drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <underley@underley.eu.org>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212         /* init args */
 213         struct file *file;
 214         struct ifreq *ifr;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static void tun_flow_init(struct tun_struct *tun);
 223 static void tun_flow_uninit(struct tun_struct *tun);
 224
 225 static int tun_napi_receive(struct napi_struct *napi, int budget)
 226 {
 227         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 228         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 229         struct sk_buff_head process_queue;
 230         struct sk_buff *skb;
 231         int received = 0;
 232
 233         __skb_queue_head_init(&process_queue);
 234
 235         spin_lock(&queue->lock);
 236         skb_queue_splice_tail_init(queue, &process_queue);
 237         spin_unlock(&queue->lock);
 238
 239         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 240                 napi_gro_receive(napi, skb);
 241                 ++received;
 242         }
 243
 244         if (!skb_queue_empty(&process_queue)) {
 245                 spin_lock(&queue->lock);
 246                 skb_queue_splice(&process_queue, queue);
 247                 spin_unlock(&queue->lock);
 248         }
 249
 250         return received;
 251 }
 252
 253 static int tun_napi_poll(struct napi_struct *napi, int budget)
 254 {
 255         unsigned int received;
 256
 257         received = tun_napi_receive(napi, budget);
 258
 259         if (received < budget)
 260                 napi_complete_done(napi, received);
 261
 262         return received;
 263 }
 264
 265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 266                           bool napi_en, bool napi_frags)
 267 {
 268         tfile->napi_enabled = napi_en;
 269         tfile->napi_frags_enabled = napi_en && napi_frags;
 270         if (napi_en) {
 271                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 272                                   NAPI_POLL_WEIGHT);
 273                 napi_enable(&tfile->napi);
 274         }
 275 }
 276
 277 static void tun_napi_enable(struct tun_file *tfile)
 278 {
 279         if (tfile->napi_enabled)
 280                 napi_enable(&tfile->napi);
 281 }
 282
 283 static void tun_napi_disable(struct tun_file *tfile)
 284 {
 285         if (tfile->napi_enabled)
 286                 napi_disable(&tfile->napi);
 287 }
 288
 289 static void tun_napi_del(struct tun_file *tfile)
 290 {
 291         if (tfile->napi_enabled)
 292                 netif_napi_del(&tfile->napi);
 293 }
 294
 295 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 296 {
 297         return tfile->napi_frags_enabled;
 298 }
 299
 300 #ifdef CONFIG_TUN_VNET_CROSS_LE
 301 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 302 {
 303         return tun->flags & TUN_VNET_BE ? false :
 304                 virtio_legacy_is_little_endian();
 305 }
 306
 307 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 308 {
 309         int be = !!(tun->flags & TUN_VNET_BE);
 310
 311         if (put_user(be, argp))
 312                 return -EFAULT;
 313
 314         return 0;
 315 }
 316
 317 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 318 {
 319         int be;
 320
 321         if (get_user(be, argp))
 322                 return -EFAULT;
 323
 324         if (be)
 325                 tun->flags |= TUN_VNET_BE;
 326         else
 327                 tun->flags &= ~TUN_VNET_BE;
 328
 329         return 0;
 330 }
 331 #else
 332 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 333 {
 334         return virtio_legacy_is_little_endian();
 335 }
 336
 337 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 338 {
 339         return -EINVAL;
 340 }
 341
 342 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 343 {
 344         return -EINVAL;
 345 }
 346 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 347
 348 static inline bool tun_is_little_endian(struct tun_struct *tun)
 349 {
 350         return tun->flags & TUN_VNET_LE ||
 351                 tun_legacy_is_little_endian(tun);
 352 }
 353
 354 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 355 {
 356         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 357 }
 358
 359 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 360 {
 361         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 362 }
 363
 364 static inline u32 tun_hashfn(u32 rxhash)
 365 {
 366         return rxhash & TUN_MASK_FLOW_ENTRIES;
 367 }
 368
 369 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 370 {
 371         struct tun_flow_entry *e;
 372
 373         hlist_for_each_entry_rcu(e, head, hash_link) {
 374                 if (e->rxhash == rxhash)
 375                         return e;
 376         }
 377         return NULL;
 378 }
 379
 380 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 381                                               struct hlist_head *head,
 382                                               u32 rxhash, u16 queue_index)
 383 {
 384         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 385
 386         if (e) {
 387                 netif_info(tun, tx_queued, tun->dev,
 388                            "create flow: hash %u index %u\n",
 389                            rxhash, queue_index);
 390                 e->updated = jiffies;
 391                 e->rxhash = rxhash;
 392                 e->rps_rxhash = 0;
 393                 e->queue_index = queue_index;
 394                 e->tun = tun;
 395                 hlist_add_head_rcu(&e->hash_link, head);
 396                 ++tun->flow_count;
 397         }
 398         return e;
 399 }
 400
 401 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 402 {
 403         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 404                    e->rxhash, e->queue_index);
 405         hlist_del_rcu(&e->hash_link);
 406         kfree_rcu(e, rcu);
 407         --tun->flow_count;
 408 }
 409
 410 static void tun_flow_flush(struct tun_struct *tun)
 411 {
 412         int i;
 413
 414         spin_lock_bh(&tun->lock);
 415         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 416                 struct tun_flow_entry *e;
 417                 struct hlist_node *n;
 418
 419                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 420                         tun_flow_delete(tun, e);
 421         }
 422         spin_unlock_bh(&tun->lock);
 423 }
 424
 425 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 426 {
 427         int i;
 428
 429         spin_lock_bh(&tun->lock);
 430         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 431                 struct tun_flow_entry *e;
 432                 struct hlist_node *n;
 433
 434                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 435                         if (e->queue_index == queue_index)
 436                                 tun_flow_delete(tun, e);
 437                 }
 438         }
 439         spin_unlock_bh(&tun->lock);
 440 }
 441
 442 static void tun_flow_cleanup(struct timer_list *t)
 443 {
 444         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 445         unsigned long delay = tun->ageing_time;
 446         unsigned long next_timer = jiffies + delay;
 447         unsigned long count = 0;
 448         int i;
 449
 450         spin_lock(&tun->lock);
 451         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 452                 struct tun_flow_entry *e;
 453                 struct hlist_node *n;
 454
 455                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 456                         unsigned long this_timer;
 457
 458                         this_timer = e->updated + delay;
 459                         if (time_before_eq(this_timer, jiffies)) {
 460                                 tun_flow_delete(tun, e);
 461                                 continue;
 462                         }
 463                         count++;
 464                         if (time_before(this_timer, next_timer))
 465                                 next_timer = this_timer;
 466                 }
 467         }
 468
 469         if (count)
 470                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 471         spin_unlock(&tun->lock);
 472 }
 473
 474 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 475                             struct tun_file *tfile)
 476 {
 477         struct hlist_head *head;
 478         struct tun_flow_entry *e;
 479         unsigned long delay = tun->ageing_time;
 480         u16 queue_index = tfile->queue_index;
 481
 482         head = &tun->flows[tun_hashfn(rxhash)];
 483
 484         rcu_read_lock();
 485
 486         e = tun_flow_find(head, rxhash);
 487         if (likely(e)) {
 488                 /* TODO: keep queueing to old queue until it's empty? */
 489                 if (READ_ONCE(e->queue_index) != queue_index)
 490                         WRITE_ONCE(e->queue_index, queue_index);
 491                 if (e->updated != jiffies)
 492                         e->updated = jiffies;
 493                 sock_rps_record_flow_hash(e->rps_rxhash);
 494         } else {
 495                 spin_lock_bh(&tun->lock);
 496                 if (!tun_flow_find(head, rxhash) &&
 497                     tun->flow_count < MAX_TAP_FLOWS)
 498                         tun_flow_create(tun, head, rxhash, queue_index);
 499
 500                 if (!timer_pending(&tun->flow_gc_timer))
 501                         mod_timer(&tun->flow_gc_timer,
 502                                   round_jiffies_up(jiffies + delay));
 503                 spin_unlock_bh(&tun->lock);
 504         }
 505
 506         rcu_read_unlock();
 507 }
 508
 509 /* Save the hash received in the stack receive path and update the
 510  * flow_hash table accordingly.
 511  */
 512 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 513 {
 514         if (unlikely(e->rps_rxhash != hash))
 515                 e->rps_rxhash = hash;
 516 }
 517
 518 /* We try to identify a flow through its rxhash. The reason that
 519  * we do not check rxq no. is because some cards(e.g 82599), chooses
 520  * the rxq based on the txq where the last packet of the flow comes. As
 521  * the userspace application move between processors, we may get a
 522  * different rxq no. here.
