drivers/net/tun.c

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