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_not_capable(struct tun_struct *tun)
 578 {
 579         const struct cred *cred = current_cred();
 580         struct net *net = dev_net(tun->dev);
 581
 582         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 583                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 584                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 585 }
 586
 587 static void tun_set_real_num_queues(struct tun_struct *tun)
 588 {
 589         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 590         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 591 }
 592
 593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 594 {
 595         tfile->detached = tun;
 596         list_add_tail(&tfile->next, &tun->disabled);
 597         ++tun->numdisabled;
 598 }
 599
 600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 601 {
 602         struct tun_struct *tun = tfile->detached;
 603
 604         tfile->detached = NULL;
 605         list_del_init(&tfile->next);
 606         --tun->numdisabled;
 607         return tun;
 608 }
 609
 610 void tun_ptr_free(void *ptr)
 611 {
 612         if (!ptr)
 613                 return;
 614         if (tun_is_xdp_frame(ptr)) {
 615                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 616
 617                 xdp_return_frame(xdpf);
 618         } else {
 619                 __skb_array_destroy_skb(ptr);
 620         }
 621 }
 622 EXPORT_SYMBOL_GPL(tun_ptr_free);
 623
 624 static void tun_queue_purge(struct tun_file *tfile)
 625 {
 626         void *ptr;
 627
 628         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 629                 tun_ptr_free(ptr);
 630
 631         skb_queue_purge(&tfile->sk.sk_write_queue);
 632         skb_queue_purge(&tfile->sk.sk_error_queue);
 633 }
 634
 635 static void __tun_detach(struct tun_file *tfile, bool clean)
 636 {
 637         struct tun_file *ntfile;
 638         struct tun_struct *tun;
 639
 640         tun = rtnl_dereference(tfile->tun);
 641
 642         if (tun && clean) {
 643                 if (!tfile->detached)
 644                         tun_napi_disable(tfile);
 645                 tun_napi_del(tfile);
 646         }
 647
 648         if (tun && !tfile->detached) {
 649                 u16 index = tfile->queue_index;
 650                 BUG_ON(index >= tun->numqueues);
 651
 652                 rcu_assign_pointer(tun->tfiles[index],
 653                                    tun->tfiles[tun->numqueues - 1]);
 654                 ntfile = rtnl_dereference(tun->tfiles[index]);
 655                 ntfile->queue_index = index;
 656                 ntfile->xdp_rxq.queue_index = index;
 657                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 658                                    NULL);
 659
 660                 --tun->numqueues;
 661                 if (clean) {
 662                         RCU_INIT_POINTER(tfile->tun, NULL);
 663                         sock_put(&tfile->sk);
 664                 } else {
 665                         tun_disable_queue(tun, tfile);
 666                         tun_napi_disable(tfile);
 667                 }
 668
 669                 synchronize_net();
 670                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 671                 /* Drop read queue */
 672                 tun_queue_purge(tfile);
 673                 tun_set_real_num_queues(tun);
 674         } else if (tfile->detached && clean) {
 675                 tun = tun_enable_queue(tfile);
 676                 sock_put(&tfile->sk);
 677         }
 678
 679         if (clean) {
 680                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 681                         netif_carrier_off(tun->dev);
 682
 683                         if (!(tun->flags & IFF_PERSIST) &&
 684                             tun->dev->reg_state == NETREG_REGISTERED)
 685                                 unregister_netdevice(tun->dev);
 686                 }
 687                 if (tun)
 688                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 690         }
 691 }
 692
 693 static void tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_struct *tun;
 696         struct net_device *dev;
 697
 698         rtnl_lock();
 699         tun = rtnl_dereference(tfile->tun);
 700         dev = tun ? tun->dev : NULL;
 701         __tun_detach(tfile, clean);
 702         if (dev)
 703                 netdev_state_change(dev);
 704         rtnl_unlock();
 705
 706         if (clean)
 707                 sock_put(&tfile->sk);
 708 }
 709
 710 static void tun_detach_all(struct net_device *dev)
 711 {
 712         struct tun_struct *tun = netdev_priv(dev);
 713         struct tun_file *tfile, *tmp;
 714         int i, n = tun->numqueues;
 715
 716         for (i = 0; i < n; i++) {
 717                 tfile = rtnl_dereference(tun->tfiles[i]);
 718                 BUG_ON(!tfile);
 719                 tun_napi_disable(tfile);
 720                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 721                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 722                 RCU_INIT_POINTER(tfile->tun, NULL);
 723                 --tun->numqueues;
 724         }
 725         list_for_each_entry(tfile, &tun->disabled, next) {
 726                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 727                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 728                 RCU_INIT_POINTER(tfile->tun, NULL);
 729         }
 730         BUG_ON(tun->numqueues != 0);
 731
 732         synchronize_net();
 733         for (i = 0; i < n; i++) {
 734                 tfile = rtnl_dereference(tun->tfiles[i]);
 735                 tun_napi_del(tfile);
 736                 /* Drop read queue */
 737                 tun_queue_purge(tfile);
 738                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 739                 sock_put(&tfile->sk);
 740         }
 741         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 742                 tun_napi_del(tfile);
 743                 tun_enable_queue(tfile);
 744                 tun_queue_purge(tfile);
 745                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 746                 sock_put(&tfile->sk);
 747         }
 748         BUG_ON(tun->numdisabled != 0);
 749
 750         if (tun->flags & IFF_PERSIST)
 751                 module_put(THIS_MODULE);
 752 }
 753
 754 static int tun_attach(struct tun_struct *tun, struct file *file,
 755                       bool skip_filter, bool napi, bool napi_frags,
 756                       bool publish_tun)
 757 {
 758         struct tun_file *tfile = file->private_data;
 759         struct net_device *dev = tun->dev;
 760         int err;
 761
 762         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 763         if (err < 0)
 764                 goto out;
 765
 766         err = -EINVAL;
 767         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 768                 goto out;
 769
 770         err = -EBUSY;
 771         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 772                 goto out;
 773
 774         err = -E2BIG;
 775         if (!tfile->detached &&
 776             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 777                 goto out;
 778
 779         err = 0;
 780
 781         /* Re-attach the filter to persist device */
 782         if (!skip_filter && (tun->filter_attached == true)) {
 783                 lock_sock(tfile->socket.sk);
 784                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 785                 release_sock(tfile->socket.sk);
 786                 if (!err)
 787                         goto out;
 788         }
 789
 790         if (!tfile->detached &&
 791             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 792                             GFP_KERNEL, tun_ptr_free)) {
 793                 err = -ENOMEM;
 794                 goto out;
 795         }
 796
 797         tfile->queue_index = tun->numqueues;
 798         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 799
 800         if (tfile->detached) {
 801                 /* Re-attach detached tfile, updating XDP queue_index */
 802                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 803
 804                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 805                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 806         } else {
 807                 /* Setup XDP RX-queue info, for new tfile getting attached */
 808                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 809                                        tun->dev, tfile->queue_index, 0);
 810                 if (err < 0)
 811                         goto out;
 812                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 813                                                  MEM_TYPE_PAGE_SHARED, NULL);
 814                 if (err < 0) {
 815                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 816                         goto out;
 817                 }
 818                 err = 0;
 819         }
 820
 821         if (tfile->detached) {
 822                 tun_enable_queue(tfile);
 823                 tun_napi_enable(tfile);
 824         } else {
 825                 sock_hold(&tfile->sk);
 826                 tun_napi_init(tun, tfile, napi, napi_frags);
 827         }
 828
 829         if (rtnl_dereference(tun->xdp_prog))
 830                 sock_set_flag(&tfile->sk, SOCK_XDP);
 831
 832         /* device is allowed to go away first, so no need to hold extra
 833          * refcnt.
 834          */
 835
 836         /* Publish tfile->tun and tun->tfiles only after we've fully
 837          * initialized tfile; otherwise we risk using half-initialized
 838          * object.
 839          */
 840         if (publish_tun)
 841                 rcu_assign_pointer(tfile->tun, tun);
 842         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 843         tun->numqueues++;
 844         tun_set_real_num_queues(tun);
 845 out:
 846         return err;
 847 }
 848
 849 static struct tun_struct *tun_get(struct tun_file *tfile)
 850 {
 851         struct tun_struct *tun;
 852
 853         rcu_read_lock();
 854         tun = rcu_dereference(tfile->tun);
 855         if (tun)
 856                 dev_hold(tun->dev);
 857         rcu_read_unlock();
 858
 859         return tun;
 860 }
 861
 862 static void tun_put(struct tun_struct *tun)
 863 {
 864         dev_put(tun->dev);
 865 }
 866
 867 /* TAP filtering */
 868 static void addr_hash_set(u32 *mask, const u8 *addr)
 869 {
 870         int n = ether_crc(ETH_ALEN, addr) >> 26;
 871         mask[n >> 5] |= (1 << (n & 31));
 872 }
 873
 874 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 875 {
 876         int n = ether_crc(ETH_ALEN, addr) >> 26;
 877         return mask[n >> 5] & (1 << (n & 31));
 878 }
 879
 880 static int update_filter(struct tap_filter *filter, void __user *arg)
 881 {
 882         struct { u8 u[ETH_ALEN]; } *addr;
 883         struct tun_filter uf;
 884         int err, alen, n, nexact;
 885
 886         if (copy_from_user(&uf, arg, sizeof(uf)))
 887                 return -EFAULT;
 888
 889         if (!uf.count) {
 890                 /* Disabled */
 891                 filter->count = 0;
 892                 return 0;
 893         }
 894
 895         alen = ETH_ALEN * uf.count;
 896         addr = memdup_user(arg + sizeof(uf), alen);
 897         if (IS_ERR(addr))
 898                 return PTR_ERR(addr);
 899
 900         /* The filter is updated without holding any locks. Which is
 901          * perfectly safe. We disable it first and in the worst
 902          * case we'll accept a few undesired packets. */
 903         filter->count = 0;
 904         wmb();
 905
 906         /* Use first set of addresses as an exact filter */
 907         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 908                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 909
 910         nexact = n;
 911
 912         /* Remaining multicast addresses are hashed,
 913          * unicast will leave the filter disabled. */
 914         memset(filter->mask, 0, sizeof(filter->mask));
 915         for (; n < uf.count; n++) {
 916                 if (!is_multicast_ether_addr(addr[n].u)) {
 917                         err = 0; /* no filter */
 918                         goto free_addr;
 919                 }
 920                 addr_hash_set(filter->mask, addr[n].u);
 921         }
 922
 923         /* For ALLMULTI just set the mask to all ones.
 924          * This overrides the mask populated above. */
 925         if ((uf.flags & TUN_FLT_ALLMULTI))
 926                 memset(filter->mask, ~0, sizeof(filter->mask));
 927
 928         /* Now enable the filter */
 929         wmb();
 930         filter->count = nexact;
 931
 932         /* Return the number of exact filters */
 933         err = nexact;
 934 free_addr:
 935         kfree(addr);
 936         return err;
 937 }
 938
 939 /* Returns: 0 - drop, !=0 - accept */
 940 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 941 {
 942         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 943          * at this point. */
 944         struct ethhdr *eh = (struct ethhdr *) skb->data;
 945         int i;
 946
 947         /* Exact match */
 948         for (i = 0; i < filter->count; i++)
 949                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 950                         return 1;
 951
 952         /* Inexact match (multicast only) */
 953         if (is_multicast_ether_addr(eh->h_dest))
 954                 return addr_hash_test(filter->mask, eh->h_dest);
 955
 956         return 0;
 957 }
 958
 959 /*
 960  * Checks whether the packet is accepted or not.