 523  */
 524 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 525 {
 526         struct tun_flow_entry *e;
 527         u32 txq = 0;
 528         u32 numqueues = 0;
 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                 /* use multiply and shift instead of expensive divide */
 539                 txq = ((u64)txq * numqueues) >> 32;
 540         }
 541
 542         return txq;
 543 }
 544
 545 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 546 {
 547         struct tun_prog *prog;
 548         u32 numqueues;
 549         u16 ret = 0;
 550
 551         numqueues = READ_ONCE(tun->numqueues);
 552         if (!numqueues)
 553                 return 0;
 554
 555         prog = rcu_dereference(tun->steering_prog);
 556         if (prog)
 557                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 558
 559         return ret % numqueues;
 560 }
 561
 562 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 563                             struct net_device *sb_dev)
 564 {
 565         struct tun_struct *tun = netdev_priv(dev);
 566         u16 ret;
 567
 568         rcu_read_lock();
 569         if (rcu_dereference(tun->steering_prog))
 570                 ret = tun_ebpf_select_queue(tun, skb);
 571         else
 572                 ret = tun_automq_select_queue(tun, skb);
 573         rcu_read_unlock();
 574
 575         return ret;
 576 }
 577
 578 static inline bool tun_not_capable(struct tun_struct *tun)
 579 {
 580         const struct cred *cred = current_cred();
 581         struct net *net = dev_net(tun->dev);
 582
 583         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 584                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 585                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 586 }
 587
 588 static void tun_set_real_num_queues(struct tun_struct *tun)
 589 {
 590         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 591         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 592 }
 593
 594 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 595 {
 596         tfile->detached = tun;
 597         list_add_tail(&tfile->next, &tun->disabled);
 598         ++tun->numdisabled;
 599 }
 600
 601 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 602 {
 603         struct tun_struct *tun = tfile->detached;
 604
 605         tfile->detached = NULL;
 606         list_del_init(&tfile->next);
 607         --tun->numdisabled;
 608         return tun;
 609 }
 610
 611 void tun_ptr_free(void *ptr)
 612 {
 613         if (!ptr)
 614                 return;
 615         if (tun_is_xdp_frame(ptr)) {
 616                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 617
 618                 xdp_return_frame(xdpf);
 619         } else {
 620                 __skb_array_destroy_skb(ptr);
 621         }
 622 }
 623 EXPORT_SYMBOL_GPL(tun_ptr_free);
 624
 625 static void tun_queue_purge(struct tun_file *tfile)
 626 {
 627         void *ptr;
 628
 629         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 630                 tun_ptr_free(ptr);
 631
 632         skb_queue_purge(&tfile->sk.sk_write_queue);
 633         skb_queue_purge(&tfile->sk.sk_error_queue);
 634 }
 635
 636 static void __tun_detach(struct tun_file *tfile, bool clean)
 637 {
 638         struct tun_file *ntfile;
 639         struct tun_struct *tun;
 640
 641         tun = rtnl_dereference(tfile->tun);
 642
 643         if (tun && clean) {
 644                 if (!tfile->detached)
 645                         tun_napi_disable(tfile);
 646                 tun_napi_del(tfile);
 647         }
 648
 649         if (tun && !tfile->detached) {
 650                 u16 index = tfile->queue_index;
 651                 BUG_ON(index >= tun->numqueues);
 652
 653                 rcu_assign_pointer(tun->tfiles[index],
 654                                    tun->tfiles[tun->numqueues - 1]);
 655                 ntfile = rtnl_dereference(tun->tfiles[index]);
 656                 ntfile->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                 sock_put(&tfile->sk);
 691         }
 692 }
 693
 694 static void tun_detach(struct tun_file *tfile, bool clean)
 695 {
 696         struct tun_struct *tun;
 697         struct net_device *dev;
 698
 699         rtnl_lock();
 700         tun = rtnl_dereference(tfile->tun);
 701         dev = tun ? tun->dev : NULL;
 702         __tun_detach(tfile, clean);
 703         if (dev)
 704                 netdev_state_change(dev);
 705         rtnl_unlock();
 706 }
 707
 708 static void tun_detach_all(struct net_device *dev)
 709 {
 710         struct tun_struct *tun = netdev_priv(dev);
 711         struct tun_file *tfile, *tmp;
 712         int i, n = tun->numqueues;
 713
 714         for (i = 0; i < n; i++) {
 715                 tfile = rtnl_dereference(tun->tfiles[i]);
 716                 BUG_ON(!tfile);
 717                 tun_napi_disable(tfile);
 718                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 719                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 720                 RCU_INIT_POINTER(tfile->tun, NULL);
 721                 --tun->numqueues;
 722         }
 723         list_for_each_entry(tfile, &tun->disabled, next) {
 724                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 725                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 726                 RCU_INIT_POINTER(tfile->tun, NULL);
 727         }
 728         BUG_ON(tun->numqueues != 0);
 729
 730         synchronize_net();
 731         for (i = 0; i < n; i++) {
 732                 tfile = rtnl_dereference(tun->tfiles[i]);
 733                 tun_napi_del(tfile);
 734                 /* Drop read queue */
 735                 tun_queue_purge(tfile);
 736                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 737                 sock_put(&tfile->sk);
 738         }
 739         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 740                 tun_napi_del(tfile);
 741                 tun_enable_queue(tfile);
 742                 tun_queue_purge(tfile);
 743                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 744                 sock_put(&tfile->sk);
 745         }
 746         BUG_ON(tun->numdisabled != 0);
 747
 748         if (tun->flags & IFF_PERSIST)
 749                 module_put(THIS_MODULE);
 750 }
 751
 752 static int tun_attach(struct tun_struct *tun, struct file *file,
 753                       bool skip_filter, bool napi, bool napi_frags,
 754                       bool publish_tun)
 755 {
 756         struct tun_file *tfile = file->private_data;
 757         struct net_device *dev = tun->dev;
 758         int err;
 759
 760         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 761         if (err < 0)
 762                 goto out;
 763
 764         err = -EINVAL;
 765         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 766                 goto out;
 767
 768         err = -EBUSY;
 769         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 770                 goto out;
 771
 772         err = -E2BIG;
 773         if (!tfile->detached &&
 774             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 775                 goto out;
 776
 777         err = 0;
 778
 779         /* Re-attach the filter to persist device */
 780         if (!skip_filter && (tun->filter_attached == true)) {
 781                 lock_sock(tfile->socket.sk);
 782                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 783                 release_sock(tfile->socket.sk);
 784                 if (!err)
 785                         goto out;
 786         }
 787
 788         if (!tfile->detached &&
 789             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 790                             GFP_KERNEL, tun_ptr_free)) {
 791                 err = -ENOMEM;
 792                 goto out;
 793         }
 794
 795         tfile->queue_index = tun->numqueues;
 796         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 797
 798         if (tfile->detached) {
 799                 /* Re-attach detached tfile, updating XDP queue_index */
 800                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 801
 802                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 803                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 804         } else {
 805                 /* Setup XDP RX-queue info, for new tfile getting attached */
 806                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 807                                        tun->dev, tfile->queue_index, 0);
 808                 if (err < 0)
 809                         goto out;
 810                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 811                                                  MEM_TYPE_PAGE_SHARED, NULL);
 812                 if (err < 0) {
 813                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 814                         goto out;
 815                 }
 816                 err = 0;
 817         }
 818
 819         if (tfile->detached) {
 820                 tun_enable_queue(tfile);
 821                 tun_napi_enable(tfile);
 822         } else {
 823                 sock_hold(&tfile->sk);
 824                 tun_napi_init(tun, tfile, napi, napi_frags);
 825         }
 826
 827         if (rtnl_dereference(tun->xdp_prog))
 828                 sock_set_flag(&tfile->sk, SOCK_XDP);
 829
 830         /* device is allowed to go away first, so no need to hold extra
 831          * refcnt.
 832          */
 833
 834         /* Publish tfile->tun and tun->tfiles only after we've fully
 835          * initialized tfile; otherwise we risk using half-initialized
 836          * object.
 837          */
 838         if (publish_tun)
 839                 rcu_assign_pointer(tfile->tun, tun);
 840         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 841         tun->numqueues++;
 842         tun_set_real_num_queues(tun);
 843 out:
 844         return err;
 845 }
 846
 847 static struct tun_struct *tun_get(struct tun_file *tfile)
 848 {
 849         struct tun_struct *tun;
 850
 851         rcu_read_lock();
 852         tun = rcu_dereference(tfile->tun);
 853         if (tun)
 854                 dev_hold(tun->dev);
 855         rcu_read_unlock();
 856
 857         return tun;
 858 }
 859
 860 static void tun_put(struct tun_struct *tun)
 861 {
 862         dev_put(tun->dev);
 863 }
 864
 865 /* TAP filtering */
 866 static void addr_hash_set(u32 *mask, const u8 *addr)
 867 {
 868         int n = ether_crc(ETH_ALEN, addr) >> 26;
 869         mask[n >> 5] |= (1 << (n & 31));
 870 }
 871
 872 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 873 {
 874         int n = ether_crc(ETH_ALEN, addr) >> 26;
 875         return mask[n >> 5] & (1 << (n & 31));
 876 }
 877
 878 static int update_filter(struct tap_filter *filter, void __user *arg)
 879 {
 880         struct { u8 u[ETH_ALEN]; } *addr;
 881         struct tun_filter uf;
 882         int err, alen, n, nexact;
 883
 884         if (copy_from_user(&uf, arg, sizeof(uf)))
 885                 return -EFAULT;
 886
 887         if (!uf.count) {
 888                 /* Disabled */
 889                 filter->count = 0;
 890                 return 0;
 891         }
 892
 893         alen = ETH_ALEN * uf.count;
 894         addr = memdup_user(arg + sizeof(uf), alen);
 895         if (IS_ERR(addr))
 896                 return PTR_ERR(addr);
 897
 898         /* The filter is updated without holding any locks. Which is
 899          * perfectly safe. We disable it first and in the worst
 900          * case we'll accept a few undesired packets. */
 901         filter->count = 0;
 902         wmb();
 903
 904         /* Use first set of addresses as an exact filter */
 905         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 906                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 907
 908         nexact = n;
 909
 910         /* Remaining multicast addresses are hashed,
 911          * unicast will leave the filter disabled. */
 912         memset(filter->mask, 0, sizeof(filter->mask));
 913         for (; n < uf.count; n++) {
 914                 if (!is_multicast_ether_addr(addr[n].u)) {
 915                         err = 0; /* no filter */
 916                         goto free_addr;
 917                 }
 918                 addr_hash_set(filter->mask, addr[n].u);
 919         }
 920
 921         /* For ALLMULTI just set the mask to all ones.
