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 <uapi/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_capable(struct tun_struct *tun)
 578 {
 579         const struct cred *cred = current_cred();
 580         struct net *net = dev_net(tun->dev);
 581
 582         if (ns_capable(net->user_ns, CAP_NET_ADMIN))
 583                 return 1;
 584         if (uid_valid(tun->owner) && uid_eq(cred->euid, tun->owner))
 585                 return 1;
 586         if (gid_valid(tun->group) && in_egroup_p(tun->group))
 587                 return 1;
 588         return 0;
 589 }
 590
 591 static void tun_set_real_num_queues(struct tun_struct *tun)
 592 {
 593         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 594         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 595 }
 596
 597 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 598 {
 599         tfile->detached = tun;
 600         list_add_tail(&tfile->next, &tun->disabled);
 601         ++tun->numdisabled;
 602 }
 603
 604 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 605 {
 606         struct tun_struct *tun = tfile->detached;
 607
 608         tfile->detached = NULL;
 609         list_del_init(&tfile->next);
 610         --tun->numdisabled;
 611         return tun;
 612 }
 613
 614 void tun_ptr_free(void *ptr)
 615 {
 616         if (!ptr)
 617                 return;
 618         if (tun_is_xdp_frame(ptr)) {
 619                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 620
 621                 xdp_return_frame(xdpf);
 622         } else {
 623                 __skb_array_destroy_skb(ptr);
 624         }
 625 }
 626 EXPORT_SYMBOL_GPL(tun_ptr_free);
 627
 628 static void tun_queue_purge(struct tun_file *tfile)
 629 {
 630         void *ptr;
 631
 632         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 633                 tun_ptr_free(ptr);
 634
 635         skb_queue_purge(&tfile->sk.sk_write_queue);
 636         skb_queue_purge(&tfile->sk.sk_error_queue);
 637 }
 638
 639 static void __tun_detach(struct tun_file *tfile, bool clean)
 640 {
 641         struct tun_file *ntfile;
 642         struct tun_struct *tun;
 643
 644         tun = rtnl_dereference(tfile->tun);
 645
 646         if (tun && clean) {
 647                 if (!tfile->detached)
 648                         tun_napi_disable(tfile);
 649                 tun_napi_del(tfile);
 650         }
 651
 652         if (tun && !tfile->detached) {
 653                 u16 index = tfile->queue_index;
 654                 BUG_ON(index >= tun->numqueues);
 655
 656                 rcu_assign_pointer(tun->tfiles[index],
 657                                    tun->tfiles[tun->numqueues - 1]);
 658                 ntfile = rtnl_dereference(tun->tfiles[index]);
 659                 ntfile->queue_index = index;
 660                 ntfile->xdp_rxq.queue_index = index;
 661                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 662                                    NULL);
 663
 664                 --tun->numqueues;
 665                 if (clean) {
 666                         RCU_INIT_POINTER(tfile->tun, NULL);
 667                         sock_put(&tfile->sk);
 668                 } else {
 669                         tun_disable_queue(tun, tfile);
 670                         tun_napi_disable(tfile);
 671                 }
 672
 673                 synchronize_net();
 674                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 675                 /* Drop read queue */
 676                 tun_queue_purge(tfile);
 677                 tun_set_real_num_queues(tun);
 678         } else if (tfile->detached && clean) {
 679                 tun = tun_enable_queue(tfile);
 680                 sock_put(&tfile->sk);
 681         }
 682
 683         if (clean) {
 684                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 685                         netif_carrier_off(tun->dev);
 686
 687                         if (!(tun->flags & IFF_PERSIST) &&
 688                             tun->dev->reg_state == NETREG_REGISTERED)
 689                                 unregister_netdevice(tun->dev);
 690                 }
 691                 if (tun)
 692                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 693                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 694         }
 695 }
 696
 697 static void tun_detach(struct tun_file *tfile, bool clean)
 698 {
 699         struct tun_struct *tun;
 700         struct net_device *dev;
 701
 702         rtnl_lock();
 703         tun = rtnl_dereference(tfile->tun);
 704         dev = tun ? tun->dev : NULL;
 705         __tun_detach(tfile, clean);
 706         if (dev)
 707                 netdev_state_change(dev);
 708         rtnl_unlock();
 709
 710         if (clean)
 711                 sock_put(&tfile->sk);
 712 }
 713
 714 static void tun_detach_all(struct net_device *dev)
 715 {
 716         struct tun_struct *tun = netdev_priv(dev);
 717         struct tun_file *tfile, *tmp;
 718         int i, n = tun->numqueues;
 719
 720         for (i = 0; i < n; i++) {
 721                 tfile = rtnl_dereference(tun->tfiles[i]);
 722                 BUG_ON(!tfile);
 723                 tun_napi_disable(tfile);
 724                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 725                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 726                 RCU_INIT_POINTER(tfile->tun, NULL);
 727                 --tun->numqueues;
 728         }
 729         list_for_each_entry(tfile, &tun->disabled, next) {
 730                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 731                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 732                 RCU_INIT_POINTER(tfile->tun, NULL);
 733         }
 734         BUG_ON(tun->numqueues != 0);
 735
 736         synchronize_net();
 737         for (i = 0; i < n; i++) {
 738                 tfile = rtnl_dereference(tun->tfiles[i]);
 739                 tun_napi_del(tfile);
 740                 /* Drop read queue */
 741                 tun_queue_purge(tfile);
 742                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 743                 sock_put(&tfile->sk);
 744         }
 745         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 746                 tun_napi_del(tfile);
 747                 tun_enable_queue(tfile);
 748                 tun_queue_purge(tfile);
 749                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 750                 sock_put(&tfile->sk);
 751         }
 752         BUG_ON(tun->numdisabled != 0);
 753
 754         if (tun->flags & IFF_PERSIST)
 755                 module_put(THIS_MODULE);
 756 }
 757
 758 static int tun_attach(struct tun_struct *tun, struct file *file,
 759                       bool skip_filter, bool napi, bool napi_frags,
 760                       bool publish_tun)
 761 {
 762         struct tun_file *tfile = file->private_data;
 763         struct net_device *dev = tun->dev;
 764         int err;
 765
 766         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 767         if (err < 0)
 768                 goto out;
 769
 770         err = -EINVAL;
 771         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 772                 goto out;
 773
 774         err = -EBUSY;
 775         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 776                 goto out;
 777
 778         err = -E2BIG;
 779         if (!tfile->detached &&
 780             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 781                 goto out;
 782
 783         err = 0;
 784
 785         /* Re-attach the filter to persist device */
 786         if (!skip_filter && (tun->filter_attached == true)) {
 787                 lock_sock(tfile->socket.sk);
 788                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 789                 release_sock(tfile->socket.sk);
 790                 if (!err)
 791                         goto out;
 792         }
 793
 794         if (!tfile->detached &&
 795             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 796                             GFP_KERNEL, tun_ptr_free)) {
 797                 err = -ENOMEM;
 798                 goto out;
 799         }
 800
 801         tfile->queue_index = tun->numqueues;
 802         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 803
 804         if (tfile->detached) {
 805                 /* Re-attach detached tfile, updating XDP queue_index */
 806                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 807
 808                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 809                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 810         } else {
 811                 /* Setup XDP RX-queue info, for new tfile getting attached */
 812                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 813                                        tun->dev, tfile->queue_index, 0);
 814                 if (err < 0)
 815                         goto out;
 816                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 817                                                  MEM_TYPE_PAGE_SHARED, NULL);
 818                 if (err < 0) {
 819                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 820                         goto out;
 821                 }
 822                 err = 0;
 823         }
 824
 825         if (tfile->detached) {
 826                 tun_enable_queue(tfile);
 827                 tun_napi_enable(tfile);
 828         } else {
 829                 sock_hold(&tfile->sk);
 830                 tun_napi_init(tun, tfile, napi, napi_frags);
 831         }
 832
 833         if (rtnl_dereference(tun->xdp_prog))
 834                 sock_set_flag(&tfile->sk, SOCK_XDP);
 835
 836         /* device is allowed to go away first, so no need to hold extra
 837          * refcnt.
 838          */
 839
 840         /* Publish tfile->tun and tun->tfiles only after we've fully
 841          * initialized tfile; otherwise we risk using half-initialized
 842          * object.
 843          */
 844         if (publish_tun)
 845                 rcu_assign_pointer(tfile->tun, tun);
 846         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 847         tun->numqueues++;
 848         tun_set_real_num_queues(tun);
 849 out:
 850         return err;
 851 }
 852
 853 static struct tun_struct *tun_get(struct tun_file *tfile)
 854 {
 855         struct tun_struct *tun;
 856
 857         rcu_read_lock();
 858         tun = rcu_dereference(tfile->tun);
 859         if (tun)
 860                 dev_hold(tun->dev);
 861         rcu_read_unlock();
 862
 863         return tun;
 864 }
 865
 866 static void tun_put(struct tun_struct *tun)
 867 {
 868         dev_put(tun->dev);
 869 }
 870
 871 /* TAP filtering */
 872 static void addr_hash_set(u32 *mask, const u8 *addr)
 873 {
 874         int n = ether_crc(ETH_ALEN, addr) >> 26;
 875         mask[n >> 5] |= (1 << (n & 31));
 876 }
 877
 878 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 879 {
 880         int n = ether_crc(ETH_ALEN, addr) >> 26;
 881         return mask[n >> 5] & (1 << (n & 31));
 882 }
 883
 884 static int update_filter(struct tap_filter *filter, void __user *arg)
 885 {
 886         struct { u8 u[ETH_ALEN]; } *addr;
 887         struct tun_filter uf;
 888         int err, alen, n, nexact;
 889
 890         if (copy_from_user(&uf, arg, sizeof(uf)))
 891                 return -EFAULT;
 892
 893         if (!uf.count) {
 894                 /* Disabled */
 895                 filter->count = 0;
 896                 return 0;
 897         }
 898
 899         alen = ETH_ALEN * uf.count;
 900         addr = memdup_user(arg + sizeof(uf), alen);
 901         if (IS_ERR(addr))
 902                 return PTR_ERR(addr);
 903
 904         /* The filter is updated without holding any locks. Which is
 905          * perfectly safe. We disable it first and in the worst
 906          * case we'll accept a few undesired packets. */
 907         filter->count = 0;
 908         wmb();
 909
 910         /* Use first set of addresses as an exact filter */
 911         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 912                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 913
 914         nexact = n;
 915
 916         /* Remaining multicast addresses are hashed,
 917          * unicast will leave the filter disabled. */
 918         memset(filter->mask, 0, sizeof(filter->mask));
 919         for (; n < uf.count; n++) {
 920                 if (!is_multicast_ether_addr(addr[n].u)) {
 921                         err = 0; /* no filter */
 922                         goto free_addr;
 923                 }
 924                 addr_hash_set(filter->mask, addr[n].u);
 925         }
 926
 927         /* For ALLMULTI just set the mask to all ones.
 928          * This overrides the mask populated above. */
 929         if ((uf.flags & TUN_FLT_ALLMULTI))
 930                 memset(filter->mask, ~0, sizeof(filter->mask));
 931
 932         /* Now enable the filter */
 933         wmb();
 934         filter->count = nexact;
 935
 936         /* Return the number of exact filters */
 937         err = nexact;
 938 free_addr:
 939         kfree(addr);
 940         return err;
 941 }
 942
 943 /* Returns: 0 - drop, !=0 - accept */
 944 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 945 {
 946         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 947          * at this point. */
 948         struct ethhdr *eh = (struct ethhdr *) skb->data;
 949         int i;
 950
 951         /* Exact match */
 952         for (i = 0; i < filter->count; i++)
 953                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 954                         return 1;
 955
 956         /* Inexact match (multicast only) */
 957         if (is_multicast_ether_addr(eh->h_dest))
 958                 return addr_hash_test(filter->mask, eh->h_dest);
 959
 960         return 0;
 961 }
 962
 963 /*
 964  * Checks whether the packet is accepted or not.