 961  * Returns: 0 - drop, !=0 - accept
 962  */
 963 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 964 {
 965         if (!filter->count)
 966                 return 1;
 967
 968         return run_filter(filter, skb);
 969 }
 970
 971 /* Network device part of the driver */
 972
 973 static const struct ethtool_ops tun_ethtool_ops;
 974
 975 static int tun_net_init(struct net_device *dev)
 976 {
 977         struct tun_struct *tun = netdev_priv(dev);
 978         struct ifreq *ifr = tun->ifr;
 979         int err;
 980
 981         spin_lock_init(&tun->lock);
 982
 983         err = security_tun_dev_alloc_security(&tun->security);
 984         if (err < 0)
 985                 return err;
 986
 987         tun_flow_init(tun);
 988
 989         dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
 990         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 991                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 992                            NETIF_F_HW_VLAN_STAG_TX;
 993         dev->features = dev->hw_features;
 994         dev->vlan_features = dev->features &
 995                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 996                                NETIF_F_HW_VLAN_STAG_TX);
 997         dev->lltx = true;
 998
 999         tun->flags = (tun->flags & ~TUN_FEATURES) |
1000                       (ifr->ifr_flags & TUN_FEATURES);
1001
1002         INIT_LIST_HEAD(&tun->disabled);
1003         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1004                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1005         if (err < 0) {
1006                 tun_flow_uninit(tun);
1007                 security_tun_dev_free_security(tun->security);
1008                 return err;
1009         }
1010         return 0;
1011 }
1012
1013 /* Net device detach from fd. */
1014 static void tun_net_uninit(struct net_device *dev)
1015 {
1016         tun_detach_all(dev);
1017 }
1018
1019 /* Net device open. */
1020 static int tun_net_open(struct net_device *dev)
1021 {
1022         netif_tx_start_all_queues(dev);
1023
1024         return 0;
1025 }
1026
1027 /* Net device close. */
1028 static int tun_net_close(struct net_device *dev)
1029 {
1030         netif_tx_stop_all_queues(dev);
1031         return 0;
1032 }
1033
1034 /* Net device start xmit */
1035 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1036 {
1037 #ifdef CONFIG_RPS
1038         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1039                 /* Select queue was not called for the skbuff, so we extract the
1040                  * RPS hash and save it into the flow_table here.
1041                  */
1042                 struct tun_flow_entry *e;
1043                 __u32 rxhash;
1044
1045                 rxhash = __skb_get_hash_symmetric(skb);
1046                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1047                 if (e)
1048                         tun_flow_save_rps_rxhash(e, rxhash);
1049         }
1050 #endif
1051 }
1052
1053 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1054                                     struct sk_buff *skb,
1055                                     int len)
1056 {
1057         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1058
1059         if (prog)
1060                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1061
1062         return len;
1063 }
1064
1065 /* Net device start xmit */
1066 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1067 {
1068         struct tun_struct *tun = netdev_priv(dev);
1069         enum skb_drop_reason drop_reason;
1070         int txq = skb->queue_mapping;
1071         struct netdev_queue *queue;
1072         struct tun_file *tfile;
1073         int len = skb->len;
1074
1075         rcu_read_lock();
1076         tfile = rcu_dereference(tun->tfiles[txq]);
1077
1078         /* Drop packet if interface is not attached */
1079         if (!tfile) {
1080                 drop_reason = SKB_DROP_REASON_DEV_READY;
1081                 goto drop;
1082         }
1083
1084         if (!rcu_dereference(tun->steering_prog))
1085                 tun_automq_xmit(tun, skb);
1086
1087         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1088
1089         /* Drop if the filter does not like it.
1090          * This is a noop if the filter is disabled.
1091          * Filter can be enabled only for the TAP devices. */
1092         if (!check_filter(&tun->txflt, skb)) {
1093                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1094                 goto drop;
1095         }
1096
1097         if (tfile->socket.sk->sk_filter &&
1098             sk_filter(tfile->socket.sk, skb)) {
1099                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1100                 goto drop;
1101         }
1102
1103         len = run_ebpf_filter(tun, skb, len);
1104         if (len == 0) {
1105                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1106                 goto drop;
1107         }
1108
1109         if (pskb_trim(skb, len)) {
1110                 drop_reason = SKB_DROP_REASON_NOMEM;
1111                 goto drop;
1112         }
1113
1114         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1115                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1116                 goto drop;
1117         }
1118
1119         skb_tx_timestamp(skb);
1120
1121         /* Orphan the skb - required as we might hang on to it
1122          * for indefinite time.
1123          */
1124         skb_orphan(skb);
1125
1126         nf_reset_ct(skb);
1127
1128         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1129                 drop_reason = SKB_DROP_REASON_FULL_RING;
1130                 goto drop;
1131         }
1132
1133         /* dev->lltx requires to do our own update of trans_start */
1134         queue = netdev_get_tx_queue(dev, txq);
1135         txq_trans_cond_update(queue);
1136
1137         /* Notify and wake up reader process */
1138         if (tfile->flags & TUN_FASYNC)
1139                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1140         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1141
1142         rcu_read_unlock();
1143         return NETDEV_TX_OK;
1144
1145 drop:
1146         dev_core_stats_tx_dropped_inc(dev);
1147         skb_tx_error(skb);
1148         kfree_skb_reason(skb, drop_reason);
1149         rcu_read_unlock();
1150         return NET_XMIT_DROP;
1151 }
1152
1153 static void tun_net_mclist(struct net_device *dev)
1154 {
1155         /*
1156          * This callback is supposed to deal with mc filter in
1157          * _rx_ path and has nothing to do with the _tx_ path.
1158          * In rx path we always accept everything userspace gives us.
1159          */
1160 }
1161
1162 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1163         netdev_features_t features)
1164 {
1165         struct tun_struct *tun = netdev_priv(dev);
1166
1167         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1168 }
1169
1170 static void tun_set_headroom(struct net_device *dev, int new_hr)
1171 {
1172         struct tun_struct *tun = netdev_priv(dev);
1173
1174         if (new_hr < NET_SKB_PAD)
1175                 new_hr = NET_SKB_PAD;
1176
1177         tun->align = new_hr;
1178 }
1179
1180 static void
1181 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1182 {
1183         struct tun_struct *tun = netdev_priv(dev);
1184
1185         dev_get_tstats64(dev, stats);
1186
1187         stats->rx_frame_errors +=
1188                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1189 }
1190
1191 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1192                        struct netlink_ext_ack *extack)
1193 {
1194         struct tun_struct *tun = netdev_priv(dev);
1195         struct tun_file *tfile;
1196         struct bpf_prog *old_prog;
1197         int i;
1198
1199         old_prog = rtnl_dereference(tun->xdp_prog);
1200         rcu_assign_pointer(tun->xdp_prog, prog);
1201         if (old_prog)
1202                 bpf_prog_put(old_prog);
1203
1204         for (i = 0; i < tun->numqueues; i++) {
1205                 tfile = rtnl_dereference(tun->tfiles[i]);
1206                 if (prog)
1207                         sock_set_flag(&tfile->sk, SOCK_XDP);
1208                 else
1209                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1210         }
1211         list_for_each_entry(tfile, &tun->disabled, next) {
1212                 if (prog)
1213                         sock_set_flag(&tfile->sk, SOCK_XDP);
1214                 else
1215                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1216         }
1217
1218         return 0;
1219 }
1220
1221 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1222 {
1223         switch (xdp->command) {
1224         case XDP_SETUP_PROG:
1225                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1226         default:
1227                 return -EINVAL;
1228         }
1229 }
1230
1231 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1232 {
1233         if (new_carrier) {
1234                 struct tun_struct *tun = netdev_priv(dev);
1235
1236                 if (!tun->numqueues)
1237                         return -EPERM;
1238
1239                 netif_carrier_on(dev);
1240         } else {
1241                 netif_carrier_off(dev);
1242         }
1243         return 0;
1244 }
1245
1246 static const struct net_device_ops tun_netdev_ops = {
1247         .ndo_init               = tun_net_init,
1248         .ndo_uninit             = tun_net_uninit,
1249         .ndo_open               = tun_net_open,
1250         .ndo_stop               = tun_net_close,
1251         .ndo_start_xmit         = tun_net_xmit,
1252         .ndo_fix_features       = tun_net_fix_features,
1253         .ndo_select_queue       = tun_select_queue,
1254         .ndo_set_rx_headroom    = tun_set_headroom,
1255         .ndo_get_stats64        = tun_net_get_stats64,
1256         .ndo_change_carrier     = tun_net_change_carrier,
1257 };
1258
1259 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1260 {
1261         /* Notify and wake up reader process */
1262         if (tfile->flags & TUN_FASYNC)
1263                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1264         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1265 }
1266
1267 static int tun_xdp_xmit(struct net_device *dev, int n,
1268                         struct xdp_frame **frames, u32 flags)
1269 {
1270         struct tun_struct *tun = netdev_priv(dev);
1271         struct tun_file *tfile;
1272         u32 numqueues;
1273         int nxmit = 0;
1274         int i;
1275
1276         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1277                 return -EINVAL;
1278
1279         rcu_read_lock();
1280
1281 resample:
1282         numqueues = READ_ONCE(tun->numqueues);
1283         if (!numqueues) {
1284                 rcu_read_unlock();
1285                 return -ENXIO; /* Caller will free/return all frames */
1286         }
1287
1288         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1289                                             numqueues]);
1290         if (unlikely(!tfile))
1291                 goto resample;
1292
1293         spin_lock(&tfile->tx_ring.producer_lock);
1294         for (i = 0; i < n; i++) {
1295                 struct xdp_frame *xdp = frames[i];
1296                 /* Encode the XDP flag into lowest bit for consumer to differ
1297                  * XDP buffer from sk_buff.
1298                  */
1299                 void *frame = tun_xdp_to_ptr(xdp);
1300
1301                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1302                         dev_core_stats_tx_dropped_inc(dev);
1303                         break;
1304                 }
1305                 nxmit++;
1306         }
1307         spin_unlock(&tfile->tx_ring.producer_lock);
1308
1309         if (flags & XDP_XMIT_FLUSH)
1310                 __tun_xdp_flush_tfile(tfile);
1311
1312         rcu_read_unlock();
1313         return nxmit;
1314 }
1315
1316 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1317 {
1318         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1319         int nxmit;
1320
1321         if (unlikely(!frame))
1322                 return -EOVERFLOW;
1323
1324         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1325         if (!nxmit)
1326                 xdp_return_frame_rx_napi(frame);
1327         return nxmit;
1328 }
1329
1330 static const struct net_device_ops tap_netdev_ops = {
1331         .ndo_init               = tun_net_init,
1332         .ndo_uninit             = tun_net_uninit,
1333         .ndo_open               = tun_net_open,
1334         .ndo_stop               = tun_net_close,
1335         .ndo_start_xmit         = tun_net_xmit,
1336         .ndo_fix_features       = tun_net_fix_features,
1337         .ndo_set_rx_mode        = tun_net_mclist,
1338         .ndo_set_mac_address    = eth_mac_addr,
1339         .ndo_validate_addr      = eth_validate_addr,
1340         .ndo_select_queue       = tun_select_queue,
1341         .ndo_features_check     = passthru_features_check,
1342         .ndo_set_rx_headroom    = tun_set_headroom,
1343         .ndo_bpf                = tun_xdp,
1344         .ndo_xdp_xmit           = tun_xdp_xmit,
1345         .ndo_change_carrier     = tun_net_change_carrier,
1346 };
1347
1348 static void tun_flow_init(struct tun_struct *tun)
1349 {
1350         int i;
1351
1352         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1353                 INIT_HLIST_HEAD(&tun->flows[i]);
1354
1355         tun->ageing_time = TUN_FLOW_EXPIRE;
1356         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1357         mod_timer(&tun->flow_gc_timer,
1358                   round_jiffies_up(jiffies + tun->ageing_time));
1359 }
1360
1361 static void tun_flow_uninit(struct tun_struct *tun)
1362 {
1363         del_timer_sync(&tun->flow_gc_timer);
1364         tun_flow_flush(tun);
1365 }
1366
1367 #define MIN_MTU 68
1368 #define MAX_MTU 65535
1369
1370 /* Initialize net device. */
1371 static void tun_net_initialize(struct net_device *dev)
1372 {
1373         struct tun_struct *tun = netdev_priv(dev);
1374
1375         switch (tun->flags & TUN_TYPE_MASK) {
1376         case IFF_TUN:
1377                 dev->netdev_ops = &tun_netdev_ops;
1378                 dev->header_ops = &ip_tunnel_header_ops;
1379
1380                 /* Point-to-Point TUN Device */
1381                 dev->hard_header_len = 0;
1382                 dev->addr_len = 0;
1383                 dev->mtu = 1500;
1384
1385                 /* Zero header length */
1386                 dev->type = ARPHRD_NONE;
1387                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1388                 break;
1389
1390         case IFF_TAP:
1391                 dev->netdev_ops = &tap_netdev_ops;
1392                 /* Ethernet TAP Device */
1393                 ether_setup(dev);
1394                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1395                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1396
1397                 eth_hw_addr_random(dev);
1398
1399                 /* Currently tun does not support XDP, only tap does. */
1400                 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1401                                     NETDEV_XDP_ACT_REDIRECT |
1402                                     NETDEV_XDP_ACT_NDO_XMIT;
1403
1404                 break;
1405         }
1406
1407         dev->min_mtu = MIN_MTU;
1408         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1409 }
1410
1411 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1412 {
1413         struct sock *sk = tfile->socket.sk;
1414
1415         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1416 }
1417
1418 /* Character device part */
1419
1420 /* Poll */
1421 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1422 {
1423         struct tun_file *tfile = file->private_data;
1424         struct tun_struct *tun = tun_get(tfile);
1425         struct sock *sk;
1426         __poll_t mask = 0;
1427
1428         if (!tun)
1429                 return EPOLLERR;
1430
1431         sk = tfile->socket.sk;
1432
1433         poll_wait(file, sk_sleep(sk), wait);
1434
1435         if (!ptr_ring_empty(&tfile->tx_ring))
1436                 mask |= EPOLLIN | EPOLLRDNORM;
1437
1438         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1439          * guarantee EPOLLOUT to be raised by either here or
1440          * tun_sock_write_space(). Then process could get notification
1441          * after it writes to a down device and meets -EIO.
