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/math.h>
  58 #include <linux/nsproxy.h>
  59 #include <linux/virtio_net.h>
  60 #include <linux/rcupdate.h>
  61 #include <net/net_namespace.h>
  62 #include <net/netns/generic.h>
  63 #include <net/rtnetlink.h>
  64 #include <net/sock.h>
  65 #include <net/xdp.h>
  66 #include <net/ip_tunnels.h>
  67 #include <linux/seq_file.h>
  68 #include <linux/uio.h>
  69 #include <linux/skb_array.h>
  70 #include <linux/bpf.h>
  71 #include <linux/bpf_trace.h>
  72 #include <linux/mutex.h>
  73 #include <linux/ieee802154.h>
  74 #include <linux/if_ltalk.h>
  75 #include <uapi/linux/if_fddi.h>
  76 #include <uapi/linux/if_hippi.h>
  77 #include <uapi/linux/if_fc.h>
  78 #include <net/ax25.h>
  79 #include <net/rose.h>
  80 #include <net/6lowpan.h>
  81 #include <net/rps.h>
  82
  83 #include <linux/uaccess.h>
  84 #include <linux/proc_fs.h>
  85
  86 static void tun_default_link_ksettings(struct net_device *dev,
  87                                        struct ethtool_link_ksettings *cmd);
  88
  89 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  90
  91 /* TUN device flags */
  92
  93 /* IFF_ATTACH_QUEUE is never stored in device flags,
  94  * overload it to mean fasync when stored there.
  95  */
  96 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  97 /* High bits in flags field are unused. */
  98 #define TUN_VNET_LE     0x80000000
  99 #define TUN_VNET_BE     0x40000000
 100
 101 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 102                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 114  * to max number of VCPUs in guest. */
 115 #define MAX_TAP_QUEUES 256
 116 #define MAX_TAP_FLOWS  4096
 117
 118 #define TUN_FLOW_EXPIRE (3 * HZ)
 119
 120 /* A tun_file connects an open character device to a tuntap netdevice. It
 121  * also contains all socket related structures (except sock_fprog and tap_filter)
 122  * to serve as one transmit queue for tuntap device. The sock_fprog and
 123  * tap_filter were kept in tun_struct since they were used for filtering for the
 124  * netdevice not for a specific queue (at least I didn't see the requirement for
 125  * this).
 126  *
 127  * RCU usage:
 128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 129  * other can only be read while rcu_read_lock or rtnl_lock is held.
 130  */
 131 struct tun_file {
 132         struct sock sk;
 133         struct socket socket;
 134         struct tun_struct __rcu *tun;
 135         struct fasync_struct *fasync;
 136         /* only used for fasnyc */
 137         unsigned int flags;
 138         union {
 139                 u16 queue_index;
 140                 unsigned int ifindex;
 141         };
 142         struct napi_struct napi;
 143         bool napi_enabled;
 144         bool napi_frags_enabled;
 145         struct mutex napi_mutex;        /* Protects access to the above napi */
 146         struct list_head next;
 147         struct tun_struct *detached;
 148         struct ptr_ring tx_ring;
 149         struct xdp_rxq_info xdp_rxq;
 150 };
 151
 152 struct tun_page {
 153         struct page *page;
 154         int count;
 155 };
 156
 157 struct tun_flow_entry {
 158         struct hlist_node hash_link;
 159         struct rcu_head rcu;
 160         struct tun_struct *tun;
 161
 162         u32 rxhash;
 163         u32 rps_rxhash;
 164         int queue_index;
 165         unsigned long updated ____cacheline_aligned_in_smp;
 166 };
 167
 168 #define TUN_NUM_FLOW_ENTRIES 1024
 169 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 170
 171 struct tun_prog {
 172         struct rcu_head rcu;
 173         struct bpf_prog *prog;
 174 };
 175
 176 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 177  * device, socket filter, sndbuf and vnet header size were restore when the
 178  * file were attached to a persist device.
 179  */
 180 struct tun_struct {
 181         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 182         unsigned int            numqueues;
 183         unsigned int            flags;
 184         kuid_t                  owner;
 185         kgid_t                  group;
 186
 187         struct net_device       *dev;
 188         netdev_features_t       set_features;
 189 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 190                           NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)
 191
 192         int                     align;
 193         int                     vnet_hdr_sz;
 194         int                     sndbuf;
 195         struct tap_filter       txflt;
 196         struct sock_fprog       fprog;
 197         /* protected by rtnl lock */
 198         bool                    filter_attached;
 199         u32                     msg_enable;
 200         spinlock_t lock;
 201         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202         struct timer_list flow_gc_timer;
 203         unsigned long ageing_time;
 204         unsigned int numdisabled;
 205         struct list_head disabled;
 206         void *security;
 207         u32 flow_count;
 208         u32 rx_batched;
 209         atomic_long_t rx_frame_errors;
 210         struct bpf_prog __rcu *xdp_prog;
 211         struct tun_prog __rcu *steering_prog;
 212         struct tun_prog __rcu *filter_prog;
 213         struct ethtool_link_ksettings link_ksettings;
 214         /* init args */
 215         struct file *file;
 216         struct ifreq *ifr;
 217 };
 218
 219 struct veth {
 220         __be16 h_vlan_proto;
 221         __be16 h_vlan_TCI;
 222 };
 223
 224 static void tun_flow_init(struct tun_struct *tun);
 225 static void tun_flow_uninit(struct tun_struct *tun);
 226
 227 static int tun_napi_receive(struct napi_struct *napi, int budget)
 228 {
 229         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 230         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 231         struct sk_buff_head process_queue;
 232         struct sk_buff *skb;
 233         int received = 0;
 234
 235         __skb_queue_head_init(&process_queue);
 236
 237         spin_lock(&queue->lock);
 238         skb_queue_splice_tail_init(queue, &process_queue);
 239         spin_unlock(&queue->lock);
 240
 241         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 242                 napi_gro_receive(napi, skb);
 243                 ++received;
 244         }
 245
 246         if (!skb_queue_empty(&process_queue)) {
 247                 spin_lock(&queue->lock);
 248                 skb_queue_splice(&process_queue, queue);
 249                 spin_unlock(&queue->lock);
 250         }
 251
 252         return received;
 253 }
 254
 255 static int tun_napi_poll(struct napi_struct *napi, int budget)
 256 {
 257         unsigned int received;
 258
 259         received = tun_napi_receive(napi, budget);
 260
 261         if (received < budget)
 262                 napi_complete_done(napi, received);
 263
 264         return received;
 265 }
 266
 267 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 268                           bool napi_en, bool napi_frags)
 269 {
 270         tfile->napi_enabled = napi_en;
 271         tfile->napi_frags_enabled = napi_en && napi_frags;
 272         if (napi_en) {
 273                 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 274                 napi_enable(&tfile->napi);
 275         }
 276 }
 277
 278 static void tun_napi_enable(struct tun_file *tfile)
 279 {
 280         if (tfile->napi_enabled)
 281                 napi_enable(&tfile->napi);
 282 }
 283
 284 static void tun_napi_disable(struct tun_file *tfile)
 285 {
 286         if (tfile->napi_enabled)
 287                 napi_disable(&tfile->napi);
 288 }
 289
 290 static void tun_napi_del(struct tun_file *tfile)
 291 {
 292         if (tfile->napi_enabled)
 293                 netif_napi_del(&tfile->napi);
 294 }
 295
 296 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 297 {
 298         return tfile->napi_frags_enabled;
 299 }
 300
 301 #ifdef CONFIG_TUN_VNET_CROSS_LE
 302 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 303 {
 304         return tun->flags & TUN_VNET_BE ? false :
 305                 virtio_legacy_is_little_endian();
 306 }
 307
 308 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 309 {
 310         int be = !!(tun->flags & TUN_VNET_BE);
 311
 312         if (put_user(be, argp))
 313                 return -EFAULT;
 314
 315         return 0;
 316 }
 317
 318 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 319 {
 320         int be;
 321
 322         if (get_user(be, argp))
 323                 return -EFAULT;
 324
 325         if (be)
 326                 tun->flags |= TUN_VNET_BE;
 327         else
 328                 tun->flags &= ~TUN_VNET_BE;
 329
 330         return 0;
 331 }
 332 #else
 333 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 334 {
 335         return virtio_legacy_is_little_endian();
 336 }
 337
 338 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 339 {
 340         return -EINVAL;
 341 }
 342
 343 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 344 {
 345         return -EINVAL;
 346 }
 347 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 348
 349 static inline bool tun_is_little_endian(struct tun_struct *tun)
 350 {
 351         return tun->flags & TUN_VNET_LE ||
 352                 tun_legacy_is_little_endian(tun);
 353 }
 354
 355 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 356 {
 357         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 358 }
 359
 360 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 361 {
 362         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 363 }
 364
 365 static inline u32 tun_hashfn(u32 rxhash)
 366 {
 367         return rxhash & TUN_MASK_FLOW_ENTRIES;
 368 }
 369
 370 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 371 {
 372         struct tun_flow_entry *e;
 373
 374         hlist_for_each_entry_rcu(e, head, hash_link) {
 375                 if (e->rxhash == rxhash)
 376                         return e;
 377         }
 378         return NULL;
 379 }
 380
 381 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 382                                               struct hlist_head *head,
 383                                               u32 rxhash, u16 queue_index)
 384 {
 385         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 386
 387         if (e) {
 388                 netif_info(tun, tx_queued, tun->dev,
 389                            "create flow: hash %u index %u\n",
 390                            rxhash, queue_index);
 391                 e->updated = jiffies;
 392                 e->rxhash = rxhash;
 393                 e->rps_rxhash = 0;
 394                 e->queue_index = queue_index;
 395                 e->tun = tun;
 396                 hlist_add_head_rcu(&e->hash_link, head);
 397                 ++tun->flow_count;
 398         }
 399         return e;
 400 }
 401
 402 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 403 {
 404         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 405                    e->rxhash, e->queue_index);
 406         hlist_del_rcu(&e->hash_link);
 407         kfree_rcu(e, rcu);
 408         --tun->flow_count;
 409 }
 410
 411 static void tun_flow_flush(struct tun_struct *tun)
 412 {
 413         int i;
 414
 415         spin_lock_bh(&tun->lock);
 416         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 417                 struct tun_flow_entry *e;
 418                 struct hlist_node *n;
 419
 420                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 421                         tun_flow_delete(tun, e);
 422         }
 423         spin_unlock_bh(&tun->lock);
 424 }
 425
 426 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 427 {
 428         int i;
 429
 430         spin_lock_bh(&tun->lock);
 431         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 432                 struct tun_flow_entry *e;
 433                 struct hlist_node *n;
 434
 435                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 436                         if (e->queue_index == queue_index)
 437                                 tun_flow_delete(tun, e);
 438                 }
 439         }
 440         spin_unlock_bh(&tun->lock);
 441 }
 442
 443 static void tun_flow_cleanup(struct timer_list *t)
 444 {
 445         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 446         unsigned long delay = tun->ageing_time;
 447         unsigned long next_timer = jiffies + delay;
 448         unsigned long count = 0;
 449         int i;
 450
 451         spin_lock(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 457                         unsigned long this_timer;
 458
 459                         this_timer = e->updated + delay;
 460                         if (time_before_eq(this_timer, jiffies)) {
 461                                 tun_flow_delete(tun, e);
 462                                 continue;
 463                         }
 464                         count++;
 465                         if (time_before(this_timer, next_timer))
 466                                 next_timer = this_timer;
 467                 }
 468         }
 469
 470         if (count)
 471                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 472         spin_unlock(&tun->lock);
 473 }
 474
 475 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 476                             struct tun_file *tfile)
 477 {
 478         struct hlist_head *head;
 479         struct tun_flow_entry *e;
 480         unsigned long delay = tun->ageing_time;
 481         u16 queue_index = tfile->queue_index;
 482
 483         head = &tun->flows[tun_hashfn(rxhash)];
 484
 485         rcu_read_lock();
 486
 487         e = tun_flow_find(head, rxhash);
 488         if (likely(e)) {
 489                 /* TODO: keep queueing to old queue until it's empty? */
 490                 if (READ_ONCE(e->queue_index) != queue_index)
 491                         WRITE_ONCE(e->queue_index, queue_index);
 492                 if (e->updated != jiffies)
 493                         e->updated = jiffies;
 494                 sock_rps_record_flow_hash(e->rps_rxhash);
 495         } else {
 496                 spin_lock_bh(&tun->lock);
 497                 if (!tun_flow_find(head, rxhash) &&
 498                     tun->flow_count < MAX_TAP_FLOWS)
 499                         tun_flow_create(tun, head, rxhash, queue_index);
 500
 501                 if (!timer_pending(&tun->flow_gc_timer))
 502                         mod_timer(&tun->flow_gc_timer,
 503                                   round_jiffies_up(jiffies + delay));
 504                 spin_unlock_bh(&tun->lock);
 505         }
 506
 507         rcu_read_unlock();
 508 }
 509
 510 /* Save the hash received in the stack receive path and update the
 511  * flow_hash table accordingly.
 512  */
 513 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 514 {
 515         if (unlikely(e->rps_rxhash != hash))
 516                 e->rps_rxhash = hash;
 517 }
 518
 519 /* We try to identify a flow through its rxhash. The reason that
 520  * we do not check rxq no. is because some cards(e.g 82599), chooses
 521  * the rxq based on the txq where the last packet of the flow comes. As
 522  * the userspace application move between processors, we may get a
 523  * different rxq no. here.