 965  * Returns: 0 - drop, !=0 - accept
 966  */
 967 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 968 {
 969         if (!filter->count)
 970                 return 1;
 971
 972         return run_filter(filter, skb);
 973 }
 974
 975 /* Network device part of the driver */
 976
 977 static const struct ethtool_ops tun_ethtool_ops;
 978
 979 static int tun_net_init(struct net_device *dev)
 980 {
 981         struct tun_struct *tun = netdev_priv(dev);
 982         struct ifreq *ifr = tun->ifr;
 983         int err;
 984
 985         spin_lock_init(&tun->lock);
 986
 987         err = security_tun_dev_alloc_security(&tun->security);
 988         if (err < 0)
 989                 return err;
 990
 991         tun_flow_init(tun);
 992
 993         dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
 994         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 995                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 996                            NETIF_F_HW_VLAN_STAG_TX;
 997         dev->features = dev->hw_features;
 998         dev->vlan_features = dev->features &
 999                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1000                                NETIF_F_HW_VLAN_STAG_TX);
1001         dev->lltx = true;
1002
1003         tun->flags = (tun->flags & ~TUN_FEATURES) |
1004                       (ifr->ifr_flags & TUN_FEATURES);
1005
1006         INIT_LIST_HEAD(&tun->disabled);
1007         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1008                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1009         if (err < 0) {
1010                 tun_flow_uninit(tun);
1011                 security_tun_dev_free_security(tun->security);
1012                 return err;
1013         }
1014         return 0;
1015 }
1016
1017 /* Net device detach from fd. */
1018 static void tun_net_uninit(struct net_device *dev)
1019 {
1020         tun_detach_all(dev);
1021 }
1022
1023 /* Net device open. */
1024 static int tun_net_open(struct net_device *dev)
1025 {
1026         netif_tx_start_all_queues(dev);
1027
1028         return 0;
1029 }
1030
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1033 {
1034         netif_tx_stop_all_queues(dev);
1035         return 0;
1036 }
1037
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 {
1041 #ifdef CONFIG_RPS
1042         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1043                 /* Select queue was not called for the skbuff, so we extract the
1044                  * RPS hash and save it into the flow_table here.
1045                  */
1046                 struct tun_flow_entry *e;
1047                 __u32 rxhash;
1048
1049                 rxhash = __skb_get_hash_symmetric(skb);
1050                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1051                 if (e)
1052                         tun_flow_save_rps_rxhash(e, rxhash);
1053         }
1054 #endif
1055 }
1056
1057 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1058                                     struct sk_buff *skb,
1059                                     int len)
1060 {
1061         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1062
1063         if (prog)
1064                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1065
1066         return len;
1067 }
1068
1069 /* Net device start xmit */
1070 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1071 {
1072         struct tun_struct *tun = netdev_priv(dev);
1073         enum skb_drop_reason drop_reason;
1074         int txq = skb->queue_mapping;
1075         struct netdev_queue *queue;
1076         struct tun_file *tfile;
1077         int len = skb->len;
1078
1079         rcu_read_lock();
1080         tfile = rcu_dereference(tun->tfiles[txq]);
1081
1082         /* Drop packet if interface is not attached */
1083         if (!tfile) {
1084                 drop_reason = SKB_DROP_REASON_DEV_READY;
1085                 goto drop;
1086         }
1087
1088         if (!rcu_dereference(tun->steering_prog))
1089                 tun_automq_xmit(tun, skb);
1090
1091         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1092
1093         /* Drop if the filter does not like it.
1094          * This is a noop if the filter is disabled.
1095          * Filter can be enabled only for the TAP devices. */
1096         if (!check_filter(&tun->txflt, skb)) {
1097                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1098                 goto drop;
1099         }
1100
1101         if (tfile->socket.sk->sk_filter &&
1102             sk_filter(tfile->socket.sk, skb)) {
1103                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1104                 goto drop;
1105         }
1106
1107         len = run_ebpf_filter(tun, skb, len);
1108         if (len == 0) {
1109                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1110                 goto drop;
1111         }
1112
1113         if (pskb_trim(skb, len)) {
1114                 drop_reason = SKB_DROP_REASON_NOMEM;
1115                 goto drop;
1116         }
1117
1118         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1119                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1120                 goto drop;
1121         }
1122
1123         skb_tx_timestamp(skb);
1124
1125         /* Orphan the skb - required as we might hang on to it
1126          * for indefinite time.
1127          */
1128         skb_orphan(skb);
1129
1130         nf_reset_ct(skb);
1131
1132         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1133                 drop_reason = SKB_DROP_REASON_FULL_RING;
1134                 goto drop;
1135         }
1136
1137         /* dev->lltx requires to do our own update of trans_start */
1138         queue = netdev_get_tx_queue(dev, txq);
1139         txq_trans_cond_update(queue);
1140
1141         /* Notify and wake up reader process */
1142         if (tfile->flags & TUN_FASYNC)
1143                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1144         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1145
1146         rcu_read_unlock();
1147         return NETDEV_TX_OK;
1148
1149 drop:
1150         dev_core_stats_tx_dropped_inc(dev);
1151         skb_tx_error(skb);
1152         kfree_skb_reason(skb, drop_reason);
1153         rcu_read_unlock();
1154         return NET_XMIT_DROP;
1155 }
1156
1157 static void tun_net_mclist(struct net_device *dev)
1158 {
1159         /*
1160          * This callback is supposed to deal with mc filter in
1161          * _rx_ path and has nothing to do with the _tx_ path.
1162          * In rx path we always accept everything userspace gives us.
1163          */
1164 }
1165
1166 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1167         netdev_features_t features)
1168 {
1169         struct tun_struct *tun = netdev_priv(dev);
1170
1171         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1172 }
1173
1174 static void tun_set_headroom(struct net_device *dev, int new_hr)
1175 {
1176         struct tun_struct *tun = netdev_priv(dev);
1177
1178         if (new_hr < NET_SKB_PAD)
1179                 new_hr = NET_SKB_PAD;
1180
1181         tun->align = new_hr;
1182 }
1183
1184 static void
1185 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1186 {
1187         struct tun_struct *tun = netdev_priv(dev);
1188
1189         dev_get_tstats64(dev, stats);
1190
1191         stats->rx_frame_errors +=
1192                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1193 }
1194
1195 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1196                        struct netlink_ext_ack *extack)
1197 {
1198         struct tun_struct *tun = netdev_priv(dev);
1199         struct tun_file *tfile;
1200         struct bpf_prog *old_prog;
1201         int i;
1202
1203         old_prog = rtnl_dereference(tun->xdp_prog);
1204         rcu_assign_pointer(tun->xdp_prog, prog);
1205         if (old_prog)
1206                 bpf_prog_put(old_prog);
1207
1208         for (i = 0; i < tun->numqueues; i++) {
1209                 tfile = rtnl_dereference(tun->tfiles[i]);
1210                 if (prog)
1211                         sock_set_flag(&tfile->sk, SOCK_XDP);
1212                 else
1213                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1214         }
1215         list_for_each_entry(tfile, &tun->disabled, next) {
1216                 if (prog)
1217                         sock_set_flag(&tfile->sk, SOCK_XDP);
1218                 else
1219                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1220         }
1221
1222         return 0;
1223 }
1224
1225 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1226 {
1227         switch (xdp->command) {
1228         case XDP_SETUP_PROG:
1229                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1230         default:
1231                 return -EINVAL;
1232         }
1233 }
1234
1235 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1236 {
1237         if (new_carrier) {
1238                 struct tun_struct *tun = netdev_priv(dev);
1239
1240                 if (!tun->numqueues)
1241                         return -EPERM;
1242
1243                 netif_carrier_on(dev);
1244         } else {
1245                 netif_carrier_off(dev);
1246         }
1247         return 0;
1248 }
1249
1250 static const struct net_device_ops tun_netdev_ops = {
1251         .ndo_init               = tun_net_init,
1252         .ndo_uninit             = tun_net_uninit,
1253         .ndo_open               = tun_net_open,
1254         .ndo_stop               = tun_net_close,
1255         .ndo_start_xmit         = tun_net_xmit,
1256         .ndo_fix_features       = tun_net_fix_features,
1257         .ndo_select_queue       = tun_select_queue,
1258         .ndo_set_rx_headroom    = tun_set_headroom,
1259         .ndo_get_stats64        = tun_net_get_stats64,
1260         .ndo_change_carrier     = tun_net_change_carrier,
1261 };
1262
1263 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1264 {
1265         /* Notify and wake up reader process */
1266         if (tfile->flags & TUN_FASYNC)
1267                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1268         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1269 }
1270
1271 static int tun_xdp_xmit(struct net_device *dev, int n,
1272                         struct xdp_frame **frames, u32 flags)
1273 {
1274         struct tun_struct *tun = netdev_priv(dev);
1275         struct tun_file *tfile;
1276         u32 numqueues;
1277         int nxmit = 0;
1278         int i;
1279
1280         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1281                 return -EINVAL;
1282
1283         rcu_read_lock();
1284
1285 resample:
1286         numqueues = READ_ONCE(tun->numqueues);
1287         if (!numqueues) {
1288                 rcu_read_unlock();
1289                 return -ENXIO; /* Caller will free/return all frames */
1290         }
1291
1292         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1293                                             numqueues]);
1294         if (unlikely(!tfile))
1295                 goto resample;
1296
1297         spin_lock(&tfile->tx_ring.producer_lock);
1298         for (i = 0; i < n; i++) {
1299                 struct xdp_frame *xdp = frames[i];
1300                 /* Encode the XDP flag into lowest bit for consumer to differ
1301                  * XDP buffer from sk_buff.
1302                  */
1303                 void *frame = tun_xdp_to_ptr(xdp);
1304
1305                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1306                         dev_core_stats_tx_dropped_inc(dev);
1307                         break;
1308                 }
1309                 nxmit++;
1310         }
1311         spin_unlock(&tfile->tx_ring.producer_lock);
1312
1313         if (flags & XDP_XMIT_FLUSH)
1314                 __tun_xdp_flush_tfile(tfile);
1315
1316         rcu_read_unlock();
1317         return nxmit;
1318 }
1319
1320 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1321 {
1322         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1323         int nxmit;
1324
1325         if (unlikely(!frame))
1326                 return -EOVERFLOW;
1327
1328         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1329         if (!nxmit)
1330                 xdp_return_frame_rx_napi(frame);
1331         return nxmit;
1332 }
1333
1334 static const struct net_device_ops tap_netdev_ops = {
1335         .ndo_init               = tun_net_init,
1336         .ndo_uninit             = tun_net_uninit,
1337         .ndo_open               = tun_net_open,
1338         .ndo_stop               = tun_net_close,
1339         .ndo_start_xmit         = tun_net_xmit,
1340         .ndo_fix_features       = tun_net_fix_features,
1341         .ndo_set_rx_mode        = tun_net_mclist,
1342         .ndo_set_mac_address    = eth_mac_addr,
1343         .ndo_validate_addr      = eth_validate_addr,
1344         .ndo_select_queue       = tun_select_queue,
1345         .ndo_features_check     = passthru_features_check,
1346         .ndo_set_rx_headroom    = tun_set_headroom,
1347         .ndo_bpf                = tun_xdp,
1348         .ndo_xdp_xmit           = tun_xdp_xmit,
1349         .ndo_change_carrier     = tun_net_change_carrier,
1350 };
1351
1352 static void tun_flow_init(struct tun_struct *tun)
1353 {
1354         int i;
1355
1356         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1357                 INIT_HLIST_HEAD(&tun->flows[i]);
1358
1359         tun->ageing_time = TUN_FLOW_EXPIRE;
1360         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1361         mod_timer(&tun->flow_gc_timer,
1362                   round_jiffies_up(jiffies + tun->ageing_time));
1363 }
1364
1365 static void tun_flow_uninit(struct tun_struct *tun)
1366 {
1367         del_timer_sync(&tun->flow_gc_timer);
1368         tun_flow_flush(tun);
1369 }
1370
1371 #define MIN_MTU 68
1372 #define MAX_MTU 65535
1373
1374 /* Initialize net device. */
1375 static void tun_net_initialize(struct net_device *dev)
1376 {
1377         struct tun_struct *tun = netdev_priv(dev);
1378
1379         switch (tun->flags & TUN_TYPE_MASK) {
1380         case IFF_TUN:
1381                 dev->netdev_ops = &tun_netdev_ops;
1382                 dev->header_ops = &ip_tunnel_header_ops;
1383
1384                 /* Point-to-Point TUN Device */
1385                 dev->hard_header_len = 0;
1386                 dev->addr_len = 0;
1387                 dev->mtu = 1500;
1388
1389                 /* Zero header length */
1390                 dev->type = ARPHRD_NONE;
1391                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1392                 break;
1393
1394         case IFF_TAP:
1395                 dev->netdev_ops = &tap_netdev_ops;
1396                 /* Ethernet TAP Device */
1397                 ether_setup(dev);
1398                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1399                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1400
1401                 eth_hw_addr_random(dev);
1402
1403                 /* Currently tun does not support XDP, only tap does. */
1404                 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1405                                     NETDEV_XDP_ACT_REDIRECT |
1406                                     NETDEV_XDP_ACT_NDO_XMIT;
1407
1408                 break;
1409         }
1410
1411         dev->min_mtu = MIN_MTU;
1412         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1413 }
1414
1415 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1416 {
1417         struct sock *sk = tfile->socket.sk;
1418
1419         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1420 }
1421
1422 /* Character device part */
1423
1424 /* Poll */
1425 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1426 {
1427         struct tun_file *tfile = file->private_data;
1428         struct tun_struct *tun = tun_get(tfile);
1429         struct sock *sk;
1430         __poll_t mask = 0;
1431
1432         if (!tun)
1433                 return EPOLLERR;
1434
1435         sk = tfile->socket.sk;
1436
1437         poll_wait(file, sk_sleep(sk), wait);
1438
1439         if (!ptr_ring_empty(&tfile->tx_ring))
1440                 mask |= EPOLLIN | EPOLLRDNORM;
1441
1442         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1443          * guarantee EPOLLOUT to be raised by either here or
1444          * tun_sock_write_space(). Then process could get notification
1445          * after it writes to a down device and meets -EIO.