1442          */
1443         if (tun_sock_writeable(tun, tfile) ||
1444             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1445              tun_sock_writeable(tun, tfile)))
1446                 mask |= EPOLLOUT | EPOLLWRNORM;
1447
1448         if (tun->dev->reg_state != NETREG_REGISTERED)
1449                 mask = EPOLLERR;
1450
1451         tun_put(tun);
1452         return mask;
1453 }
1454
1455 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1456                                             size_t len,
1457                                             const struct iov_iter *it)
1458 {
1459         struct sk_buff *skb;
1460         size_t linear;
1461         int err;
1462         int i;
1463
1464         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1465             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1466                 return ERR_PTR(-EMSGSIZE);
1467
1468         local_bh_disable();
1469         skb = napi_get_frags(&tfile->napi);
1470         local_bh_enable();
1471         if (!skb)
1472                 return ERR_PTR(-ENOMEM);
1473
1474         linear = iov_iter_single_seg_count(it);
1475         err = __skb_grow(skb, linear);
1476         if (err)
1477                 goto free;
1478
1479         skb->len = len;
1480         skb->data_len = len - linear;
1481         skb->truesize += skb->data_len;
1482
1483         for (i = 1; i < it->nr_segs; i++) {
1484                 const struct iovec *iov = iter_iov(it) + i;
1485                 size_t fragsz = iov->iov_len;
1486                 struct page *page;
1487                 void *frag;
1488
1489                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1490                         err = -EINVAL;
1491                         goto free;
1492                 }
1493                 frag = netdev_alloc_frag(fragsz);
1494                 if (!frag) {
1495                         err = -ENOMEM;
1496                         goto free;
1497                 }
1498                 page = virt_to_head_page(frag);
1499                 skb_fill_page_desc(skb, i - 1, page,
1500                                    frag - page_address(page), fragsz);
1501         }
1502
1503         return skb;
1504 free:
1505         /* frees skb and all frags allocated with napi_alloc_frag() */
1506         napi_free_frags(&tfile->napi);
1507         return ERR_PTR(err);
1508 }
1509
1510 /* prepad is the amount to reserve at front.  len is length after that.
1511  * linear is a hint as to how much to copy (usually headers). */
1512 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1513                                      size_t prepad, size_t len,
1514                                      size_t linear, int noblock)
1515 {
1516         struct sock *sk = tfile->socket.sk;
1517         struct sk_buff *skb;
1518         int err;
1519
1520         /* Under a page?  Don't bother with paged skb. */
1521         if (prepad + len < PAGE_SIZE)
1522                 linear = len;
1523
1524         if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1525                 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1526         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1527                                    &err, PAGE_ALLOC_COSTLY_ORDER);
1528         if (!skb)
1529                 return ERR_PTR(err);
1530
1531         skb_reserve(skb, prepad);
1532         skb_put(skb, linear);
1533         skb->data_len = len - linear;
1534         skb->len += len - linear;
1535
1536         return skb;
1537 }
1538
1539 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1540                            struct sk_buff *skb, int more)
1541 {
1542         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1543         struct sk_buff_head process_queue;
1544         u32 rx_batched = tun->rx_batched;
1545         bool rcv = false;
1546
1547         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1548                 local_bh_disable();
1549                 skb_record_rx_queue(skb, tfile->queue_index);
1550                 netif_receive_skb(skb);
1551                 local_bh_enable();
1552                 return;
1553         }
1554
1555         spin_lock(&queue->lock);
1556         if (!more || skb_queue_len(queue) == rx_batched) {
1557                 __skb_queue_head_init(&process_queue);
1558                 skb_queue_splice_tail_init(queue, &process_queue);
1559                 rcv = true;
1560         } else {
1561                 __skb_queue_tail(queue, skb);
1562         }
1563         spin_unlock(&queue->lock);
1564
1565         if (rcv) {
1566                 struct sk_buff *nskb;
1567
1568                 local_bh_disable();
1569                 while ((nskb = __skb_dequeue(&process_queue))) {
1570                         skb_record_rx_queue(nskb, tfile->queue_index);
1571                         netif_receive_skb(nskb);
1572                 }
1573                 skb_record_rx_queue(skb, tfile->queue_index);
1574                 netif_receive_skb(skb);
1575                 local_bh_enable();
1576         }
1577 }
1578
1579 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1580                               int len, int noblock, bool zerocopy)
1581 {
1582         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1583                 return false;
1584
1585         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1586                 return false;
1587
1588         if (!noblock)
1589                 return false;
1590
1591         if (zerocopy)
1592                 return false;
1593
1594         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1595             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1596                 return false;
1597
1598         return true;
1599 }
1600
1601 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1602                                        struct page_frag *alloc_frag, char *buf,
1603                                        int buflen, int len, int pad)
1604 {
1605         struct sk_buff *skb = build_skb(buf, buflen);
1606
1607         if (!skb)
1608                 return ERR_PTR(-ENOMEM);
1609
1610         skb_reserve(skb, pad);
1611         skb_put(skb, len);
1612         skb_set_owner_w(skb, tfile->socket.sk);
1613
1614         get_page(alloc_frag->page);
1615         alloc_frag->offset += buflen;
1616
1617         return skb;
1618 }
1619
1620 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1621                        struct xdp_buff *xdp, u32 act)
1622 {
1623         int err;
1624
1625         switch (act) {
1626         case XDP_REDIRECT:
1627                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1628                 if (err) {
1629                         dev_core_stats_rx_dropped_inc(tun->dev);
1630                         return err;
1631                 }
1632                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1633                 break;
1634         case XDP_TX:
1635                 err = tun_xdp_tx(tun->dev, xdp);
1636                 if (err < 0) {
1637                         dev_core_stats_rx_dropped_inc(tun->dev);
1638                         return err;
1639                 }
1640                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1641                 break;
1642         case XDP_PASS:
1643                 break;
1644         default:
1645                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1646                 fallthrough;
1647         case XDP_ABORTED:
1648                 trace_xdp_exception(tun->dev, xdp_prog, act);
1649                 fallthrough;
1650         case XDP_DROP:
1651                 dev_core_stats_rx_dropped_inc(tun->dev);
1652                 break;
1653         }
1654
1655         return act;
1656 }
1657
1658 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1659                                      struct tun_file *tfile,
1660                                      struct iov_iter *from,
1661                                      struct virtio_net_hdr *hdr,
1662                                      int len, int *skb_xdp)
1663 {
1664         struct page_frag *alloc_frag = &current->task_frag;
1665         struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
1666         struct bpf_prog *xdp_prog;
1667         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1668         char *buf;
1669         size_t copied;
1670         int pad = TUN_RX_PAD;
1671         int err = 0;
1672
1673         rcu_read_lock();
1674         xdp_prog = rcu_dereference(tun->xdp_prog);
1675         if (xdp_prog)
1676                 pad += XDP_PACKET_HEADROOM;
1677         buflen += SKB_DATA_ALIGN(len + pad);
1678         rcu_read_unlock();
1679
1680         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1681         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1682                 return ERR_PTR(-ENOMEM);
1683
1684         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1685         copied = copy_page_from_iter(alloc_frag->page,
1686                                      alloc_frag->offset + pad,
1687                                      len, from);
1688         if (copied != len)
1689                 return ERR_PTR(-EFAULT);
1690
1691         /* There's a small window that XDP may be set after the check
1692          * of xdp_prog above, this should be rare and for simplicity
1693          * we do XDP on skb in case the headroom is not enough.
1694          */
1695         if (hdr->gso_type || !xdp_prog) {
1696                 *skb_xdp = 1;
1697                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1698                                        pad);
1699         }
1700
1701         *skb_xdp = 0;
1702
1703         local_bh_disable();
1704         rcu_read_lock();
1705         bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
1706         xdp_prog = rcu_dereference(tun->xdp_prog);
1707         if (xdp_prog) {
1708                 struct xdp_buff xdp;
1709                 u32 act;
1710
1711                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1712                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1713
1714                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1715                 if (act == XDP_REDIRECT || act == XDP_TX) {
1716                         get_page(alloc_frag->page);
1717                         alloc_frag->offset += buflen;
1718                 }
1719                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1720                 if (err < 0) {
1721                         if (act == XDP_REDIRECT || act == XDP_TX)
1722                                 put_page(alloc_frag->page);
1723                         goto out;
1724                 }
1725
1726                 if (err == XDP_REDIRECT)
1727                         xdp_do_flush();
1728                 if (err != XDP_PASS)
1729                         goto out;
1730
1731                 pad = xdp.data - xdp.data_hard_start;
1732                 len = xdp.data_end - xdp.data;
1733         }
1734         bpf_net_ctx_clear(bpf_net_ctx);
1735         rcu_read_unlock();
1736         local_bh_enable();
1737
1738         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1739
1740 out:
1741         bpf_net_ctx_clear(bpf_net_ctx);
1742         rcu_read_unlock();
1743         local_bh_enable();
1744         return NULL;
1745 }
1746
1747 /* Get packet from user space buffer */
1748 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1749                             void *msg_control, struct iov_iter *from,
1750                             int noblock, bool more)
1751 {
1752         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1753         struct sk_buff *skb;
1754         size_t total_len = iov_iter_count(from);
1755         size_t len = total_len, align = tun->align, linear;
1756         struct virtio_net_hdr gso = { 0 };
1757         int good_linear;
1758         int copylen;
1759         bool zerocopy = false;
1760         int err;
1761         u32 rxhash = 0;
1762         int skb_xdp = 1;
1763         bool frags = tun_napi_frags_enabled(tfile);
1764         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1765
1766         if (!(tun->flags & IFF_NO_PI)) {
1767                 if (len < sizeof(pi))
1768                         return -EINVAL;
1769                 len -= sizeof(pi);
1770
1771                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1772                         return -EFAULT;
1773         }
1774
1775         if (tun->flags & IFF_VNET_HDR) {
1776                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1777
1778                 if (len < vnet_hdr_sz)
1779                         return -EINVAL;
1780                 len -= vnet_hdr_sz;
1781
1782                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1783                         return -EFAULT;
1784
1785                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1786                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1787                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1788
1789                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1790                         return -EINVAL;
1791                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1792         }
1793
1794         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1795                 align += NET_IP_ALIGN;
1796                 if (unlikely(len < ETH_HLEN ||
1797                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1798                         return -EINVAL;
1799         }
1800
1801         good_linear = SKB_MAX_HEAD(align);
1802
1803         if (msg_control) {
1804                 struct iov_iter i = *from;
1805
1806                 /* There are 256 bytes to be copied in skb, so there is
1807                  * enough room for skb expand head in case it is used.