 922          * This overrides the mask populated above. */
 923         if ((uf.flags & TUN_FLT_ALLMULTI))
 924                 memset(filter->mask, ~0, sizeof(filter->mask));
 925
 926         /* Now enable the filter */
 927         wmb();
 928         filter->count = nexact;
 929
 930         /* Return the number of exact filters */
 931         err = nexact;
 932 free_addr:
 933         kfree(addr);
 934         return err;
 935 }
 936
 937 /* Returns: 0 - drop, !=0 - accept */
 938 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 939 {
 940         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 941          * at this point. */
 942         struct ethhdr *eh = (struct ethhdr *) skb->data;
 943         int i;
 944
 945         /* Exact match */
 946         for (i = 0; i < filter->count; i++)
 947                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 948                         return 1;
 949
 950         /* Inexact match (multicast only) */
 951         if (is_multicast_ether_addr(eh->h_dest))
 952                 return addr_hash_test(filter->mask, eh->h_dest);
 953
 954         return 0;
 955 }
 956
 957 /*
 958  * Checks whether the packet is accepted or not.
 959  * Returns: 0 - drop, !=0 - accept
 960  */
 961 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 962 {
 963         if (!filter->count)
 964                 return 1;
 965
 966         return run_filter(filter, skb);
 967 }
 968
 969 /* Network device part of the driver */
 970
 971 static const struct ethtool_ops tun_ethtool_ops;
 972
 973 static int tun_net_init(struct net_device *dev)
 974 {
 975         struct tun_struct *tun = netdev_priv(dev);
 976         struct ifreq *ifr = tun->ifr;
 977         int err;
 978
 979         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 980         if (!dev->tstats)
 981                 return -ENOMEM;
 982
 983         spin_lock_init(&tun->lock);
 984
 985         err = security_tun_dev_alloc_security(&tun->security);
 986         if (err < 0) {
 987                 free_percpu(dev->tstats);
 988                 return err;
 989         }
 990
 991         tun_flow_init(tun);
 992
 993         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 994                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 995                            NETIF_F_HW_VLAN_STAG_TX;
 996         dev->features = dev->hw_features | NETIF_F_LLTX;
 997         dev->vlan_features = dev->features &
 998                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 999                                NETIF_F_HW_VLAN_STAG_TX);
1000
1001         tun->flags = (tun->flags & ~TUN_FEATURES) |
1002                       (ifr->ifr_flags & TUN_FEATURES);
1003
1004         INIT_LIST_HEAD(&tun->disabled);
1005         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1006                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1007         if (err < 0) {
1008                 tun_flow_uninit(tun);
1009                 security_tun_dev_free_security(tun->security);
1010                 free_percpu(dev->tstats);
1011                 return err;
1012         }
1013         return 0;
1014 }
1015
1016 /* Net device detach from fd. */
1017 static void tun_net_uninit(struct net_device *dev)
1018 {
1019         tun_detach_all(dev);
1020 }
1021
1022 /* Net device open. */
1023 static int tun_net_open(struct net_device *dev)
1024 {
1025         netif_tx_start_all_queues(dev);
1026
1027         return 0;
1028 }
1029
1030 /* Net device close. */
1031 static int tun_net_close(struct net_device *dev)
1032 {
1033         netif_tx_stop_all_queues(dev);
1034         return 0;
1035 }
1036
1037 /* Net device start xmit */
1038 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1039 {
1040 #ifdef CONFIG_RPS
1041         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1042                 /* Select queue was not called for the skbuff, so we extract the
1043                  * RPS hash and save it into the flow_table here.
1044                  */
1045                 struct tun_flow_entry *e;
1046                 __u32 rxhash;
1047
1048                 rxhash = __skb_get_hash_symmetric(skb);
1049                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1050                 if (e)
1051                         tun_flow_save_rps_rxhash(e, rxhash);
1052         }
1053 #endif
1054 }
1055
1056 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1057                                     struct sk_buff *skb,
1058                                     int len)
1059 {
1060         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1061
1062         if (prog)
1063                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1064
1065         return len;
1066 }
1067
1068 /* Net device start xmit */
1069 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1070 {
1071         struct tun_struct *tun = netdev_priv(dev);
1072         int txq = skb->queue_mapping;
1073         struct netdev_queue *queue;
1074         struct tun_file *tfile;
1075         int len = skb->len;
1076
1077         rcu_read_lock();
1078         tfile = rcu_dereference(tun->tfiles[txq]);
1079
1080         /* Drop packet if interface is not attached */
1081         if (!tfile)
1082                 goto drop;
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                 goto drop;
1094
1095         if (tfile->socket.sk->sk_filter &&
1096             sk_filter(tfile->socket.sk, skb))
1097                 goto drop;
1098
1099         len = run_ebpf_filter(tun, skb, len);
1100         if (len == 0)
1101                 goto drop;
1102
1103         if (pskb_trim(skb, len))
1104                 goto drop;
1105
1106         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1107                 goto drop;
1108
1109         skb_tx_timestamp(skb);
1110
1111         /* Orphan the skb - required as we might hang on to it
1112          * for indefinite time.
1113          */
1114         skb_orphan(skb);
1115
1116         nf_reset_ct(skb);
1117
1118         if (ptr_ring_produce(&tfile->tx_ring, skb))
1119                 goto drop;
1120
1121         /* NETIF_F_LLTX requires to do our own update of trans_start */
1122         queue = netdev_get_tx_queue(dev, txq);
1123         queue->trans_start = jiffies;
1124
1125         /* Notify and wake up reader process */
1126         if (tfile->flags & TUN_FASYNC)
1127                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1128         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1129
1130         rcu_read_unlock();
1131         return NETDEV_TX_OK;
1132
1133 drop:
1134         atomic_long_inc(&dev->tx_dropped);
1135         skb_tx_error(skb);
1136         kfree_skb(skb);
1137         rcu_read_unlock();
1138         return NET_XMIT_DROP;
1139 }
1140
1141 static void tun_net_mclist(struct net_device *dev)
1142 {
1143         /*
1144          * This callback is supposed to deal with mc filter in
1145          * _rx_ path and has nothing to do with the _tx_ path.
1146          * In rx path we always accept everything userspace gives us.