1446          */
1447         if (tun_sock_writeable(tun, tfile) ||
1448             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1449              tun_sock_writeable(tun, tfile)))
1450                 mask |= EPOLLOUT | EPOLLWRNORM;
1451
1452         if (tun->dev->reg_state != NETREG_REGISTERED)
1453                 mask = EPOLLERR;
1454
1455         tun_put(tun);
1456         return mask;
1457 }
1458
1459 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1460                                             size_t len,
1461                                             const struct iov_iter *it)
1462 {
1463         struct sk_buff *skb;
1464         size_t linear;
1465         int err;
1466         int i;
1467
1468         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1469             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1470                 return ERR_PTR(-EMSGSIZE);
1471
1472         local_bh_disable();
1473         skb = napi_get_frags(&tfile->napi);
1474         local_bh_enable();
1475         if (!skb)
1476                 return ERR_PTR(-ENOMEM);
1477
1478         linear = iov_iter_single_seg_count(it);
1479         err = __skb_grow(skb, linear);
1480         if (err)
1481                 goto free;
1482
1483         skb->len = len;
1484         skb->data_len = len - linear;
1485         skb->truesize += skb->data_len;
1486
1487         for (i = 1; i < it->nr_segs; i++) {
1488                 const struct iovec *iov = iter_iov(it) + i;
1489                 size_t fragsz = iov->iov_len;
1490                 struct page *page;
1491                 void *frag;
1492
1493                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1494                         err = -EINVAL;
1495                         goto free;
1496                 }
1497                 frag = netdev_alloc_frag(fragsz);
1498                 if (!frag) {
1499                         err = -ENOMEM;
1500                         goto free;
1501                 }
1502                 page = virt_to_head_page(frag);
1503                 skb_fill_page_desc(skb, i - 1, page,
1504                                    frag - page_address(page), fragsz);
1505         }
1506
1507         return skb;
1508 free:
1509         /* frees skb and all frags allocated with napi_alloc_frag() */
1510         napi_free_frags(&tfile->napi);
1511         return ERR_PTR(err);
1512 }
1513
1514 /* prepad is the amount to reserve at front.  len is length after that.
1515  * linear is a hint as to how much to copy (usually headers). */
1516 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1517                                      size_t prepad, size_t len,
1518                                      size_t linear, int noblock)
1519 {
1520         struct sock *sk = tfile->socket.sk;
1521         struct sk_buff *skb;
1522         int err;
1523
1524         /* Under a page?  Don't bother with paged skb. */
1525         if (prepad + len < PAGE_SIZE)
1526                 linear = len;
1527
1528         if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1529                 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1530         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1531                                    &err, PAGE_ALLOC_COSTLY_ORDER);
1532         if (!skb)
1533                 return ERR_PTR(err);
1534
1535         skb_reserve(skb, prepad);
1536         skb_put(skb, linear);
1537         skb->data_len = len - linear;
1538         skb->len += len - linear;
1539
1540         return skb;
1541 }
1542
1543 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1544                            struct sk_buff *skb, int more)
1545 {
1546         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1547         struct sk_buff_head process_queue;
1548         u32 rx_batched = tun->rx_batched;
1549         bool rcv = false;
1550
1551         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1552                 local_bh_disable();
1553                 skb_record_rx_queue(skb, tfile->queue_index);
1554                 netif_receive_skb(skb);
1555                 local_bh_enable();
1556                 return;
1557         }
1558
1559         spin_lock(&queue->lock);
1560         if (!more || skb_queue_len(queue) == rx_batched) {
1561                 __skb_queue_head_init(&process_queue);
1562                 skb_queue_splice_tail_init(queue, &process_queue);
1563                 rcv = true;
1564         } else {
1565                 __skb_queue_tail(queue, skb);
1566         }
1567         spin_unlock(&queue->lock);
1568
1569         if (rcv) {
1570                 struct sk_buff *nskb;
1571
1572                 local_bh_disable();
1573                 while ((nskb = __skb_dequeue(&process_queue))) {
1574                         skb_record_rx_queue(nskb, tfile->queue_index);
1575                         netif_receive_skb(nskb);
1576                 }
1577                 skb_record_rx_queue(skb, tfile->queue_index);
1578                 netif_receive_skb(skb);
1579                 local_bh_enable();
1580         }
1581 }
1582
1583 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1584                               int len, int noblock, bool zerocopy)
1585 {
1586         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1587                 return false;
1588
1589         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1590                 return false;
1591
1592         if (!noblock)
1593                 return false;
1594
1595         if (zerocopy)
1596                 return false;
1597
1598         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1599             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1600                 return false;
1601
1602         return true;
1603 }
1604
1605 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1606                                        struct page_frag *alloc_frag, char *buf,
1607                                        int buflen, int len, int pad)
1608 {
1609         struct sk_buff *skb = build_skb(buf, buflen);
1610
1611         if (!skb)
1612                 return ERR_PTR(-ENOMEM);
1613
1614         skb_reserve(skb, pad);
1615         skb_put(skb, len);
1616         skb_set_owner_w(skb, tfile->socket.sk);
1617
1618         get_page(alloc_frag->page);
1619         alloc_frag->offset += buflen;
1620
1621         return skb;
1622 }
1623
1624 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1625                        struct xdp_buff *xdp, u32 act)
1626 {
1627         int err;
1628
1629         switch (act) {
1630         case XDP_REDIRECT:
1631                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1632                 if (err) {
1633                         dev_core_stats_rx_dropped_inc(tun->dev);
1634                         return err;
1635                 }
1636                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1637                 break;
1638         case XDP_TX:
1639                 err = tun_xdp_tx(tun->dev, xdp);
1640                 if (err < 0) {
1641                         dev_core_stats_rx_dropped_inc(tun->dev);
1642                         return err;
1643                 }
1644                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1645                 break;
1646         case XDP_PASS:
1647                 break;
1648         default:
1649                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1650                 fallthrough;
1651         case XDP_ABORTED:
1652                 trace_xdp_exception(tun->dev, xdp_prog, act);
1653                 fallthrough;
1654         case XDP_DROP:
1655                 dev_core_stats_rx_dropped_inc(tun->dev);
1656                 break;
1657         }
1658
1659         return act;
1660 }
1661
1662 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1663                                      struct tun_file *tfile,
1664                                      struct iov_iter *from,
1665                                      struct virtio_net_hdr *hdr,
1666                                      int len, int *skb_xdp)
1667 {
1668         struct page_frag *alloc_frag = &current->task_frag;
1669         struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
1670         struct bpf_prog *xdp_prog;
1671         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1672         char *buf;
1673         size_t copied;
1674         int pad = TUN_RX_PAD;
1675         int err = 0;
1676
1677         rcu_read_lock();
1678         xdp_prog = rcu_dereference(tun->xdp_prog);
1679         if (xdp_prog)
1680                 pad += XDP_PACKET_HEADROOM;
1681         buflen += SKB_DATA_ALIGN(len + pad);
1682         rcu_read_unlock();
1683
1684         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1685         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1686                 return ERR_PTR(-ENOMEM);
1687
1688         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1689         copied = copy_page_from_iter(alloc_frag->page,
1690                                      alloc_frag->offset + pad,
1691                                      len, from);
1692         if (copied != len)
1693                 return ERR_PTR(-EFAULT);
1694
1695         /* There's a small window that XDP may be set after the check
1696          * of xdp_prog above, this should be rare and for simplicity
1697          * we do XDP on skb in case the headroom is not enough.
1698          */
1699         if (hdr->gso_type || !xdp_prog) {
1700                 *skb_xdp = 1;
1701                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1702                                        pad);
1703         }
1704
1705         *skb_xdp = 0;
1706
1707         local_bh_disable();
1708         rcu_read_lock();
1709         bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
1710         xdp_prog = rcu_dereference(tun->xdp_prog);
1711         if (xdp_prog) {
1712                 struct xdp_buff xdp;
1713                 u32 act;
1714
1715                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1716                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1717
1718                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1719                 if (act == XDP_REDIRECT || act == XDP_TX) {
1720                         get_page(alloc_frag->page);
1721                         alloc_frag->offset += buflen;
1722                 }
1723                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1724                 if (err < 0) {
1725                         if (act == XDP_REDIRECT || act == XDP_TX)
1726                                 put_page(alloc_frag->page);
1727                         goto out;
1728                 }
1729
1730                 if (err == XDP_REDIRECT)
1731                         xdp_do_flush();
1732                 if (err != XDP_PASS)
1733                         goto out;
1734
1735                 pad = xdp.data - xdp.data_hard_start;
1736                 len = xdp.data_end - xdp.data;
1737         }
1738         bpf_net_ctx_clear(bpf_net_ctx);
1739         rcu_read_unlock();
1740         local_bh_enable();
1741
1742         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1743
1744 out:
1745         bpf_net_ctx_clear(bpf_net_ctx);
1746         rcu_read_unlock();
1747         local_bh_enable();
1748         return NULL;
1749 }
1750
1751 /* Get packet from user space buffer */
1752 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1753                             void *msg_control, struct iov_iter *from,
1754                             int noblock, bool more)
1755 {
1756         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1757         struct sk_buff *skb;
1758         size_t total_len = iov_iter_count(from);
1759         size_t len = total_len, align = tun->align, linear;
1760         struct virtio_net_hdr gso = { 0 };
1761         int good_linear;
1762         int copylen;
1763         bool zerocopy = false;
1764         int err;
1765         u32 rxhash = 0;
1766         int skb_xdp = 1;
1767         bool frags = tun_napi_frags_enabled(tfile);
1768         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1769
1770         if (!(tun->flags & IFF_NO_PI)) {
1771                 if (len < sizeof(pi))
1772                         return -EINVAL;
1773                 len -= sizeof(pi);
1774
1775                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1776                         return -EFAULT;
1777         }
1778
1779         if (tun->flags & IFF_VNET_HDR) {
1780                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1781
1782                 if (len < vnet_hdr_sz)
1783                         return -EINVAL;
1784                 len -= vnet_hdr_sz;
1785
1786                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1787                         return -EFAULT;
1788
1789                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1790                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1791                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1792
1793                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1794                         return -EINVAL;
1795                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1796         }
1797
1798         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1799                 align += NET_IP_ALIGN;
1800                 if (unlikely(len < ETH_HLEN ||
1801                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1802                         return -EINVAL;
1803         }
1804
1805         good_linear = SKB_MAX_HEAD(align);
1806
1807         if (msg_control) {
1808                 struct iov_iter i = *from;
1809
1810                 /* There are 256 bytes to be copied in skb, so there is
1811                  * enough room for skb expand head in case it is used.