1808                  * The rest of the buffer is mapped from userspace.
1809                  */
1810                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1811                 if (copylen > good_linear)
1812                         copylen = good_linear;
1813                 linear = copylen;
1814                 iov_iter_advance(&i, copylen);
1815                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1816                         zerocopy = true;
1817         }
1818
1819         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1820                 /* For the packet that is not easy to be processed
1821                  * (e.g gso or jumbo packet), we will do it at after
1822                  * skb was created with generic XDP routine.
1823                  */
1824                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1825                 err = PTR_ERR_OR_ZERO(skb);
1826                 if (err)
1827                         goto drop;
1828                 if (!skb)
1829                         return total_len;
1830         } else {
1831                 if (!zerocopy) {
1832                         copylen = len;
1833                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1834                                 linear = good_linear;
1835                         else
1836                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1837                 }
1838
1839                 if (frags) {
1840                         mutex_lock(&tfile->napi_mutex);
1841                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1842                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1843                          * If zerocopy is enabled, then this layout will be
1844                          * overwritten by zerocopy_sg_from_iter().
1845                          */
1846                         zerocopy = false;
1847                 } else {
1848                         if (!linear)
1849                                 linear = min_t(size_t, good_linear, copylen);
1850
1851                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1852                                             noblock);
1853                 }
1854
1855                 err = PTR_ERR_OR_ZERO(skb);
1856                 if (err)
1857                         goto drop;
1858
1859                 if (zerocopy)
1860                         err = zerocopy_sg_from_iter(skb, from);
1861                 else
1862                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1863
1864                 if (err) {
1865                         err = -EFAULT;
1866                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1867                         goto drop;
1868                 }
1869         }
1870
1871         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1872                 atomic_long_inc(&tun->rx_frame_errors);
1873                 err = -EINVAL;
1874                 goto free_skb;
1875         }
1876
1877         switch (tun->flags & TUN_TYPE_MASK) {
1878         case IFF_TUN:
1879                 if (tun->flags & IFF_NO_PI) {
1880                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1881
1882                         switch (ip_version) {
1883                         case 4:
1884                                 pi.proto = htons(ETH_P_IP);
1885                                 break;
1886                         case 6:
1887                                 pi.proto = htons(ETH_P_IPV6);
1888                                 break;
1889                         default:
1890                                 err = -EINVAL;
1891                                 goto drop;
1892                         }
1893                 }
1894
1895                 skb_reset_mac_header(skb);
1896                 skb->protocol = pi.proto;
1897                 skb->dev = tun->dev;
1898                 break;
1899         case IFF_TAP:
1900                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1901                         err = -ENOMEM;
1902                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1903                         goto drop;
1904                 }
1905                 skb->protocol = eth_type_trans(skb, tun->dev);
1906                 break;
1907         }
1908
1909         /* copy skb_ubuf_info for callback when skb has no error */
1910         if (zerocopy) {
1911                 skb_zcopy_init(skb, msg_control);
1912         } else if (msg_control) {
1913                 struct ubuf_info *uarg = msg_control;
1914                 uarg->ops->complete(NULL, uarg, false);
1915         }
1916
1917         skb_reset_network_header(skb);
1918         skb_probe_transport_header(skb);
1919         skb_record_rx_queue(skb, tfile->queue_index);
1920
1921         if (skb_xdp) {
1922                 struct bpf_prog *xdp_prog;
1923                 int ret;
1924
1925                 local_bh_disable();
1926                 rcu_read_lock();
1927                 xdp_prog = rcu_dereference(tun->xdp_prog);
1928                 if (xdp_prog) {
1929                         ret = do_xdp_generic(xdp_prog, &skb);
1930                         if (ret != XDP_PASS) {
1931                                 rcu_read_unlock();
1932                                 local_bh_enable();
1933                                 goto unlock_frags;
1934                         }
1935                 }
1936                 rcu_read_unlock();
1937                 local_bh_enable();
1938         }
1939
1940         /* Compute the costly rx hash only if needed for flow updates.
1941          * We may get a very small possibility of OOO during switching, not
1942          * worth to optimize.
1943          */
1944         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1945             !tfile->detached)
1946                 rxhash = __skb_get_hash_symmetric(skb);
1947
1948         rcu_read_lock();
1949         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1950                 err = -EIO;
1951                 rcu_read_unlock();
1952                 drop_reason = SKB_DROP_REASON_DEV_READY;
1953                 goto drop;
1954         }
1955
1956         if (frags) {
1957                 u32 headlen;
1958
1959                 /* Exercise flow dissector code path. */
1960                 skb_push(skb, ETH_HLEN);
1961                 headlen = eth_get_headlen(tun->dev, skb->data,
1962                                           skb_headlen(skb));
1963
1964                 if (unlikely(headlen > skb_headlen(skb))) {
1965                         WARN_ON_ONCE(1);
1966                         err = -ENOMEM;
1967                         dev_core_stats_rx_dropped_inc(tun->dev);
1968 napi_busy:
1969                         napi_free_frags(&tfile->napi);
1970                         rcu_read_unlock();
1971                         mutex_unlock(&tfile->napi_mutex);
1972                         return err;
1973                 }
1974
1975                 if (likely(napi_schedule_prep(&tfile->napi))) {
1976                         local_bh_disable();
1977                         napi_gro_frags(&tfile->napi);
1978                         napi_complete(&tfile->napi);
1979                         local_bh_enable();
1980                 } else {
1981                         err = -EBUSY;
1982                         goto napi_busy;
1983                 }
1984                 mutex_unlock(&tfile->napi_mutex);
1985         } else if (tfile->napi_enabled) {
1986                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1987                 int queue_len;
1988
1989                 spin_lock_bh(&queue->lock);
1990
1991                 if (unlikely(tfile->detached)) {
1992                         spin_unlock_bh(&queue->lock);
1993                         rcu_read_unlock();
1994                         err = -EBUSY;
1995                         goto free_skb;
1996                 }
1997
1998                 __skb_queue_tail(queue, skb);
1999                 queue_len = skb_queue_len(queue);
2000                 spin_unlock(&queue->lock);
2001
2002                 if (!more || queue_len > NAPI_POLL_WEIGHT)
2003                         napi_schedule(&tfile->napi);
2004
2005                 local_bh_enable();
2006         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2007                 tun_rx_batched(tun, tfile, skb, more);
2008         } else {
2009                 netif_rx(skb);
2010         }
2011         rcu_read_unlock();
2012
2013         preempt_disable();
2014         dev_sw_netstats_rx_add(tun->dev, len);
2015         preempt_enable();
2016
2017         if (rxhash)
2018                 tun_flow_update(tun, rxhash, tfile);
2019
2020         return total_len;
2021
2022 drop:
2023         if (err != -EAGAIN)
2024                 dev_core_stats_rx_dropped_inc(tun->dev);
2025
2026 free_skb:
2027         if (!IS_ERR_OR_NULL(skb))
2028                 kfree_skb_reason(skb, drop_reason);
2029
2030 unlock_frags:
2031         if (frags) {
2032                 tfile->napi.skb = NULL;
2033                 mutex_unlock(&tfile->napi_mutex);
2034         }
2035
2036         return err ?: total_len;
2037 }
2038
2039 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2040 {
2041         struct file *file = iocb->ki_filp;
2042         struct tun_file *tfile = file->private_data;
2043         struct tun_struct *tun = tun_get(tfile);
2044         ssize_t result;
2045         int noblock = 0;
2046
2047         if (!tun)
2048                 return -EBADFD;
2049
2050         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2051                 noblock = 1;
2052
2053         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2054
2055         tun_put(tun);
2056         return result;
2057 }
2058
2059 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2060                                 struct tun_file *tfile,
2061                                 struct xdp_frame *xdp_frame,
2062                                 struct iov_iter *iter)
2063 {
2064         int vnet_hdr_sz = 0;
2065         size_t size = xdp_frame->len;
2066         size_t ret;
2067
2068         if (tun->flags & IFF_VNET_HDR) {
2069                 struct virtio_net_hdr gso = { 0 };
2070
2071                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2072                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2073                         return -EINVAL;
2074                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2075                              sizeof(gso)))
2076                         return -EFAULT;
2077                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2078         }
2079
2080         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2081
2082         preempt_disable();
2083         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2084         preempt_enable();
2085
2086         return ret;
2087 }
2088
2089 /* Put packet to the user space buffer */
2090 static ssize_t tun_put_user(struct tun_struct *tun,
2091                             struct tun_file *tfile,
2092                             struct sk_buff *skb,
2093                             struct iov_iter *iter)
2094 {
2095         struct tun_pi pi = { 0, skb->protocol };
2096         ssize_t total;
2097         int vlan_offset = 0;
2098         int vlan_hlen = 0;
2099         int vnet_hdr_sz = 0;
2100
2101         if (skb_vlan_tag_present(skb))
2102                 vlan_hlen = VLAN_HLEN;
2103
2104         if (tun->flags & IFF_VNET_HDR)
2105                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2106
2107         total = skb->len + vlan_hlen + vnet_hdr_sz;
2108
2109         if (!(tun->flags & IFF_NO_PI)) {
2110                 if (iov_iter_count(iter) < sizeof(pi))
2111                         return -EINVAL;
2112
2113                 total += sizeof(pi);
2114                 if (iov_iter_count(iter) < total) {
2115                         /* Packet will be striped */
2116                         pi.flags |= TUN_PKT_STRIP;
2117                 }
2118
2119                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2120                         return -EFAULT;
2121         }
2122
2123         if (vnet_hdr_sz) {
2124                 struct virtio_net_hdr gso;
2125
2126                 if (iov_iter_count(iter) < vnet_hdr_sz)
2127                         return -EINVAL;
2128
2129                 if (virtio_net_hdr_from_skb(skb, &gso,
2130                                             tun_is_little_endian(tun), true,
2131                                             vlan_hlen)) {
2132                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2133
2134                         if (net_ratelimit()) {
2135                                 netdev_err(tun->dev, "unexpected GSO type: 0x%x, gso_size %d, hdr_len %d\n",
2136                                            sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2137                                            tun16_to_cpu(tun, gso.hdr_len));
2138                                 print_hex_dump(KERN_ERR, "tun: ",
2139                                                DUMP_PREFIX_NONE,
2140                                                16, 1, skb->head,
2141                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2142                         }
2143                         WARN_ON_ONCE(1);
2144                         return -EINVAL;
2145                 }
2146
2147                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2148                         return -EFAULT;
2149
2150                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2151         }
2152
2153         if (vlan_hlen) {
2154                 int ret;
2155                 struct veth veth;
2156
2157                 veth.h_vlan_proto = skb->vlan_proto;
2158                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2159
2160                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2161
2162                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2163                 if (ret || !iov_iter_count(iter))
2164                         goto done;
2165
2166                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2167                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2168                         goto done;
2169         }
2170
2171         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2172
2173 done:
2174         /* caller is in process context, */
2175         preempt_disable();
2176         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2177         preempt_enable();
2178
2179         return total;
2180 }
2181
2182 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2183 {
2184         DECLARE_WAITQUEUE(wait, current);
2185         void *ptr = NULL;
2186         int error = 0;
2187
2188         ptr = ptr_ring_consume(&tfile->tx_ring);
2189         if (ptr)
2190                 goto out;
2191         if (noblock) {
2192                 error = -EAGAIN;
2193                 goto out;
2194         }
2195
2196         add_wait_queue(&tfile->socket.wq.wait, &wait);
2197
2198         while (1) {
2199                 set_current_state(TASK_INTERRUPTIBLE);
2200                 ptr = ptr_ring_consume(&tfile->tx_ring);
2201                 if (ptr)
2202                         break;
2203                 if (signal_pending(current)) {
2204                         error = -ERESTARTSYS;
2205                         break;
2206                 }
2207                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2208                         error = -EFAULT;
2209                         break;
2210                 }
2211
2212                 schedule();
2213         }
2214
2215         __set_current_state(TASK_RUNNING);
2216         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2217
2218 out:
2219         *err = error;
2220         return ptr;
2221 }
2222
2223 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2224                            struct iov_iter *to,
2225                            int noblock, void *ptr)
2226 {
2227         ssize_t ret;
2228         int err;
2229
2230         if (!iov_iter_count(to)) {
2231                 tun_ptr_free(ptr);
2232                 return 0;
2233         }
2234
2235         if (!ptr) {
2236                 /* Read frames from ring */
2237                 ptr = tun_ring_recv(tfile, noblock, &err);
2238                 if (!ptr)
2239                         return err;
2240         }
2241
2242         if (tun_is_xdp_frame(ptr)) {
2243                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2244