1147          */
1148 }
1149
1150 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1151         netdev_features_t features)
1152 {
1153         struct tun_struct *tun = netdev_priv(dev);
1154
1155         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1156 }
1157
1158 static void tun_set_headroom(struct net_device *dev, int new_hr)
1159 {
1160         struct tun_struct *tun = netdev_priv(dev);
1161
1162         if (new_hr < NET_SKB_PAD)
1163                 new_hr = NET_SKB_PAD;
1164
1165         tun->align = new_hr;
1166 }
1167
1168 static void
1169 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1170 {
1171         struct tun_struct *tun = netdev_priv(dev);
1172
1173         dev_get_tstats64(dev, stats);
1174
1175         stats->rx_frame_errors +=
1176                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1177 }
1178
1179 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1180                        struct netlink_ext_ack *extack)
1181 {
1182         struct tun_struct *tun = netdev_priv(dev);
1183         struct tun_file *tfile;
1184         struct bpf_prog *old_prog;
1185         int i;
1186
1187         old_prog = rtnl_dereference(tun->xdp_prog);
1188         rcu_assign_pointer(tun->xdp_prog, prog);
1189         if (old_prog)
1190                 bpf_prog_put(old_prog);
1191
1192         for (i = 0; i < tun->numqueues; i++) {
1193                 tfile = rtnl_dereference(tun->tfiles[i]);
1194                 if (prog)
1195                         sock_set_flag(&tfile->sk, SOCK_XDP);
1196                 else
1197                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1198         }
1199         list_for_each_entry(tfile, &tun->disabled, next) {
1200                 if (prog)
1201                         sock_set_flag(&tfile->sk, SOCK_XDP);
1202                 else
1203                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1204         }
1205
1206         return 0;
1207 }
1208
1209 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1210 {
1211         switch (xdp->command) {
1212         case XDP_SETUP_PROG:
1213                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1214         default:
1215                 return -EINVAL;
1216         }
1217 }
1218
1219 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1220 {
1221         if (new_carrier) {
1222                 struct tun_struct *tun = netdev_priv(dev);
1223
1224                 if (!tun->numqueues)
1225                         return -EPERM;
1226
1227                 netif_carrier_on(dev);
1228         } else {
1229                 netif_carrier_off(dev);
1230         }
1231         return 0;
1232 }
1233
1234 static const struct net_device_ops tun_netdev_ops = {
1235         .ndo_init               = tun_net_init,
1236         .ndo_uninit             = tun_net_uninit,
1237         .ndo_open               = tun_net_open,
1238         .ndo_stop               = tun_net_close,
1239         .ndo_start_xmit         = tun_net_xmit,
1240         .ndo_fix_features       = tun_net_fix_features,
1241         .ndo_select_queue       = tun_select_queue,
1242         .ndo_set_rx_headroom    = tun_set_headroom,
1243         .ndo_get_stats64        = tun_net_get_stats64,
1244         .ndo_change_carrier     = tun_net_change_carrier,
1245 };
1246
1247 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1248 {
1249         /* Notify and wake up reader process */
1250         if (tfile->flags & TUN_FASYNC)
1251                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1252         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1253 }
1254
1255 static int tun_xdp_xmit(struct net_device *dev, int n,
1256                         struct xdp_frame **frames, u32 flags)
1257 {
1258         struct tun_struct *tun = netdev_priv(dev);
1259         struct tun_file *tfile;
1260         u32 numqueues;
1261         int nxmit = 0;
1262         int i;
1263
1264         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1265                 return -EINVAL;
1266
1267         rcu_read_lock();
1268
1269 resample:
1270         numqueues = READ_ONCE(tun->numqueues);
1271         if (!numqueues) {
1272                 rcu_read_unlock();
1273                 return -ENXIO; /* Caller will free/return all frames */
1274         }
1275
1276         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1277                                             numqueues]);
1278         if (unlikely(!tfile))
1279                 goto resample;
1280
1281         spin_lock(&tfile->tx_ring.producer_lock);
1282         for (i = 0; i < n; i++) {
1283                 struct xdp_frame *xdp = frames[i];
1284                 /* Encode the XDP flag into lowest bit for consumer to differ
1285                  * XDP buffer from sk_buff.
1286                  */
1287                 void *frame = tun_xdp_to_ptr(xdp);
1288
1289                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1290                         atomic_long_inc(&dev->tx_dropped);
1291                         break;
1292                 }
1293                 nxmit++;
1294         }
1295         spin_unlock(&tfile->tx_ring.producer_lock);
1296
1297         if (flags & XDP_XMIT_FLUSH)
1298                 __tun_xdp_flush_tfile(tfile);
1299
1300         rcu_read_unlock();
1301         return nxmit;
1302 }
1303
1304 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1305 {
1306         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1307         int nxmit;
1308
1309         if (unlikely(!frame))
1310                 return -EOVERFLOW;
1311
1312         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1313         if (!nxmit)
1314                 xdp_return_frame_rx_napi(frame);
1315         return nxmit;
1316 }
1317
1318 static const struct net_device_ops tap_netdev_ops = {
1319         .ndo_init               = tun_net_init,
1320         .ndo_uninit             = tun_net_uninit,
1321         .ndo_open               = tun_net_open,
1322         .ndo_stop               = tun_net_close,
1323         .ndo_start_xmit         = tun_net_xmit,
1324         .ndo_fix_features       = tun_net_fix_features,
1325         .ndo_set_rx_mode        = tun_net_mclist,
1326         .ndo_set_mac_address    = eth_mac_addr,
1327         .ndo_validate_addr      = eth_validate_addr,
1328         .ndo_select_queue       = tun_select_queue,
1329         .ndo_features_check     = passthru_features_check,
1330         .ndo_set_rx_headroom    = tun_set_headroom,
1331         .ndo_get_stats64        = dev_get_tstats64,
1332         .ndo_bpf                = tun_xdp,
1333         .ndo_xdp_xmit           = tun_xdp_xmit,
1334         .ndo_change_carrier     = tun_net_change_carrier,
1335 };
1336
1337 static void tun_flow_init(struct tun_struct *tun)
1338 {
1339         int i;
1340
1341         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1342                 INIT_HLIST_HEAD(&tun->flows[i]);
1343
1344         tun->ageing_time = TUN_FLOW_EXPIRE;
1345         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1346         mod_timer(&tun->flow_gc_timer,
1347                   round_jiffies_up(jiffies + tun->ageing_time));
1348 }
1349
1350 static void tun_flow_uninit(struct tun_struct *tun)
1351 {
1352         del_timer_sync(&tun->flow_gc_timer);
1353         tun_flow_flush(tun);
1354 }
1355
1356 #define MIN_MTU 68
1357 #define MAX_MTU 65535
1358
1359 /* Initialize net device. */
1360 static void tun_net_initialize(struct net_device *dev)
1361 {
1362         struct tun_struct *tun = netdev_priv(dev);
1363
1364         switch (tun->flags & TUN_TYPE_MASK) {
1365         case IFF_TUN:
1366                 dev->netdev_ops = &tun_netdev_ops;
1367                 dev->header_ops = &ip_tunnel_header_ops;
1368
1369                 /* Point-to-Point TUN Device */
1370                 dev->hard_header_len = 0;
1371                 dev->addr_len = 0;
1372                 dev->mtu = 1500;
1373
1374                 /* Zero header length */
1375                 dev->type = ARPHRD_NONE;
1376                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1377                 break;
1378
1379         case IFF_TAP:
1380                 dev->netdev_ops = &tap_netdev_ops;
1381                 /* Ethernet TAP Device */
1382                 ether_setup(dev);
1383                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1384                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1385
1386                 eth_hw_addr_random(dev);
1387
1388                 break;
1389         }
1390
1391         dev->min_mtu = MIN_MTU;
1392         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1393 }
1394
1395 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1396 {
1397         struct sock *sk = tfile->socket.sk;
1398
1399         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1400 }
1401
1402 /* Character device part */
1403
1404 /* Poll */
1405 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1406 {
1407         struct tun_file *tfile = file->private_data;
1408         struct tun_struct *tun = tun_get(tfile);
1409         struct sock *sk;
1410         __poll_t mask = 0;
1411
1412         if (!tun)
1413                 return EPOLLERR;
1414
1415         sk = tfile->socket.sk;
1416
1417         poll_wait(file, sk_sleep(sk), wait);
1418
1419         if (!ptr_ring_empty(&tfile->tx_ring))
1420                 mask |= EPOLLIN | EPOLLRDNORM;
1421
1422         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1423          * guarantee EPOLLOUT to be raised by either here or
1424          * tun_sock_write_space(). Then process could get notification
1425          * after it writes to a down device and meets -EIO.
1426          */
1427         if (tun_sock_writeable(tun, tfile) ||
1428             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1429              tun_sock_writeable(tun, tfile)))
1430                 mask |= EPOLLOUT | EPOLLWRNORM;
1431
1432         if (tun->dev->reg_state != NETREG_REGISTERED)
1433                 mask = EPOLLERR;
1434
1435         tun_put(tun);
1436         return mask;
1437 }
1438
1439 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1440                                             size_t len,
1441                                             const struct iov_iter *it)
1442 {
1443         struct sk_buff *skb;
1444         size_t linear;
1445         int err;
1446         int i;
1447
1448         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1449             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1450                 return ERR_PTR(-EMSGSIZE);
1451
1452         local_bh_disable();
1453         skb = napi_get_frags(&tfile->napi);
1454         local_bh_enable();
1455         if (!skb)
1456                 return ERR_PTR(-ENOMEM);
1457
1458         linear = iov_iter_single_seg_count(it);
1459         err = __skb_grow(skb, linear);
1460         if (err)
1461                 goto free;
1462
1463         skb->len = len;
1464         skb->data_len = len - linear;
1465         skb->truesize += skb->data_len;
1466
1467         for (i = 1; i < it->nr_segs; i++) {
1468                 size_t fragsz = it->iov[i].iov_len;
1469                 struct page *page;
1470                 void *frag;
1471
1472                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1473                         err = -EINVAL;
1474                         goto free;
1475                 }
1476                 frag = netdev_alloc_frag(fragsz);
1477                 if (!frag) {
1478                         err = -ENOMEM;
1479                         goto free;
1480                 }
1481                 page = virt_to_head_page(frag);
1482                 skb_fill_page_desc(skb, i - 1, page,
1483                                    frag - page_address(page), fragsz);
1484         }
1485
1486         return skb;
1487 free:
1488         /* frees skb and all frags allocated with napi_alloc_frag() */
1489         napi_free_frags(&tfile->napi);
1490         return ERR_PTR(err);
1491 }
1492
1493 /* prepad is the amount to reserve at front.  len is length after that.
