wait);
3374
3375         tfile->socket.file = file;
3376         tfile->socket.ops = &tun_socket_ops;
3377
3378         sock_init_data(&tfile->socket, &tfile->sk);
3379
3380         tfile->sk.sk_write_space = tun_sock_write_space;
3381         tfile->sk.sk_sndbuf = INT_MAX;
3382
3383         file->private_data = tfile;
3384         INIT_LIST_HEAD(&tfile->next);
3385
3386         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3387
3388         return 0;
3389 }
3390
3391 static int tun_chr_close(struct inode *inode, struct file *file)
3392 {
3393         struct tun_file *tfile = file->private_data;
3394
3395         tun_detach(tfile, true);
3396
3397         return 0;
3398 }
3399
3400 #ifdef CONFIG_PROC_FS
3401 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3402 {
3403         struct tun_file *tfile = file->private_data;
3404         struct tun_struct *tun;
3405         struct ifreq ifr;
3406
3407         memset(&ifr, 0, sizeof(ifr));
3408
3409         rtnl_lock();
3410         tun = tun_get(tfile);
3411         if (tun)
3412                 tun_get_iff(tun, &ifr);
3413         rtnl_unlock();
3414
3415         if (tun)
3416                 tun_put(tun);
3417
3418         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3419 }
3420 #endif
3421
3422 static const struct file_operations tun_fops = {
3423         .owner  = THIS_MODULE,
3424         .llseek = no_llseek,
3425         .read_iter  = tun_chr_read_iter,
3426         .write_iter = tun_chr_write_iter,
3427         .poll   = tun_chr_poll,
3428         .unlocked_ioctl = tun_chr_ioctl,
3429 #ifdef CONFIG_COMPAT
3430         .compat_ioctl = tun_chr_compat_ioctl,
3431 #endif
3432         .open   = tun_chr_open,
3433         .release = tun_chr_close,
3434         .fasync = tun_chr_fasync,
3435 #ifdef CONFIG_PROC_FS
3436         .show_fdinfo = tun_chr_show_fdinfo,
3437 #endif
3438 };
3439
3440 static struct miscdevice tun_miscdev = {
3441         .minor = TUN_MINOR,
3442         .name = "tun",
3443         .nodename = "net/tun",
3444         .fops = &tun_fops,
3445 };
3446
3447 /* ethtool interface */
3448
3449 static void tun_default_link_ksettings(struct net_device *dev,
3450                                        struct ethtool_link_ksettings *cmd)
3451 {
3452         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3453         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3454         cmd->base.speed         = SPEED_10;
3455         cmd->base.duplex        = DUPLEX_FULL;
3456         cmd->base.port          = PORT_TP;
3457         cmd->base.phy_address   = 0;
3458         cmd->base.autoneg       = AUTONEG_DISABLE;
3459 }
3460
3461 static int tun_get_link_ksettings(struct net_device *dev,
3462                                   struct ethtool_link_ksettings *cmd)
3463 {
3464         struct tun_struct *tun = netdev_priv(dev);
3465
3466         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3467         return 0;
3468 }
3469
3470 static int tun_set_link_ksettings(struct net_device *dev,
3471                                   const struct ethtool_link_ksettings *cmd)
3472 {
3473         struct tun_struct *tun = netdev_priv(dev);
3474
3475         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3476         return 0;
3477 }
3478
3479 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3480 {
3481         struct tun_struct *tun = netdev_priv(dev);
3482
3483         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3484         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3485
3486         switch (tun->flags & TUN_TYPE_MASK) {
3487         case IFF_TUN:
3488                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3489                 break;
3490         case IFF_TAP:
3491                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3492                 break;
3493         }
3494 }
3495
3496 static u32 tun_get_msglevel(struct net_device *dev)
3497 {
3498         struct tun_struct *tun = netdev_priv(dev);
3499
3500         return tun->msg_enable;
3501 }
3502
3503 static void tun_set_msglevel(struct net_device *dev, u32 value)
3504 {
3505         struct tun_struct *tun = netdev_priv(dev);
3506
3507         tun->msg_enable = value;
3508 }
3509
3510 static int tun_get_coalesce(struct net_device *dev,
3511                             struct ethtool_coalesce *ec)
3512 {
3513         struct tun_struct *tun = netdev_priv(dev);
3514
3515         ec->rx_max_coalesced_frames = tun->rx_batched;
3516
3517         return 0;
3518 }
3519
3520 static int tun_set_coalesce(struct net_device *dev,
3521                             struct ethtool_coalesce *ec)
3522 {
3523         struct tun_struct *tun = netdev_priv(dev);
3524
3525         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3526                 tun->rx_batched = NAPI_POLL_WEIGHT;
3527         else
3528                 tun->rx_batched = ec->rx_max_coalesced_frames;
3529
3530         return 0;
3531 }
3532
3533 static const struct ethtool_ops tun_ethtool_ops = {
3534         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3535         .get_drvinfo    = tun_get_drvinfo,