2245                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2246                 xdp_return_frame(xdpf);
2247         } else {
2248                 struct sk_buff *skb = ptr;
2249
2250                 ret = tun_put_user(tun, tfile, skb, to);
2251                 if (unlikely(ret < 0))
2252                         kfree_skb(skb);
2253                 else
2254                         consume_skb(skb);
2255         }
2256
2257         return ret;
2258 }
2259
2260 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2261 {
2262         struct file *file = iocb->ki_filp;
2263         struct tun_file *tfile = file->private_data;
2264         struct tun_struct *tun = tun_get(tfile);
2265         ssize_t len = iov_iter_count(to), ret;
2266         int noblock = 0;
2267
2268         if (!tun)
2269                 return -EBADFD;
2270
2271         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2272                 noblock = 1;
2273
2274         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2275         ret = min_t(ssize_t, ret, len);
2276         if (ret > 0)
2277                 iocb->ki_pos = ret;
2278         tun_put(tun);
2279         return ret;
2280 }
2281
2282 static void tun_prog_free(struct rcu_head *rcu)
2283 {
2284         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2285
2286         bpf_prog_destroy(prog->prog);
2287         kfree(prog);
2288 }
2289
2290 static int __tun_set_ebpf(struct tun_struct *tun,
2291                           struct tun_prog __rcu **prog_p,
2292                           struct bpf_prog *prog)
2293 {
2294         struct tun_prog *old, *new = NULL;
2295
2296         if (prog) {
2297                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2298                 if (!new)
2299                         return -ENOMEM;
2300                 new->prog = prog;
2301         }
2302
2303         spin_lock_bh(&tun->lock);
2304         old = rcu_dereference_protected(*prog_p,
2305                                         lockdep_is_held(&tun->lock));
2306         rcu_assign_pointer(*prog_p, new);
2307         spin_unlock_bh(&tun->lock);
2308
2309         if (old)
2310                 call_rcu(&old->rcu, tun_prog_free);
2311
2312         return 0;
2313 }
2314
2315 static void tun_free_netdev(struct net_device *dev)
2316 {
2317         struct tun_struct *tun = netdev_priv(dev);
2318
2319         BUG_ON(!(list_empty(&tun->disabled)));
2320
2321         tun_flow_uninit(tun);
2322         security_tun_dev_free_security(tun->security);
2323         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2324         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2325 }
2326
2327 static void tun_setup(struct net_device *dev)
2328 {
2329         struct tun_struct *tun = netdev_priv(dev);
2330
2331         tun->owner = INVALID_UID;
2332         tun->group = INVALID_GID;
2333         tun_default_link_ksettings(dev, &tun->link_ksettings);
2334
2335         dev->ethtool_ops = &tun_ethtool_ops;
2336         dev->needs_free_netdev = true;
2337         dev->priv_destructor = tun_free_netdev;
2338         /* We prefer our own queue length */
2339         dev->tx_queue_len = TUN_READQ_SIZE;
2340 }
2341
2342 /* Trivial set of netlink ops to allow deleting tun or tap
2343  * device with netlink.
2344  */
2345 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2346                         struct netlink_ext_ack *extack)
2347 {
2348         NL_SET_ERR_MSG(extack,
2349                        "tun/tap creation via rtnetlink is not supported.");
2350         return -EOPNOTSUPP;
2351 }
2352
2353 static size_t tun_get_size(const struct net_device *dev)
2354 {
2355         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2356         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2357
2358         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2359                nla_total_size(sizeof(gid_t)) + /* GROUP */
2360                nla_total_size(sizeof(u8)) + /* TYPE */
2361                nla_total_size(sizeof(u8)) + /* PI */
2362                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2363                nla_total_size(sizeof(u8)) + /* PERSIST */
2364                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2365                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2366                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2367                0;
2368 }
2369
2370 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2371 {
2372         struct tun_struct *tun = netdev_priv(dev);
2373
2374         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2375                 goto nla_put_failure;
2376         if (uid_valid(tun->owner) &&
2377             nla_put_u32(skb, IFLA_TUN_OWNER,
2378                         from_kuid_munged(current_user_ns(), tun->owner)))
2379                 goto nla_put_failure;
2380         if (gid_valid(tun->group) &&
2381             nla_put_u32(skb, IFLA_TUN_GROUP,
2382                         from_kgid_munged(current_user_ns(), tun->group)))
2383                 goto nla_put_failure;
2384         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2385                 goto nla_put_failure;
2386         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2387                 goto nla_put_failure;
2388         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2389                 goto nla_put_failure;
2390         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2391                        !!(tun->flags & IFF_MULTI_QUEUE)))
2392                 goto nla_put_failure;
2393         if (tun->flags & IFF_MULTI_QUEUE) {
2394                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2395                         goto nla_put_failure;
2396                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2397                                 tun->numdisabled))
2398                         goto nla_put_failure;
2399         }
2400
2401         return 0;
2402
2403 nla_put_failure:
2404         return -EMSGSIZE;
2405 }
2406
2407 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2408         .kind           = DRV_NAME,
2409         .priv_size      = sizeof(struct tun_struct),
2410         .setup          = tun_setup,
2411         .validate       = tun_validate,
2412         .get_size       = tun_get_size,
2413         .fill_info      = tun_fill_info,
2414 };
2415
2416 static void tun_sock_write_space(struct sock *sk)
2417 {
2418         struct tun_file *tfile;
2419         wait_queue_head_t *wqueue;
2420
2421         if (!sock_writeable(sk))
2422                 return;
2423
2424         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2425                 return;
2426
2427         wqueue = sk_sleep(sk);
2428         if (wqueue && waitqueue_active(wqueue))
2429                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2430                                                 EPOLLWRNORM | EPOLLWRBAND);
2431
2432         tfile = container_of(sk, struct tun_file, sk);
2433         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2434 }
2435
2436 static void tun_put_page(struct tun_page *tpage)
2437 {
2438         if (tpage->page)
2439                 __page_frag_cache_drain(tpage->page, tpage->count);
2440 }
2441
2442 static int tun_xdp_one(struct tun_struct *tun,
2443                        struct tun_file *tfile,
2444                        struct xdp_buff *xdp, int *flush,
2445                        struct tun_page *tpage)
2446 {
2447         unsigned int datasize = xdp->data_end - xdp->data;
2448         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2449         struct virtio_net_hdr *gso = &hdr->gso;
2450         struct bpf_prog *xdp_prog;
2451         struct sk_buff *skb = NULL;
2452         struct sk_buff_head *queue;
2453         u32 rxhash = 0, act;
2454         int buflen = hdr->buflen;
2455         int ret = 0;
2456         bool skb_xdp = false;
2457         struct page *page;
2458
2459         if (unlikely(datasize < ETH_HLEN))
2460                 return -EINVAL;
2461
2462         xdp_prog = rcu_dereference(tun->xdp_prog);
2463         if (xdp_prog) {
2464                 if (gso->gso_type) {
2465                         skb_xdp = true;
2466                         goto build;
2467                 }
2468
2469                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2470                 xdp_set_data_meta_invalid(xdp);
2471
2472                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2473                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2474                 if (ret < 0) {
2475                         put_page(virt_to_head_page(xdp->data));
2476                         return ret;
2477                 }
2478
2479                 switch (ret) {
2480                 case XDP_REDIRECT:
2481                         *flush = true;
2482                         fallthrough;
2483                 case XDP_TX:
2484                         return 0;
2485                 case XDP_PASS:
2486                         break;
2487                 default:
2488                         page = virt_to_head_page(xdp->data);
2489                         if (tpage->page == page) {
2490                                 ++tpage->count;
2491                         } else {
2492                                 tun_put_page(tpage);
2493                                 tpage->page = page;
2494                                 tpage->count = 1;
2495                         }
2496                         return 0;
2497                 }
2498         }
2499
2500 build:
2501         skb = build_skb(xdp->data_hard_start, buflen);
2502         if (!skb) {
2503                 ret = -ENOMEM;
2504                 goto out;
2505         }
2506
2507         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2508         skb_put(skb, xdp->data_end - xdp->data);
2509
2510         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2511                 atomic_long_inc(&tun->rx_frame_errors);
2512                 kfree_skb(skb);
2513                 ret = -EINVAL;
2514                 goto out;
2515         }
2516
2517         skb->protocol = eth_type_trans(skb, tun->dev);
2518         skb_reset_network_header(skb);
2519         skb_probe_transport_header(skb);
2520         skb_record_rx_queue(skb, tfile->queue_index);
2521
2522         if (skb_xdp) {
2523                 ret = do_xdp_generic(xdp_prog, &skb);
2524                 if (ret != XDP_PASS) {
2525                         ret = 0;
2526                         goto out;
2527                 }
2528         }
2529
2530         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2531             !tfile->detached)
2532                 rxhash = __skb_get_hash_symmetric(skb);
2533
2534         if (tfile->napi_enabled) {
2535                 queue = &tfile->sk.sk_write_queue;
2536                 spin_lock(&queue->lock);
2537
2538                 if (unlikely(tfile->detached)) {
2539                         spin_unlock(&queue->lock);
2540                         kfree_skb(skb);
2541                         return -EBUSY;
2542                 }
2543
2544                 __skb_queue_tail(queue, skb);
2545                 spin_unlock(&queue->lock);
2546                 ret = 1;
2547         } else {
2548                 netif_receive_skb(skb);
2549                 ret = 0;
2550         }
2551
2552         /* No need to disable preemption here since this function is
2553          * always called with bh disabled
2554          */
2555         dev_sw_netstats_rx_add(tun->dev, datasize);
2556
2557         if (rxhash)
2558                 tun_flow_update(tun, rxhash, tfile);
2559
2560 out:
2561         return ret;
2562 }
2563
2564 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2565 {
2566         int ret, i;
2567         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2568         struct tun_struct *tun = tun_get(tfile);
2569         struct tun_msg_ctl *ctl = m->msg_control;
2570         struct xdp_buff *xdp;
2571
2572         if (!tun)
2573                 return -EBADFD;
2574
2575         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2576             ctl && ctl->type == TUN_MSG_PTR) {
2577                 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
2578                 struct tun_page tpage;
2579                 int n = ctl->num;
2580                 int flush = 0, queued = 0;
2581
2582                 memset(&tpage, 0, sizeof(tpage));
2583
2584                 local_bh_disable();
2585                 rcu_read_lock();
2586                 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
2587
2588                 for (i = 0; i < n; i++) {
2589                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2590                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2591                         if (ret > 0)
2592                                 queued += ret;
2593                 }
2594
2595                 if (flush)
2596                         xdp_do_flush();
2597
2598                 if (tfile->napi_enabled && queued > 0)
2599                         napi_schedule(&tfile->napi);
2600
2601                 bpf_net_ctx_clear(bpf_net_ctx);
2602                 rcu_read_unlock();
2603                 local_bh_enable();
2604
2605                 tun_put_page(&tpage);
2606
2607                 ret = total_len;
2608                 goto out;
2609         }
2610
2611         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2612                            m->msg_flags & MSG_DONTWAIT,
2613                            m->msg_flags & MSG_MORE);
2614 out:
2615         tun_put(tun);
2616         return ret;
2617 }
2618
2619 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2620                        int flags)
2621 {
2622         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2623         struct tun_struct *tun = tun_get(tfile);
2624         void *ptr = m->msg_control;
2625         int ret;
2626
2627         if (!tun) {
2628                 ret = -EBADFD;
2629                 goto out_free;
2630         }
2631
2632         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2633                 ret = -EINVAL;
2634                 goto out_put_tun;
2635         }
2636         if (flags & MSG_ERRQUEUE) {
2637                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2638                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2639                 goto out;
2640         }
2641         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2642         if (ret > (ssize_t)total_len) {
2643                 m->msg_flags |= MSG_TRUNC;
2644                 ret = flags & MSG_TRUNC ? ret : total_len;
2645         }
2646 out:
2647         tun_put(tun);
2648         return ret;
2649
2650 out_put_tun:
2651         tun_put(tun);
2652 out_free:
2653         tun_ptr_free(ptr);
2654         return ret;
2655 }
2656
2657 static int tun_ptr_peek_len(void *ptr)
2658 {
2659         if (likely(ptr)) {
2660                 if (tun_is_xdp_frame(ptr)) {
2661                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2662
2663                         return xdpf->len;
2664                 }
2665                 return __skb_array_len_with_tag(ptr);
2666         } else {
2667                 return 0;
2668         }
2669 }
2670
2671 static int tun_peek_len(struct socket *sock)
2672 {
2673         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2674         struct tun_struct *tun;
2675         int ret = 0;
2676
2677         tun = tun_get(tfile);
2678         if (!tun)
2679                 return 0;
2680
2681         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2682         tun_put(tun);
2683
2684         return ret;
2685 }
2686
2687 /* Ops structure to mimic raw sockets with tun */
2688 static const struct proto_ops tun_socket_ops = {
2689         .peek_len = tun_peek_len,
2690         .sendmsg = tun_sendmsg,
2691         .recvmsg = tun_recvmsg,
2692 };
2693
2694 static struct proto tun_proto = {
2695         .name           = "tun",
2696         .owner          = THIS_MODULE,
2697         .obj_size       = sizeof(struct tun_file),
2698 };
2699
2700 static int tun_flags(struct tun_struct *tun)
2701 {
2702         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2703 }
2704
2705 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2706                               char *buf)
2707 {
2708         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2709         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2710 }
2711
2712 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2713                           char *buf)
2714 {
2715         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2716         return uid_valid(tun->owner)?