1494  * linear is a hint as to how much to copy (usually headers). */
1495 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1496                                      size_t prepad, size_t len,
1497                                      size_t linear, int noblock)
1498 {
1499         struct sock *sk = tfile->socket.sk;
1500         struct sk_buff *skb;
1501         int err;
1502
1503         /* Under a page?  Don't bother with paged skb. */
1504         if (prepad + len < PAGE_SIZE || !linear)
1505                 linear = len;
1506
1507         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1508                                    &err, 0);
1509         if (!skb)
1510                 return ERR_PTR(err);
1511
1512         skb_reserve(skb, prepad);
1513         skb_put(skb, linear);
1514         skb->data_len = len - linear;
1515         skb->len += len - linear;
1516
1517         return skb;
1518 }
1519
1520 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1521                            struct sk_buff *skb, int more)
1522 {
1523         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1524         struct sk_buff_head process_queue;
1525         u32 rx_batched = tun->rx_batched;
1526         bool rcv = false;
1527
1528         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1529                 local_bh_disable();
1530                 skb_record_rx_queue(skb, tfile->queue_index);
1531                 netif_receive_skb(skb);
1532                 local_bh_enable();
1533                 return;
1534         }
1535
1536         spin_lock(&queue->lock);
1537         if (!more || skb_queue_len(queue) == rx_batched) {
1538                 __skb_queue_head_init(&process_queue);
1539                 skb_queue_splice_tail_init(queue, &process_queue);
1540                 rcv = true;
1541         } else {
1542                 __skb_queue_tail(queue, skb);
1543         }
1544         spin_unlock(&queue->lock);
1545
1546         if (rcv) {
1547                 struct sk_buff *nskb;
1548
1549                 local_bh_disable();
1550                 while ((nskb = __skb_dequeue(&process_queue))) {
1551                         skb_record_rx_queue(nskb, tfile->queue_index);
1552                         netif_receive_skb(nskb);
1553                 }
1554                 skb_record_rx_queue(skb, tfile->queue_index);
1555                 netif_receive_skb(skb);
1556                 local_bh_enable();
1557         }
1558 }
1559
1560 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1561                               int len, int noblock, bool zerocopy)
1562 {
1563         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1564                 return false;
1565
1566         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1567                 return false;
1568
1569         if (!noblock)
1570                 return false;
1571
1572         if (zerocopy)
1573                 return false;
1574
1575         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1576             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1577                 return false;
1578
1579         return true;
1580 }
1581
1582 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1583                                        struct page_frag *alloc_frag, char *buf,
1584                                        int buflen, int len, int pad)
1585 {
1586         struct sk_buff *skb = build_skb(buf, buflen);
1587
1588         if (!skb)
1589                 return ERR_PTR(-ENOMEM);
1590
1591         skb_reserve(skb, pad);
1592         skb_put(skb, len);
1593         skb_set_owner_w(skb, tfile->socket.sk);
1594
1595         get_page(alloc_frag->page);
1596         alloc_frag->offset += buflen;
1597
1598         return skb;
1599 }
1600
1601 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1602                        struct xdp_buff *xdp, u32 act)
1603 {
1604         int err;
1605
1606         switch (act) {
1607         case XDP_REDIRECT:
1608                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1609                 if (err)
1610                         return err;
1611                 break;
1612         case XDP_TX:
1613                 err = tun_xdp_tx(tun->dev, xdp);
1614                 if (err < 0)
1615                         return err;
1616                 break;
1617         case XDP_PASS:
1618                 break;
1619         default:
1620                 bpf_warn_invalid_xdp_action(act);
1621                 fallthrough;
1622         case XDP_ABORTED:
1623                 trace_xdp_exception(tun->dev, xdp_prog, act);
1624                 fallthrough;
1625         case XDP_DROP:
1626                 atomic_long_inc(&tun->dev->rx_dropped);
1627                 break;
1628         }
1629
1630         return act;
1631 }
1632
1633 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1634                                      struct tun_file *tfile,
1635                                      struct iov_iter *from,
1636                                      struct virtio_net_hdr *hdr,
1637                                      int len, int *skb_xdp)
1638 {
1639         struct page_frag *alloc_frag = &current->task_frag;
1640         struct bpf_prog *xdp_prog;
1641         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1642         char *buf;
1643         size_t copied;
1644         int pad = TUN_RX_PAD;
1645         int err = 0;
1646
1647         rcu_read_lock();
1648         xdp_prog = rcu_dereference(tun->xdp_prog);
1649         if (xdp_prog)
1650                 pad += XDP_PACKET_HEADROOM;
1651         buflen += SKB_DATA_ALIGN(len + pad);
1652         rcu_read_unlock();
1653
1654         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1655         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1656                 return ERR_PTR(-ENOMEM);
1657
1658         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1659         copied = copy_page_from_iter(alloc_frag->page,
1660                                      alloc_frag->offset + pad,
1661                                      len, from);
1662         if (copied != len)
1663                 return ERR_PTR(-EFAULT);
1664
1665         /* There's a small window that XDP may be set after the check
1666          * of xdp_prog above, this should be rare and for simplicity
1667          * we do XDP on skb in case the headroom is not enough.
1668          */
1669         if (hdr->gso_type || !xdp_prog) {
1670                 *skb_xdp = 1;
1671                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1672                                        pad);
1673         }
1674
1675         *skb_xdp = 0;
1676
1677         local_bh_disable();
1678         rcu_read_lock();
1679         xdp_prog = rcu_dereference(tun->xdp_prog);
1680         if (xdp_prog) {
1681                 struct xdp_buff xdp;
1682                 u32 act;
1683
1684                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1685                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1686
1687                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1688                 if (act == XDP_REDIRECT || act == XDP_TX) {
1689                         get_page(alloc_frag->page);
1690                         alloc_frag->offset += buflen;
1691                 }
1692                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1693                 if (err < 0) {
1694                         if (act == XDP_REDIRECT || act == XDP_TX)
1695                                 put_page(alloc_frag->page);
1696                         goto out;
1697                 }
1698
1699                 if (err == XDP_REDIRECT)
1700                         xdp_do_flush();
1701                 if (err != XDP_PASS)
1702                         goto out;
1703
1704                 pad = xdp.data - xdp.data_hard_start;
1705                 len = xdp.data_end - xdp.data;
1706         }
1707         rcu_read_unlock();
1708         local_bh_enable();
1709
1710         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1711
1712 out:
1713         rcu_read_unlock();
1714         local_bh_enable();
1715         return NULL;
1716 }
1717
1718 /* Get packet from user space buffer */
1719 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1720                             void *msg_control, struct iov_iter *from,
1721                             int noblock, bool more)
1722 {
1723         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1724         struct sk_buff *skb;
1725         size_t total_len = iov_iter_count(from);
1726         size_t len = total_len, align = tun->align, linear;
1727         struct virtio_net_hdr gso = { 0 };
1728         int good_linear;
1729         int copylen;
1730         bool zerocopy = false;
1731         int err;
1732         u32 rxhash = 0;
1733         int skb_xdp = 1;
1734         bool frags = tun_napi_frags_enabled(tfile);
1735
1736         if (!(tun->flags & IFF_NO_PI)) {
1737                 if (len < sizeof(pi))
1738                         return -EINVAL;
1739                 len -= sizeof(pi);
1740
1741                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1742                         return -EFAULT;
1743         }
1744
1745         if (tun->flags & IFF_VNET_HDR) {
1746                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1747
1748                 if (len < vnet_hdr_sz)
1749                         return -EINVAL;
1750                 len -= vnet_hdr_sz;
1751
1752                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1753                         return -EFAULT;
1754
1755                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1756                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1757                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1758
1759                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1760                         return -EINVAL;
1761                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1762         }
1763
1764         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1765                 align += NET_IP_ALIGN;
1766                 if (unlikely(len < ETH_HLEN ||
1767                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1768                         return -EINVAL;
1769         }
1770
1771         good_linear = SKB_MAX_HEAD(align);
1772
1773         if (msg_control) {
1774                 struct iov_iter i = *from;
1775
1776                 /* There are 256 bytes to be copied in skb, so there is
1777                  * enough room for skb expand head in case it is used.
1778                  * The rest of the buffer is mapped from userspace.
1779                  */
1780                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1781                 if (copylen > good_linear)
1782                         copylen = good_linear;
1783                 linear = copylen;
1784                 iov_iter_advance(&i, copylen);
1785                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1786                         zerocopy = true;
1787         }
1788
1789         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1790                 /* For the packet that is not easy to be processed
1791                  * (e.g gso or jumbo packet), we will do it at after
1792                  * skb was created with generic XDP routine.
1793                  */
1794                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1795                 if (IS_ERR(skb)) {
1796                         atomic_long_inc(&tun->dev->rx_dropped);
1797                         return PTR_ERR(skb);
1798                 }
1799                 if (!skb)
1800                         return total_len;
1801         } else {
1802                 if (!zerocopy) {
1803                         copylen = len;
1804                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1805                                 linear = good_linear;
1806                         else
1807                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1808                 }
1809
1810                 if (frags) {
1811                         mutex_lock(&tfile->napi_mutex);
1812                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1813                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1814                          * If zerocopy is enabled, then this layout will be
1815                          * overwritten by zerocopy_sg_from_iter().