1812                  * The rest of the buffer is mapped from userspace.
1813                  */
1814                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1815                 if (copylen > good_linear)
1816                         copylen = good_linear;
1817                 linear = copylen;
1818                 iov_iter_advance(&i, copylen);
1819                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1820                         zerocopy = true;
1821         }
1822
1823         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1824                 /* For the packet that is not easy to be processed
1825                  * (e.g gso or jumbo packet), we will do it at after
1826                  * skb was created with generic XDP routine.
1827                  */
1828                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1829                 err = PTR_ERR_OR_ZERO(skb);
1830                 if (err)
1831                         goto drop;
1832                 if (!skb)
1833                         return total_len;
1834         } else {
1835                 if (!zerocopy) {
1836                         copylen = len;
1837                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1838                                 linear = good_linear;
1839                         else
1840                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1841                 }
1842
1843                 if (frags) {
1844                         mutex_lock(&tfile->napi_mutex);
1845                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1846                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1847                          * If zerocopy is enabled, then this layout will be
1848                          * overwritten by zerocopy_sg_from_iter().
1849                          */
1850                         zerocopy = false;
1851                 } else {
1852                         if (!linear)
1853                                 linear = min_t(size_t, good_linear, copylen);
1854
1855                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1856                                             noblock);
1857                 }
1858
1859                 err = PTR_ERR_OR_ZERO(skb);
1860                 if (err)
1861                         goto drop;
1862
1863                 if (zerocopy)
1864                         err = zerocopy_sg_from_iter(skb, from);
1865                 else
1866                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1867
1868                 if (err) {
1869                         err = -EFAULT;
1870                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1871                         goto drop;
1872                 }
1873         }
1874
1875         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1876                 atomic_long_inc(&tun->rx_frame_errors);
1877                 err = -EINVAL;
1878                 goto free_skb;
1879         }
1880
1881         switch (tun->flags & TUN_TYPE_MASK) {
1882         case IFF_TUN:
1883                 if (tun->flags & IFF_NO_PI) {
1884                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1885
1886                         switch (ip_version) {
1887                         case 4:
1888                                 pi.proto = htons(ETH_P_IP);
1889                                 break;
1890                         case 6:
1891                                 pi.proto = htons(ETH_P_IPV6);
1892                                 break;
1893                         default:
1894                                 err = -EINVAL;
1895                                 goto drop;
1896                         }
1897                 }
1898
1899                 skb_reset_mac_header(skb);
1900                 skb->protocol = pi.proto;
1901                 skb->dev = tun->dev;
1902                 break;
1903         case IFF_TAP:
1904                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1905                         err = -ENOMEM;
1906                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1907                         goto drop;
1908                 }
1909                 skb->protocol = eth_type_trans(skb, tun->dev);
1910                 break;
1911         }
1912
1913         /* copy skb_ubuf_info for callback when skb has no error */
1914         if (zerocopy) {
1915                 skb_zcopy_init(skb, msg_control);
1916         } else if (msg_control) {
1917                 struct ubuf_info *uarg = msg_control;
1918                 uarg->ops->complete(NULL, uarg, false);
1919         }
1920
1921         skb_reset_network_header(skb);
1922         skb_probe_transport_header(skb);
1923         skb_record_rx_queue(skb, tfile->queue_index);
1924
1925         if (skb_xdp) {
1926                 struct bpf_prog *xdp_prog;
1927                 int ret;
1928
1929                 local_bh_disable();
1930                 rcu_read_lock();
1931                 xdp_prog = rcu_dereference(tun->xdp_prog);
1932                 if (xdp_prog) {
1933                         ret = do_xdp_generic(xdp_prog, &skb);
1934                         if (ret != XDP_PASS) {
1935                                 rcu_read_unlock();
1936                                 local_bh_enable();
1937                                 goto unlock_frags;
1938                         }
1939                 }
1940                 rcu_read_unlock();
1941                 local_bh_enable();
1942         }
1943
1944         /* Compute the costly rx hash only if needed for flow updates.
1945          * We may get a very small possibility of OOO during switching, not
1946          * worth to optimize.
1947          */
1948         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1949             !tfile->detached)
1950                 rxhash = __skb_get_hash_symmetric(skb);
1951
1952         rcu_read_lock();
1953         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1954                 err = -EIO;
1955                 rcu_read_unlock();
1956                 drop_reason = SKB_DROP_REASON_DEV_READY;
1957                 goto drop;
1958         }
1959
1960         if (frags) {
1961                 u32 headlen;
1962
1963                 /* Exercise flow dissector code path. */
1964                 skb_push(skb, ETH_HLEN);
1965                 headlen = eth_get_headlen(tun->dev, skb->data,
1966                                           skb_headlen(skb));
1967
1968                 if (unlikely(headlen > skb_headlen(skb))) {
1969                         WARN_ON_ONCE(1);
1970                         err = -ENOMEM;
1971                         dev_core_stats_rx_dropped_inc(tun->dev);
1972 napi_busy:
1973                         napi_free_frags(&tfile->napi);
1974                         rcu_read_unlock();
1975                         mutex_unlock(&tfile->napi_mutex);
1976                         return err;
1977                 }
1978
1979                 if (likely(napi_schedule_prep(&tfile->napi))) {
1980                         local_bh_disable();
1981                         napi_gro_frags(&tfile->napi);
1982                         napi_complete(&tfile->napi);
1983                         local_bh_enable();
1984                 } else {
1985                         err = -EBUSY;
1986                         goto napi_busy;
1987                 }
1988                 mutex_unlock(&tfile->napi_mutex);
1989         } else if (tfile->napi_enabled) {
1990                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1991                 int queue_len;
1992
1993                 spin_lock_bh(&queue->lock);
1994
1995                 if (unlikely(tfile->detached)) {
1996                         spin_unlock_bh(&queue->lock);
1997                         rcu_read_unlock();
1998                         err = -EBUSY;
1999                         goto free_skb;
2000                 }
2001
2002                 __skb_queue_tail(queue, skb);
2003                 queue_len = skb_queue_len(queue);
2004                 spin_unlock(&queue->lock);
2005
2006                 if (!more || queue_len > NAPI_POLL_WEIGHT)
2007                         napi_schedule(&tfile->napi);
2008
2009                 local_bh_enable();
2010         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2011                 tun_rx_batched(tun, tfile, skb, more);
2012         } else {
2013                 netif_rx(skb);
2014         }
2015         rcu_read_unlock();
2016
2017         preempt_disable();
2018         dev_sw_netstats_rx_add(tun->dev, len);
2019         preempt_enable();
2020
2021         if (rxhash)
2022                 tun_flow_update(tun, rxhash, tfile);
2023
2024         return total_len;
2025
2026 drop:
2027         if (err != -EAGAIN)
2028                 dev_core_stats_rx_dropped_inc(tun->dev);
2029
2030 free_skb:
2031         if (!IS_ERR_OR_NULL(skb))
2032                 kfree_skb_reason(skb, drop_reason);
2033
2034 unlock_frags:
2035         if (frags) {
2036                 tfile->napi.skb = NULL;
2037                 mutex_unlock(&tfile->napi_mutex);
2038         }
2039
2040         return err ?: total_len;
2041 }
2042
2043 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2044 {
2045         struct file *file = iocb->ki_filp;
2046         struct tun_file *tfile = file->private_data;
2047         struct tun_struct *tun = tun_get(tfile);
2048         ssize_t result;
2049         int noblock = 0;
2050
2051         if (!tun)
2052                 return -EBADFD;
2053
2054         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2055                 noblock = 1;
2056
2057         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2058
2059         tun_put(tun);
2060         return result;
2061 }
2062
2063 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2064                                 struct tun_file *tfile,
2065                                 struct xdp_frame *xdp_frame,
2066                                 struct iov_iter *iter)
2067 {
2068         int vnet_hdr_sz = 0;
2069         size_t size = xdp_frame->len;
2070         size_t ret;
2071
2072         if (tun->flags & IFF_VNET_HDR) {
2073                 struct virtio_net_hdr gso = { 0 };
2074
2075                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2076                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2077                         return -EINVAL;
2078                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2079                              sizeof(gso)))
2080                         return -EFAULT;
2081                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2082         }
2083
2084         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2085
2086         preempt_disable();
2087         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2088         preempt_enable();
2089
2090         return ret;
2091 }
2092
2093 /* Put packet to the user space buffer */
2094 static ssize_t tun_put_user(struct tun_struct *tun,
2095                             struct tun_file *tfile,
2096                             struct sk_buff *skb,
2097                             struct iov_iter *iter)
2098 {
2099         struct tun_pi pi = { 0, skb->protocol };
2100         ssize_t total;
2101         int vlan_offset = 0;
2102         int vlan_hlen = 0;
2103         int vnet_hdr_sz = 0;
2104
2105         if (skb_vlan_tag_present(skb))
2106                 vlan_hlen = VLAN_HLEN;
2107
2108         if (tun->flags & IFF_VNET_HDR)
2109                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2110
2111         total = skb->len + vlan_hlen + vnet_hdr_sz;
2112
2113         if (!(tun->flags & IFF_NO_PI)) {
2114                 if (iov_iter_count(iter) < sizeof(pi))
2115                         return -EINVAL;
2116
2117                 total += sizeof(pi);
2118                 if (iov_iter_count(iter) < total) {
2119                         /* Packet will be striped */
2120                         pi.flags |= TUN_PKT_STRIP;
2121                 }
2122
2123                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2124                         return -EFAULT;
2125         }
2126
2127         if (vnet_hdr_sz) {
2128                 struct virtio_net_hdr gso;
2129
2130                 if (iov_iter_count(iter) < vnet_hdr_sz)
2131                         return -EINVAL;
2132
2133                 if (virtio_net_hdr_from_skb(skb, &gso,
2134                                             tun_is_little_endian(tun), true,
2135                                             vlan_hlen)) {
2136                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2137
2138                         if (net_ratelimit()) {
2139                                 netdev_err(tun->dev, "unexpected GSO type: 0x%x, gso_size %d, hdr_len %d\n",
2140                                            sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2141                                            tun16_to_cpu(tun, gso.hdr_len));
2142                                 print_hex_dump(KERN_ERR, "tun: ",
2143                                                DUMP_PREFIX_NONE,
2144                                                16, 1, skb->head,
2145                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2146                         }
2147                         WARN_ON_ONCE(1);
2148                         return -EINVAL;
2149                 }
2150
2151                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2152                         return -EFAULT;
2153
2154                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2155         }
2156
2157         if (vlan_hlen) {
2158                 int ret;
2159                 struct veth veth;
2160
2161                 veth.h_vlan_proto = skb->vlan_proto;
2162                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2163
2164                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2165
2166                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2167                 if (ret || !iov_iter_count(iter))
2168                         goto done;
2169
2170                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2171                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2172                         goto done;
2173         }
2174
2175         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2176
2177 done:
2178         /* caller is in process context, */
2179         preempt_disable();
2180         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2181         preempt_enable();
2182
2183         return total;
2184 }
2185
2186 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2187 {
2188         DECLARE_WAITQUEUE(wait, current);
2189         void *ptr = NULL;
2190         int error = 0;
2191
2192         ptr = ptr_ring_consume(&tfile->tx_ring);
2193         if (ptr)
2194                 goto out;
2195         if (noblock) {
2196                 error = -EAGAIN;
2197                 goto out;
2198         }
2199
2200         add_wait_queue(&tfile->socket.wq.wait, &wait);
2201
2202         while (1) {
2203                 set_current_state(TASK_INTERRUPTIBLE);
2204                 ptr = ptr_ring_consume(&tfile->tx_ring);
2205                 if (ptr)
2206                         break;
2207                 if (signal_pending(current)) {
2208                         error = -ERESTARTSYS;
2209                         break;
2210                 }
2211                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2212                         error = -EFAULT;
2213                         break;
2214                 }
2215
2216                 schedule();
2217         }
2218
2219         __set_current_state(TASK_RUNNING);
2220         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2221
2222 out:
2223         *err = error;
2224         return ptr;
2225 }
2226
2227 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2228                            struct iov_iter *to,
2229                            int noblock, void *ptr)
2230 {
2231         ssize_t ret;
2232         int err;
2233
2234         if (!iov_iter_count(to)) {
2235                 tun_ptr_free(ptr);
2236                 return 0;
2237         }
2238
2239         if (!ptr) {
2240                 /* Read frames from ring */
2241                 ptr = tun_ring_recv(tfile, noblock, &err);
2242                 if (!ptr)
2243                         return err;
2244         }
2245
2246         if (tun_is_xdp_frame(ptr)) {
2247                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2248