2717                 sysfs_emit(buf, "%u\n",
2718                            from_kuid_munged(current_user_ns(), tun->owner)) :
2719                 sysfs_emit(buf, "-1\n");
2720 }
2721
2722 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2723                           char *buf)
2724 {
2725         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2726         return gid_valid(tun->group) ?
2727                 sysfs_emit(buf, "%u\n",
2728                            from_kgid_munged(current_user_ns(), tun->group)) :
2729                 sysfs_emit(buf, "-1\n");
2730 }
2731
2732 static DEVICE_ATTR_RO(tun_flags);
2733 static DEVICE_ATTR_RO(owner);
2734 static DEVICE_ATTR_RO(group);
2735
2736 static struct attribute *tun_dev_attrs[] = {
2737         &dev_attr_tun_flags.attr,
2738         &dev_attr_owner.attr,
2739         &dev_attr_group.attr,
2740         NULL
2741 };
2742
2743 static const struct attribute_group tun_attr_group = {
2744         .attrs = tun_dev_attrs
2745 };
2746
2747 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2748 {
2749         struct tun_struct *tun;
2750         struct tun_file *tfile = file->private_data;
2751         struct net_device *dev;
2752         int err;
2753
2754         if (tfile->detached)
2755                 return -EINVAL;
2756
2757         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2758                 if (!capable(CAP_NET_ADMIN))
2759                         return -EPERM;
2760
2761                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2762                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2763                         return -EINVAL;
2764         }
2765
2766         dev = __dev_get_by_name(net, ifr->ifr_name);
2767         if (dev) {
2768                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2769                         return -EBUSY;
2770                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2771                         tun = netdev_priv(dev);
2772                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2773                         tun = netdev_priv(dev);
2774                 else
2775                         return -EINVAL;
2776
2777                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2778                     !!(tun->flags & IFF_MULTI_QUEUE))
2779                         return -EINVAL;
2780
2781                 if (tun_not_capable(tun))
2782                         return -EPERM;
2783                 err = security_tun_dev_open(tun->security);
2784                 if (err < 0)
2785                         return err;
2786
2787                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2788                                  ifr->ifr_flags & IFF_NAPI,
2789                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2790                 if (err < 0)
2791                         return err;
2792
2793                 if (tun->flags & IFF_MULTI_QUEUE &&
2794                     (tun->numqueues + tun->numdisabled > 1)) {
2795                         /* One or more queue has already been attached, no need
2796                          * to initialize the device again.
2797                          */
2798                         netdev_state_change(dev);
2799                         return 0;
2800                 }
2801
2802                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2803                               (ifr->ifr_flags & TUN_FEATURES);
2804
2805                 netdev_state_change(dev);
2806         } else {
2807                 char *name;
2808                 unsigned long flags = 0;
2809                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2810                              MAX_TAP_QUEUES : 1;
2811
2812                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2813                         return -EPERM;
2814                 err = security_tun_dev_create();
2815                 if (err < 0)
2816                         return err;
2817
2818                 /* Set dev type */
2819                 if (ifr->ifr_flags & IFF_TUN) {
2820                         /* TUN device */
2821                         flags |= IFF_TUN;
2822                         name = "tun%d";
2823                 } else if (ifr->ifr_flags & IFF_TAP) {
2824                         /* TAP device */
2825                         flags |= IFF_TAP;
2826                         name = "tap%d";
2827                 } else
2828                         return -EINVAL;
2829
2830                 if (*ifr->ifr_name)
2831                         name = ifr->ifr_name;
2832
2833                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2834                                        NET_NAME_UNKNOWN, tun_setup, queues,
2835                                        queues);
2836
2837                 if (!dev)
2838                         return -ENOMEM;
2839
2840                 dev_net_set(dev, net);
2841                 dev->rtnl_link_ops = &tun_link_ops;
2842                 dev->ifindex = tfile->ifindex;
2843                 dev->sysfs_groups[0] = &tun_attr_group;
2844
2845                 tun = netdev_priv(dev);
2846                 tun->dev = dev;
2847                 tun->flags = flags;
2848                 tun->txflt.count = 0;
2849                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2850
2851                 tun->align = NET_SKB_PAD;
2852                 tun->filter_attached = false;
2853                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2854                 tun->rx_batched = 0;
2855                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2856
2857                 tun->ifr = ifr;
2858                 tun->file = file;
2859
2860                 tun_net_initialize(dev);
2861
2862                 err = register_netdevice(tun->dev);
2863                 if (err < 0) {
2864                         free_netdev(dev);
2865                         return err;
2866                 }
2867                 /* free_netdev() won't check refcnt, to avoid race
2868                  * with dev_put() we need publish tun after registration.
2869                  */
2870                 rcu_assign_pointer(tfile->tun, tun);
2871         }
2872
2873         if (ifr->ifr_flags & IFF_NO_CARRIER)
2874                 netif_carrier_off(tun->dev);
2875         else
2876                 netif_carrier_on(tun->dev);
2877
2878         /* Make sure persistent devices do not get stuck in
2879          * xoff state.
2880          */
2881         if (netif_running(tun->dev))
2882                 netif_tx_wake_all_queues(tun->dev);
2883
2884         strcpy(ifr->ifr_name, tun->dev->name);
2885         return 0;
2886 }
2887
2888 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2889 {
2890         strcpy(ifr->ifr_name, tun->dev->name);
2891
2892         ifr->ifr_flags = tun_flags(tun);
2893
2894 }
2895
2896 /* This is like a cut-down ethtool ops, except done via tun fd so no
2897  * privs required. */
2898 static int set_offload(struct tun_struct *tun, unsigned long arg)
2899 {
2900         netdev_features_t features = 0;
2901
2902         if (arg & TUN_F_CSUM) {
2903                 features |= NETIF_F_HW_CSUM;
2904                 arg &= ~TUN_F_CSUM;
2905
2906                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2907                         if (arg & TUN_F_TSO_ECN) {
2908                                 features |= NETIF_F_TSO_ECN;
2909                                 arg &= ~TUN_F_TSO_ECN;
2910                         }
2911                         if (arg & TUN_F_TSO4)
2912                                 features |= NETIF_F_TSO;
2913                         if (arg & TUN_F_TSO6)
2914                                 features |= NETIF_F_TSO6;
2915                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2916                 }
2917
2918                 arg &= ~TUN_F_UFO;
2919
2920                 /* TODO: for now USO4 and USO6 should work simultaneously */
2921                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2922                         features |= NETIF_F_GSO_UDP_L4;
2923                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2924                 }
2925         }
2926
2927         /* This gives the user a way to test for new features in future by
2928          * trying to set them. */
2929         if (arg)
2930                 return -EINVAL;
2931
2932         tun->set_features = features;
2933         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2934         tun->dev->wanted_features |= features;
2935         netdev_update_features(tun->dev);
2936
2937         return 0;
2938 }
2939
2940 static void tun_detach_filter(struct tun_struct *tun, int n)
2941 {
2942         int i;
2943         struct tun_file *tfile;
2944
2945         for (i = 0; i < n; i++) {
2946                 tfile = rtnl_dereference(tun->tfiles[i]);
2947                 lock_sock(tfile->socket.sk);
2948                 sk_detach_filter(tfile->socket.sk);
2949                 release_sock(tfile->socket.sk);
2950         }
2951
2952         tun->filter_attached = false;
2953 }
2954
2955 static int tun_attach_filter(struct tun_struct *tun)
2956 {
2957         int i, ret = 0;
2958         struct tun_file *tfile;
2959
2960         for (i = 0; i < tun->numqueues; i++) {
2961                 tfile = rtnl_dereference(tun->tfiles[i]);
2962                 lock_sock(tfile->socket.sk);
2963                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2964                 release_sock(tfile->socket.sk);
2965                 if (ret) {
2966                         tun_detach_filter(tun, i);
2967                         return ret;
2968                 }
2969         }
2970
2971         tun->filter_attached = true;
2972         return ret;
2973 }
2974
2975 static void tun_set_sndbuf(struct tun_struct *tun)
2976 {
2977         struct tun_file *tfile;
2978         int i;
2979
2980         for (i = 0; i < tun->numqueues; i++) {
2981                 tfile = rtnl_dereference(tun->tfiles[i]);
2982                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2983         }
2984 }
2985
2986 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2987 {
2988         struct tun_file *tfile = file->private_data;
2989         struct tun_struct *tun;
2990         int ret = 0;
2991
2992         rtnl_lock();
2993
2994         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2995                 tun = tfile->detached;
2996                 if (!tun) {
2997                         ret = -EINVAL;
2998                         goto unlock;
2999                 }
3000                 ret = security_tun_dev_attach_queue(tun->security);
3001                 if (ret < 0)
3002                         goto unlock;
3003                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
3004                                  tun->flags & IFF_NAPI_FRAGS, true);
3005         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
3006                 tun = rtnl_dereference(tfile->tun);
3007                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3008                         ret = -EINVAL;
3009                 else
3010                         __tun_detach(tfile, false);
3011         } else
3012                 ret = -EINVAL;
3013
3014         if (ret >= 0)
3015                 netdev_state_change(tun->dev);
3016
3017 unlock:
3018         rtnl_unlock();
3019         return ret;
3020 }
3021
3022 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3023                         void __user *data)
3024 {
3025         struct bpf_prog *prog;
3026         int fd;
3027
3028         if (copy_from_user(&fd, data, sizeof(fd)))
3029                 return -EFAULT;
3030
3031         if (fd == -1) {
3032                 prog = NULL;
3033         } else {
3034                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3035                 if (IS_ERR(prog))
3036                         return PTR_ERR(prog);
3037         }
3038
3039         return __tun_set_ebpf(tun, prog_p, prog);
3040 }
3041
3042 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3043 static unsigned char tun_get_addr_len(unsigned short type)
3044 {
3045         switch (type) {
3046         case ARPHRD_IP6GRE:
3047         case ARPHRD_TUNNEL6:
3048                 return sizeof(struct in6_addr);
3049         case ARPHRD_IPGRE:
3050         case ARPHRD_TUNNEL:
3051         case ARPHRD_SIT:
3052                 return 4;
3053         case ARPHRD_ETHER:
3054                 return ETH_ALEN;
3055         case ARPHRD_IEEE802154:
3056         case ARPHRD_IEEE802154_MONITOR:
3057                 return IEEE802154_EXTENDED_ADDR_LEN;
3058         case ARPHRD_PHONET_PIPE:
3059         case ARPHRD_PPP:
3060         case ARPHRD_NONE:
3061                 return 0;
3062         case ARPHRD_6LOWPAN:
3063                 return EUI64_ADDR_LEN;
3064         case ARPHRD_FDDI:
3065                 return FDDI_K_ALEN;
3066         case ARPHRD_HIPPI:
3067                 return HIPPI_ALEN;
3068         case ARPHRD_IEEE802:
3069                 return FC_ALEN;
3070         case ARPHRD_ROSE:
3071                 return ROSE_ADDR_LEN;
3072         case ARPHRD_NETROM:
3073                 return AX25_ADDR_LEN;
3074         case ARPHRD_LOCALTLK:
3075                 return LTALK_ALEN;
3076         default:
3077                 return 0;
3078         }
3079 }
3080
3081 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3082                             unsigned long arg, int ifreq_len)
3083 {
3084         struct tun_file *tfile = file->private_data;
3085         struct net *net = sock_net(&tfile->sk);
3086         struct tun_struct *tun;
3087         void __user* argp = (void __user*)arg;
3088         unsigned int carrier;
3089         struct ifreq ifr;
3090         kuid_t owner;
3091         kgid_t group;
3092         int ifindex;
3093         int sndbuf;
3094         int vnet_hdr_sz;
3095         int le;
3096         int ret;
3097         bool do_notify = false;
3098
3099         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3100             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3101                 if (copy_from_user(&ifr, argp, ifreq_len))
3102                         return -EFAULT;
3103         } else {
3104                 memset(&ifr, 0, sizeof(ifr));
3105         }
3106         if (cmd == TUNGETFEATURES) {
3107                 /* Currently this just means: "what IFF flags are valid?".