 524  */
 525 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 526 {
 527         struct tun_flow_entry *e;
 528         u32 txq, numqueues;
 529
 530         numqueues = READ_ONCE(tun->numqueues);
 531
 532         txq = __skb_get_hash_symmetric(skb);
 533         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 534         if (e) {
 535                 tun_flow_save_rps_rxhash(e, txq);
 536                 txq = e->queue_index;
 537         } else {
 538                 txq = reciprocal_scale(txq, numqueues);
 539         }
 540
 541         return txq;
 542 }
 543
 544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 545 {
 546         struct tun_prog *prog;
 547         u32 numqueues;
 548         u16 ret = 0;
 549
 550         numqueues = READ_ONCE(tun->numqueues);
 551         if (!numqueues)
 552                 return 0;
 553
 554         prog = rcu_dereference(tun->steering_prog);
 555         if (prog)
 556                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 557
 558         return ret % numqueues;
 559 }
 560
 561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 562                             struct net_device *sb_dev)
 563 {
 564         struct tun_struct *tun = netdev_priv(dev);
 565         u16 ret;
 566
 567         rcu_read_lock();
 568         if (rcu_dereference(tun->steering_prog))
 569                 ret = tun_ebpf_select_queue(tun, skb);
 570         else
 571                 ret = tun_automq_select_queue(tun, skb);
 572         rcu_read_unlock();
 573
 574         return ret;
 575 }
 576
 577 static inline bool tun_not_capable(struct tun_struct *tun)
 578 {
 579         const struct cred *cred = current_cred();
 580         struct net *net = dev_net(tun->dev);
 581
 582         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 583                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 584                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 585 }
 586
 587 static void tun_set_real_num_queues(struct tun_struct *tun)
 588 {
 589         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 590         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 591 }
 592
 593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 594 {
 595         tfile->detached = tun;
 596         list_add_tail(&tfile->next, &tun->disabled);
 597         ++tun->numdisabled;
 598 }
 599
 600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 601 {
 602         struct tun_struct *tun = tfile->detached;
 603
 604         tfile->detached = NULL;
 605         list_del_init(&tfile->next);
 606         --tun->numdisabled;
 607         return tun;
 608 }
 609
 610 void tun_ptr_free(void *ptr)
 611 {
 612         if (!ptr)
 613                 return;
 614         if (tun_is_xdp_frame(ptr)) {
 615                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 616
 617                 xdp_return_frame(xdpf);
 618         } else {
 619                 __skb_array_destroy_skb(ptr);
 620         }
 621 }
 622 EXPORT_SYMBOL_GPL(tun_ptr_free);
 623
 624 static void tun_queue_purge(struct tun_file *tfile)
 625 {
 626         void *ptr;
 627
 628         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 629                 tun_ptr_free(ptr);
 630
 631         skb_queue_purge(&tfile->sk.sk_write_queue);
 632         skb_queue_purge(&tfile->sk.sk_error_queue);
 633 }
 634
 635 static void __tun_detach(struct tun_file *tfile, bool clean)
 636 {
 637         struct tun_file *ntfile;
 638         struct tun_struct *tun;
 639
 640         tun = rtnl_dereference(tfile->tun);
 641
 642         if (tun && clean) {
 643                 if (!tfile->detached)
 644                         tun_napi_disable(tfile);
 645                 tun_napi_del(tfile);
 646         }
 647
 648         if (tun && !tfile->detached) {
 649                 u16 index = tfile->queue_index;
 650                 BUG_ON(index >= tun->numqueues);
 651
 652                 rcu_assign_pointer(tun->tfiles[index],
 653                                    tun->tfiles[tun->numqueues - 1]);
 654                 ntfile = rtnl_dereference(tun->tfiles[index]);
 655                 ntfile->queue_index = index;
 656                 ntfile->xdp_rxq.queue_index = index;
 657                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 658                                    NULL);
 659
 660                 --tun->numqueues;
 661                 if (clean) {
 662                         RCU_INIT_POINTER(tfile->tun, NULL);
 663                         sock_put(&tfile->sk);
 664                 } else {
 665                         tun_disable_queue(tun, tfile);
 666                         tun_napi_disable(tfile);
 667                 }
 668
 669                 synchronize_net();
 670                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 671                 /* Drop read queue */
 672                 tun_queue_purge(tfile);
 673                 tun_set_real_num_queues(tun);
 674         } else if (tfile->detached && clean) {
 675                 tun = tun_enable_queue(tfile);
 676                 sock_put(&tfile->sk);
 677         }
 678
 679         if (clean) {
 680                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 681                         netif_carrier_off(tun->dev);
 682
 683                         if (!(tun->flags & IFF_PERSIST) &&
 684                             tun->dev->reg_state == NETREG_REGISTERED)
 685                                 unregister_netdevice(tun->dev);
 686                 }
 687                 if (tun)
 688                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 690         }
 691 }
 692
 693 static void tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_struct *tun;
 696         struct net_device *dev;
 697
 698         rtnl_lock();
 699         tun = rtnl_dereference(tfile->tun);
 700         dev = tun ? tun->dev : NULL;
 701         __tun_detach(tfile, clean);
 702         if (dev)
 703                 netdev_state_change(dev);
 704         rtnl_unlock();
 705
 706         if (clean)
 707                 sock_put(&tfile->sk);
 708 }
 709
 710 static void tun_detach_all(struct net_device *dev)
 711 {
 712         struct tun_struct *tun = netdev_priv(dev);
 713         struct tun_file *tfile, *tmp;
 714         int i, n = tun->numqueues;
 715
 716         for (i = 0; i < n; i++) {
 717                 tfile = rtnl_dereference(tun->tfiles[i]);
 718                 BUG_ON(!tfile);
 719                 tun_napi_disable(tfile);
 720                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 721                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 722                 RCU_INIT_POINTER(tfile->tun, NULL);
 723                 --tun->numqueues;
 724         }
 725         list_for_each_entry(tfile, &tun->disabled, next) {
 726                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 727                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 728                 RCU_INIT_POINTER(tfile->tun, NULL);
 729         }
 730         BUG_ON(tun->numqueues != 0);
 731
 732         synchronize_net();
 733         for (i = 0; i < n; i++) {
 734                 tfile = rtnl_dereference(tun->tfiles[i]);
 735                 tun_napi_del(tfile);
 736                 /* Drop read queue */
 737                 tun_queue_purge(tfile);
 738                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 739                 sock_put(&tfile->sk);
 740         }
 741         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 742                 tun_napi_del(tfile);
 743                 tun_enable_queue(tfile);
 744                 tun_queue_purge(tfile);
 745                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 746                 sock_put(&tfile->sk);
 747         }
 748         BUG_ON(tun->numdisabled != 0);
 749
 750         if (tun->flags & IFF_PERSIST)
 751                 module_put(THIS_MODULE);
 752 }
 753
 754 static int tun_attach(struct tun_struct *tun, struct file *file,
 755                       bool skip_filter, bool napi, bool napi_frags,
 756                       bool publish_tun)
 757 {
 758         struct tun_file *tfile = file->private_data;
 759         struct net_device *dev = tun->dev;
 760         int err;
 761
 762         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 763         if (err < 0)
 764                 goto out;
 765
 766         err = -EINVAL;
 767         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 768                 goto out;
 769
 770         err = -EBUSY;
 771         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 772                 goto out;
 773
 774         err = -E2BIG;
 775         if (!tfile->detached &&
 776             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 777                 goto out;
 778
 779         err = 0;
 780
 781         /* Re-attach the filter to persist device */
 782         if (!skip_filter && (tun->filter_attached == true)) {
 783                 lock_sock(tfile->socket.sk);
 784                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 785                 release_sock(tfile->socket.sk);
 786                 if (!err)
 787                         goto out;
 788         }
 789
 790         if (!tfile->detached &&
 791             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 792                             GFP_KERNEL, tun_ptr_free)) {
 793                 err = -ENOMEM;
 794                 goto out;
 795         }
 796
 797         tfile->queue_index = tun->numqueues;
 798         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 799
 800         if (tfile->detached) {
 801                 /* Re-attach detached tfile, updating XDP queue_index */
 802                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 803
 804                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 805                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 806         } else {
 807                 /* Setup XDP RX-queue info, for new tfile getting attached */
 808                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 809                                        tun->dev, tfile->queue_index, 0);
 810                 if (err < 0)
 811                         goto out;
 812                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 813                                                  MEM_TYPE_PAGE_SHARED, NULL);
 814                 if (err < 0) {
 815                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 816                         goto out;
 817                 }
 818                 err = 0;
 819         }
 820
 821         if (tfile->detached) {
 822                 tun_enable_queue(tfile);
 823                 tun_napi_enable(tfile);
 824         } else {
 825                 sock_hold(&tfile->sk);
 826                 tun_napi_init(tun, tfile, napi, napi_frags);
 827         }
 828
 829         if (rtnl_dereference(tun->xdp_prog))
 830                 sock_set_flag(&tfile->sk, SOCK_XDP);
 831
 832         /* device is allowed to go away first, so no need to hold extra
 833          * refcnt.
 834          */
 835
 836         /* Publish tfile->tun and tun->tfiles only after we've fully
 837          * initialized tfile; otherwise we risk using half-initialized
 838          * object.
 839          */
 840         if (publish_tun)
 841                 rcu_assign_pointer(tfile->tun, tun);
 842         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 843         tun->numqueues++;
 844         tun_set_real_num_queues(tun);
 845 out:
 846         return err;
 847 }
 848
 849 static struct tun_struct *tun_get(struct tun_file *tfile)
 850 {
 851         struct tun_struct *tun;
 852
 853         rcu_read_lock();
 854         tun = rcu_dereference(tfile->tun);
 855         if (tun)
 856                 dev_hold(tun->dev);
 857         rcu_read_unlock();
 858
 859         return tun;
 860 }
 861
 862 static void tun_put(struct tun_struct *tun)
 863 {
 864         dev_put(tun->dev);
 865 }
 866
 867 /* TAP filtering */
 868 static void addr_hash_set(u32 *mask, const u8 *addr)
 869 {
 870         int n = ether_crc(ETH_ALEN, addr) >> 26;
 871         mask[n >> 5] |= (1 << (n & 31));
 872 }
 873
 874 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 875 {
 876         int n = ether_crc(ETH_ALEN, addr) >> 26;
 877         return mask[n >> 5] & (1 << (n & 31));
 878 }
 879
 880 static int update_filter(struct tap_filter *filter, void __user *arg)
 881 {
 882         struct { u8 u[ETH_ALEN]; } *addr;
 883         struct tun_filter uf;
 884         int err, alen, n, nexact;
 885
 886         if (copy_from_user(&uf, arg, sizeof(uf)))
 887                 return -EFAULT;
 888
 889         if (!uf.count) {
 890                 /* Disabled */
 891                 filter->count = 0;
 892                 return 0;
 893         }
 894
 895         alen = ETH_ALEN * uf.count;
 896         addr = memdup_user(arg + sizeof(uf), alen);
 897         if (IS_ERR(addr))
 898                 return PTR_ERR(addr);
 899
 900         /* The filter is updated without holding any locks. Which is
 901          * perfectly safe. We disable it first and in the worst
 902          * case we'll accept a few undesired packets. */
 903         filter->count = 0;
 904         wmb();
 905
 906         /* Use first set of addresses as an exact filter */
 907         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 908                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 909
 910         nexact = n;
 911
 912         /* Remaining multicast addresses are hashed,
 913          * unicast will leave the filter disabled. */
 914         memset(filter->mask, 0, sizeof(filter->mask));
 915         for (; n < uf.count; n++) {
 916                 if (!is_multicast_ether_addr(addr[n].u)) {
 917                         err = 0; /* no filter */
 918                         goto free_addr;
 919                 }
 920                 addr_hash_set(filter->mask, addr[n].u);
 921         }
 922
 923         /* For ALLMULTI just set the mask to all ones.
 924          * This overrides the mask populated above. */
 925         if ((uf.flags & TUN_FLT_ALLMULTI))
 926                 memset(filter->mask, ~0, sizeof(filter->mask));
 927
 928         /* Now enable the filter */
 929         wmb();
 930         filter->count = nexact;
 931
 932         /* Return the number of exact filters */
 933         err = nexact;
 934 free_addr:
 935         kfree(addr);
 936         return err;
 937 }
 938
 939 /* Returns: 0 - drop, !=0 - accept */
 940 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 941 {
 942         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 943          * at this point. */
 944         struct ethhdr *eh = (struct ethhdr *) skb->data;
 945         int i;
 946
 947         /* Exact match */
 948         for (i = 0; i < filter->count; i++)
 949                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 950                         return 1;
 951
 952         /* Inexact match (multicast only) */
 953         if (is_multicast_ether_addr(eh->h_dest))
 954                 return addr_hash_test(filter->mask, eh->h_dest);
 955
 956         return 0;
 957 }
 958
 959 /*
 960  * Checks whether the packet is accepted or not.
 961  * Returns: 0 - drop, !=0 - accept
 962  */
 963 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 964 {
 965         if (!filter->count)
 966                 return 1;
 967
 968         return run_filter(filter, skb);
 969 }
 970
 971 /* Network device part of the driver */
 972
 973 static const struct ethtool_ops tun_ethtool_ops;
 974
 975 static int tun_net_init(struct net_device *dev)
 976 {
 977         struct tun_struct *tun = netdev_priv(dev);
 978         struct ifreq *ifr = tun->ifr;
 979         int err;
 980
 981         spin_lock_init(&tun->lock);
 982
 983         err = security_tun_dev_alloc_security(&tun->security);
 984         if (err < 0)
 985                 return err;
 986
 987         tun_flow_init(tun);
 988
 989         dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
 990         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 991                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 992                            NETIF_F_HW_VLAN_STAG_TX;
 993         dev->features = dev->hw_features | NETIF_F_LLTX;
 994         dev->vlan_features = dev->features &
 995                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 996                                NETIF_F_HW_VLAN_STAG_TX);
 997
 998         tun->flags = (tun->flags & ~TUN_FEATURES) |
 999                       (ifr->ifr_flags & TUN_FEATURES);
1000
1001         INIT_LIST_HEAD(&tun->disabled);
1002         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1003                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1004         if (err < 0) {
1005                 tun_flow_uninit(tun);
1006                 security_tun_dev_free_security(tun->security);
1007                 return err;
1008         }
1009         return 0;
1010 }
1011
1012 /* Net device detach from fd. */
1013 static void tun_net_uninit(struct net_device *dev)
1014 {
1015         tun_detach_all(dev);
1016 }
1017
1018 /* Net device open. */
1019 static int tun_net_open(struct net_device *dev)
1020 {
1021         netif_tx_start_all_queues(dev);
1022
1023         return 0;
1024 }
1025
1026 /* Net device close. */
1027 static int tun_net_close(struct net_device *dev)
1028 {
1029         netif_tx_stop_all_queues(dev);
1030         return 0;
1031 }
1032
1033 /* Net device start xmit */
1034 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1035 {
1036 #ifdef CONFIG_RPS
1037         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1038                 /* Select queue was not called for the skbuff, so we extract the
1039                  * RPS hash and save it into the flow_table here.