2249                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2250                 xdp_return_frame(xdpf);
2251         } else {
2252                 struct sk_buff *skb = ptr;
2253
2254                 ret = tun_put_user(tun, tfile, skb, to);
2255                 if (unlikely(ret < 0))
2256                         kfree_skb(skb);
2257                 else
2258                         consume_skb(skb);
2259         }
2260
2261         return ret;
2262 }
2263
2264 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2265 {
2266         struct file *file = iocb->ki_filp;
2267         struct tun_file *tfile = file->private_data;
2268         struct tun_struct *tun = tun_get(tfile);
2269         ssize_t len = iov_iter_count(to), ret;
2270         int noblock = 0;
2271
2272         if (!tun)
2273                 return -EBADFD;
2274
2275         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2276                 noblock = 1;
2277
2278         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2279         ret = min_t(ssize_t, ret, len);
2280         if (ret > 0)
2281                 iocb->ki_pos = ret;
2282         tun_put(tun);
2283         return ret;
2284 }
2285
2286 static void tun_prog_free(struct rcu_head *rcu)
2287 {
2288         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2289
2290         bpf_prog_destroy(prog->prog);
2291         kfree(prog);
2292 }
2293
2294 static int __tun_set_ebpf(struct tun_struct *tun,
2295                           struct tun_prog __rcu **prog_p,
2296                           struct bpf_prog *prog)
2297 {
2298         struct tun_prog *old, *new = NULL;
2299
2300         if (prog) {
2301                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2302                 if (!new)
2303                         return -ENOMEM;
2304                 new->prog = prog;
2305         }
2306
2307         spin_lock_bh(&tun->lock);
2308         old = rcu_dereference_protected(*prog_p,
2309                                         lockdep_is_held(&tun->lock));
2310         rcu_assign_pointer(*prog_p, new);
2311         spin_unlock_bh(&tun->lock);
2312
2313         if (old)
2314                 call_rcu(&old->rcu, tun_prog_free);
2315
2316         return 0;
2317 }
2318
2319 static void tun_free_netdev(struct net_device *dev)
2320 {
2321         struct tun_struct *tun = netdev_priv(dev);
2322
2323         BUG_ON(!(list_empty(&tun->disabled)));
2324
2325         tun_flow_uninit(tun);
2326         security_tun_dev_free_security(tun->security);
2327         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2328         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2329 }
2330
2331 static void tun_setup(struct net_device *dev)
2332 {
2333         struct tun_struct *tun = netdev_priv(dev);
2334
2335         tun->owner = INVALID_UID;
2336         tun->group = INVALID_GID;
2337         tun_default_link_ksettings(dev, &tun->link_ksettings);
2338
2339         dev->ethtool_ops = &tun_ethtool_ops;
2340         dev->needs_free_netdev = true;
2341         dev->priv_destructor = tun_free_netdev;
2342         /* We prefer our own queue length */
2343         dev->tx_queue_len = TUN_READQ_SIZE;
2344 }
2345
2346 /* Trivial set of netlink ops to allow deleting tun or tap
2347  * device with netlink.
2348  */
2349 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2350                         struct netlink_ext_ack *extack)
2351 {
2352         NL_SET_ERR_MSG(extack,
2353                        "tun/tap creation via rtnetlink is not supported.");
2354         return -EOPNOTSUPP;
2355 }
2356
2357 static size_t tun_get_size(const struct net_device *dev)
2358 {
2359         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2360         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2361
2362         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2363                nla_total_size(sizeof(gid_t)) + /* GROUP */
2364                nla_total_size(sizeof(u8)) + /* TYPE */
2365                nla_total_size(sizeof(u8)) + /* PI */
2366                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2367                nla_total_size(sizeof(u8)) + /* PERSIST */
2368                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2369                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2370                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2371                0;
2372 }
2373
2374 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2375 {
2376         struct tun_struct *tun = netdev_priv(dev);
2377
2378         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2379                 goto nla_put_failure;
2380         if (uid_valid(tun->owner) &&
2381             nla_put_u32(skb, IFLA_TUN_OWNER,
2382                         from_kuid_munged(current_user_ns(), tun->owner)))
2383                 goto nla_put_failure;
2384         if (gid_valid(tun->group) &&
2385             nla_put_u32(skb, IFLA_TUN_GROUP,
2386                         from_kgid_munged(current_user_ns(), tun->group)))
2387                 goto nla_put_failure;
2388         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2389                 goto nla_put_failure;
2390         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2391                 goto nla_put_failure;
2392         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2393                 goto nla_put_failure;
2394         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2395                        !!(tun->flags & IFF_MULTI_QUEUE)))
2396                 goto nla_put_failure;
2397         if (tun->flags & IFF_MULTI_QUEUE) {
2398                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2399                         goto nla_put_failure;
2400                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2401                                 tun->numdisabled))
2402                         goto nla_put_failure;
2403         }
2404
2405         return 0;
2406
2407 nla_put_failure:
2408         return -EMSGSIZE;
2409 }
2410
2411 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2412         .kind           = DRV_NAME,
2413         .priv_size      = sizeof(struct tun_struct),
2414         .setup          = tun_setup,
2415         .validate       = tun_validate,
2416         .get_size       = tun_get_size,
2417         .fill_info      = tun_fill_info,
2418 };
2419
2420 static void tun_sock_write_space(struct sock *sk)
2421 {
2422         struct tun_file *tfile;
2423         wait_queue_head_t *wqueue;
2424
2425         if (!sock_writeable(sk))
2426                 return;
2427
2428         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2429                 return;
2430
2431         wqueue = sk_sleep(sk);
2432         if (wqueue && waitqueue_active(wqueue))
2433                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2434                                                 EPOLLWRNORM | EPOLLWRBAND);
2435
2436         tfile = container_of(sk, struct tun_file, sk);
2437         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2438 }
2439
2440 static void tun_put_page(struct tun_page *tpage)
2441 {
2442         if (tpage->page)
2443                 __page_frag_cache_drain(tpage->page, tpage->count);
2444 }
2445
2446 static int tun_xdp_one(struct tun_struct *tun,
2447                        struct tun_file *tfile,
2448                        struct xdp_buff *xdp, int *flush,
2449                        struct tun_page *tpage)
2450 {
2451         unsigned int datasize = xdp->data_end - xdp->data;
2452         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2453         struct virtio_net_hdr *gso = &hdr->gso;
2454         struct bpf_prog *xdp_prog;
2455         struct sk_buff *skb = NULL;
2456         struct sk_buff_head *queue;
2457         u32 rxhash = 0, act;
2458         int buflen = hdr->buflen;
2459         int ret = 0;
2460         bool skb_xdp = false;
2461         struct page *page;
2462
2463         if (unlikely(datasize < ETH_HLEN))
2464                 return -EINVAL;
2465
2466         xdp_prog = rcu_dereference(tun->xdp_prog);
2467         if (xdp_prog) {
2468                 if (gso->gso_type) {
2469                         skb_xdp = true;
2470                         goto build;
2471                 }
2472
2473                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2474                 xdp_set_data_meta_invalid(xdp);
2475
2476                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2477                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2478                 if (ret < 0) {
2479                         put_page(virt_to_head_page(xdp->data));
2480                         return ret;
2481                 }
2482
2483                 switch (ret) {
2484                 case XDP_REDIRECT:
2485                         *flush = true;
2486                         fallthrough;
2487                 case XDP_TX:
2488                         return 0;
2489                 case XDP_PASS:
2490                         break;
2491                 default:
2492                         page = virt_to_head_page(xdp->data);
2493                         if (tpage->page == page) {
2494                                 ++tpage->count;
2495                         } else {
2496                                 tun_put_page(tpage);
2497                                 tpage->page = page;
2498                                 tpage->count = 1;
2499                         }
2500                         return 0;
2501                 }
2502         }
2503
2504 build:
2505         skb = build_skb(xdp->data_hard_start, buflen);
2506         if (!skb) {
2507                 ret = -ENOMEM;
2508                 goto out;
2509         }
2510
2511         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2512         skb_put(skb, xdp->data_end - xdp->data);
2513
2514         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2515                 atomic_long_inc(&tun->rx_frame_errors);
2516                 kfree_skb(skb);
2517                 ret = -EINVAL;
2518                 goto out;
2519         }
2520
2521         skb->protocol = eth_type_trans(skb, tun->dev);
2522         skb_reset_network_header(skb);
2523         skb_probe_transport_header(skb);
2524         skb_record_rx_queue(skb, tfile->queue_index);
2525
2526         if (skb_xdp) {
2527                 ret = do_xdp_generic(xdp_prog, &skb);
2528                 if (ret != XDP_PASS) {
2529                         ret = 0;
2530                         goto out;
2531                 }
2532         }
2533
2534         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2535             !tfile->detached)
2536                 rxhash = __skb_get_hash_symmetric(skb);
2537
2538         if (tfile->napi_enabled) {
2539                 queue = &tfile->sk.sk_write_queue;
2540                 spin_lock(&queue->lock);
2541
2542                 if (unlikely(tfile->detached)) {
2543                         spin_unlock(&queue->lock);
2544                         kfree_skb(skb);
2545                         return -EBUSY;
2546                 }
2547
2548                 __skb_queue_tail(queue, skb);
2549                 spin_unlock(&queue->lock);
2550                 ret = 1;
2551         } else {
2552                 netif_receive_skb(skb);
2553                 ret = 0;
2554         }
2555
2556         /* No need to disable preemption here since this function is
2557          * always called with bh disabled
2558          */
2559         dev_sw_netstats_rx_add(tun->dev, datasize);
2560
2561         if (rxhash)
2562                 tun_flow_update(tun, rxhash, tfile);
2563
2564 out:
2565         return ret;
2566 }
2567
2568 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2569 {
2570         int ret, i;
2571         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2572         struct tun_struct *tun = tun_get(tfile);
2573         struct tun_msg_ctl *ctl = m->msg_control;
2574         struct xdp_buff *xdp;
2575
2576         if (!tun)
2577                 return -EBADFD;
2578
2579         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2580             ctl && ctl->type == TUN_MSG_PTR) {
2581                 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
2582                 struct tun_page tpage;
2583                 int n = ctl->num;
2584                 int flush = 0, queued = 0;
2585
2586                 memset(&tpage, 0, sizeof(tpage));
2587
2588                 local_bh_disable();
2589                 rcu_read_lock();
2590                 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
2591
2592                 for (i = 0; i < n; i++) {
2593                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2594                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2595                         if (ret > 0)
2596                                 queued += ret;
2597                 }
2598
2599                 if (flush)
2600                         xdp_do_flush();
2601
2602                 if (tfile->napi_enabled && queued > 0)
2603                         napi_schedule(&tfile->napi);
2604
2605                 bpf_net_ctx_clear(bpf_net_ctx);
2606                 rcu_read_unlock();
2607                 local_bh_enable();
2608
2609                 tun_put_page(&tpage);
2610
2611                 ret = total_len;
2612                 goto out;
2613         }
2614
2615         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2616                            m->msg_flags & MSG_DONTWAIT,
2617                            m->msg_flags & MSG_MORE);
2618 out:
2619         tun_put(tun);
2620         return ret;
2621 }
2622
2623 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2624                        int flags)
2625 {
2626         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2627         struct tun_struct *tun = tun_get(tfile);
2628         void *ptr = m->msg_control;
2629         int ret;
2630
2631         if (!tun) {
2632                 ret = -EBADFD;
2633                 goto out_free;
2634         }
2635
2636         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2637                 ret = -EINVAL;
2638                 goto out_put_tun;
2639         }
2640         if (flags & MSG_ERRQUEUE) {
2641                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2642                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2643                 goto out;
2644         }
2645         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2646         if (ret > (ssize_t)total_len) {
2647                 m->msg_flags |= MSG_TRUNC;
2648                 ret = flags & MSG_TRUNC ? ret : total_len;
2649         }
2650 out:
2651         tun_put(tun);
2652         return ret;
2653
2654 out_put_tun:
2655         tun_put(tun);
2656 out_free:
2657         tun_ptr_free(ptr);
2658         return ret;
2659 }
2660
2661 static int tun_ptr_peek_len(void *ptr)
2662 {
2663         if (likely(ptr)) {
2664                 if (tun_is_xdp_frame(ptr)) {
2665                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2666
2667                         return xdpf->len;
2668                 }
2669                 return __skb_array_len_with_tag(ptr);
2670         } else {
2671                 return 0;
2672         }
2673 }
2674
2675 static int tun_peek_len(struct socket *sock)
2676 {
2677         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2678         struct tun_struct *tun;
2679         int ret = 0;
2680
2681         tun = tun_get(tfile);
2682         if (!tun)
2683                 return 0;
2684
2685         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2686         tun_put(tun);
2687
2688         return ret;
2689 }
2690
2691 /* Ops structure to mimic raw sockets with tun */
2692 static const struct proto_ops tun_socket_ops = {
2693         .peek_len = tun_peek_len,
2694         .sendmsg = tun_sendmsg,
2695         .recvmsg = tun_recvmsg,
2696 };
2697
2698 static struct proto tun_proto = {
2699         .name           = "tun",
2700         .owner          = THIS_MODULE,
2701         .obj_size       = sizeof(struct tun_file),
2702 };
2703
2704 static int tun_flags(struct tun_struct *tun)
2705 {
2706         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2707 }
2708
2709 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2710                               char *buf)
2711 {
2712         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2713         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2714 }
2715
2716 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2717                           char *buf)
2718 {
2719         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2720         return uid_valid(tun->owner)?