3536         .get_msglevel   = tun_get_msglevel,
3537         .set_msglevel   = tun_set_msglevel,
3538         .get_link       = ethtool_op_get_link,
3539         .get_ts_info    = ethtool_op_get_ts_info,
3540         .get_coalesce   = tun_get_coalesce,
3541         .set_coalesce   = tun_set_coalesce,
3542         .get_link_ksettings = tun_get_link_ksettings,
3543         .set_link_ksettings = tun_set_link_ksettings,
3544 };
3545
3546 static int tun_queue_resize(struct tun_struct *tun)
3547 {
3548         struct net_device *dev = tun->dev;
3549         struct tun_file *tfile;
3550         struct ptr_ring **rings;
3551         int n = tun->numqueues + tun->numdisabled;
3552         int ret, i;
3553
3554         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3555         if (!rings)
3556                 return -ENOMEM;
3557
3558         for (i = 0; i < tun->numqueues; i++) {
3559                 tfile = rtnl_dereference(tun->tfiles[i]);
3560                 rings[i] = &tfile->tx_ring;
3561         }
3562         list_for_each_entry(tfile, &tun->disabled, next)
3563                 rings[i++] = &tfile->tx_ring;
3564
3565         ret = ptr_ring_resize_multiple(rings, n,
3566                                        dev->tx_queue_len, GFP_KERNEL,
3567                                        tun_ptr_free);
3568
3569         kfree(rings);
3570         return ret;
3571 }
3572
3573 static int tun_device_event(struct notifier_block *unused,
3574                             unsigned long event, void *ptr)
3575 {
3576         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3577         struct tun_struct *tun = netdev_priv(dev);
3578         int i;
3579
3580         if (dev->rtnl_link_ops != &tun_link_ops)
3581                 return NOTIFY_DONE;
3582
3583         switch (event) {
3584         case NETDEV_CHANGE_TX_QUEUE_LEN:
3585                 if (tun_queue_resize(tun))
3586                         return NOTIFY_BAD;
3587                 break;
3588         case NETDEV_UP:
3589                 for (i = 0; i < tun->numqueues; i++) {
3590                         struct tun_file *tfile;
3591
3592                         tfile = rtnl_dereference(tun->tfiles[i]);
3593                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3594                 }
3595                 break;
3596         default:
3597                 break;
3598         }
3599
3600         return NOTIFY_DONE;
3601 }
3602
3603 static struct notifier_block tun_notifier_block __read_mostly = {
3604         .notifier_call  = tun_device_event,
3605 };
3606
3607 static int __init tun_init(void)
3608 {
3609         int ret = 0;
3610
3611         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3612
3613         ret = rtnl_link_register(&tun_link_ops);
3614         if (ret) {
3615                 pr_err("Can't register link_ops\n");
3616                 goto err_linkops;
3617         }
3618
3619         ret = misc_register(&tun_miscdev);
3620         if (ret) {
3621                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3622                 goto err_misc;
3623         }
3624
3625         ret = register_netdevice_notifier(&tun_notifier_block);
3626         if (ret) {
3627                 pr_err("Can't register netdevice notifier\n");
3628                 goto err_notifier;
3629         }
3630
3631         return  0;
3632
3633 err_notifier:
3634         misc_deregister(&tun_miscdev);
3635 err_misc:
3636         rtnl_link_unregister(&tun_link_ops);
3637 err_linkops:
3638         return ret;
3639 }
3640
3641 static void tun_cleanup(void)
3642 {
3643         misc_deregister(&tun_miscdev);
3644         rtnl_link_unregister(&tun_link_ops);
3645         unregister_netdevice_notifier(&tun_notifier_block);
3646 }
3647
3648 /* Get an underlying socket object from tun file.  Returns error unless file is
3649  * attached to a device.  The returned object works like a packet socket, it
3650  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3651  * holding a reference to the file for as long as the socket is in use. */
3652 struct socket *tun_get_socket(struct file *file)
3653 {
3654         struct tun_file *tfile;
3655         if (file->f_op != &tun_fops)
3656                 return ERR_PTR(-EINVAL);
3657         tfile = file->private_data;
3658         if (!tfile)
3659                 return ERR_PTR(-EBADFD);
3660         return &tfile->socket;
3661 }
3662 EXPORT_SYMBOL_GPL(tun_get_socket);
3663
3664 struct ptr_ring *tun_get_tx_ring(struct file *file)
3665 {
3666         struct tun_file *tfile;
3667
3668         if (file->f_op != &tun_fops)
3669                 return ERR_PTR(-EINVAL);
3670         tfile = file->private_data;
3671         if (!tfile)
3672                 return ERR_PTR(-EBADFD);
3673         return &tfile->tx_ring;
3674 }
3675 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3676
3677 module_init(tun_init);
3678 module_exit(tun_cleanup);
3679 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3680 MODULE_AUTHOR(DRV_COPYRIGHT);
3681 MODULE_LICENSE("GPL");
3682 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3683 MODULE_ALIAS("devname:net/tun");