1040                  */
1041                 struct tun_flow_entry *e;
1042                 __u32 rxhash;
1043
1044                 rxhash = __skb_get_hash_symmetric(skb);
1045                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1046                 if (e)
1047                         tun_flow_save_rps_rxhash(e, rxhash);
1048         }
1049 #endif
1050 }
1051
1052 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1053                                     struct sk_buff *skb,
1054                                     int len)
1055 {
1056         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1057
1058         if (prog)
1059                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1060
1061         return len;
1062 }
1063
1064 /* Net device start xmit */
1065 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1066 {
1067         struct tun_struct *tun = netdev_priv(dev);
1068         enum skb_drop_reason drop_reason;
1069         int txq = skb->queue_mapping;
1070         struct netdev_queue *queue;
1071         struct tun_file *tfile;
1072         int len = skb->len;
1073
1074         rcu_read_lock();
1075         tfile = rcu_dereference(tun->tfiles[txq]);
1076
1077         /* Drop packet if interface is not attached */
1078         if (!tfile) {
1079                 drop_reason = SKB_DROP_REASON_DEV_READY;
1080                 goto drop;
1081         }
1082
1083         if (!rcu_dereference(tun->steering_prog))
1084                 tun_automq_xmit(tun, skb);
1085
1086         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1087
1088         /* Drop if the filter does not like it.
1089          * This is a noop if the filter is disabled.
1090          * Filter can be enabled only for the TAP devices. */
1091         if (!check_filter(&tun->txflt, skb)) {
1092                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1093                 goto drop;
1094         }
1095
1096         if (tfile->socket.sk->sk_filter &&
1097             sk_filter(tfile->socket.sk, skb)) {
1098                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1099                 goto drop;
1100         }
1101
1102         len = run_ebpf_filter(tun, skb, len);
1103         if (len == 0) {
1104                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1105                 goto drop;
1106         }
1107
1108         if (pskb_trim(skb, len)) {
1109                 drop_reason = SKB_DROP_REASON_NOMEM;
1110                 goto drop;
1111         }
1112
1113         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1114                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1115                 goto drop;
1116         }
1117
1118         skb_tx_timestamp(skb);
1119
1120         /* Orphan the skb - required as we might hang on to it
1121          * for indefinite time.
1122          */
1123         skb_orphan(skb);
1124
1125         nf_reset_ct(skb);
1126
1127         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1128                 drop_reason = SKB_DROP_REASON_FULL_RING;
1129                 goto drop;
1130         }
1131
1132         /* NETIF_F_LLTX requires to do our own update of trans_start */
1133         queue = netdev_get_tx_queue(dev, txq);
1134         txq_trans_cond_update(queue);
1135
1136         /* Notify and wake up reader process */
1137         if (tfile->flags & TUN_FASYNC)
1138                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1139         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1140
1141         rcu_read_unlock();
1142         return NETDEV_TX_OK;
1143
1144 drop:
1145         dev_core_stats_tx_dropped_inc(dev);
1146         skb_tx_error(skb);
1147         kfree_skb_reason(skb, drop_reason);
1148         rcu_read_unlock();
1149         return NET_XMIT_DROP;
1150 }
1151
1152 static void tun_net_mclist(struct net_device *dev)
1153 {
1154         /*
1155          * This callback is supposed to deal with mc filter in
1156          * _rx_ path and has nothing to do with the _tx_ path.
1157          * In rx path we always accept everything userspace gives us.
1158          */
1159 }
1160
1161 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1162         netdev_features_t features)
1163 {
1164         struct tun_struct *tun = netdev_priv(dev);
1165
1166         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1167 }
1168
1169 static void tun_set_headroom(struct net_device *dev, int new_hr)
1170 {
1171         struct tun_struct *tun = netdev_priv(dev);
1172
1173         if (new_hr < NET_SKB_PAD)
1174                 new_hr = NET_SKB_PAD;
1175
1176         tun->align = new_hr;
1177 }
1178
1179 static void
1180 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1181 {
1182         struct tun_struct *tun = netdev_priv(dev);
1183
1184         dev_get_tstats64(dev, stats);
1185
1186         stats->rx_frame_errors +=
1187                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1188 }
1189
1190 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1191                        struct netlink_ext_ack *extack)
1192 {
1193         struct tun_struct *tun = netdev_priv(dev);
1194         struct tun_file *tfile;
1195         struct bpf_prog *old_prog;
1196         int i;
1197
1198         old_prog = rtnl_dereference(tun->xdp_prog);
1199         rcu_assign_pointer(tun->xdp_prog, prog);
1200         if (old_prog)
1201                 bpf_prog_put(old_prog);
1202
1203         for (i = 0; i < tun->numqueues; i++) {
1204                 tfile = rtnl_dereference(tun->tfiles[i]);
1205                 if (prog)
1206                         sock_set_flag(&tfile->sk, SOCK_XDP);
1207                 else
1208                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1209         }
1210         list_for_each_entry(tfile, &tun->disabled, next) {
1211                 if (prog)
1212                         sock_set_flag(&tfile->sk, SOCK_XDP);
1213                 else
1214                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1215         }
1216
1217         return 0;
1218 }
1219
1220 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1221 {
1222         switch (xdp->command) {
1223         case XDP_SETUP_PROG:
1224                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1225         default:
1226                 return -EINVAL;
1227         }
1228 }
1229
1230 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1231 {
1232         if (new_carrier) {
1233                 struct tun_struct *tun = netdev_priv(dev);
1234
1235                 if (!tun->numqueues)
1236                         return -EPERM;
1237
1238                 netif_carrier_on(dev);
1239         } else {
1240                 netif_carrier_off(dev);
1241         }
1242         return 0;
1243 }
1244
1245 static const struct net_device_ops tun_netdev_ops = {
1246         .ndo_init               = tun_net_init,
1247         .ndo_uninit             = tun_net_uninit,
1248         .ndo_open               = tun_net_open,
1249         .ndo_stop               = tun_net_close,
1250         .ndo_start_xmit         = tun_net_xmit,
1251         .ndo_fix_features       = tun_net_fix_features,
1252         .ndo_select_queue       = tun_select_queue,
1253         .ndo_set_rx_headroom    = tun_set_headroom,
1254         .ndo_get_stats64        = tun_net_get_stats64,
1255         .ndo_change_carrier     = tun_net_change_carrier,
1256 };
1257
1258 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1259 {
1260         /* Notify and wake up reader process */
1261         if (tfile->flags & TUN_FASYNC)
1262                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1263         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1264 }
1265
1266 static int tun_xdp_xmit(struct net_device *dev, int n,
1267                         struct xdp_frame **frames, u32 flags)
1268 {
1269         struct tun_struct *tun = netdev_priv(dev);
1270         struct tun_file *tfile;
1271         u32 numqueues;
1272         int nxmit = 0;
1273         int i;
1274
1275         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1276                 return -EINVAL;
1277
1278         rcu_read_lock();
1279
1280 resample:
1281         numqueues = READ_ONCE(tun->numqueues);
1282         if (!numqueues) {
1283                 rcu_read_unlock();
1284                 return -ENXIO; /* Caller will free/return all frames */
1285         }
1286
1287         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1288                                             numqueues]);
1289         if (unlikely(!tfile))
1290                 goto resample;
1291
1292         spin_lock(&tfile->tx_ring.producer_lock);
1293         for (i = 0; i < n; i++) {
1294                 struct xdp_frame *xdp = frames[i];
1295                 /* Encode the XDP flag into lowest bit for consumer to differ
1296                  * XDP buffer from sk_buff.
1297                  */
1298                 void *frame = tun_xdp_to_ptr(xdp);
1299
1300                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1301                         dev_core_stats_tx_dropped_inc(dev);
1302                         break;
1303                 }
1304                 nxmit++;
1305         }
1306         spin_unlock(&tfile->tx_ring.producer_lock);
1307
1308         if (flags & XDP_XMIT_FLUSH)
1309                 __tun_xdp_flush_tfile(tfile);
1310
1311         rcu_read_unlock();
1312         return nxmit;
1313 }
1314
1315 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1316 {
1317         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1318         int nxmit;
1319
1320         if (unlikely(!frame))
1321                 return -EOVERFLOW;
1322
1323         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1324         if (!nxmit)
1325                 xdp_return_frame_rx_napi(frame);
1326         return nxmit;
1327 }
1328
1329 static const struct net_device_ops tap_netdev_ops = {
1330         .ndo_init               = tun_net_init,
1331         .ndo_uninit             = tun_net_uninit,
1332         .ndo_open               = tun_net_open,
1333         .ndo_stop               = tun_net_close,
1334         .ndo_start_xmit         = tun_net_xmit,
1335         .ndo_fix_features       = tun_net_fix_features,
1336         .ndo_set_rx_mode        = tun_net_mclist,
1337         .ndo_set_mac_address    = eth_mac_addr,
1338         .ndo_validate_addr      = eth_validate_addr,
1339         .ndo_select_queue       = tun_select_queue,
1340         .ndo_features_check     = passthru_features_check,
1341         .ndo_set_rx_headroom    = tun_set_headroom,
1342         .ndo_bpf                = tun_xdp,
1343         .ndo_xdp_xmit           = tun_xdp_xmit,
1344         .ndo_change_carrier     = tun_net_change_carrier,
1345 };
1346
1347 static void tun_flow_init(struct tun_struct *tun)
1348 {
1349         int i;
1350
1351         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1352                 INIT_HLIST_HEAD(&tun->flows[i]);
1353
1354         tun->ageing_time = TUN_FLOW_EXPIRE;
1355         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1356         mod_timer(&tun->flow_gc_timer,
1357                   round_jiffies_up(jiffies + tun->ageing_time));
1358 }
1359
1360 static void tun_flow_uninit(struct tun_struct *tun)
1361 {
1362         del_timer_sync(&tun->flow_gc_timer);
1363         tun_flow_flush(tun);
1364 }
1365
1366 #define MIN_MTU 68
1367 #define MAX_MTU 65535
1368
1369 /* Initialize net device. */
1370 static void tun_net_initialize(struct net_device *dev)
1371 {
1372         struct tun_struct *tun = netdev_priv(dev);
1373
1374         switch (tun->flags & TUN_TYPE_MASK) {
1375         case IFF_TUN:
1376                 dev->netdev_ops = &tun_netdev_ops;
1377                 dev->header_ops = &ip_tunnel_header_ops;
1378
1379                 /* Point-to-Point TUN Device */
1380                 dev->hard_header_len = 0;
1381                 dev->addr_len = 0;
1382                 dev->mtu = 1500;
1383
1384                 /* Zero header length */
1385                 dev->type = ARPHRD_NONE;
1386                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1387                 break;
1388
1389         case IFF_TAP:
1390                 dev->netdev_ops = &tap_netdev_ops;
1391                 /* Ethernet TAP Device */
1392                 ether_setup(dev);
1393                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1394                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1395
1396                 eth_hw_addr_random(dev);
1397
1398                 /* Currently tun does not support XDP, only tap does. */
1399                 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1400                                     NETDEV_XDP_ACT_REDIRECT |
1401                                     NETDEV_XDP_ACT_NDO_XMIT;
1402
1403                 break;
1404         }
1405
1406         dev->min_mtu = MIN_MTU;
1407         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1408 }
1409
1410 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1411 {
1412         struct sock *sk = tfile->socket.sk;
1413
1414         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1415 }
1416
1417 /* Character device part */
1418
1419 /* Poll */
1420 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1421 {
1422         struct tun_file *tfile = file->private_data;
1423         struct tun_struct *tun = tun_get(tfile);
1424         struct sock *sk;
1425         __poll_t mask = 0;
1426
1427         if (!tun)
1428                 return EPOLLERR;
1429
1430         sk = tfile->socket.sk;
1431
1432         poll_wait(file, sk_sleep(sk), wait);
1433
1434         if (!ptr_ring_empty(&tfile->tx_ring))
1435                 mask |= EPOLLIN | EPOLLRDNORM;
1436
1437         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1438          * guarantee EPOLLOUT to be raised by either here or
1439          * tun_sock_write_space(). Then process could get notification
1440          * after it writes to a down device and meets -EIO.