1816                          */
1817                         zerocopy = false;
1818                 } else {
1819                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1820                                             noblock);
1821                 }
1822
1823                 if (IS_ERR(skb)) {
1824                         if (PTR_ERR(skb) != -EAGAIN)
1825                                 atomic_long_inc(&tun->dev->rx_dropped);
1826                         if (frags)
1827                                 mutex_unlock(&tfile->napi_mutex);
1828                         return PTR_ERR(skb);
1829                 }
1830
1831                 if (zerocopy)
1832                         err = zerocopy_sg_from_iter(skb, from);
1833                 else
1834                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1835
1836                 if (err) {
1837                         err = -EFAULT;
1838 drop:
1839                         atomic_long_inc(&tun->dev->rx_dropped);
1840                         kfree_skb(skb);
1841                         if (frags) {
1842                                 tfile->napi.skb = NULL;
1843                                 mutex_unlock(&tfile->napi_mutex);
1844                         }
1845
1846                         return err;
1847                 }
1848         }
1849
1850         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1851                 atomic_long_inc(&tun->rx_frame_errors);
1852                 kfree_skb(skb);
1853                 if (frags) {
1854                         tfile->napi.skb = NULL;
1855                         mutex_unlock(&tfile->napi_mutex);
1856                 }
1857
1858                 return -EINVAL;
1859         }
1860
1861         switch (tun->flags & TUN_TYPE_MASK) {
1862         case IFF_TUN:
1863                 if (tun->flags & IFF_NO_PI) {
1864                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1865
1866                         switch (ip_version) {
1867                         case 4:
1868                                 pi.proto = htons(ETH_P_IP);
1869                                 break;
1870                         case 6:
1871                                 pi.proto = htons(ETH_P_IPV6);
1872                                 break;
1873                         default:
1874                                 atomic_long_inc(&tun->dev->rx_dropped);
1875                                 kfree_skb(skb);
1876                                 return -EINVAL;
1877                         }
1878                 }
1879
1880                 skb_reset_mac_header(skb);
1881                 skb->protocol = pi.proto;
1882                 skb->dev = tun->dev;
1883                 break;
1884         case IFF_TAP:
1885                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1886                         err = -ENOMEM;
1887                         goto drop;
1888                 }
1889                 skb->protocol = eth_type_trans(skb, tun->dev);
1890                 break;
1891         }
1892
1893         /* copy skb_ubuf_info for callback when skb has no error */
1894         if (zerocopy) {
1895                 skb_zcopy_init(skb, msg_control);
1896         } else if (msg_control) {
1897                 struct ubuf_info *uarg = msg_control;
1898                 uarg->callback(NULL, uarg, false);
1899         }
1900
1901         skb_reset_network_header(skb);
1902         skb_probe_transport_header(skb);
1903         skb_record_rx_queue(skb, tfile->queue_index);
1904
1905         if (skb_xdp) {
1906                 struct bpf_prog *xdp_prog;
1907                 int ret;
1908
1909                 local_bh_disable();
1910                 rcu_read_lock();
1911                 xdp_prog = rcu_dereference(tun->xdp_prog);
1912                 if (xdp_prog) {
1913                         ret = do_xdp_generic(xdp_prog, skb);
1914                         if (ret != XDP_PASS) {
1915                                 rcu_read_unlock();
1916                                 local_bh_enable();
1917                                 if (frags) {
1918                                         tfile->napi.skb = NULL;
1919                                         mutex_unlock(&tfile->napi_mutex);
1920                                 }
1921                                 return total_len;
1922                         }
1923                 }
1924                 rcu_read_unlock();
1925                 local_bh_enable();
1926         }
1927
1928         /* Compute the costly rx hash only if needed for flow updates.
1929          * We may get a very small possibility of OOO during switching, not
1930          * worth to optimize.
1931          */
1932         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1933             !tfile->detached)
1934                 rxhash = __skb_get_hash_symmetric(skb);
1935
1936         rcu_read_lock();
1937         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1938                 err = -EIO;
1939                 rcu_read_unlock();
1940                 goto drop;
1941         }
1942
1943         if (frags) {
1944                 u32 headlen;
1945
1946                 /* Exercise flow dissector code path. */
1947                 skb_push(skb, ETH_HLEN);
1948                 headlen = eth_get_headlen(tun->dev, skb->data,
1949                                           skb_headlen(skb));
1950
1951                 if (unlikely(headlen > skb_headlen(skb))) {
1952                         WARN_ON_ONCE(1);
1953                         err = -ENOMEM;
1954                         atomic_long_inc(&tun->dev->rx_dropped);
1955 napi_busy:
1956                         napi_free_frags(&tfile->napi);
1957                         rcu_read_unlock();
1958                         mutex_unlock(&tfile->napi_mutex);
1959                         return err;
1960                 }
1961
1962                 if (likely(napi_schedule_prep(&tfile->napi))) {
1963                         local_bh_disable();
1964                         napi_gro_frags(&tfile->napi);
1965                         napi_complete(&tfile->napi);
1966                         local_bh_enable();
1967                 } else {
1968                         err = -EBUSY;
1969                         goto napi_busy;
1970                 }
1971                 mutex_unlock(&tfile->napi_mutex);
1972         } else if (tfile->napi_enabled) {
1973                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1974                 int queue_len;
1975
1976                 spin_lock_bh(&queue->lock);
1977                 __skb_queue_tail(queue, skb);
1978                 queue_len = skb_queue_len(queue);
1979                 spin_unlock(&queue->lock);
1980
1981                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1982                         napi_schedule(&tfile->napi);
1983
1984                 local_bh_enable();
1985         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1986                 tun_rx_batched(tun, tfile, skb, more);
1987         } else {
1988                 netif_rx_ni(skb);
1989         }
1990         rcu_read_unlock();
1991
1992         preempt_disable();
1993         dev_sw_netstats_rx_add(tun->dev, len);
1994         preempt_enable();
1995
1996         if (rxhash)
1997                 tun_flow_update(tun, rxhash, tfile);
1998
1999         return total_len;
2000 }
2001
2002 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2003 {
2004         struct file *file = iocb->ki_filp;
2005         struct tun_file *tfile = file->private_data;
2006         struct tun_struct *tun = tun_get(tfile);
2007         ssize_t result;
2008         int noblock = 0;
2009
2010         if (!tun)
2011                 return -EBADFD;
2012
2013         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2014                 noblock = 1;
2015
2016         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2017
2018         tun_put(tun);
2019         return result;
2020 }
2021
2022 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2023                                 struct tun_file *tfile,
2024                                 struct xdp_frame *xdp_frame,
2025                                 struct iov_iter *iter)
2026 {
2027         int vnet_hdr_sz = 0;
2028         size_t size = xdp_frame->len;
2029         size_t ret;
2030
2031         if (tun->flags & IFF_VNET_HDR) {
2032                 struct virtio_net_hdr gso = { 0 };
2033
2034                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2035                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2036                         return -EINVAL;
2037                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2038                              sizeof(gso)))
2039                         return -EFAULT;
2040                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2041         }
2042
2043         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2044
2045         preempt_disable();
2046         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2047         preempt_enable();
2048
2049         return ret;
2050 }
2051
2052 /* Put packet to the user space buffer */
2053 static ssize_t tun_put_user(struct tun_struct *tun,
2054                             struct tun_file *tfile,
2055                             struct sk_buff *skb,
2056                             struct iov_iter *iter)
2057 {
2058         struct tun_pi pi = { 0, skb->protocol };
2059         ssize_t total;
2060         int vlan_offset = 0;
2061         int vlan_hlen = 0;
2062         int vnet_hdr_sz = 0;
2063
2064         if (skb_vlan_tag_present(skb))
2065                 vlan_hlen = VLAN_HLEN;
2066
2067         if (tun->flags & IFF_VNET_HDR)
2068                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2069
2070         total = skb->len + vlan_hlen + vnet_hdr_sz;
2071
2072         if (!(tun->flags & IFF_NO_PI)) {
2073                 if (iov_iter_count(iter) < sizeof(pi))
2074                         return -EINVAL;
2075
2076                 total += sizeof(pi);
2077                 if (iov_iter_count(iter) < total) {
2078                         /* Packet will be striped */
2079                         pi.flags |= TUN_PKT_STRIP;
2080                 }
2081
2082                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2083                         return -EFAULT;
2084         }
2085
2086         if (vnet_hdr_sz) {
2087                 struct virtio_net_hdr gso;
2088
2089                 if (iov_iter_count(iter) < vnet_hdr_sz)
2090                         return -EINVAL;
2091
2092                 if (virtio_net_hdr_from_skb(skb, &gso,
2093                                             tun_is_little_endian(tun), true,
2094                                             vlan_hlen)) {
2095                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2096                         pr_err("unexpected GSO type: "
2097                                "0x%x, gso_size %d, hdr_len %d\n",
2098                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2099                                tun16_to_cpu(tun, gso.hdr_len));
2100                         print_hex_dump(KERN_ERR, "tun: ",
2101                                        DUMP_PREFIX_NONE,
2102                                        16, 1, skb->head,
2103                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2104                         WARN_ON_ONCE(1);
2105                         return -EINVAL;
2106                 }
2107
2108                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2109                         return -EFAULT;
2110
2111                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2112         }
2113
2114         if (vlan_hlen) {