2721                 sysfs_emit(buf, "%u\n",
2722                            from_kuid_munged(current_user_ns(), tun->owner)) :
2723                 sysfs_emit(buf, "-1\n");
2724 }
2725
2726 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2727                           char *buf)
2728 {
2729         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2730         return gid_valid(tun->group) ?
2731                 sysfs_emit(buf, "%u\n",
2732                            from_kgid_munged(current_user_ns(), tun->group)) :
2733                 sysfs_emit(buf, "-1\n");
2734 }
2735
2736 static DEVICE_ATTR_RO(tun_flags);
2737 static DEVICE_ATTR_RO(owner);
2738 static DEVICE_ATTR_RO(group);
2739
2740 static struct attribute *tun_dev_attrs[] = {
2741         &dev_attr_tun_flags.attr,
2742         &dev_attr_owner.attr,
2743         &dev_attr_group.attr,
2744         NULL
2745 };
2746
2747 static const struct attribute_group tun_attr_group = {
2748         .attrs = tun_dev_attrs
2749 };
2750
2751 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2752 {
2753         struct tun_struct *tun;
2754         struct tun_file *tfile = file->private_data;
2755         struct net_device *dev;
2756         int err;
2757
2758         if (tfile->detached)
2759                 return -EINVAL;
2760
2761         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2762                 if (!capable(CAP_NET_ADMIN))
2763                         return -EPERM;
2764
2765                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2766                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2767                         return -EINVAL;
2768         }
2769
2770         dev = __dev_get_by_name(net, ifr->ifr_name);
2771         if (dev) {
2772                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2773                         return -EBUSY;
2774                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2775                         tun = netdev_priv(dev);
2776                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2777                         tun = netdev_priv(dev);
2778                 else
2779                         return -EINVAL;
2780
2781                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2782                     !!(tun->flags & IFF_MULTI_QUEUE))
2783                         return -EINVAL;
2784
2785                 if (!tun_capable(tun))
2786                         return -EPERM;
2787                 err = security_tun_dev_open(tun->security);
2788                 if (err < 0)
2789                         return err;
2790
2791                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2792                                  ifr->ifr_flags & IFF_NAPI,
2793                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2794                 if (err < 0)
2795                         return err;
2796
2797                 if (tun->flags & IFF_MULTI_QUEUE &&
2798                     (tun->numqueues + tun->numdisabled > 1)) {
2799                         /* One or more queue has already been attached, no need
2800                          * to initialize the device again.
2801                          */
2802                         netdev_state_change(dev);
2803                         return 0;
2804                 }
2805
2806                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2807                               (ifr->ifr_flags & TUN_FEATURES);
2808
2809                 netdev_state_change(dev);
2810         } else {
2811                 char *name;
2812                 unsigned long flags = 0;
2813                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2814                              MAX_TAP_QUEUES : 1;
2815
2816                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2817                         return -EPERM;
2818                 err = security_tun_dev_create();
2819                 if (err < 0)
2820                         return err;
2821
2822                 /* Set dev type */
2823                 if (ifr->ifr_flags & IFF_TUN) {
2824                         /* TUN device */
2825                         flags |= IFF_TUN;
2826                         name = "tun%d";
2827                 } else if (ifr->ifr_flags & IFF_TAP) {
2828                         /* TAP device */
2829                         flags |= IFF_TAP;
2830                         name = "tap%d";
2831                 } else
2832                         return -EINVAL;
2833
2834                 if (*ifr->ifr_name)
2835                         name = ifr->ifr_name;
2836
2837                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2838                                        NET_NAME_UNKNOWN, tun_setup, queues,
2839                                        queues);
2840
2841                 if (!dev)
2842                         return -ENOMEM;
2843
2844                 dev_net_set(dev, net);
2845                 dev->rtnl_link_ops = &tun_link_ops;
2846                 dev->ifindex = tfile->ifindex;
2847                 dev->sysfs_groups[0] = &tun_attr_group;
2848
2849                 tun = netdev_priv(dev);
2850                 tun->dev = dev;
2851                 tun->flags = flags;
2852                 tun->txflt.count = 0;
2853                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2854
2855                 tun->align = NET_SKB_PAD;
2856                 tun->filter_attached = false;
2857                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2858                 tun->rx_batched = 0;
2859                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2860
2861                 tun->ifr = ifr;
2862                 tun->file = file;
2863
2864                 tun_net_initialize(dev);
2865
2866                 err = register_netdevice(tun->dev);
2867                 if (err < 0) {
2868                         free_netdev(dev);
2869                         return err;
2870                 }
2871                 /* free_netdev() won't check refcnt, to avoid race
2872                  * with dev_put() we need publish tun after registration.
2873                  */
2874                 rcu_assign_pointer(tfile->tun, tun);
2875         }
2876
2877         if (ifr->ifr_flags & IFF_NO_CARRIER)
2878                 netif_carrier_off(tun->dev);
2879         else
2880                 netif_carrier_on(tun->dev);
2881
2882         /* Make sure persistent devices do not get stuck in
2883          * xoff state.
2884          */
2885         if (netif_running(tun->dev))
2886                 netif_tx_wake_all_queues(tun->dev);
2887
2888         strcpy(ifr->ifr_name, tun->dev->name);
2889         return 0;
2890 }
2891
2892 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2893 {
2894         strcpy(ifr->ifr_name, tun->dev->name);
2895
2896         ifr->ifr_flags = tun_flags(tun);
2897
2898 }
2899
2900 /* This is like a cut-down ethtool ops, except done via tun fd so no
2901  * privs required. */
2902 static int set_offload(struct tun_struct *tun, unsigned long arg)
2903 {
2904         netdev_features_t features = 0;
2905
2906         if (arg & TUN_F_CSUM) {
2907                 features |= NETIF_F_HW_CSUM;
2908                 arg &= ~TUN_F_CSUM;
2909
2910                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2911                         if (arg & TUN_F_TSO_ECN) {
2912                                 features |= NETIF_F_TSO_ECN;
2913                                 arg &= ~TUN_F_TSO_ECN;
2914                         }
2915                         if (arg & TUN_F_TSO4)
2916                                 features |= NETIF_F_TSO;
2917                         if (arg & TUN_F_TSO6)
2918                                 features |= NETIF_F_TSO6;
2919                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2920                 }
2921
2922                 arg &= ~TUN_F_UFO;
2923
2924                 /* TODO: for now USO4 and USO6 should work simultaneously */
2925                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2926                         features |= NETIF_F_GSO_UDP_L4;
2927                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2928                 }
2929         }
2930
2931         /* This gives the user a way to test for new features in future by
2932          * trying to set them. */
2933         if (arg)
2934                 return -EINVAL;
2935
2936         tun->set_features = features;
2937         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2938         tun->dev->wanted_features |= features;
2939         netdev_update_features(tun->dev);
2940
2941         return 0;
2942 }
2943
2944 static void tun_detach_filter(struct tun_struct *tun, int n)
2945 {
2946         int i;
2947         struct tun_file *tfile;
2948
2949         for (i = 0; i < n; i++) {
2950                 tfile = rtnl_dereference(tun->tfiles[i]);
2951                 lock_sock(tfile->socket.sk);
2952                 sk_detach_filter(tfile->socket.sk);
2953                 release_sock(tfile->socket.sk);
2954         }
2955
2956         tun->filter_attached = false;
2957 }
2958
2959 static int tun_attach_filter(struct tun_struct *tun)
2960 {
2961         int i, ret = 0;
2962         struct tun_file *tfile;
2963
2964         for (i = 0; i < tun->numqueues; i++) {
2965                 tfile = rtnl_dereference(tun->tfiles[i]);
2966                 lock_sock(tfile->socket.sk);
2967                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2968                 release_sock(tfile->socket.sk);
2969                 if (ret) {
2970                         tun_detach_filter(tun, i);
2971                         return ret;
2972                 }
2973         }
2974
2975         tun->filter_attached = true;
2976         return ret;
2977 }
2978
2979 static void tun_set_sndbuf(struct tun_struct *tun)
2980 {
2981         struct tun_file *tfile;
2982         int i;
2983
2984         for (i = 0; i < tun->numqueues; i++) {
2985                 tfile = rtnl_dereference(tun->tfiles[i]);
2986                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2987         }
2988 }
2989
2990 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2991 {
2992         struct tun_file *tfile = file->private_data;
2993         struct tun_struct *tun;
2994         int ret = 0;
2995
2996         rtnl_lock();
2997
2998         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2999                 tun = tfile->detached;
3000                 if (!tun) {
3001                         ret = -EINVAL;
3002                         goto unlock;
3003                 }
3004                 ret = security_tun_dev_attach_queue(tun->security);
3005                 if (ret < 0)
3006                         goto unlock;
3007                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3008                                  tun->flags & IFF_NAPI_FRAGS, true);
3009         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3010                 tun = rtnl_dereference(tfile->tun);
3011                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3012                         ret = -EINVAL;
3013                 else
3014                         __tun_detach(tfile, false);
3015         } else
3016                 ret = -EINVAL;
3017
3018         if (ret >= 0)
3019                 netdev_state_change(tun->dev);
3020
3021 unlock:
3022         rtnl_unlock();
3023         return ret;
3024 }
3025
3026 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3027                         void __user *data)
3028 {
3029         struct bpf_prog *prog;
3030         int fd;
3031
3032         if (copy_from_user(&fd, data, sizeof(fd)))
3033                 return -EFAULT;
3034
3035         if (fd == -1) {
3036                 prog = NULL;
3037         } else {
3038                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3039                 if (IS_ERR(prog))
3040                         return PTR_ERR(prog);
3041         }
3042
3043         return __tun_set_ebpf(tun, prog_p, prog);
3044 }
3045
3046 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3047 static unsigned char tun_get_addr_len(unsigned short type)
3048 {
3049         switch (type) {
3050         case ARPHRD_IP6GRE:
3051         case ARPHRD_TUNNEL6:
3052                 return sizeof(struct in6_addr);
3053         case ARPHRD_IPGRE:
3054         case ARPHRD_TUNNEL:
3055         case ARPHRD_SIT:
3056                 return 4;
3057         case ARPHRD_ETHER:
3058                 return ETH_ALEN;
3059         case ARPHRD_IEEE802154:
3060         case ARPHRD_IEEE802154_MONITOR:
3061                 return IEEE802154_EXTENDED_ADDR_LEN;
3062         case ARPHRD_PHONET_PIPE:
3063         case ARPHRD_PPP:
3064         case ARPHRD_NONE:
3065                 return 0;
3066         case ARPHRD_6LOWPAN:
3067                 return EUI64_ADDR_LEN;
3068         case ARPHRD_FDDI:
3069                 return FDDI_K_ALEN;
3070         case ARPHRD_HIPPI:
3071                 return HIPPI_ALEN;
3072         case ARPHRD_IEEE802:
3073                 return FC_ALEN;
3074         case ARPHRD_ROSE:
3075                 return ROSE_ADDR_LEN;
3076         case ARPHRD_NETROM:
3077                 return AX25_ADDR_LEN;
3078         case ARPHRD_LOCALTLK:
3079                 return LTALK_ALEN;
3080         default:
3081                 return 0;
3082         }
3083 }
3084
3085 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3086                             unsigned long arg, int ifreq_len)
3087 {
3088         struct tun_file *tfile = file->private_data;
3089         struct net *net = sock_net(&tfile->sk);
3090         struct tun_struct *tun;
3091         void __user* argp = (void __user*)arg;
3092         unsigned int carrier;
3093         struct ifreq ifr;
3094         kuid_t owner;
3095         kgid_t group;
3096         int ifindex;
3097         int sndbuf;
3098         int vnet_hdr_sz;
3099         int le;
3100         int ret;
3101         bool do_notify = false;
3102
3103         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3104             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3105                 if (copy_from_user(&ifr, argp, ifreq_len))
3106                         return -EFAULT;
3107         } else {
3108                 memset(&ifr, 0, sizeof(ifr));
3109         }
3110         if (cmd == TUNGETFEATURES) {
3111                 /* Currently this just means: "what IFF flags are valid?".