1441          */
1442         if (tun_sock_writeable(tun, tfile) ||
1443             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1444              tun_sock_writeable(tun, tfile)))
1445                 mask |= EPOLLOUT | EPOLLWRNORM;
1446
1447         if (tun->dev->reg_state != NETREG_REGISTERED)
1448                 mask = EPOLLERR;
1449
1450         tun_put(tun);
1451         return mask;
1452 }
1453
1454 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1455                                             size_t len,
1456                                             const struct iov_iter *it)
1457 {
1458         struct sk_buff *skb;
1459         size_t linear;
1460         int err;
1461         int i;
1462
1463         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1464             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1465                 return ERR_PTR(-EMSGSIZE);
1466
1467         local_bh_disable();
1468         skb = napi_get_frags(&tfile->napi);
1469         local_bh_enable();
1470         if (!skb)
1471                 return ERR_PTR(-ENOMEM);
1472
1473         linear = iov_iter_single_seg_count(it);
1474         err = __skb_grow(skb, linear);
1475         if (err)
1476                 goto free;
1477
1478         skb->len = len;
1479         skb->data_len = len - linear;
1480         skb->truesize += skb->data_len;
1481
1482         for (i = 1; i < it->nr_segs; i++) {
1483                 const struct iovec *iov = iter_iov(it);
1484                 size_t fragsz = iov->iov_len;
1485                 struct page *page;
1486                 void *frag;
1487
1488                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1489                         err = -EINVAL;
1490                         goto free;
1491                 }
1492                 frag = netdev_alloc_frag(fragsz);
1493                 if (!frag) {
1494                         err = -ENOMEM;
1495                         goto free;
1496                 }
1497                 page = virt_to_head_page(frag);
1498                 skb_fill_page_desc(skb, i - 1, page,
1499                                    frag - page_address(page), fragsz);
1500         }
1501
1502         return skb;
1503 free:
1504         /* frees skb and all frags allocated with napi_alloc_frag() */
1505         napi_free_frags(&tfile->napi);
1506         return ERR_PTR(err);
1507 }
1508
1509 /* prepad is the amount to reserve at front.  len is length after that.
1510  * linear is a hint as to how much to copy (usually headers). */
1511 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1512                                      size_t prepad, size_t len,
1513                                      size_t linear, int noblock)
1514 {
1515         struct sock *sk = tfile->socket.sk;
1516         struct sk_buff *skb;
1517         int err;
1518
1519         /* Under a page?  Don't bother with paged skb. */
1520         if (prepad + len < PAGE_SIZE)
1521                 linear = len;
1522
1523         if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1524                 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1525         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1526                                    &err, PAGE_ALLOC_COSTLY_ORDER);
1527         if (!skb)
1528                 return ERR_PTR(err);
1529
1530         skb_reserve(skb, prepad);
1531         skb_put(skb, linear);
1532         skb->data_len = len - linear;
1533         skb->len += len - linear;
1534
1535         return skb;
1536 }
1537
1538 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1539                            struct sk_buff *skb, int more)
1540 {
1541         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1542         struct sk_buff_head process_queue;
1543         u32 rx_batched = tun->rx_batched;
1544         bool rcv = false;
1545
1546         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1547                 local_bh_disable();
1548                 skb_record_rx_queue(skb, tfile->queue_index);
1549                 netif_receive_skb(skb);
1550                 local_bh_enable();
1551                 return;
1552         }
1553
1554         spin_lock(&queue->lock);
1555         if (!more || skb_queue_len(queue) == rx_batched) {
1556                 __skb_queue_head_init(&process_queue);
1557                 skb_queue_splice_tail_init(queue, &process_queue);
1558                 rcv = true;
1559         } else {
1560                 __skb_queue_tail(queue, skb);
1561         }
1562         spin_unlock(&queue->lock);
1563
1564         if (rcv) {
1565                 struct sk_buff *nskb;
1566
1567                 local_bh_disable();
1568                 while ((nskb = __skb_dequeue(&process_queue))) {
1569                         skb_record_rx_queue(nskb, tfile->queue_index);
1570                         netif_receive_skb(nskb);
1571                 }
1572                 skb_record_rx_queue(skb, tfile->queue_index);
1573                 netif_receive_skb(skb);
1574                 local_bh_enable();
1575         }
1576 }
1577
1578 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1579                               int len, int noblock, bool zerocopy)
1580 {
1581         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1582                 return false;
1583
1584         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1585                 return false;
1586
1587         if (!noblock)
1588                 return false;
1589
1590         if (zerocopy)
1591                 return false;
1592
1593         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1594             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1595                 return false;
1596
1597         return true;
1598 }
1599
1600 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1601                                        struct page_frag *alloc_frag, char *buf,
1602                                        int buflen, int len, int pad)
1603 {
1604         struct sk_buff *skb = build_skb(buf, buflen);
1605
1606         if (!skb)
1607                 return ERR_PTR(-ENOMEM);
1608
1609         skb_reserve(skb, pad);
1610         skb_put(skb, len);
1611         skb_set_owner_w(skb, tfile->socket.sk);
1612
1613         get_page(alloc_frag->page);
1614         alloc_frag->offset += buflen;
1615
1616         return skb;
1617 }
1618
1619 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1620                        struct xdp_buff *xdp, u32 act)
1621 {
1622         int err;
1623
1624         switch (act) {
1625         case XDP_REDIRECT:
1626                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1627                 if (err) {
1628                         dev_core_stats_rx_dropped_inc(tun->dev);
1629                         return err;
1630                 }
1631                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1632                 break;
1633         case XDP_TX:
1634                 err = tun_xdp_tx(tun->dev, xdp);
1635                 if (err < 0) {
1636                         dev_core_stats_rx_dropped_inc(tun->dev);
1637                         return err;
1638                 }
1639                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1640                 break;
1641         case XDP_PASS:
1642                 break;
1643         default:
1644                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1645                 fallthrough;
1646         case XDP_ABORTED:
1647                 trace_xdp_exception(tun->dev, xdp_prog, act);
1648                 fallthrough;
1649         case XDP_DROP:
1650                 dev_core_stats_rx_dropped_inc(tun->dev);
1651                 break;
1652         }
1653
1654         return act;
1655 }
1656
1657 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1658                                      struct tun_file *tfile,
1659                                      struct iov_iter *from,
1660                                      struct virtio_net_hdr *hdr,
1661                                      int len, int *skb_xdp)
1662 {
1663         struct page_frag *alloc_frag = &current->task_frag;
1664         struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
1665         struct bpf_prog *xdp_prog;
1666         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1667         char *buf;
1668         size_t copied;
1669         int pad = TUN_RX_PAD;
1670         int err = 0;
1671
1672         rcu_read_lock();
1673         xdp_prog = rcu_dereference(tun->xdp_prog);
1674         if (xdp_prog)
1675                 pad += XDP_PACKET_HEADROOM;
1676         buflen += SKB_DATA_ALIGN(len + pad);
1677         rcu_read_unlock();
1678
1679         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1680         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1681                 return ERR_PTR(-ENOMEM);
1682
1683         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1684         copied = copy_page_from_iter(alloc_frag->page,
1685                                      alloc_frag->offset + pad,
1686                                      len, from);
1687         if (copied != len)
1688                 return ERR_PTR(-EFAULT);
1689
1690         /* There's a small window that XDP may be set after the check
1691          * of xdp_prog above, this should be rare and for simplicity
1692          * we do XDP on skb in case the headroom is not enough.
1693          */
1694         if (hdr->gso_type || !xdp_prog) {
1695                 *skb_xdp = 1;
1696                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1697                                        pad);
1698         }
1699
1700         *skb_xdp = 0;
1701
1702         local_bh_disable();
1703         rcu_read_lock();
1704         bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
1705         xdp_prog = rcu_dereference(tun->xdp_prog);
1706         if (xdp_prog) {
1707                 struct xdp_buff xdp;
1708                 u32 act;
1709
1710                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1711                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1712
1713                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1714                 if (act == XDP_REDIRECT || act == XDP_TX) {
1715                         get_page(alloc_frag->page);
1716                         alloc_frag->offset += buflen;
1717                 }
1718                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1719                 if (err < 0) {
1720                         if (act == XDP_REDIRECT || act == XDP_TX)
1721                                 put_page(alloc_frag->page);
1722                         goto out;
1723                 }
1724
1725                 if (err == XDP_REDIRECT)
1726                         xdp_do_flush();
1727                 if (err != XDP_PASS)
1728                         goto out;
1729
1730                 pad = xdp.data - xdp.data_hard_start;
1731                 len = xdp.data_end - xdp.data;
1732         }
1733         bpf_net_ctx_clear(bpf_net_ctx);
1734         rcu_read_unlock();
1735         local_bh_enable();
1736
1737         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1738
1739 out:
1740         bpf_net_ctx_clear(bpf_net_ctx);
1741         rcu_read_unlock();
1742         local_bh_enable();
1743         return NULL;
1744 }
1745
1746 /* Get packet from user space buffer */
1747 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1748                             void *msg_control, struct iov_iter *from,
1749                             int noblock, bool more)
1750 {
1751         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1752         struct sk_buff *skb;
1753         size_t total_len = iov_iter_count(from);
1754         size_t len = total_len, align = tun->align, linear;
1755         struct virtio_net_hdr gso = { 0 };
1756         int good_linear;
1757         int copylen;
1758         bool zerocopy = false;
1759         int err;
1760         u32 rxhash = 0;
1761         int skb_xdp = 1;
1762         bool frags = tun_napi_frags_enabled(tfile);
1763         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1764
1765         if (!(tun->flags & IFF_NO_PI)) {
1766                 if (len < sizeof(pi))
1767                         return -EINVAL;
1768                 len -= sizeof(pi);
1769
1770                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1771                         return -EFAULT;
1772         }
1773
1774         if (tun->flags & IFF_VNET_HDR) {
1775                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1776
1777                 if (len < vnet_hdr_sz)
1778                         return -EINVAL;
1779                 len -= vnet_hdr_sz;
1780
1781                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1782                         return -EFAULT;
1783
1784                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1785                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1786                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1787
1788                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1789                         return -EINVAL;
1790                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1791         }
1792
1793         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1794                 align += NET_IP_ALIGN;
1795                 if (unlikely(len < ETH_HLEN ||
1796                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1797                         return -EINVAL;
1798         }
1799
1800         good_linear = SKB_MAX_HEAD(align);
1801
1802         if (msg_control) {
1803                 struct iov_iter i = *from;
1804
1805                 /* There are 256 bytes to be copied in skb, so there is
1806                  * enough room for skb expand head in case it is used.
1807                  * The rest of the buffer is mapped from userspace.
1808                  */
1809                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1810                 if (copylen > good_linear)
1811                         copylen = good_linear;
1812                 linear = copylen;
1813                 iov_iter_advance(&i, copylen);
1814                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1815                         zerocopy = true;
1816         }
1817
1818         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1819                 /* For the packet that is not easy to be processed
1820                  * (e.g gso or jumbo packet), we will do it at after
1821                  * skb was created with generic XDP routine.
1822                  */
1823                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1824                 err = PTR_ERR_OR_ZERO(skb);
1825                 if (err)
1826                         goto drop;
1827                 if (!skb)
1828                         return total_len;
1829         } else {
1830                 if (!zerocopy) {
1831                         copylen = len;
1832                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1833                                 linear = good_linear;
1834                         else
1835                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1836                 }
1837
1838                 if (frags) {
1839                         mutex_lock(&tfile->napi_mutex);
1840                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1841                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1842                          * If zerocopy is enabled, then this layout will be
1843                          * overwritten by zerocopy_sg_from_iter().
1844                          */
1845                         zerocopy = false;
1846                 } else {
1847                         if (!linear)
1848                                 linear = min_t(size_t, good_linear, copylen);
1849
1850                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1851                                             noblock);
1852                 }
1853
1854                 err = PTR_ERR_OR_ZERO(skb);
1855                 if (err)
1856                         goto drop;
1857
1858                 if (zerocopy)
1859                         err = zerocopy_sg_from_iter(skb, from);
1860                 else
1861                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1862
1863                 if (err) {
1864                         err = -EFAULT;
1865                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1866                         goto drop;
1867                 }
1868         }
1869
1870         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1871                 atomic_long_inc(&tun->rx_frame_errors);
1872                 err = -EINVAL;
1873                 goto free_skb;
1874         }
1875
1876         switch (tun->flags & TUN_TYPE_MASK) {
1877         case IFF_TUN:
1878                 if (tun->flags & IFF_NO_PI) {
1879                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1880
1881                         switch (ip_version) {
1882                         case 4:
1883                                 pi.proto = htons(ETH_P_IP);
1884                                 break;
1885                         case 6:
1886                                 pi.proto = htons(ETH_P_IPV6);
1887                                 break;
1888                         default:
1889                                 err = -EINVAL;
1890                                 goto drop;
1891                         }
1892                 }
1893
1894                 skb_reset_mac_header(skb);
1895                 skb->protocol = pi.proto;
1896                 skb->dev = tun->dev;
1897                 break;
1898         case IFF_TAP:
1899                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1900                         err = -ENOMEM;
1901                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1902                         goto drop;
1903                 }
1904                 skb->protocol = eth_type_trans(skb, tun->dev);
1905                 break;
1906         }
1907
1908         /* copy skb_ubuf_info for callback when skb has no error */
1909         if (zerocopy) {
1910                 skb_zcopy_init(skb, msg_control);
1911         } else if (msg_control) {
1912                 struct ubuf_info *uarg = msg_control;
1913                 uarg->ops->complete(NULL, uarg, false);
1914         }
1915
1916         skb_reset_network_header(skb);
1917         skb_probe_transport_header(skb);
1918         skb_record_rx_queue(skb, tfile->queue_index);
1919
1920         if (skb_xdp) {
1921                 struct bpf_prog *xdp_prog;
1922                 int ret;
1923
1924                 local_bh_disable();
1925                 rcu_read_lock();
1926                 xdp_prog = rcu_dereference(tun->xdp_prog);
1927                 if (xdp_prog) {
1928                         ret = do_xdp_generic(xdp_prog, &skb);
1929                         if (ret != XDP_PASS) {
1930                                 rcu_read_unlock();
1931                                 local_bh_enable();
1932                                 goto unlock_frags;
1933                         }
1934                 }
1935                 rcu_read_unlock();
1936                 local_bh_enable();
1937         }
1938
1939         /* Compute the costly rx hash only if needed for flow updates.
1940          * We may get a very small possibility of OOO during switching, not
1941          * worth to optimize.