3108                  * This is needed because we never checked for invalid flags on
3109                  * TUNSETIFF.
3110                  */
3111                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3112                                 TUN_FEATURES, (unsigned int __user*)argp);
3113         } else if (cmd == TUNSETQUEUE) {
3114                 return tun_set_queue(file, &ifr);
3115         } else if (cmd == SIOCGSKNS) {
3116                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3117                         return -EPERM;
3118                 return open_related_ns(&net->ns, get_net_ns);
3119         }
3120
3121         rtnl_lock();
3122
3123         tun = tun_get(tfile);
3124         if (cmd == TUNSETIFF) {
3125                 ret = -EEXIST;
3126                 if (tun)
3127                         goto unlock;
3128
3129                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3130
3131                 ret = tun_set_iff(net, file, &ifr);
3132
3133                 if (ret)
3134                         goto unlock;
3135
3136                 if (copy_to_user(argp, &ifr, ifreq_len))
3137                         ret = -EFAULT;
3138                 goto unlock;
3139         }
3140         if (cmd == TUNSETIFINDEX) {
3141                 ret = -EPERM;
3142                 if (tun)
3143                         goto unlock;
3144
3145                 ret = -EFAULT;
3146                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3147                         goto unlock;
3148                 ret = -EINVAL;
3149                 if (ifindex < 0)
3150                         goto unlock;
3151                 ret = 0;
3152                 tfile->ifindex = ifindex;
3153                 goto unlock;
3154         }
3155
3156         ret = -EBADFD;
3157         if (!tun)
3158                 goto unlock;
3159
3160         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3161
3162         net = dev_net(tun->dev);
3163         ret = 0;
3164         switch (cmd) {
3165         case TUNGETIFF:
3166                 tun_get_iff(tun, &ifr);
3167
3168                 if (tfile->detached)
3169                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3170                 if (!tfile->socket.sk->sk_filter)
3171                         ifr.ifr_flags |= IFF_NOFILTER;
3172
3173                 if (copy_to_user(argp, &ifr, ifreq_len))
3174                         ret = -EFAULT;
3175                 break;
3176
3177         case TUNSETNOCSUM:
3178                 /* Disable/Enable checksum */
3179
3180                 /* [unimplemented] */
3181                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3182                            arg ? "disabled" : "enabled");
3183                 break;
3184
3185         case TUNSETPERSIST:
3186                 /* Disable/Enable persist mode. Keep an extra reference to the
3187                  * module to prevent the module being unprobed.
3188                  */
3189                 if (arg && !(tun->flags & IFF_PERSIST)) {
3190                         tun->flags |= IFF_PERSIST;
3191                         __module_get(THIS_MODULE);
3192                         do_notify = true;
3193                 }
3194                 if (!arg && (tun->flags & IFF_PERSIST)) {
3195                         tun->flags &= ~IFF_PERSIST;
3196                         module_put(THIS_MODULE);
3197                         do_notify = true;
3198                 }
3199
3200                 netif_info(tun, drv, tun->dev, "persist %s\n",
3201                            arg ? "enabled" : "disabled");
3202                 break;
3203
3204         case TUNSETOWNER:
3205                 /* Set owner of the device */
3206                 owner = make_kuid(current_user_ns(), arg);
3207                 if (!uid_valid(owner)) {
3208                         ret = -EINVAL;
3209                         break;
3210                 }
3211                 tun->owner = owner;
3212                 do_notify = true;
3213                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3214                            from_kuid(&init_user_ns, tun->owner));
3215                 break;
3216
3217         case TUNSETGROUP:
3218                 /* Set group of the device */
3219                 group = make_kgid(current_user_ns(), arg);
3220                 if (!gid_valid(group)) {
3221                         ret = -EINVAL;
3222                         break;
3223                 }
3224                 tun->group = group;
3225                 do_notify = true;
3226                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3227                            from_kgid(&init_user_ns, tun->group));
3228                 break;
3229
3230         case TUNSETLINK:
3231                 /* Only allow setting the type when the interface is down */
3232                 if (tun->dev->flags & IFF_UP) {
3233                         netif_info(tun, drv, tun->dev,
3234                                    "Linktype set failed because interface is up\n");
3235                         ret = -EBUSY;
3236                 } else {
3237                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3238                                                        tun->dev);
3239                         ret = notifier_to_errno(ret);
3240                         if (ret) {
3241                                 netif_info(tun, drv, tun->dev,
3242                                            "Refused to change device type\n");
3243                                 break;
3244                         }
3245                         tun->dev->type = (int) arg;
3246                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3247                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3248                                    tun->dev->type);
3249                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3250                                                  tun->dev);
3251                 }
3252                 break;
3253
3254         case TUNSETDEBUG:
3255                 tun->msg_enable = (u32)arg;
3256                 break;
3257
3258         case TUNSETOFFLOAD:
3259                 ret = set_offload(tun, arg);
3260                 break;
3261
3262         case TUNSETTXFILTER:
3263                 /* Can be set only for TAPs */
3264                 ret = -EINVAL;
3265                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3266                         break;
3267                 ret = update_filter(&tun->txflt, (void __user *)arg);
3268                 break;
3269
3270         case SIOCGIFHWADDR:
3271                 /* Get hw address */
3272                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3273                 if (copy_to_user(argp, &ifr, ifreq_len))
3274                         ret = -EFAULT;
3275                 break;
3276
3277         case SIOCSIFHWADDR:
3278                 /* Set hw address */
3279                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3280                 break;
3281
3282         case TUNGETSNDBUF:
3283                 sndbuf = tfile->socket.sk->sk_sndbuf;
3284                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3285                         ret = -EFAULT;
3286                 break;
3287
3288         case TUNSETSNDBUF:
3289                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3290                         ret = -EFAULT;
3291                         break;
3292                 }
3293                 if (sndbuf <= 0) {
3294                         ret = -EINVAL;
3295                         break;
3296                 }
3297
3298                 tun->sndbuf = sndbuf;
3299                 tun_set_sndbuf(tun);
3300                 break;
3301
3302         case TUNGETVNETHDRSZ:
3303                 vnet_hdr_sz = tun->vnet_hdr_sz;
3304                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3305                         ret = -EFAULT;
3306                 break;
3307
3308         case TUNSETVNETHDRSZ:
3309                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3310                         ret = -EFAULT;
3311                         break;
3312                 }
3313                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3314                         ret = -EINVAL;
3315                         break;
3316                 }
3317
3318                 tun->vnet_hdr_sz = vnet_hdr_sz;
3319                 break;
3320
3321         case TUNGETVNETLE:
3322                 le = !!(tun->flags & TUN_VNET_LE);
3323                 if (put_user(le, (int __user *)argp))
3324                         ret = -EFAULT;
3325                 break;
3326
3327         case TUNSETVNETLE:
3328                 if (get_user(le, (int __user *)argp)) {
3329                         ret = -EFAULT;
3330                         break;
3331                 }
3332                 if (le)
3333                         tun->flags |= TUN_VNET_LE;
3334                 else
3335                         tun->flags &= ~TUN_VNET_LE;
3336                 break;
3337
3338         case TUNGETVNETBE:
3339                 ret = tun_get_vnet_be(tun, argp);
3340                 break;
3341
3342         case TUNSETVNETBE:
3343                 ret = tun_set_vnet_be(tun, argp);
3344                 break;
3345
3346         case TUNATTACHFILTER:
3347                 /* Can be set only for TAPs */
3348                 ret = -EINVAL;
3349                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3350                         break;
3351                 ret = -EFAULT;
3352                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3353                         break;
3354
3355                 ret = tun_attach_filter(tun);
3356                 break;
3357
3358         case TUNDETACHFILTER:
3359                 /* Can be set only for TAPs */
3360                 ret = -EINVAL;
3361                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3362                         break;
3363                 ret = 0;
3364                 tun_detach_filter(tun, tun->numqueues);
3365                 break;
3366
3367         case TUNGETFILTER:
3368                 ret = -EINVAL;
3369                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3370                         break;
3371                 ret = -EFAULT;
3372                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3373                         break;
3374                 ret = 0;
3375                 break;
3376
3377         case TUNSETSTEERINGEBPF:
3378                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3379                 break;
3380
3381         case TUNSETFILTEREBPF:
3382                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3383                 break;
3384
3385         case TUNSETCARRIER:
3386                 ret = -EFAULT;
3387                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3388                         goto unlock;
3389
3390                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3391                 break;
3392
3393         case TUNGETDEVNETNS:
3394                 ret = -EPERM;
3395                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3396                         goto unlock;
3397                 ret = open_related_ns(&net->ns, get_net_ns);
3398                 break;
3399
3400         default:
3401                 ret = -EINVAL;
3402                 break;
3403         }
3404
3405         if (do_notify)
3406                 netdev_state_change(tun->dev);
3407
3408 unlock:
3409         rtnl_unlock();
3410         if (tun)
3411                 tun_put(tun);
3412         return ret;
3413 }
3414
3415 static long tun_chr_ioctl(struct file *file,
3416                           unsigned int cmd, unsigned long arg)
3417 {
3418         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3419 }
3420
3421 #ifdef CONFIG_COMPAT
3422 static long tun_chr_compat_ioctl(struct file *file,
3423                          unsigned int cmd, unsigned long arg)
3424 {
3425         switch (cmd) {
3426         case TUNSETIFF:
3427         case TUNGETIFF:
3428         case TUNSETTXFILTER:
3429         case TUNGETSNDBUF:
3430         case TUNSETSNDBUF:
3431         case SIOCGIFHWADDR:
3432         case SIOCSIFHWADDR:
3433                 arg = (unsigned long)compat_ptr(arg);
3434                 break;
3435         default:
3436                 arg = (compat_ulong_t)arg;
3437                 break;
3438         }
3439
3440         /*
3441          * compat_ifreq is shorter than ifreq, so we must not access beyond
3442          * the end of that structure. All fields that are used in this
3443          * driver are compatible though, we don't need to convert the
3444          * contents.