3112                  * This is needed because we never checked for invalid flags on
3113                  * TUNSETIFF.
3114                  */
3115                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3116                                 TUN_FEATURES, (unsigned int __user*)argp);
3117         } else if (cmd == TUNSETQUEUE) {
3118                 return tun_set_queue(file, &ifr);
3119         } else if (cmd == SIOCGSKNS) {
3120                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3121                         return -EPERM;
3122                 return open_related_ns(&net->ns, get_net_ns);
3123         }
3124
3125         rtnl_lock();
3126
3127         tun = tun_get(tfile);
3128         if (cmd == TUNSETIFF) {
3129                 ret = -EEXIST;
3130                 if (tun)
3131                         goto unlock;
3132
3133                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3134
3135                 ret = tun_set_iff(net, file, &ifr);
3136
3137                 if (ret)
3138                         goto unlock;
3139
3140                 if (copy_to_user(argp, &ifr, ifreq_len))
3141                         ret = -EFAULT;
3142                 goto unlock;
3143         }
3144         if (cmd == TUNSETIFINDEX) {
3145                 ret = -EPERM;
3146                 if (tun)
3147                         goto unlock;
3148
3149                 ret = -EFAULT;
3150                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3151                         goto unlock;
3152                 ret = -EINVAL;
3153                 if (ifindex < 0)
3154                         goto unlock;
3155                 ret = 0;
3156                 tfile->ifindex = ifindex;
3157                 goto unlock;
3158         }
3159
3160         ret = -EBADFD;
3161         if (!tun)
3162                 goto unlock;
3163
3164         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3165
3166         net = dev_net(tun->dev);
3167         ret = 0;
3168         switch (cmd) {
3169         case TUNGETIFF:
3170                 tun_get_iff(tun, &ifr);
3171
3172                 if (tfile->detached)
3173                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3174                 if (!tfile->socket.sk->sk_filter)
3175                         ifr.ifr_flags |= IFF_NOFILTER;
3176
3177                 if (copy_to_user(argp, &ifr, ifreq_len))
3178                         ret = -EFAULT;
3179                 break;
3180
3181         case TUNSETNOCSUM:
3182                 /* Disable/Enable checksum */
3183
3184                 /* [unimplemented] */
3185                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3186                            arg ? "disabled" : "enabled");
3187                 break;
3188
3189         case TUNSETPERSIST:
3190                 /* Disable/Enable persist mode. Keep an extra reference to the
3191                  * module to prevent the module being unprobed.
3192                  */
3193                 if (arg && !(tun->flags & IFF_PERSIST)) {
3194                         tun->flags |= IFF_PERSIST;
3195                         __module_get(THIS_MODULE);
3196                         do_notify = true;
3197                 }
3198                 if (!arg && (tun->flags & IFF_PERSIST)) {
3199                         tun->flags &= ~IFF_PERSIST;
3200                         module_put(THIS_MODULE);
3201                         do_notify = true;
3202                 }
3203
3204                 netif_info(tun, drv, tun->dev, "persist %s\n",
3205                            arg ? "enabled" : "disabled");
3206                 break;
3207
3208         case TUNSETOWNER:
3209                 /* Set owner of the device */
3210                 owner = make_kuid(current_user_ns(), arg);
3211                 if (!uid_valid(owner)) {
3212                         ret = -EINVAL;
3213                         break;
3214                 }
3215                 tun->owner = owner;
3216                 do_notify = true;
3217                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3218                            from_kuid(&init_user_ns, tun->owner));
3219                 break;
3220
3221         case TUNSETGROUP:
3222                 /* Set group of the device */
3223                 group = make_kgid(current_user_ns(), arg);
3224                 if (!gid_valid(group)) {
3225                         ret = -EINVAL;
3226                         break;
3227                 }
3228                 tun->group = group;
3229                 do_notify = true;
3230                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3231                            from_kgid(&init_user_ns, tun->group));
3232                 break;
3233
3234         case TUNSETLINK:
3235                 /* Only allow setting the type when the interface is down */
3236                 if (tun->dev->flags & IFF_UP) {
3237                         netif_info(tun, drv, tun->dev,
3238                                    "Linktype set failed because interface is up\n");
3239                         ret = -EBUSY;
3240                 } else {
3241                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3242                                                        tun->dev);
3243                         ret = notifier_to_errno(ret);
3244                         if (ret) {
3245                                 netif_info(tun, drv, tun->dev,
3246                                            "Refused to change device type\n");
3247                                 break;
3248                         }
3249                         tun->dev->type = (int) arg;
3250                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3251                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3252                                    tun->dev->type);
3253                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3254                                                  tun->dev);
3255                 }
3256                 break;
3257
3258         case TUNSETDEBUG:
3259                 tun->msg_enable = (u32)arg;
3260                 break;
3261
3262         case TUNSETOFFLOAD:
3263                 ret = set_offload(tun, arg);
3264                 break;
3265
3266         case TUNSETTXFILTER:
3267                 /* Can be set only for TAPs */
3268                 ret = -EINVAL;
3269                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3270                         break;
3271                 ret = update_filter(&tun->txflt, (void __user *)arg);
3272                 break;
3273
3274         case SIOCGIFHWADDR:
3275                 /* Get hw address */
3276                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3277                 if (copy_to_user(argp, &ifr, ifreq_len))
3278                         ret = -EFAULT;
3279                 break;
3280
3281         case SIOCSIFHWADDR:
3282                 /* Set hw address */
3283                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3284                 break;
3285
3286         case TUNGETSNDBUF:
3287                 sndbuf = tfile->socket.sk->sk_sndbuf;
3288                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3289                         ret = -EFAULT;
3290                 break;
3291
3292         case TUNSETSNDBUF:
3293                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3294                         ret = -EFAULT;
3295                         break;
3296                 }
3297                 if (sndbuf <= 0) {
3298                         ret = -EINVAL;
3299                         break;
3300                 }
3301
3302                 tun->sndbuf = sndbuf;
3303                 tun_set_sndbuf(tun);
3304                 break;
3305
3306         case TUNGETVNETHDRSZ:
3307                 vnet_hdr_sz = tun->vnet_hdr_sz;
3308                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3309                         ret = -EFAULT;
3310                 break;
3311
3312         case TUNSETVNETHDRSZ:
3313                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3314                         ret = -EFAULT;
3315                         break;
3316                 }
3317                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3318                         ret = -EINVAL;
3319                         break;
3320                 }
3321
3322                 tun->vnet_hdr_sz = vnet_hdr_sz;
3323                 break;
3324
3325         case TUNGETVNETLE:
3326                 le = !!(tun->flags & TUN_VNET_LE);
3327                 if (put_user(le, (int __user *)argp))
3328                         ret = -EFAULT;
3329                 break;
3330
3331         case TUNSETVNETLE:
3332                 if (get_user(le, (int __user *)argp)) {
3333                         ret = -EFAULT;
3334                         break;
3335                 }
3336                 if (le)
3337                         tun->flags |= TUN_VNET_LE;
3338                 else
3339                         tun->flags &= ~TUN_VNET_LE;
3340                 break;
3341
3342         case TUNGETVNETBE:
3343                 ret = tun_get_vnet_be(tun, argp);
3344                 break;
3345
3346         case TUNSETVNETBE:
3347                 ret = tun_set_vnet_be(tun, argp);
3348                 break;
3349
3350         case TUNATTACHFILTER:
3351                 /* Can be set only for TAPs */
3352                 ret = -EINVAL;
3353                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3354                         break;
3355                 ret = -EFAULT;
3356                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3357                         break;
3358
3359                 ret = tun_attach_filter(tun);
3360                 break;
3361
3362         case TUNDETACHFILTER:
3363                 /* Can be set only for TAPs */
3364                 ret = -EINVAL;
3365                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3366                         break;
3367                 ret = 0;
3368                 tun_detach_filter(tun, tun->numqueues);
3369                 break;
3370
3371         case TUNGETFILTER:
3372                 ret = -EINVAL;
3373                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3374                         break;
3375                 ret = -EFAULT;
3376                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3377                         break;
3378                 ret = 0;
3379                 break;
3380
3381         case TUNSETSTEERINGEBPF:
3382                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3383                 break;
3384
3385         case TUNSETFILTEREBPF:
3386                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3387                 break;
3388
3389         case TUNSETCARRIER:
3390                 ret = -EFAULT;
3391                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3392                         goto unlock;
3393
3394                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3395                 break;
3396
3397         case TUNGETDEVNETNS:
3398                 ret = -EPERM;
3399                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3400                         goto unlock;
3401                 ret = open_related_ns(&net->ns, get_net_ns);
3402                 break;
3403
3404         default:
3405                 ret = -EINVAL;
3406                 break;
3407         }
3408
3409         if (do_notify)
3410                 netdev_state_change(tun->dev);
3411
3412 unlock:
3413         rtnl_unlock();
3414         if (tun)
3415                 tun_put(tun);
3416         return ret;
3417 }
3418
3419 static long tun_chr_ioctl(struct file *file,
3420                           unsigned int cmd, unsigned long arg)
3421 {
3422         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3423 }
3424
3425 #ifdef CONFIG_COMPAT
3426 static long tun_chr_compat_ioctl(struct file *file,
3427                          unsigned int cmd, unsigned long arg)
3428 {
3429         switch (cmd) {
3430         case TUNSETIFF:
3431         case TUNGETIFF:
3432         case TUNSETTXFILTER:
3433         case TUNGETSNDBUF:
3434         case TUNSETSNDBUF:
3435         case SIOCGIFHWADDR:
3436         case SIOCSIFHWADDR:
3437                 arg = (unsigned long)compat_ptr(arg);
3438                 break;
3439         default:
3440                 arg = (compat_ulong_t)arg;
3441                 break;
3442         }
3443
3444         /*
3445          * compat_ifreq is shorter than ifreq, so we must not access beyond
3446          * the end of that structure. All fields that are used in this
3447          * driver are compatible though, we don't need to convert the
3448          * contents.