1942          */
1943         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1944             !tfile->detached)
1945                 rxhash = __skb_get_hash_symmetric(skb);
1946
1947         rcu_read_lock();
1948         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1949                 err = -EIO;
1950                 rcu_read_unlock();
1951                 drop_reason = SKB_DROP_REASON_DEV_READY;
1952                 goto drop;
1953         }
1954
1955         if (frags) {
1956                 u32 headlen;
1957
1958                 /* Exercise flow dissector code path. */
1959                 skb_push(skb, ETH_HLEN);
1960                 headlen = eth_get_headlen(tun->dev, skb->data,
1961                                           skb_headlen(skb));
1962
1963                 if (unlikely(headlen > skb_headlen(skb))) {
1964                         WARN_ON_ONCE(1);
1965                         err = -ENOMEM;
1966                         dev_core_stats_rx_dropped_inc(tun->dev);
1967 napi_busy:
1968                         napi_free_frags(&tfile->napi);
1969                         rcu_read_unlock();
1970                         mutex_unlock(&tfile->napi_mutex);
1971                         return err;
1972                 }
1973
1974                 if (likely(napi_schedule_prep(&tfile->napi))) {
1975                         local_bh_disable();
1976                         napi_gro_frags(&tfile->napi);
1977                         napi_complete(&tfile->napi);
1978                         local_bh_enable();
1979                 } else {
1980                         err = -EBUSY;
1981                         goto napi_busy;
1982                 }
1983                 mutex_unlock(&tfile->napi_mutex);
1984         } else if (tfile->napi_enabled) {
1985                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1986                 int queue_len;
1987
1988                 spin_lock_bh(&queue->lock);
1989
1990                 if (unlikely(tfile->detached)) {
1991                         spin_unlock_bh(&queue->lock);
1992                         rcu_read_unlock();
1993                         err = -EBUSY;
1994                         goto free_skb;
1995                 }
1996
1997                 __skb_queue_tail(queue, skb);
1998                 queue_len = skb_queue_len(queue);
1999                 spin_unlock(&queue->lock);
2000
2001                 if (!more || queue_len > NAPI_POLL_WEIGHT)
2002                         napi_schedule(&tfile->napi);
2003
2004                 local_bh_enable();
2005         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2006                 tun_rx_batched(tun, tfile, skb, more);
2007         } else {
2008                 netif_rx(skb);
2009         }
2010         rcu_read_unlock();
2011
2012         preempt_disable();
2013         dev_sw_netstats_rx_add(tun->dev, len);
2014         preempt_enable();
2015
2016         if (rxhash)
2017                 tun_flow_update(tun, rxhash, tfile);
2018
2019         return total_len;
2020
2021 drop:
2022         if (err != -EAGAIN)
2023                 dev_core_stats_rx_dropped_inc(tun->dev);
2024
2025 free_skb:
2026         if (!IS_ERR_OR_NULL(skb))
2027                 kfree_skb_reason(skb, drop_reason);
2028
2029 unlock_frags:
2030         if (frags) {
2031                 tfile->napi.skb = NULL;
2032                 mutex_unlock(&tfile->napi_mutex);
2033         }
2034
2035         return err ?: total_len;
2036 }
2037
2038 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2039 {
2040         struct file *file = iocb->ki_filp;
2041         struct tun_file *tfile = file->private_data;
2042         struct tun_struct *tun = tun_get(tfile);
2043         ssize_t result;
2044         int noblock = 0;
2045
2046         if (!tun)
2047                 return -EBADFD;
2048
2049         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2050                 noblock = 1;
2051
2052         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2053
2054         tun_put(tun);
2055         return result;
2056 }
2057
2058 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2059                                 struct tun_file *tfile,
2060                                 struct xdp_frame *xdp_frame,
2061                                 struct iov_iter *iter)
2062 {
2063         int vnet_hdr_sz = 0;
2064         size_t size = xdp_frame->len;
2065         size_t ret;
2066
2067         if (tun->flags & IFF_VNET_HDR) {
2068                 struct virtio_net_hdr gso = { 0 };
2069
2070                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2071                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2072                         return -EINVAL;
2073                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2074                              sizeof(gso)))
2075                         return -EFAULT;
2076                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2077         }
2078
2079         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2080
2081         preempt_disable();
2082         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2083         preempt_enable();
2084
2085         return ret;
2086 }
2087
2088 /* Put packet to the user space buffer */
2089 static ssize_t tun_put_user(struct tun_struct *tun,
2090                             struct tun_file *tfile,
2091                             struct sk_buff *skb,
2092                             struct iov_iter *iter)
2093 {
2094         struct tun_pi pi = { 0, skb->protocol };
2095         ssize_t total;
2096         int vlan_offset = 0;
2097         int vlan_hlen = 0;
2098         int vnet_hdr_sz = 0;
2099
2100         if (skb_vlan_tag_present(skb))
2101                 vlan_hlen = VLAN_HLEN;
2102
2103         if (tun->flags & IFF_VNET_HDR)
2104                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2105
2106         total = skb->len + vlan_hlen + vnet_hdr_sz;
2107
2108         if (!(tun->flags & IFF_NO_PI)) {
2109                 if (iov_iter_count(iter) < sizeof(pi))
2110                         return -EINVAL;
2111
2112                 total += sizeof(pi);
2113                 if (iov_iter_count(iter) < total) {
2114                         /* Packet will be striped */
2115                         pi.flags |= TUN_PKT_STRIP;
2116                 }
2117
2118                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2119                         return -EFAULT;
2120         }
2121
2122         if (vnet_hdr_sz) {
2123                 struct virtio_net_hdr gso;
2124
2125                 if (iov_iter_count(iter) < vnet_hdr_sz)
2126                         return -EINVAL;
2127
2128                 if (virtio_net_hdr_from_skb(skb, &gso,
2129                                             tun_is_little_endian(tun), true,
2130                                             vlan_hlen)) {
2131                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2132
2133                         if (net_ratelimit()) {
2134                                 netdev_err(tun->dev, "unexpected GSO type: 0x%x, gso_size %d, hdr_len %d\n",
2135                                            sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2136                                            tun16_to_cpu(tun, gso.hdr_len));
2137                                 print_hex_dump(KERN_ERR, "tun: ",
2138                                                DUMP_PREFIX_NONE,
2139                                                16, 1, skb->head,
2140                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2141                         }
2142                         WARN_ON_ONCE(1);
2143                         return -EINVAL;
2144                 }
2145
2146                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2147                         return -EFAULT;
2148
2149                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2150         }
2151
2152         if (vlan_hlen) {
2153                 int ret;
2154                 struct veth veth;
2155
2156                 veth.h_vlan_proto = skb->vlan_proto;
2157                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2158
2159                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2160
2161                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2162                 if (ret || !iov_iter_count(iter))
2163                         goto done;
2164
2165                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2166                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2167                         goto done;
2168         }
2169
2170         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2171
2172 done:
2173         /* caller is in process context, */
2174         preempt_disable();
2175         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2176         preempt_enable();
2177
2178         return total;
2179 }
2180
2181 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2182 {
2183         DECLARE_WAITQUEUE(wait, current);
2184         void *ptr = NULL;
2185         int error = 0;
2186
2187         ptr = ptr_ring_consume(&tfile->tx_ring);
2188         if (ptr)
2189                 goto out;
2190         if (noblock) {
2191                 error = -EAGAIN;
2192                 goto out;
2193         }
2194
2195         add_wait_queue(&tfile->socket.wq.wait, &wait);
2196
2197         while (1) {
2198                 set_current_state(TASK_INTERRUPTIBLE);
2199                 ptr = ptr_ring_consume(&tfile->tx_ring);
2200                 if (ptr)
2201                         break;
2202                 if (signal_pending(current)) {
2203                         error = -ERESTARTSYS;
2204                         break;
2205                 }
2206                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2207                         error = -EFAULT;
2208                         break;
2209                 }
2210
2211                 schedule();
2212         }
2213
2214         __set_current_state(TASK_RUNNING);
2215         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2216
2217 out:
2218         *err = error;
2219         return ptr;
2220 }
2221
2222 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2223                            struct iov_iter *to,
2224                            int noblock, void *ptr)
2225 {
2226         ssize_t ret;
2227         int err;
2228
2229         if (!iov_iter_count(to)) {
2230                 tun_ptr_free(ptr);
2231                 return 0;
2232         }
2233
2234         if (!ptr) {
2235                 /* Read frames from ring */
2236                 ptr = tun_ring_recv(tfile, noblock, &err);
2237                 if (!ptr)
2238                         return err;
2239         }
2240
2241         if (tun_is_xdp_frame(ptr)) {
2242                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2243
2244                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2245                 xdp_return_frame(xdpf);
2246         } else {
2247                 struct sk_buff *skb = ptr;
2248
2249                 ret = tun_put_user(tun, tfile, skb, to);
2250                 if (unlikely(ret < 0))
2251                         kfree_skb(skb);
2252                 else
2253                         consume_skb(skb);
2254         }
2255
2256         return ret;
2257 }
2258
2259 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2260 {
2261         struct file *file = iocb->ki_filp;
2262         struct tun_file *tfile = file->private_data;
2263         struct tun_struct *tun = tun_get(tfile);
2264         ssize_t len = iov_iter_count(to), ret;
2265         int noblock = 0;
2266
2267         if (!tun)
2268                 return -EBADFD;
2269
2270         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2271                 noblock = 1;
2272
2273         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2274         ret = min_t(ssize_t, ret, len);
2275         if (ret > 0)
2276                 iocb->ki_pos = ret;
2277         tun_put(tun);
2278         return ret;
2279 }
2280
2281 static void tun_prog_free(struct rcu_head *rcu)
2282 {
2283         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2284
2285         bpf_prog_destroy(prog->prog);
2286         kfree(prog);
2287 }
2288
2289 static int __tun_set_ebpf(struct tun_struct *tun,
2290                           struct tun_prog __rcu **prog_p,
2291                           struct bpf_prog *prog)
2292 {
2293         struct tun_prog *old, *new = NULL;
2294
2295         if (prog) {
2296                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2297                 if (!new)
2298                         return -ENOMEM;
2299                 new->prog = prog;
2300         }
2301
2302         spin_lock_bh(&tun->lock);
2303         old = rcu_dereference_protected(*prog_p,
2304                                         lockdep_is_held(&tun->lock));
2305         rcu_assign_pointer(*prog_p, new);
2306         spin_unlock_bh(&tun->lock);
2307
2308         if (old)
2309                 call_rcu(&old->rcu, tun_prog_free);
2310
2311         return 0;
2312 }
2313
2314 static void tun_free_netdev(struct net_device *dev)
2315 {
2316         struct tun_struct *tun = netdev_priv(dev);
2317
2318         BUG_ON(!(list_empty(&tun->disabled)));
2319
2320         tun_flow_uninit(tun);
2321         security_tun_dev_free_security(tun->security);
2322         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2323         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2324 }
2325
2326 static void tun_setup(struct net_device *dev)
2327 {
2328         struct tun_struct *tun = netdev_priv(dev);
2329
2330         tun->owner = INVALID_UID;
2331         tun->group = INVALID_GID;
2332         tun_default_link_ksettings(dev, &tun->link_ksettings);
2333
2334         dev->ethtool_ops = &tun_ethtool_ops;
2335         dev->needs_free_netdev = true;
2336         dev->priv_destructor = tun_free_netdev;
2337         /* We prefer our own queue length */
2338         dev->tx_queue_len = TUN_READQ_SIZE;
2339 }
2340
2341 /* Trivial set of netlink ops to allow deleting tun or tap
2342  * device with netlink.