2115                 int ret;
2116                 struct veth veth;
2117
2118                 veth.h_vlan_proto = skb->vlan_proto;
2119                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2120
2121                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2122
2123                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2124                 if (ret || !iov_iter_count(iter))
2125                         goto done;
2126
2127                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2128                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2129                         goto done;
2130         }
2131
2132         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2133
2134 done:
2135         /* caller is in process context, */
2136         preempt_disable();
2137         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2138         preempt_enable();
2139
2140         return total;
2141 }
2142
2143 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2144 {
2145         DECLARE_WAITQUEUE(wait, current);
2146         void *ptr = NULL;
2147         int error = 0;
2148
2149         ptr = ptr_ring_consume(&tfile->tx_ring);
2150         if (ptr)
2151                 goto out;
2152         if (noblock) {
2153                 error = -EAGAIN;
2154                 goto out;
2155         }
2156
2157         add_wait_queue(&tfile->socket.wq.wait, &wait);
2158
2159         while (1) {
2160                 set_current_state(TASK_INTERRUPTIBLE);
2161                 ptr = ptr_ring_consume(&tfile->tx_ring);
2162                 if (ptr)
2163                         break;
2164                 if (signal_pending(current)) {
2165                         error = -ERESTARTSYS;
2166                         break;
2167                 }
2168                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2169                         error = -EFAULT;
2170                         break;
2171                 }
2172
2173                 schedule();
2174         }
2175
2176         __set_current_state(TASK_RUNNING);
2177         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2178
2179 out:
2180         *err = error;
2181         return ptr;
2182 }
2183
2184 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2185                            struct iov_iter *to,
2186                            int noblock, void *ptr)
2187 {
2188         ssize_t ret;
2189         int err;
2190
2191         if (!iov_iter_count(to)) {
2192                 tun_ptr_free(ptr);
2193                 return 0;
2194         }
2195
2196         if (!ptr) {
2197                 /* Read frames from ring */
2198                 ptr = tun_ring_recv(tfile, noblock, &err);
2199                 if (!ptr)
2200                         return err;
2201         }
2202
2203         if (tun_is_xdp_frame(ptr)) {
2204                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2205
2206                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2207                 xdp_return_frame(xdpf);
2208         } else {
2209                 struct sk_buff *skb = ptr;
2210
2211                 ret = tun_put_user(tun, tfile, skb, to);
2212                 if (unlikely(ret < 0))
2213                         kfree_skb(skb);
2214                 else
2215                         consume_skb(skb);
2216         }
2217
2218         return ret;
2219 }
2220
2221 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2222 {
2223         struct file *file = iocb->ki_filp;
2224         struct tun_file *tfile = file->private_data;
2225         struct tun_struct *tun = tun_get(tfile);
2226         ssize_t len = iov_iter_count(to), ret;
2227         int noblock = 0;
2228
2229         if (!tun)
2230                 return -EBADFD;
2231
2232         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2233                 noblock = 1;
2234
2235         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2236         ret = min_t(ssize_t, ret, len);
2237         if (ret > 0)
2238                 iocb->ki_pos = ret;
2239         tun_put(tun);
2240         return ret;
2241 }
2242
2243 static void tun_prog_free(struct rcu_head *rcu)
2244 {
2245         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2246
2247         bpf_prog_destroy(prog->prog);
2248         kfree(prog);
2249 }
2250
2251 static int __tun_set_ebpf(struct tun_struct *tun,
2252                           struct tun_prog __rcu **prog_p,
2253                           struct bpf_prog *prog)
2254 {
2255         struct tun_prog *old, *new = NULL;
2256
2257         if (prog) {
2258                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2259                 if (!new)
2260                         return -ENOMEM;
2261                 new->prog = prog;
2262         }
2263
2264         spin_lock_bh(&tun->lock);
2265         old = rcu_dereference_protected(*prog_p,
2266                                         lockdep_is_held(&tun->lock));
2267         rcu_assign_pointer(*prog_p, new);
2268         spin_unlock_bh(&tun->lock);
2269
2270         if (old)
2271                 call_rcu(&old->rcu, tun_prog_free);
2272
2273         return 0;
2274 }
2275
2276 static void tun_free_netdev(struct net_device *dev)
2277 {
2278         struct tun_struct *tun = netdev_priv(dev);
2279
2280         BUG_ON(!(list_empty(&tun->disabled)));
2281
2282         free_percpu(dev->tstats);
2283         tun_flow_uninit(tun);
2284         security_tun_dev_free_security(tun->security);
2285         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2286         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2287 }
2288
2289 static void tun_setup(struct net_device *dev)
2290 {
2291         struct tun_struct *tun = netdev_priv(dev);
2292
2293         tun->owner = INVALID_UID;
2294         tun->group = INVALID_GID;
2295         tun_default_link_ksettings(dev, &tun->link_ksettings);
2296
2297         dev->ethtool_ops = &tun_ethtool_ops;
2298         dev->needs_free_netdev = true;
2299         dev->priv_destructor = tun_free_netdev;
2300         /* We prefer our own queue length */
2301         dev->tx_queue_len = TUN_READQ_SIZE;
2302 }
2303
2304 /* Trivial set of netlink ops to allow deleting tun or tap
2305  * device with netlink.
2306  */
2307 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2308                         struct netlink_ext_ack *extack)
2309 {
2310         NL_SET_ERR_MSG(extack,
2311                        "tun/tap creation via rtnetlink is not supported.");
2312         return -EOPNOTSUPP;
2313 }
2314
2315 static size_t tun_get_size(const struct net_device *dev)
2316 {
2317         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2318         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2319
2320         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2321                nla_total_size(sizeof(gid_t)) + /* GROUP */
2322                nla_total_size(sizeof(u8)) + /* TYPE */
2323                nla_total_size(sizeof(u8)) + /* PI */
2324                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2325                nla_total_size(sizeof(u8)) + /* PERSIST */
2326                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2327                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2328                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2329                0;
2330 }
2331
2332 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2333 {
2334         struct tun_struct *tun = netdev_priv(dev);
2335
2336         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2337                 goto nla_put_failure;
2338         if (uid_valid(tun->owner) &&
2339             nla_put_u32(skb, IFLA_TUN_OWNER,
2340                         from_kuid_munged(current_user_ns(), tun->owner)))
2341                 goto nla_put_failure;
2342         if (gid_valid(tun->group) &&
2343             nla_put_u32(skb, IFLA_TUN_GROUP,
2344                         from_kgid_munged(current_user_ns(), tun->group)))
2345                 goto nla_put_failure;
2346         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2347                 goto nla_put_failure;
2348         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2349                 goto nla_put_failure;
2350         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2351                 goto nla_put_failure;
2352         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2353                        !!(tun->flags & IFF_MULTI_QUEUE)))
2354                 goto nla_put_failure;
2355         if (tun->flags & IFF_MULTI_QUEUE) {
2356                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2357                         goto nla_put_failure;
2358                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2359                                 tun->numdisabled))
2360                         goto nla_put_failure;
2361         }
2362
2363         return 0;
2364
2365 nla_put_failure:
2366         return -EMSGSIZE;
2367 }
2368
2369 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2370         .kind           = DRV_NAME,
2371         .priv_size      = sizeof(struct tun_struct),
2372         .setup          = tun_setup,
2373         .validate       = tun_validate,
2374         .get_size       = tun_get_size,
2375         .fill_info      = tun_fill_info,
2376 };
2377
2378 static void tun_sock_write_space(struct sock *sk)
2379 {
2380         struct tun_file *tfile;
2381         wait_queue_head_t *wqueue;
2382
2383         if (!sock_writeable(sk))
2384                 return;
2385
2386         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2387                 return;
2388
2389         wqueue = sk_sleep(sk);
2390         if (wqueue && waitqueue_active(wqueue))
2391                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2392                                                 EPOLLWRNORM | EPOLLWRBAND);
2393
2394         tfile = container_of(sk, struct tun_file, sk);
2395         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2396 }
2397
2398 static void tun_put_page(struct tun_page *tpage)
2399 {
2400         if (tpage->page)
2401                 __page_frag_cache_drain(tpage->page, tpage->count);
2402 }
2403
2404 static int tun_xdp_one(struct tun_struct *tun,
2405                        struct tun_file *tfile,
2406                        struct xdp_buff *xdp, int *flush,
2407                        struct tun_page *tpage)
2408 {
2409         unsigned int datasize = xdp->data_end - xdp->data;
2410         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2411         struct virtio_net_hdr *gso = &hdr->gso;
2412         struct bpf_prog *xdp_prog;
2413         struct sk_buff *skb = NULL;
2414         u32 rxhash = 0, act;
2415         int buflen = hdr->buflen;
2416         int err = 0;
2417         bool skb_xdp = false;
2418         struct page *page;
2419
2420         xdp_prog = rcu_dereference(tun->xdp_prog);
2421         if (xdp_prog) {
2422                 if (gso->gso_type) {
2423                         skb_xdp = true;
2424                         goto build;
2425                 }
2426
2427                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2428                 xdp_set_data_meta_invalid(xdp);
2429
2430                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2431                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2432                 if (err < 0) {
2433                         put_page(virt_to_head_page(xdp->data));
2434                         return err;
2435                 }
2436
2437                 switch (err) {
2438                 case XDP_REDIRECT:
2439                         *flush = true;
2440                         fallthrough;