3449          */
3450         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3451 }
3452 #endif /* CONFIG_COMPAT */
3453
3454 static int tun_chr_fasync(int fd, struct file *file, int on)
3455 {
3456         struct tun_file *tfile = file->private_data;
3457         int ret;
3458
3459         if (on) {
3460                 ret = file_f_owner_allocate(file);
3461                 if (ret)
3462                         goto out;
3463         }
3464
3465         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3466                 goto out;
3467
3468         if (on) {
3469                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3470                 tfile->flags |= TUN_FASYNC;
3471         } else
3472                 tfile->flags &= ~TUN_FASYNC;
3473         ret = 0;
3474 out:
3475         return ret;
3476 }
3477
3478 static int tun_chr_open(struct inode *inode, struct file * file)
3479 {
3480         struct net *net = current->nsproxy->net_ns;
3481         struct tun_file *tfile;
3482
3483         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3484                                             &tun_proto, 0);
3485         if (!tfile)
3486                 return -ENOMEM;
3487         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3488                 sk_free(&tfile->sk);
3489                 return -ENOMEM;
3490         }
3491
3492         mutex_init(&tfile->napi_mutex);
3493         RCU_INIT_POINTER(tfile->tun, NULL);
3494         tfile->flags = 0;
3495         tfile->ifindex = 0;
3496
3497         init_waitqueue_head(&tfile->socket.wq.wait);
3498
3499         tfile->socket.file = file;
3500         tfile->socket.ops = &tun_socket_ops;
3501
3502         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3503
3504         tfile->sk.sk_write_space = tun_sock_write_space;
3505         tfile->sk.sk_sndbuf = INT_MAX;
3506
3507         file->private_data = tfile;
3508         INIT_LIST_HEAD(&tfile->next);
3509
3510         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3511
3512         /* tun groks IOCB_NOWAIT just fine, mark it as such */
3513         file->f_mode |= FMODE_NOWAIT;
3514         return 0;
3515 }
3516
3517 static int tun_chr_close(struct inode *inode, struct file *file)
3518 {
3519         struct tun_file *tfile = file->private_data;
3520
3521         tun_detach(tfile, true);
3522
3523         return 0;
3524 }
3525
3526 #ifdef CONFIG_PROC_FS
3527 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3528 {
3529         struct tun_file *tfile = file->private_data;
3530         struct tun_struct *tun;
3531         struct ifreq ifr;
3532
3533         memset(&ifr, 0, sizeof(ifr));
3534
3535         rtnl_lock();
3536         tun = tun_get(tfile);
3537         if (tun)
3538                 tun_get_iff(tun, &ifr);
3539         rtnl_unlock();
3540
3541         if (tun)
3542                 tun_put(tun);
3543
3544         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3545 }
3546 #endif
3547
3548 static const struct file_operations tun_fops = {
3549         .owner  = THIS_MODULE,
3550         .read_iter  = tun_chr_read_iter,
3551         .write_iter = tun_chr_write_iter,
3552         .poll   = tun_chr_poll,
3553         .unlocked_ioctl = tun_chr_ioctl,
3554 #ifdef CONFIG_COMPAT
3555         .compat_ioctl = tun_chr_compat_ioctl,
3556 #endif
3557         .open   = tun_chr_open,
3558         .release = tun_chr_close,
3559         .fasync = tun_chr_fasync,
3560 #ifdef CONFIG_PROC_FS
3561         .show_fdinfo = tun_chr_show_fdinfo,
3562 #endif
3563 };
3564
3565 static struct miscdevice tun_miscdev = {
3566         .minor = TUN_MINOR,
3567         .name = "tun",
3568         .nodename = "net/tun",
3569         .fops = &tun_fops,
3570 };
3571
3572 /* ethtool interface */
3573
3574 static void tun_default_link_ksettings(struct net_device *dev,
3575                                        struct ethtool_link_ksettings *cmd)
3576 {
3577         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3578         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3579         cmd->base.speed         = SPEED_10000;
3580         cmd->base.duplex        = DUPLEX_FULL;
3581         cmd->base.port          = PORT_TP;
3582         cmd->base.phy_address   = 0;
3583         cmd->base.autoneg       = AUTONEG_DISABLE;
3584 }
3585
3586 static int tun_get_link_ksettings(struct net_device *dev,
3587                                   struct ethtool_link_ksettings *cmd)
3588 {
3589         struct tun_struct *tun = netdev_priv(dev);
3590
3591         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3592         return 0;
3593 }
3594
3595 static int tun_set_link_ksettings(struct net_device *dev,
3596                                   const struct ethtool_link_ksettings *cmd)
3597 {
3598         struct tun_struct *tun = netdev_priv(dev);
3599
3600         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3601         return 0;
3602 }
3603
3604 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3605 {
3606         struct tun_struct *tun = netdev_priv(dev);
3607
3608         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3609         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3610
3611         switch (tun->flags & TUN_TYPE_MASK) {
3612         case IFF_TUN:
3613                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3614                 break;
3615         case IFF_TAP:
3616                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3617                 break;
3618         }
3619 }
3620
3621 static u32 tun_get_msglevel(struct net_device *dev)
3622 {
3623         struct tun_struct *tun = netdev_priv(dev);
3624
3625         return tun->msg_enable;
3626 }
3627
3628 static void tun_set_msglevel(struct net_device *dev, u32 value)
3629 {
3630         struct tun_struct *tun = netdev_priv(dev);
3631
3632         tun->msg_enable = value;
3633 }
3634
3635 static int tun_get_coalesce(struct net_device *dev,
3636                             struct ethtool_coalesce *ec,
3637                             struct kernel_ethtool_coalesce *kernel_coal,
3638                             struct netlink_ext_ack *extack)
3639 {
3640         struct tun_struct *tun = netdev_priv(dev);
3641
3642         ec->rx_max_coalesced_frames = tun->rx_batched;
3643
3644         return 0;
3645 }
3646
3647 static int tun_set_coalesce(struct net_device *dev,
3648                             struct ethtool_coalesce *ec,
3649                             struct kernel_ethtool_coalesce *kernel_coal,
3650                             struct netlink_ext_ack *extack)
3651 {
3652         struct tun_struct *tun = netdev_priv(dev);
3653
3654         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3655                 tun->rx_batched = NAPI_POLL_WEIGHT;
3656         else
3657                 tun->rx_batched = ec->rx_max_coalesced_frames;
3658
3659         return 0;
3660 }
3661
3662 static void tun_get_channels(struct net_device *dev,
3663                              struct ethtool_channels *channels)
3664 {
3665         struct tun_struct *tun = netdev_priv(dev);
3666
3667         channels->combined_count = tun->numqueues;
3668         channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3669 }
3670
3671 static const struct ethtool_ops tun_ethtool_ops = {
3672         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3673         .get_drvinfo    = tun_get_drvinfo,
3674         .get_msglevel   = tun_get_msglevel,
3675         .set_msglevel   = tun_set_msglevel,
3676         .get_link       = ethtool_op_get_link,
3677         .get_channels   = tun_get_channels,
3678         .get_ts_info    = ethtool_op_get_ts_info,
3679         .get_coalesce   = tun_get_coalesce,
3680         .set_coalesce   = tun_set_coalesce,
3681         .get_link_ksettings = tun_get_link_ksettings,
3682         .set_link_ksettings = tun_set_link_ksettings,
3683 };
3684
3685 static int tun_queue_resize(struct tun_struct *tun)
3686 {
3687         struct net_device *dev = tun->dev;
3688         struct tun_file *tfile;
3689         struct ptr_ring **rings;
3690         int n = tun->numqueues + tun->numdisabled;
3691         int ret, i;
3692
3693         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3694         if (!rings)
3695                 return -ENOMEM;
3696
3697         for (i = 0; i < tun->numqueues; i++) {
3698                 tfile = rtnl_dereference(tun->tfiles[i]);
3699                 rings[i] = &tfile->tx_ring;
3700         }
3701         list_for_each_entry(tfile, &tun->disabled, next)
3702                 rings[i++] = &tfile->tx_ring;
3703
3704         ret = ptr_ring_resize_multiple_bh(rings, n,
3705                                           dev->tx_queue_len, GFP_KERNEL,
3706                                           tun_ptr_free);
3707
3708         kfree(rings);
3709         return ret;
3710 }
3711
3712 static int tun_device_event(struct notifier_block *unused,
3713                             unsigned long event, void *ptr)
3714 {
3715         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3716         struct tun_struct *tun = netdev_priv(dev);
3717         int i;
3718
3719         if (dev->rtnl_link_ops != &tun_link_ops)
3720                 return NOTIFY_DONE;
3721
3722         switch (event) {
3723         case NETDEV_CHANGE_TX_QUEUE_LEN:
3724                 if (tun_queue_resize(tun))
3725                         return NOTIFY_BAD;
3726                 break;
3727         case NETDEV_UP:
3728                 for (i = 0; i < tun->numqueues; i++) {
3729                         struct tun_file *tfile;
3730
3731                         tfile = rtnl_dereference(tun->tfiles[i]);
3732                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3733                 }
3734                 break;
3735         default:
3736                 break;
3737         }
3738
3739         return NOTIFY_DONE;
3740 }
3741
3742 static struct notifier_block tun_notifier_block __read_mostly = {
3743         .notifier_call  = tun_device_event,
3744 };
3745
3746 static int __init tun_init(void)
3747 {
3748         int ret = 0;
3749
3750         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3751
3752         ret = rtnl_link_register(&tun_link_ops);
3753         if (ret) {
3754                 pr_err("Can't register link_ops\n");
3755                 goto err_linkops;
3756         }
3757
3758         ret = misc_register(&tun_miscdev);
3759         if (ret) {
3760                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3761                 goto err_misc;
3762         }
3763
3764         ret = register_netdevice_notifier(&tun_notifier_block);
3765         if (ret) {
3766                 pr_err("Can't register netdevice notifier\n");
3767                 goto err_notifier;
3768         }
3769
3770         return  0;
3771
3772 err_notifier:
3773         misc_deregister(&tun_miscdev);
3774 err_misc:
3775         rtnl_link_unregister(&tun_link_ops);
3776 err_linkops:
3777         return ret;
3778 }
3779
3780 static void __exit tun_cleanup(void)
3781 {
3782         misc_deregister(&tun_miscdev);
3783         rtnl_link_unregister(&tun_link_ops);
3784         unregister_netdevice_notifier(&tun_notifier_block);
3785 }
3786
3787 /* Get an underlying socket object from tun file.  Returns error unless file is
3788  * attached to a device.  The returned object works like a packet socket, it
3789  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3790  * holding a reference to the file for as long as the socket is in use. */
3791 struct socket *tun_get_socket(struct file *file)
3792 {
3793         struct tun_file *tfile;
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->socket;
3800 }
3801 EXPORT_SYMBOL_GPL(tun_get_socket);
3802
3803 struct ptr_ring *tun_get_tx_ring(struct file *file)
3804 {
3805         struct tun_file *tfile;
3806
3807         if (file->f_op != &tun_fops)
3808                 return ERR_PTR(-EINVAL);
3809         tfile = file->private_data;
3810         if (!tfile)
3811                 return ERR_PTR(-EBADFD);
3812         return &tfile->tx_ring;
3813 }
3814 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3815
3816 module_init(tun_init);
3817 module_exit(tun_cleanup);
3818 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3819 MODULE_AUTHOR(DRV_COPYRIGHT);
3820 MODULE_LICENSE("GPL");
3821 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3822 MODULE_ALIAS("devname:net/tun");