2343  */
2344 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2345                         struct netlink_ext_ack *extack)
2346 {
2347         NL_SET_ERR_MSG(extack,
2348                        "tun/tap creation via rtnetlink is not supported.");
2349         return -EOPNOTSUPP;
2350 }
2351
2352 static size_t tun_get_size(const struct net_device *dev)
2353 {
2354         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2355         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2356
2357         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2358                nla_total_size(sizeof(gid_t)) + /* GROUP */
2359                nla_total_size(sizeof(u8)) + /* TYPE */
2360                nla_total_size(sizeof(u8)) + /* PI */
2361                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2362                nla_total_size(sizeof(u8)) + /* PERSIST */
2363                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2364                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2365                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2366                0;
2367 }
2368
2369 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2370 {
2371         struct tun_struct *tun = netdev_priv(dev);
2372
2373         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2374                 goto nla_put_failure;
2375         if (uid_valid(tun->owner) &&
2376             nla_put_u32(skb, IFLA_TUN_OWNER,
2377                         from_kuid_munged(current_user_ns(), tun->owner)))
2378                 goto nla_put_failure;
2379         if (gid_valid(tun->group) &&
2380             nla_put_u32(skb, IFLA_TUN_GROUP,
2381                         from_kgid_munged(current_user_ns(), tun->group)))
2382                 goto nla_put_failure;
2383         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2384                 goto nla_put_failure;
2385         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2386                 goto nla_put_failure;
2387         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2388                 goto nla_put_failure;
2389         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2390                        !!(tun->flags & IFF_MULTI_QUEUE)))
2391                 goto nla_put_failure;
2392         if (tun->flags & IFF_MULTI_QUEUE) {
2393                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2394                         goto nla_put_failure;
2395                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2396                                 tun->numdisabled))
2397                         goto nla_put_failure;
2398         }
2399
2400         return 0;
2401
2402 nla_put_failure:
2403         return -EMSGSIZE;
2404 }
2405
2406 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2407         .kind           = DRV_NAME,
2408         .priv_size      = sizeof(struct tun_struct),
2409         .setup          = tun_setup,
2410         .validate       = tun_validate,
2411         .get_size       = tun_get_size,
2412         .fill_info      = tun_fill_info,
2413 };
2414
2415 static void tun_sock_write_space(struct sock *sk)
2416 {
2417         struct tun_file *tfile;
2418         wait_queue_head_t *wqueue;
2419
2420         if (!sock_writeable(sk))
2421                 return;
2422
2423         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2424                 return;
2425
2426         wqueue = sk_sleep(sk);
2427         if (wqueue && waitqueue_active(wqueue))
2428                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2429                                                 EPOLLWRNORM | EPOLLWRBAND);
2430
2431         tfile = container_of(sk, struct tun_file, sk);
2432         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2433 }
2434
2435 static void tun_put_page(struct tun_page *tpage)
2436 {
2437         if (tpage->page)
2438                 __page_frag_cache_drain(tpage->page, tpage->count);
2439 }
2440
2441 static int tun_xdp_one(struct tun_struct *tun,
2442                        struct tun_file *tfile,
2443                        struct xdp_buff *xdp, int *flush,
2444                        struct tun_page *tpage)
2445 {
2446         unsigned int datasize = xdp->data_end - xdp->data;
2447         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2448         struct virtio_net_hdr *gso = &hdr->gso;
2449         struct bpf_prog *xdp_prog;
2450         struct sk_buff *skb = NULL;
2451         struct sk_buff_head *queue;
2452         u32 rxhash = 0, act;
2453         int buflen = hdr->buflen;
2454         int ret = 0;
2455         bool skb_xdp = false;
2456         struct page *page;
2457
2458         xdp_prog = rcu_dereference(tun->xdp_prog);
2459         if (xdp_prog) {
2460                 if (gso->gso_type) {
2461                         skb_xdp = true;
2462                         goto build;
2463                 }
2464
2465                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2466                 xdp_set_data_meta_invalid(xdp);
2467
2468                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2469                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2470                 if (ret < 0) {
2471                         put_page(virt_to_head_page(xdp->data));
2472                         return ret;
2473                 }
2474
2475                 switch (ret) {
2476                 case XDP_REDIRECT:
2477                         *flush = true;
2478                         fallthrough;
2479                 case XDP_TX:
2480                         return 0;
2481                 case XDP_PASS:
2482                         break;
2483                 default:
2484                         page = virt_to_head_page(xdp->data);
2485                         if (tpage->page == page) {
2486                                 ++tpage->count;
2487                         } else {
2488                                 tun_put_page(tpage);
2489                                 tpage->page = page;
2490                                 tpage->count = 1;
2491                         }
2492                         return 0;
2493                 }
2494         }
2495
2496 build:
2497         skb = build_skb(xdp->data_hard_start, buflen);
2498         if (!skb) {
2499                 ret = -ENOMEM;
2500                 goto out;
2501         }
2502
2503         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2504         skb_put(skb, xdp->data_end - xdp->data);
2505
2506         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2507                 atomic_long_inc(&tun->rx_frame_errors);
2508                 kfree_skb(skb);
2509                 ret = -EINVAL;
2510                 goto out;
2511         }
2512
2513         skb->protocol = eth_type_trans(skb, tun->dev);
2514         skb_reset_network_header(skb);
2515         skb_probe_transport_header(skb);
2516         skb_record_rx_queue(skb, tfile->queue_index);
2517
2518         if (skb_xdp) {
2519                 ret = do_xdp_generic(xdp_prog, &skb);
2520                 if (ret != XDP_PASS) {
2521                         ret = 0;
2522                         goto out;
2523                 }
2524         }
2525
2526         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2527             !tfile->detached)
2528                 rxhash = __skb_get_hash_symmetric(skb);
2529
2530         if (tfile->napi_enabled) {
2531                 queue = &tfile->sk.sk_write_queue;
2532                 spin_lock(&queue->lock);
2533
2534                 if (unlikely(tfile->detached)) {
2535                         spin_unlock(&queue->lock);
2536                         kfree_skb(skb);
2537                         return -EBUSY;
2538                 }
2539
2540                 __skb_queue_tail(queue, skb);
2541                 spin_unlock(&queue->lock);
2542                 ret = 1;
2543         } else {
2544                 netif_receive_skb(skb);
2545                 ret = 0;
2546         }
2547
2548         /* No need to disable preemption here since this function is
2549          * always called with bh disabled
2550          */
2551         dev_sw_netstats_rx_add(tun->dev, datasize);
2552
2553         if (rxhash)
2554                 tun_flow_update(tun, rxhash, tfile);
2555
2556 out:
2557         return ret;
2558 }
2559
2560 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2561 {
2562         int ret, i;
2563         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2564         struct tun_struct *tun = tun_get(tfile);
2565         struct tun_msg_ctl *ctl = m->msg_control;
2566         struct xdp_buff *xdp;
2567
2568         if (!tun)
2569                 return -EBADFD;
2570
2571         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2572             ctl && ctl->type == TUN_MSG_PTR) {
2573                 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
2574                 struct tun_page tpage;
2575                 int n = ctl->num;
2576                 int flush = 0, queued = 0;
2577
2578                 memset(&tpage, 0, sizeof(tpage));
2579
2580                 local_bh_disable();
2581                 rcu_read_lock();
2582                 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
2583
2584                 for (i = 0; i < n; i++) {
2585                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2586                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2587                         if (ret > 0)
2588                                 queued += ret;
2589                 }
2590
2591                 if (flush)
2592                         xdp_do_flush();
2593
2594                 if (tfile->napi_enabled && queued > 0)
2595                         napi_schedule(&tfile->napi);
2596
2597                 bpf_net_ctx_clear(bpf_net_ctx);
2598                 rcu_read_unlock();
2599                 local_bh_enable();
2600
2601                 tun_put_page(&tpage);
2602
2603                 ret = total_len;
2604                 goto out;
2605         }
2606
2607         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2608                            m->msg_flags & MSG_DONTWAIT,
2609                            m->msg_flags & MSG_MORE);
2610 out:
2611         tun_put(tun);
2612         return ret;
2613 }
2614
2615 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2616                        int flags)
2617 {
2618         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2619         struct tun_struct *tun = tun_get(tfile);
2620         void *ptr = m->msg_control;
2621         int ret;
2622
2623         if (!tun) {
2624                 ret = -EBADFD;
2625                 goto out_free;
2626         }
2627
2628         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2629                 ret = -EINVAL;
2630                 goto out_put_tun;
2631         }
2632         if (flags & MSG_ERRQUEUE) {
2633                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2634                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2635                 goto out;
2636         }
2637         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2638         if (ret > (ssize_t)total_len) {
2639                 m->msg_flags |= MSG_TRUNC;
2640                 ret = flags & MSG_TRUNC ? ret : total_len;
2641         }
2642 out:
2643         tun_put(tun);
2644         return ret;
2645
2646 out_put_tun:
2647         tun_put(tun);
2648 out_free:
2649         tun_ptr_free(ptr);
2650         return ret;
2651 }
2652
2653 static int tun_ptr_peek_len(void *ptr)
2654 {
2655         if (likely(ptr)) {
2656                 if (tun_is_xdp_frame(ptr)) {
2657                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2658
2659                         return xdpf->len;
2660                 }
2661                 return __skb_array_len_with_tag(ptr);
2662         } else {
2663                 return 0;
2664         }
2665 }
2666
2667 static int tun_peek_len(struct socket *sock)
2668 {
2669         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2670         struct tun_struct *tun;
2671         int ret = 0;
2672
2673         tun = tun_get(tfile);
2674         if (!tun)
2675                 return 0;
2676
2677         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2678         tun_put(tun);
2679
2680         return ret;
2681 }
2682
2683 /* Ops structure to mimic raw sockets with tun */
2684 static const struct proto_ops tun_socket_ops = {
2685         .peek_len = tun_peek_len,
2686         .sendmsg = tun_sendmsg,
2687         .recvmsg = tun_recvmsg,
2688 };
2689
2690 static struct proto tun_proto = {
2691         .name           = "tun",
2692         .owner          = THIS_MODULE,
2693         .obj_size       = sizeof(struct tun_file),
2694 };
2695
2696 static int tun_flags(struct tun_struct *tun)
2697 {
2698         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2699 }
2700
2701 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2702                               char *buf)
2703 {
2704         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2705         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2706 }
2707
2708 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2709                           char *buf)
2710 {
2711         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2712         return uid_valid(tun->owner)?
2713                 sysfs_emit(buf, "%u\n",
2714                            from_kuid_munged(current_user_ns(), tun->owner)) :
2715                 sysfs_emit(buf, "-1\n");
2716 }
2717
2718 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2719                           char *buf)
2720 {
2721         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2722         return gid_valid(tun->group) ?
2723                 sysfs_emit(buf, "%u\n",
2724                            from_kgid_munged(current_user_ns(), tun->group)) :
2725                 sysfs_emit(buf, "-1\n");
2726 }
2727
2728 static DEVICE_ATTR_RO(tun_flags);
2729 static DEVICE_ATTR_RO(owner);
2730 static DEVICE_ATTR_RO(group);
2731
2732 static struct attribute *tun_dev_attrs[] = {
2733         &dev_attr_tun_flags.attr,
2734         &dev_attr_owner.attr,
2735         &dev_attr_group.attr,
2736         NULL
2737 };
2738
2739 static const struct attribute_group tun_attr_group = {
2740         .attrs = tun_dev_attrs
2741 };
2742
2743 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2744 {
2745         struct tun_struct *tun;
2746         struct tun_file *tfile = file->private_data;
2747         struct net_device *dev;
2748         int err;
2749
2750         if (tfile->detached)
2751                 return -EINVAL;
2752
2753         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2754                 if (!capable(CAP_NET_ADMIN))
2755                         return -EPERM;
2756
2757                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2758                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2759                         return -EINVAL;
2760         }
2761
2762         dev = __dev_get_by_name(net, ifr->ifr_name);
2763         if (dev) {
2764                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2765                         return -EBUSY;
2766                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2767                         tun = netdev_priv(dev);
2768                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2769                         tun = netdev_priv(dev);
2770                 else
2771                         return -EINVAL;
2772
2773                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2774                     !!(tun->flags & IFF_MULTI_QUEUE))
2775                         return -EINVAL;
2776
2777                 if (tun_not_capable(tun))
2778                         return -EPERM;
2779                 err = security_tun_dev_open(tun->security);
2780                 if (err < 0)
2781                         return err;
2782
2783                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2784                                  ifr->ifr_flags & IFF_NAPI,
2785                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2786                 if (err < 0)
2787                         return err;
2788
2789                 if (tun->flags & IFF_MULTI_QUEUE &&
2790                     (tun->numqueues + tun->numdisabled > 1)) {
2791                         /* One or more queue has already been attached, no need
2792                          * to initialize the device again.
2793                          */
2794                         netdev_state_change(dev);
2795                         return 0;
2796                 }
2797
2798                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2799                               (ifr->ifr_flags & TUN_FEATURES);
2800
2801                 netdev_state_change(dev);
2802         } else {
2803                 char *name;
2804                 unsigned long flags = 0;
2805                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2806                              MAX_TAP_QUEUES : 1;
2807
2808                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2809                         return -EPERM;
2810                 err = security_tun_dev_create();
2811                 if (err < 0)
2812                         return err;
2813
2814                 /* Set dev type */
2815                 if (ifr->ifr_flags & IFF_TUN) {
2816                         /* TUN device */
2817                         flags |= IFF_TUN;
2818                         name = "tun%d";
2819                 } else if (ifr->ifr_flags & IFF_TAP) {
2820                         /* TAP device */
2821                         flags |= IFF_TAP;
2822                         name = "tap%d";
2823                 } else
2824                         return -EINVAL;
2825
2826                 if (*ifr->ifr_name)
2827                         name = ifr->ifr_name;
2828
2829                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2830                                        NET_NAME_UNKNOWN, tun_setup, queues,
2831                                        queues);
2832
2833                 if (!dev)
2834                         return -ENOMEM;
2835
2836                 dev_net_set(dev, net);
2837                 dev->rtnl_link_ops = &tun_link_ops;
2838                 dev->ifindex = tfile->ifindex;
2839                 dev->sysfs_groups[0] = &tun_attr_group;
2840
2841                 tun = netdev_priv(dev);
2842                 tun->dev = dev;
2843                 tun->flags = flags;
2844                 tun->txflt.count = 0;
2845                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2846
2847                 tun->align = NET_SKB_PAD;
2848                 tun->filter_attached = false;
2849                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2850                 tun->rx_batched = 0;
2851                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2852
2853                 tun->ifr = ifr;
2854                 tun->file = file;
2855
2856                 tun_net_initialize(dev);
2857
2858                 err = register_netdevice(tun->dev);
2859                 if (err < 0) {
2860                         free_netdev(dev);
2861                         return err;
2862                 }
2863                 /* free_netdev() won't check refcnt, to avoid race
2864                  * with dev_put() we need publish tun after registration.
2865                  */
2866                 rcu_assign_pointer(tfile->tun, tun);
2867         }
2868
2869         if (ifr->ifr_flags & IFF_NO_CARRIER)
2870                 netif_carrier_off(tun->dev);
2871         else
2872                 netif_carrier_on(tun->dev);
2873
2874         /* Make sure persistent devices do not get stuck in
2875          * xoff state.