3445          */
3446         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3447 }
3448 #endif /* CONFIG_COMPAT */
3449
3450 static int tun_chr_fasync(int fd, struct file *file, int on)
3451 {
3452         struct tun_file *tfile = file->private_data;
3453         int ret;
3454
3455         if (on) {
3456                 ret = file_f_owner_allocate(file);
3457                 if (ret)
3458                         goto out;
3459         }
3460
3461         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3462                 goto out;
3463
3464         if (on) {
3465                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3466                 tfile->flags |= TUN_FASYNC;
3467         } else
3468                 tfile->flags &= ~TUN_FASYNC;
3469         ret = 0;
3470 out:
3471         return ret;
3472 }
3473
3474 static int tun_chr_open(struct inode *inode, struct file * file)
3475 {
3476         struct net *net = current->nsproxy->net_ns;
3477         struct tun_file *tfile;
3478
3479         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3480                                             &tun_proto, 0);
3481         if (!tfile)
3482                 return -ENOMEM;
3483         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3484                 sk_free(&tfile->sk);
3485                 return -ENOMEM;
3486         }
3487
3488         mutex_init(&tfile->napi_mutex);
3489         RCU_INIT_POINTER(tfile->tun, NULL);
3490         tfile->flags = 0;
3491         tfile->ifindex = 0;
3492
3493         init_waitqueue_head(&tfile->socket.wq.wait);
3494
3495         tfile->socket.file = file;
3496         tfile->socket.ops = &tun_socket_ops;
3497
3498         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3499
3500         tfile->sk.sk_write_space = tun_sock_write_space;
3501         tfile->sk.sk_sndbuf = INT_MAX;
3502
3503         file->private_data = tfile;
3504         INIT_LIST_HEAD(&tfile->next);
3505
3506         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3507
3508         /* tun groks IOCB_NOWAIT just fine, mark it as such */
3509         file->f_mode |= FMODE_NOWAIT;
3510         return 0;
3511 }
3512
3513 static int tun_chr_close(struct inode *inode, struct file *file)
3514 {
3515         struct tun_file *tfile = file->private_data;
3516
3517         tun_detach(tfile, true);
3518
3519         return 0;
3520 }
3521
3522 #ifdef CONFIG_PROC_FS
3523 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3524 {
3525         struct tun_file *tfile = file->private_data;
3526         struct tun_struct *tun;
3527         struct ifreq ifr;
3528
3529         memset(&ifr, 0, sizeof(ifr));
3530
3531         rtnl_lock();
3532         tun = tun_get(tfile);
3533         if (tun)
3534                 tun_get_iff(tun, &ifr);
3535         rtnl_unlock();
3536
3537         if (tun)
3538                 tun_put(tun);
3539
3540         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3541 }
3542 #endif
3543
3544 static const struct file_operations tun_fops = {
3545         .owner  = THIS_MODULE,
3546         .read_iter  = tun_chr_read_iter,
3547         .write_iter = tun_chr_write_iter,
3548         .poll   = tun_chr_poll,
3549         .unlocked_ioctl = tun_chr_ioctl,
3550 #ifdef CONFIG_COMPAT
3551         .compat_ioctl = tun_chr_compat_ioctl,
3552 #endif
3553         .open   = tun_chr_open,
3554         .release = tun_chr_close,
3555         .fasync = tun_chr_fasync,
3556 #ifdef CONFIG_PROC_FS
3557         .show_fdinfo = tun_chr_show_fdinfo,
3558 #endif
3559 };
3560
3561 static struct miscdevice tun_miscdev = {
3562         .minor = TUN_MINOR,
3563         .name = "tun",
3564         .nodename = "net/tun",
3565         .fops = &tun_fops,
3566 };
3567
3568 /* ethtool interface */
3569
3570 static void tun_default_link_ksettings(struct net_device *dev,
3571                                        struct ethtool_link_ksettings *cmd)
3572 {
3573         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3574         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3575         cmd->base.speed         = SPEED_10000;
3576         cmd->base.duplex        = DUPLEX_FULL;
3577         cmd->base.port          = PORT_TP;
3578         cmd->base.phy_address   = 0;
3579         cmd->base.autoneg       = AUTONEG_DISABLE;
3580 }
3581
3582 static int tun_get_link_ksettings(struct net_device *dev,
3583                                   struct ethtool_link_ksettings *cmd)
3584 {
3585         struct tun_struct *tun = netdev_priv(dev);
3586
3587         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3588         return 0;
3589 }
3590
3591 static int tun_set_link_ksettings(struct net_device *dev,
3592                                   const struct ethtool_link_ksettings *cmd)
3593 {
3594         struct tun_struct *tun = netdev_priv(dev);
3595
3596         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3597         return 0;
3598 }
3599
3600 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3601 {
3602         struct tun_struct *tun = netdev_priv(dev);
3603
3604         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3605         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3606
3607         switch (tun->flags & TUN_TYPE_MASK) {
3608         case IFF_TUN:
3609                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3610                 break;
3611         case IFF_TAP:
3612                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3613                 break;
3614         }
3615 }
3616
3617 static u32 tun_get_msglevel(struct net_device *dev)
3618 {
3619         struct tun_struct *tun = netdev_priv(dev);
3620
3621         return tun->msg_enable;
3622 }
3623
3624 static void tun_set_msglevel(struct net_device *dev, u32 value)
3625 {
3626         struct tun_struct *tun = netdev_priv(dev);
3627
3628         tun->msg_enable = value;
3629 }
3630
3631 static int tun_get_coalesce(struct net_device *dev,
3632                             struct ethtool_coalesce *ec,
3633                             struct kernel_ethtool_coalesce *kernel_coal,
3634                             struct netlink_ext_ack *extack)
3635 {
3636         struct tun_struct *tun = netdev_priv(dev);
3637
3638         ec->rx_max_coalesced_frames = tun->rx_batched;
3639
3640         return 0;
3641 }
3642
3643 static int tun_set_coalesce(struct net_device *dev,
3644                             struct ethtool_coalesce *ec,
3645                             struct kernel_ethtool_coalesce *kernel_coal,
3646                             struct netlink_ext_ack *extack)
3647 {
3648         struct tun_struct *tun = netdev_priv(dev);
3649
3650         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3651                 tun->rx_batched = NAPI_POLL_WEIGHT;
3652         else
3653                 tun->rx_batched = ec->rx_max_coalesced_frames;
3654
3655         return 0;
3656 }
3657
3658 static void tun_get_channels(struct net_device *dev,
3659                              struct ethtool_channels *channels)
3660 {
3661         struct tun_struct *tun = netdev_priv(dev);
3662
3663         channels->combined_count = tun->numqueues;
3664         channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3665 }
3666
3667 static const struct ethtool_ops tun_ethtool_ops = {
3668         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3669         .get_drvinfo    = tun_get_drvinfo,
3670         .get_msglevel   = tun_get_msglevel,
3671         .set_msglevel   = tun_set_msglevel,
3672         .get_link       = ethtool_op_get_link,
3673         .get_channels   = tun_get_channels,
3674         .get_ts_info    = ethtool_op_get_ts_info,
3675         .get_coalesce   = tun_get_coalesce,
3676         .set_coalesce   = tun_set_coalesce,
3677         .get_link_ksettings = tun_get_link_ksettings,
3678         .set_link_ksettings = tun_set_link_ksettings,
3679 };
3680
3681 static int tun_queue_resize(struct tun_struct *tun)
3682 {
3683         struct net_device *dev = tun->dev;
3684         struct tun_file *tfile;
3685         struct ptr_ring **rings;
3686         int n = tun->numqueues + tun->numdisabled;
3687         int ret, i;
3688
3689         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3690         if (!rings)
3691                 return -ENOMEM;
3692
3693         for (i = 0; i < tun->numqueues; i++) {
3694                 tfile = rtnl_dereference(tun->tfiles[i]);
3695                 rings[i] = &tfile->tx_ring;
3696         }
3697         list_for_each_entry(tfile, &tun->disabled, next)
3698                 rings[i++] = &tfile->tx_ring;
3699
3700         ret = ptr_ring_resize_multiple(rings, n,
3701                                        dev->tx_queue_len, GFP_KERNEL,
3702                                        tun_ptr_free);
3703
3704         kfree(rings);
3705         return ret;
3706 }
3707
3708 static int tun_device_event(struct notifier_block *unused,
3709                             unsigned long event, void *ptr)
3710 {
3711         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3712         struct tun_struct *tun = netdev_priv(dev);
3713         int i;
3714
3715         if (dev->rtnl_link_ops != &tun_link_ops)
3716                 return NOTIFY_DONE;
3717
3718         switch (event) {
3719         case NETDEV_CHANGE_TX_QUEUE_LEN:
3720                 if (tun_queue_resize(tun))
3721                         return NOTIFY_BAD;
3722                 break;
3723         case NETDEV_UP:
3724                 for (i = 0; i < tun->numqueues; i++) {
3725                         struct tun_file *tfile;
3726
3727                         tfile = rtnl_dereference(tun->tfiles[i]);
3728                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3729                 }
3730                 break;
3731         default:
3732                 break;
3733         }
3734
3735         return NOTIFY_DONE;
3736 }
3737
3738 static struct notifier_block tun_notifier_block __read_mostly = {
3739         .notifier_call  = tun_device_event,
3740 };
3741
3742 static int __init tun_init(void)
3743 {
3744         int ret = 0;
3745
3746         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3747
3748         ret = rtnl_link_register(&tun_link_ops);
3749         if (ret) {
3750                 pr_err("Can't register link_ops\n");
3751                 goto err_linkops;
3752         }
3753
3754         ret = misc_register(&tun_miscdev);
3755         if (ret) {
3756                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3757                 goto err_misc;
3758         }
3759
3760         ret = register_netdevice_notifier(&tun_notifier_block);
3761         if (ret) {
3762                 pr_err("Can't register netdevice notifier\n");
3763                 goto err_notifier;
3764         }
3765
3766         return  0;
3767
3768 err_notifier:
3769         misc_deregister(&tun_miscdev);
3770 err_misc:
3771         rtnl_link_unregister(&tun_link_ops);
3772 err_linkops:
3773         return ret;
3774 }
3775
3776 static void __exit tun_cleanup(void)
3777 {
3778         misc_deregister(&tun_miscdev);
3779         rtnl_link_unregister(&tun_link_ops);
3780         unregister_netdevice_notifier(&tun_notifier_block);
3781 }
3782
3783 /* Get an underlying socket object from tun file.  Returns error unless file is
3784  * attached to a device.  The returned object works like a packet socket, it
3785  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3786  * holding a reference to the file for as long as the socket is in use. */
3787 struct socket *tun_get_socket(struct file *file)
3788 {
3789         struct tun_file *tfile;
3790         if (file->f_op != &tun_fops)
3791                 return ERR_PTR(-EINVAL);
3792         tfile = file->private_data;
3793         if (!tfile)
3794                 return ERR_PTR(-EBADFD);
3795         return &tfile->socket;
3796 }
3797 EXPORT_SYMBOL_GPL(tun_get_socket);
3798
3799 struct ptr_ring *tun_get_tx_ring(struct file *file)
3800 {
3801         struct tun_file *tfile;
3802
3803         if (file->f_op != &tun_fops)
3804                 return ERR_PTR(-EINVAL);
3805         tfile = file->private_data;
3806         if (!tfile)
3807                 return ERR_PTR(-EBADFD);
3808         return &tfile->tx_ring;
3809 }
3810 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3811
3812 module_init(tun_init);
3813 module_exit(tun_cleanup);
3814 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3815 MODULE_AUTHOR(DRV_COPYRIGHT);
3816 MODULE_LICENSE("GPL");
3817 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3818 MODULE_ALIAS("devname:net/tun");