2876          */
2877         if (netif_running(tun->dev))
2878                 netif_tx_wake_all_queues(tun->dev);
2879
2880         strcpy(ifr->ifr_name, tun->dev->name);
2881         return 0;
2882 }
2883
2884 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2885 {
2886         strcpy(ifr->ifr_name, tun->dev->name);
2887
2888         ifr->ifr_flags = tun_flags(tun);
2889
2890 }
2891
2892 /* This is like a cut-down ethtool ops, except done via tun fd so no
2893  * privs required. */
2894 static int set_offload(struct tun_struct *tun, unsigned long arg)
2895 {
2896         netdev_features_t features = 0;
2897
2898         if (arg & TUN_F_CSUM) {
2899                 features |= NETIF_F_HW_CSUM;
2900                 arg &= ~TUN_F_CSUM;
2901
2902                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2903                         if (arg & TUN_F_TSO_ECN) {
2904                                 features |= NETIF_F_TSO_ECN;
2905                                 arg &= ~TUN_F_TSO_ECN;
2906                         }
2907                         if (arg & TUN_F_TSO4)
2908                                 features |= NETIF_F_TSO;
2909                         if (arg & TUN_F_TSO6)
2910                                 features |= NETIF_F_TSO6;
2911                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2912                 }
2913
2914                 arg &= ~TUN_F_UFO;
2915
2916                 /* TODO: for now USO4 and USO6 should work simultaneously */
2917                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2918                         features |= NETIF_F_GSO_UDP_L4;
2919                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2920                 }
2921         }
2922
2923         /* This gives the user a way to test for new features in future by
2924          * trying to set them. */
2925         if (arg)
2926                 return -EINVAL;
2927
2928         tun->set_features = features;
2929         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2930         tun->dev->wanted_features |= features;
2931         netdev_update_features(tun->dev);
2932
2933         return 0;
2934 }
2935
2936 static void tun_detach_filter(struct tun_struct *tun, int n)
2937 {
2938         int i;
2939         struct tun_file *tfile;
2940
2941         for (i = 0; i < n; i++) {
2942                 tfile = rtnl_dereference(tun->tfiles[i]);
2943                 lock_sock(tfile->socket.sk);
2944                 sk_detach_filter(tfile->socket.sk);
2945                 release_sock(tfile->socket.sk);
2946         }
2947
2948         tun->filter_attached = false;
2949 }
2950
2951 static int tun_attach_filter(struct tun_struct *tun)
2952 {
2953         int i, ret = 0;
2954         struct tun_file *tfile;
2955
2956         for (i = 0; i < tun->numqueues; i++) {
2957                 tfile = rtnl_dereference(tun->tfiles[i]);
2958                 lock_sock(tfile->socket.sk);
2959                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2960                 release_sock(tfile->socket.sk);
2961                 if (ret) {
2962                         tun_detach_filter(tun, i);
2963                         return ret;
2964                 }
2965         }
2966
2967         tun->filter_attached = true;
2968         return ret;
2969 }
2970
2971 static void tun_set_sndbuf(struct tun_struct *tun)
2972 {
2973         struct tun_file *tfile;
2974         int i;
2975
2976         for (i = 0; i < tun->numqueues; i++) {
2977                 tfile = rtnl_dereference(tun->tfiles[i]);
2978                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2979         }
2980 }
2981
2982 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2983 {
2984         struct tun_file *tfile = file->private_data;
2985         struct tun_struct *tun;
2986         int ret = 0;
2987
2988         rtnl_lock();
2989
2990         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2991                 tun = tfile->detached;
2992                 if (!tun) {
2993                         ret = -EINVAL;
2994                         goto unlock;
2995                 }
2996                 ret = security_tun_dev_attach_queue(tun->security);
2997                 if (ret < 0)
2998                         goto unlock;
2999                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3000                                  tun->flags & IFF_NAPI_FRAGS, true);
3001         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3002                 tun = rtnl_dereference(tfile->tun);
3003                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3004                         ret = -EINVAL;
3005                 else
3006                         __tun_detach(tfile, false);
3007         } else
3008                 ret = -EINVAL;
3009
3010         if (ret >= 0)
3011                 netdev_state_change(tun->dev);
3012
3013 unlock:
3014         rtnl_unlock();
3015         return ret;
3016 }
3017
3018 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3019                         void __user *data)
3020 {
3021         struct bpf_prog *prog;
3022         int fd;
3023
3024         if (copy_from_user(&fd, data, sizeof(fd)))
3025                 return -EFAULT;
3026
3027         if (fd == -1) {
3028                 prog = NULL;
3029         } else {
3030                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3031                 if (IS_ERR(prog))
3032                         return PTR_ERR(prog);
3033         }
3034
3035         return __tun_set_ebpf(tun, prog_p, prog);
3036 }
3037
3038 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3039 static unsigned char tun_get_addr_len(unsigned short type)
3040 {
3041         switch (type) {
3042         case ARPHRD_IP6GRE:
3043         case ARPHRD_TUNNEL6:
3044                 return sizeof(struct in6_addr);
3045         case ARPHRD_IPGRE:
3046         case ARPHRD_TUNNEL:
3047         case ARPHRD_SIT:
3048                 return 4;
3049         case ARPHRD_ETHER:
3050                 return ETH_ALEN;
3051         case ARPHRD_IEEE802154:
3052         case ARPHRD_IEEE802154_MONITOR:
3053                 return IEEE802154_EXTENDED_ADDR_LEN;
3054         case ARPHRD_PHONET_PIPE:
3055         case ARPHRD_PPP:
3056         case ARPHRD_NONE:
3057                 return 0;
3058         case ARPHRD_6LOWPAN:
3059                 return EUI64_ADDR_LEN;
3060         case ARPHRD_FDDI:
3061                 return FDDI_K_ALEN;
3062         case ARPHRD_HIPPI:
3063                 return HIPPI_ALEN;
3064         case ARPHRD_IEEE802:
3065                 return FC_ALEN;
3066         case ARPHRD_ROSE:
3067                 return ROSE_ADDR_LEN;
3068         case ARPHRD_NETROM:
3069                 return AX25_ADDR_LEN;
3070         case ARPHRD_LOCALTLK:
3071                 return LTALK_ALEN;
3072         default:
3073                 return 0;
3074         }
3075 }
3076
3077 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3078                             unsigned long arg, int ifreq_len)
3079 {
3080         struct tun_file *tfile = file->private_data;
3081         struct net *net = sock_net(&tfile->sk);
3082         struct tun_struct *tun;
3083         void __user* argp = (void __user*)arg;
3084         unsigned int carrier;
3085         struct ifreq ifr;
3086         kuid_t owner;
3087         kgid_t group;
3088         int ifindex;
3089         int sndbuf;
3090         int vnet_hdr_sz;
3091         int le;
3092         int ret;
3093         bool do_notify = false;
3094
3095         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3096             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3097                 if (copy_from_user(&ifr, argp, ifreq_len))
3098                         return -EFAULT;
3099         } else {
3100                 memset(&ifr, 0, sizeof(ifr));
3101         }
3102         if (cmd == TUNGETFEATURES) {
3103                 /* Currently this just means: "what IFF flags are valid?".
3104                  * This is needed because we never checked for invalid flags on
3105                  * TUNSETIFF.
3106                  */
3107                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3108                                 TUN_FEATURES, (unsigned int __user*)argp);
3109         } else if (cmd == TUNSETQUEUE) {
3110                 return tun_set_queue(file, &ifr);
3111         } else if (cmd == SIOCGSKNS) {
3112                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3113                         return -EPERM;
3114                 return open_related_ns(&net->ns, get_net_ns);
3115         }
3116
3117         rtnl_lock();
3118
3119         tun = tun_get(tfile);
3120         if (cmd == TUNSETIFF) {
3121                 ret = -EEXIST;
3122                 if (tun)
3123                         goto unlock;
3124
3125                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3126
3127                 ret = tun_set_iff(net, file, &ifr);
3128
3129                 if (ret)
3130                         goto unlock;
3131
3132                 if (copy_to_user(argp, &ifr, ifreq_len))
3133                         ret = -EFAULT;
3134                 goto unlock;
3135         }
3136         if (cmd == TUNSETIFINDEX) {
3137                 ret = -EPERM;
3138                 if (tun)
3139                         goto unlock;
3140
3141                 ret = -EFAULT;
3142                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3143                         goto unlock;
3144                 ret = -EINVAL;
3145                 if (ifindex < 0)
3146                         goto unlock;
3147                 ret = 0;
3148                 tfile->ifindex = ifindex;
3149                 goto unlock;
3150         }
3151
3152         ret = -EBADFD;
3153         if (!tun)
3154                 goto unlock;
3155
3156         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3157
3158         net = dev_net(tun->dev);
3159         ret = 0;
3160         switch (cmd) {
3161         case TUNGETIFF:
3162                 tun_get_iff(tun, &ifr);
3163
3164                 if (tfile->detached)
3165                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3166                 if (!tfile->socket.sk->sk_filter)
3167                         ifr.ifr_flags |= IFF_NOFILTER;
3168
3169                 if (copy_to_user(argp, &ifr, ifreq_len))
3170                         ret = -EFAULT;
3171                 break;
3172
3173         case TUNSETNOCSUM:
3174                 /* Disable/Enable checksum */
3175
3176                 /* [unimplemented] */
3177                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3178                            arg ? "disabled" : "enabled");
3179                 break;
3180
3181         case TUNSETPERSIST:
3182                 /* Disable/Enable persist mode. Keep an extra reference to the
3183                  * module to prevent the module being unprobed.
3184                  */
3185                 if (arg && !(tun->flags & IFF_PERSIST)) {
3186                         tun->flags |= IFF_PERSIST;
3187                         __module_get(THIS_MODULE);
3188                         do_notify = true;
3189                 }
3190                 if (!arg && (tun->flags & IFF_PERSIST)) {
3191                         tun->flags &= ~IFF_PERSIST;
3192                         module_put(THIS_MODULE);
3193                         do_notify = true;
3194                 }
3195
3196                 netif_info(tun, drv, tun->dev, "persist %s\n",
3197                            arg ? "enabled" : "disabled");
3198                 break;
3199
3200         case TUNSETOWNER:
3201                 /* Set owner of the device */
3202                 owner = make_kuid(current_user_ns(), arg);
3203                 if (!uid_valid(owner)) {
3204                         ret = -EINVAL;
3205                         break;
3206                 }
3207                 tun->owner = owner;
3208                 do_notify = true;
3209                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3210                            from_kuid(&init_user_ns, tun->owner));
3211                 break;
3212
3213         case TUNSETGROUP:
3214                 /* Set group of the device */
3215                 group = make_kgid(current_user_ns(), arg);
3216                 if (!gid_valid(group)) {
3217                         ret = -EINVAL;
3218                         break;
3219                 }
3220                 tun->group = group;
3221                 do_notify = true;
3222                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3223                            from_kgid(&init_user_ns, tun->group));
3224                 break;
3225
3226         case TUNSETLINK:
3227                 /* Only allow setting the type when the interface is down */
3228                 if (tun->dev->flags & IFF_UP) {
3229                         netif_info(tun, drv, tun->dev,
3230                                    "Linktype set failed because interface is up\n");
3231                         ret = -EBUSY;
3232                 } else {
3233                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3234                                                        tun->dev);
3235                         ret = notifier_to_errno(ret);
3236                         if (ret) {
3237                                 netif_info(tun, drv, tun->dev,
3238                                            "Refused to change device type\n");
3239                                 break;
3240                         }
3241                         tun->dev->type = (int) arg;
3242                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3243                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3244                                    tun->dev->type);
3245                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3246                                                  tun->dev);
3247                 }
3248                 break;
3249
3250         case TUNSETDEBUG:
3251                 tun->msg_enable = (u32)arg;
3252                 break;
3253
3254         case TUNSETOFFLOAD:
3255                 ret = set_offload(tun, arg);
3256                 break;
3257
3258         case TUNSETTXFILTER:
3259                 /* Can be set only for TAPs */
3260                 ret = -EINVAL;
3261                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3262                         break;
3263                 ret = update_filter(&tun->txflt, (void __user *)arg);
3264                 break;
3265
3266         case SIOCGIFHWADDR:
3267                 /* Get hw address */
3268                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3269                 if (copy_to_user(argp, &ifr, ifreq_len))
3270                         ret = -EFAULT;
3271                 break;
3272
3273         case SIOCSIFHWADDR:
3274                 /* Set hw address */
3275                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3276                 break;
3277
3278         case TUNGETSNDBUF:
3279                 sndbuf = tfile->socket.sk->sk_sndbuf;
3280                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3281                         ret = -EFAULT;
3282                 break;
3283
3284         case TUNSETSNDBUF:
3285                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3286                         ret = -EFAULT;
3287                         break;
3288                 }
3289                 if (sndbuf <= 0) {
3290                         ret = -EINVAL;
3291                         break;
3292                 }
3293
3294                 tun->sndbuf = sndbuf;
3295                 tun_set_sndbuf(tun);
3296                 break;
3297
3298         case TUNGETVNETHDRSZ:
3299                 vnet_hdr_sz = tun->vnet_hdr_sz;
3300                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3301                         ret = -EFAULT;
3302                 break;
3303
3304         case TUNSETVNETHDRSZ:
3305                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3306                         ret = -EFAULT;
3307                         break;
3308                 }
3309                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3310                         ret = -EINVAL;
3311                         break;
3312                 }
3313
3314                 tun->vnet_hdr_sz = vnet_hdr_sz;
3315                 break;
3316
3317         case TUNGETVNETLE:
3318                 le = !!(tun->flags & TUN_VNET_LE);
3319                 if (put_user(le, (int __user *)argp))
3320                         ret = -EFAULT;
3321                 break;
3322
3323         case TUNSETVNETLE:
3324                 if (get_user(le, (int __user *)argp)) {
3325                         ret = -EFAULT;
3326                         break;
3327                 }
3328                 if (le)
3329                         tun->flags |= TUN_VNET_LE;
3330                 else
3331                         tun->flags &= ~TUN_VNET_LE;
3332                 break;
3333
3334         case TUNGETVNETBE:
3335                 ret = tun_get_vnet_be(tun, argp);
3336                 break;
3337
3338         case TUNSETVNETBE:
3339                 ret = tun_set_vnet_be(tun, argp);
3340                 break;
3341
3342         case TUNATTACHFILTER:
3343                 /* Can be set only for TAPs */
3344                 ret = -EINVAL;
3345                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3346                         break;
3347                 ret = -EFAULT;
3348                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3349                         break;
3350
3351                 ret = tun_attach_filter(tun);
3352                 break;
3353
3354         case TUNDETACHFILTER:
3355                 /* Can be set only for TAPs */
3356                 ret = -EINVAL;
3357                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3358                         break;
3359                 ret = 0;
3360                 tun_detach_filter(tun, tun->numqueues);
3361                 break;
3362
3363         case TUNGETFILTER:
3364                 ret = -EINVAL;
3365                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3366                         break;
3367                 ret = -EFAULT;
3368                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3369                         break;
3370                 ret = 0;
3371                 break;
3372
3373         case TUNSETSTEERINGEBPF:
3374                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3375                 break;
3376
3377         case TUNSETFILTEREBPF:
3378                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3379                 break;
3380
3381         case TUNSETCARRIER:
3382                 ret = -EFAULT;
3383                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3384                         goto unlock;
3385
3386                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3387                 break;
3388
3389         case TUNGETDEVNETNS:
3390                 ret = -EPERM;
3391                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3392                         goto unlock;
3393                 ret = open_related_ns(&net->ns, get_net_ns);
3394                 break;
3395
3396         default:
3397                 ret = -EINVAL;
3398                 break;
3399         }
3400
3401         if (do_notify)
3402                 netdev_state_change(tun->dev);
3403
3404 unlock:
3405         rtnl_unlock();
3406         if (tun)
3407                 tun_put(tun);
3408         return ret;
3409 }
3410
3411 static long tun_chr_ioctl(struct file *file,
3412                           unsigned int cmd, unsigned long arg)
3413 {
3414         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3415 }
3416
3417 #ifdef CONFIG_COMPAT
3418 static long tun_chr_compat_ioctl(struct file *file,
3419                          unsigned int cmd, unsigned long arg)
3420 {
3421         switch (cmd) {
3422         case TUNSETIFF:
3423         case TUNGETIFF:
3424         case TUNSETTXFILTER:
3425         case TUNGETSNDBUF:
3426         case TUNSETSNDBUF:
3427         case SIOCGIFHWADDR:
3428         case SIOCSIFHWADDR:
3429                 arg = (unsigned long)compat_ptr(arg);
3430                 break;
3431         default:
3432                 arg = (compat_ulong_t)arg;
3433                 break;
3434         }
3435
3436         /*
3437          * compat_ifreq is shorter than ifreq, so we must not access beyond
3438          * the end of that structure. All fields that are used in this
3439          * driver are compatible though, we don't need to convert the
3440          * contents.
3441          */
3442         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3443 }
3444 #endif /* CONFIG_COMPAT */
3445
3446 static int tun_chr_fasync(int fd, struct file *file, int on)
3447 {
3448         struct tun_file *tfile = file->private_data;
3449         int ret;
3450
3451         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3452                 goto out;
3453
3454         if (on) {
3455                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3456                 tfile->flags |= TUN_FASYNC;
3457         } else
3458                 tfile->flags &= ~TUN_FASYNC;
3459         ret = 0;
3460 out:
3461         return ret;
3462 }
3463
3464 static int tun_chr_open(struct inode *inode, struct file * file)
3465 {
3466         struct net *net = current->nsproxy->net_ns;
3467         struct tun_file *tfile;
3468
3469         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3470                                             &tun_proto, 0);
3471         if (!tfile)
3472                 return -ENOMEM;
3473         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3474                 sk_free(&tfile->sk);
3475                 return -ENOMEM;
3476         }
3477
3478         mutex_init(&tfile->napi_mutex);
3479         RCU_INIT_POINTER(tfile->tun, NULL);
3480         tfile->flags = 0;
3481         tfile->ifindex = 0;
3482
3483         init_waitqueue_head(&tfile->socket.wq.wait);
3484
3485         tfile->socket.file = file;
3486         tfile->socket.ops = &tun_socket_ops;
3487
3488         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3489
3490         tfile->sk.sk_write_space = tun_sock_write_space;
3491         tfile->sk.sk_sndbuf = INT_MAX;
3492
3493         file->private_data = tfile;
3494         INIT_LIST_HEAD(&tfile->next);
3495
3496         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3497
3498         /* tun groks IOCB_NOWAIT just fine, mark it as such */
3499         file->f_mode |= FMODE_NOWAIT;
3500         return 0;
3501 }
3502
3503 static int tun_chr_close(struct inode *inode, struct file *file)
3504 {
3505         struct tun_file *tfile = file->private_data;
3506
3507         tun_detach(tfile, true);
3508
3509         return 0;
3510 }
3511
3512 #ifdef CONFIG_PROC_FS
3513 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3514 {
3515         struct tun_file *tfile = file->private_data;
3516         struct tun_struct *tun;
3517         struct ifreq ifr;
3518
3519         memset(&ifr, 0, sizeof(ifr));
3520
3521         rtnl_lock();
3522         tun = tun_get(tfile);
3523         if (tun)
3524                 tun_get_iff(tun, &ifr);
3525         rtnl_unlock();
3526
3527         if (tun)
3528                 tun_put(tun);
3529
3530         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3531 }
3532 #endif
3533
3534 static const struct file_operations tun_fops = {
3535         .owner  = THIS_MODULE,
3536         .llseek = no_llseek,
3537         .read_iter  = tun_chr_read_iter,
3538         .write_iter = tun_chr_write_iter,
3539         .poll   = tun_chr_poll,
3540         .unlocked_ioctl = tun_chr_ioctl,
3541 #ifdef CONFIG_COMPAT
3542         .compat_ioctl = tun_chr_compat_ioctl,
3543 #endif
3544         .open   = tun_chr_open,
3545         .release = tun_chr_close,
3546         .fasync = tun_chr_fasync,
3547 #ifdef CONFIG_PROC_FS
3548         .show_fdinfo = tun_chr_show_fdinfo,
3549 #endif
3550 };
3551
3552 static struct miscdevice tun_miscdev = {
3553         .minor = TUN_MINOR,
3554         .name = "tun",
3555         .nodename = "net/tun",
3556         .fops = &tun_fops,
3557 };
3558
3559 /* ethtool interface */
3560
3561 static void tun_default_link_ksettings(struct net_device *dev,
3562                                        struct ethtool_link_ksettings *cmd)
3563 {
3564         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3565         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3566         cmd->base.speed         = SPEED_10000;
3567         cmd->base.duplex        = DUPLEX_FULL;
3568         cmd->base.port          = PORT_TP;
3569         cmd->base.phy_address   = 0;
3570         cmd->base.autoneg       = AUTONEG_DISABLE;
3571 }
3572
3573 static int tun_get_link_ksettings(struct net_device *dev,
3574                                   struct ethtool_link_ksettings *cmd)
3575 {
3576         struct tun_struct *tun = netdev_priv(dev);
3577
3578         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3579         return 0;
3580 }
3581
3582 static int tun_set_link_ksettings(struct net_device *dev,
3583                                   const struct ethtool_link_ksettings *cmd)
3584 {
3585         struct tun_struct *tun = netdev_priv(dev);
3586
3587         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3588         return 0;
3589 }
3590
3591 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3592 {
3593         struct tun_struct *tun = netdev_priv(dev);
3594
3595         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3596         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3597
3598         switch (tun->flags & TUN_TYPE_MASK) {
3599         case IFF_TUN:
3600                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3601                 break;
3602         case IFF_TAP:
3603                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3604                 break;
3605         }
3606 }
3607
3608 static u32 tun_get_msglevel(struct net_device *dev)
3609 {
3610         struct tun_struct *tun = netdev_priv(dev);
3611
3612         return tun->msg_enable;
3613 }
3614
3615 static void tun_set_msglevel(struct net_device *dev, u32 value)
3616 {
3617         struct tun_struct *tun = netdev_priv(dev);
3618
3619         tun->msg_enable = value;
3620 }
3621
3622 static int tun_get_coalesce(struct net_device *dev,
3623                             struct ethtool_coalesce *ec,
3624                             struct kernel_ethtool_coalesce *kernel_coal,
3625                             struct netlink_ext_ack *extack)
3626 {
3627         struct tun_struct *tun = netdev_priv(dev);
3628
3629         ec->rx_max_coalesced_frames = tun->rx_batched;
3630
3631         return 0;
3632 }
3633
3634 static int tun_set_coalesce(struct net_device *dev,
3635                             struct ethtool_coalesce *ec,
3636                             struct kernel_ethtool_coalesce *kernel_coal,
3637                             struct netlink_ext_ack *extack)
3638 {
3639         struct tun_struct *tun = netdev_priv(dev);
3640
3641         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3642                 tun->rx_batched = NAPI_POLL_WEIGHT;
3643         else
3644                 tun->rx_batched = ec->rx_max_coalesced_frames;
3645
3646         return 0;
3647 }
3648
3649 static void tun_get_channels(struct net_device *dev,
3650                              struct ethtool_channels *channels)
3651 {
3652         struct tun_struct *tun = netdev_priv(dev);
3653
3654         channels->combined_count = tun->numqueues;
3655         channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3656 }
3657
3658 static const struct ethtool_ops tun_ethtool_ops = {
3659         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3660         .get_drvinfo    = tun_get_drvinfo,
3661         .get_msglevel   = tun_get_msglevel,
3662         .set_msglevel   = tun_set_msglevel,
3663         .get_link       = ethtool_op_get_link,
3664         .get_channels   = tun_get_channels,
3665         .get_ts_info    = ethtool_op_get_ts_info,
3666         .get_coalesce   = tun_get_coalesce,
3667         .set_coalesce   = tun_set_coalesce,
3668         .get_link_ksettings = tun_get_link_ksettings,
3669         .set_link_ksettings = tun_set_link_ksettings,
3670 };
3671
3672 static int tun_queue_resize(struct tun_struct *tun)
3673 {
3674         struct net_device *dev = tun->dev;
3675         struct tun_file *tfile;
3676         struct ptr_ring **rings;
3677         int n = tun->numqueues + tun->numdisabled;
3678         int ret, i;
3679
3680         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3681         if (!rings)
3682                 return -ENOMEM;
3683
3684         for (i = 0; i < tun->numqueues; i++) {
3685                 tfile = rtnl_dereference(tun->tfiles[i]);
3686                 rings[i] = &tfile->tx_ring;
3687         }
3688         list_for_each_entry(tfile, &tun->disabled, next)
3689                 rings[i++] = &tfile->tx_ring;
3690
3691         ret = ptr_ring_resize_multiple(rings, n,
3692                                        dev->tx_queue_len, GFP_KERNEL,
3693                                        tun_ptr_free);
3694
3695         kfree(rings);
3696         return ret;
3697 }
3698
3699 static int tun_device_event(struct notifier_block *unused,
3700                             unsigned long event, void *ptr)
3701 {
3702         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3703         struct tun_struct *tun = netdev_priv(dev);
3704         int i;
3705
3706         if (dev->rtnl_link_ops != &tun_link_ops)
3707                 return NOTIFY_DONE;
3708
3709         switch (event) {
3710         case NETDEV_CHANGE_TX_QUEUE_LEN:
3711                 if (tun_queue_resize(tun))
3712                         return NOTIFY_BAD;
3713                 break;
3714         case NETDEV_UP:
3715                 for (i = 0; i < tun->numqueues; i++) {
3716                         struct tun_file *tfile;
3717
3718                         tfile = rtnl_dereference(tun->tfiles[i]);
3719                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3720                 }
3721                 break;
3722         default:
3723                 break;
3724         }
3725
3726         return NOTIFY_DONE;
3727 }
3728
3729 static struct notifier_block tun_notifier_block __read_mostly = {
3730         .notifier_call  = tun_device_event,
3731 };
3732
3733 static int __init tun_init(void)
3734 {
3735         int ret = 0;
3736
3737         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3738
3739         ret = rtnl_link_register(&tun_link_ops);
3740         if (ret) {
3741                 pr_err("Can't register link_ops\n");
3742                 goto err_linkops;
3743         }
3744
3745         ret = misc_register(&tun_miscdev);
3746         if (ret) {
3747                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3748                 goto err_misc;
3749         }
3750
3751         ret = register_netdevice_notifier(&tun_notifier_block);
3752         if (ret) {
3753                 pr_err("Can't register netdevice notifier\n");
3754                 goto err_notifier;
3755         }
3756
3757         return  0;
3758
3759 err_notifier:
3760         misc_deregister(&tun_miscdev);
3761 err_misc:
3762         rtnl_link_unregister(&tun_link_ops);
3763 err_linkops:
3764         return ret;
3765 }
3766
3767 static void __exit tun_cleanup(void)
3768 {
3769         misc_deregister(&tun_miscdev);
3770         rtnl_link_unregister(&tun_link_ops);
3771         unregister_netdevice_notifier(&tun_notifier_block);
3772 }
3773
3774 /* Get an underlying socket object from tun file.  Returns error unless file is
3775  * attached to a device.  The returned object works like a packet socket, it
3776  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3777  * holding a reference to the file for as long as the socket is in use. */
3778 struct socket *tun_get_socket(struct file *file)
3779 {
3780         struct tun_file *tfile;
3781         if (file->f_op != &tun_fops)
3782                 return ERR_PTR(-EINVAL);
3783         tfile = file->private_data;
3784         if (!tfile)
3785                 return ERR_PTR(-EBADFD);
3786         return &tfile->socket;
3787 }
3788 EXPORT_SYMBOL_GPL(tun_get_socket);
3789
3790 struct ptr_ring *tun_get_tx_ring(struct file *file)
3791 {
3792         struct tun_file *tfile;
3793
3794         if (file->f_op != &tun_fops)
3795                 return ERR_PTR(-EINVAL);
3796         tfile = file->private_data;
3797         if (!tfile)
3798                 return ERR_PTR(-EBADFD);
3799         return &tfile->tx_ring;
3800 }
3801 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3802
3803 module_init(tun_init);
3804 module_exit(tun_cleanup);
3805 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3806 MODULE_AUTHOR(DRV_COPYRIGHT);
3807 MODULE_LICENSE("GPL");
3808 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3809 MODULE_ALIAS("devname:net/tun");