2441                 case XDP_TX:
2442                         return 0;
2443                 case XDP_PASS:
2444                         break;
2445                 default:
2446                         page = virt_to_head_page(xdp->data);
2447                         if (tpage->page == page) {
2448                                 ++tpage->count;
2449                         } else {
2450                                 tun_put_page(tpage);
2451                                 tpage->page = page;
2452                                 tpage->count = 1;
2453                         }
2454                         return 0;
2455                 }
2456         }
2457
2458 build:
2459         skb = build_skb(xdp->data_hard_start, buflen);
2460         if (!skb) {
2461                 err = -ENOMEM;
2462                 goto out;
2463         }
2464
2465         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2466         skb_put(skb, xdp->data_end - xdp->data);
2467
2468         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2469                 atomic_long_inc(&tun->rx_frame_errors);
2470                 kfree_skb(skb);
2471                 err = -EINVAL;
2472                 goto out;
2473         }
2474
2475         skb->protocol = eth_type_trans(skb, tun->dev);
2476         skb_reset_network_header(skb);
2477         skb_probe_transport_header(skb);
2478         skb_record_rx_queue(skb, tfile->queue_index);
2479
2480         if (skb_xdp) {
2481                 err = do_xdp_generic(xdp_prog, skb);
2482                 if (err != XDP_PASS)
2483                         goto out;
2484         }
2485
2486         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2487             !tfile->detached)
2488                 rxhash = __skb_get_hash_symmetric(skb);
2489
2490         netif_receive_skb(skb);
2491
2492         /* No need to disable preemption here since this function is
2493          * always called with bh disabled
2494          */
2495         dev_sw_netstats_rx_add(tun->dev, datasize);
2496
2497         if (rxhash)
2498                 tun_flow_update(tun, rxhash, tfile);
2499
2500 out:
2501         return err;
2502 }
2503
2504 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2505 {
2506         int ret, i;
2507         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2508         struct tun_struct *tun = tun_get(tfile);
2509         struct tun_msg_ctl *ctl = m->msg_control;
2510         struct xdp_buff *xdp;
2511
2512         if (!tun)
2513                 return -EBADFD;
2514
2515         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2516             ctl && ctl->type == TUN_MSG_PTR) {
2517                 struct tun_page tpage;
2518                 int n = ctl->num;
2519                 int flush = 0;
2520
2521                 memset(&tpage, 0, sizeof(tpage));
2522
2523                 local_bh_disable();
2524                 rcu_read_lock();
2525
2526                 for (i = 0; i < n; i++) {
2527                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2528                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2529                 }
2530
2531                 if (flush)
2532                         xdp_do_flush();
2533
2534                 rcu_read_unlock();
2535                 local_bh_enable();
2536
2537                 tun_put_page(&tpage);
2538
2539                 ret = total_len;
2540                 goto out;
2541         }
2542
2543         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2544                            m->msg_flags & MSG_DONTWAIT,
2545                            m->msg_flags & MSG_MORE);
2546 out:
2547         tun_put(tun);
2548         return ret;
2549 }
2550
2551 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2552                        int flags)
2553 {
2554         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2555         struct tun_struct *tun = tun_get(tfile);
2556         void *ptr = m->msg_control;
2557         int ret;
2558
2559         if (!tun) {
2560                 ret = -EBADFD;
2561                 goto out_free;
2562         }
2563
2564         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2565                 ret = -EINVAL;
2566                 goto out_put_tun;
2567         }
2568         if (flags & MSG_ERRQUEUE) {
2569                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2570                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2571                 goto out;
2572         }
2573         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2574         if (ret > (ssize_t)total_len) {
2575                 m->msg_flags |= MSG_TRUNC;
2576                 ret = flags & MSG_TRUNC ? ret : total_len;
2577         }
2578 out:
2579         tun_put(tun);
2580         return ret;
2581
2582 out_put_tun:
2583         tun_put(tun);
2584 out_free:
2585         tun_ptr_free(ptr);
2586         return ret;
2587 }
2588
2589 static int tun_ptr_peek_len(void *ptr)
2590 {
2591         if (likely(ptr)) {
2592                 if (tun_is_xdp_frame(ptr)) {
2593                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2594
2595                         return xdpf->len;
2596                 }
2597                 return __skb_array_len_with_tag(ptr);
2598         } else {
2599                 return 0;
2600         }
2601 }
2602
2603 static int tun_peek_len(struct socket *sock)
2604 {
2605         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2606         struct tun_struct *tun;
2607         int ret = 0;
2608
2609         tun = tun_get(tfile);
2610         if (!tun)
2611                 return 0;
2612
2613         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2614         tun_put(tun);
2615
2616         return ret;
2617 }
2618
2619 /* Ops structure to mimic raw sockets with tun */
2620 static const struct proto_ops tun_socket_ops = {
2621         .peek_len = tun_peek_len,
2622         .sendmsg = tun_sendmsg,
2623         .recvmsg = tun_recvmsg,
2624 };
2625
2626 static struct proto tun_proto = {
2627         .name           = "tun",
2628         .owner          = THIS_MODULE,
2629         .obj_size       = sizeof(struct tun_file),
2630 };
2631
2632 static int tun_flags(struct tun_struct *tun)
2633 {
2634         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2635 }
2636
2637 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2638                               char *buf)
2639 {
2640         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2641         return sprintf(buf, "0x%x\n", tun_flags(tun));
2642 }
2643
2644 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2645                           char *buf)
2646 {
2647         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2648         return uid_valid(tun->owner)?
2649                 sprintf(buf, "%u\n",
2650                         from_kuid_munged(current_user_ns(), tun->owner)):
2651                 sprintf(buf, "-1\n");
2652 }
2653
2654 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2655                           char *buf)
2656 {
2657         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2658         return gid_valid(tun->group) ?
2659                 sprintf(buf, "%u\n",
2660                         from_kgid_munged(current_user_ns(), tun->group)):
2661                 sprintf(buf, "-1\n");
2662 }
2663
2664 static DEVICE_ATTR_RO(tun_flags);
2665 static DEVICE_ATTR_RO(owner);
2666 static DEVICE_ATTR_RO(group);
2667
2668 static struct attribute *tun_dev_attrs[] = {
2669         &dev_attr_tun_flags.attr,
2670         &dev_attr_owner.attr,
2671         &dev_attr_group.attr,
2672         NULL
2673 };
2674
2675 static const struct attribute_group tun_attr_group = {
2676         .attrs = tun_dev_attrs
2677 };
2678
2679 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2680 {
2681         struct tun_struct *tun;
2682         struct tun_file *tfile = file->private_data;
2683         struct net_device *dev;
2684         int err;
2685
2686         if (tfile->detached)
2687                 return -EINVAL;
2688
2689         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2690                 if (!capable(CAP_NET_ADMIN))
2691                         return -EPERM;
2692
2693                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2694                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2695                         return -EINVAL;
2696         }
2697
2698         dev = __dev_get_by_name(net, ifr->ifr_name);
2699         if (dev) {
2700                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2701                         return -EBUSY;
2702                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2703                         tun = netdev_priv(dev);
2704                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2705                         tun = netdev_priv(dev);
2706                 else
2707                         return -EINVAL;
2708
2709                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2710                     !!(tun->flags & IFF_MULTI_QUEUE))
2711                         return -EINVAL;
2712
2713                 if (tun_not_capable(tun))
2714                         return -EPERM;
2715                 err = security_tun_dev_open(tun->security);
2716                 if (err < 0)
2717                         return err;
2718
2719                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2720                                  ifr->ifr_flags & IFF_NAPI,
2721                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2722                 if (err < 0)
2723                         return err;
2724
2725                 if (tun->flags & IFF_MULTI_QUEUE &&
2726                     (tun->numqueues + tun->numdisabled > 1)) {
2727                         /* One or more queue has already been attached, no need
2728                          * to initialize the device again.
2729                          */
2730                         netdev_state_change(dev);
2731                         return 0;
2732                 }
2733
2734                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2735                               (ifr->ifr_flags & TUN_FEATURES);
2736
2737                 netdev_state_change(dev);
2738         } else {
2739                 char *name;
2740                 unsigned long flags = 0;
2741                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2742                              MAX_TAP_QUEUES : 1;
2743
2744                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2745                         return -EPERM;
2746                 err = security_tun_dev_create();
2747                 if (err < 0)
2748                         return err;
2749
2750                 /* Set dev type */
2751                 if (ifr->ifr_flags & IFF_TUN) {
2752                         /* TUN device */
2753                         flags |= IFF_TUN;
2754                         name = "tun%d";
2755                 } else if (ifr->ifr_flags & IFF_TAP) {
2756                         /* TAP device */
2757                         flags |= IFF_TAP;
2758                         name = "tap%d";
2759                 } else
2760                         return -EINVAL;
2761
2762                 if (*ifr->ifr_name)
2763                         name = ifr->ifr_name;
2764
2765                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2766                                        NET_NAME_UNKNOWN, tun_setup, queues,
2767                                        queues);
2768
2769                 if (!dev)
2770                         return -ENOMEM;
2771
2772                 dev_net_set(dev, net);
2773                 dev->rtnl_link_ops = &tun_link_ops;
2774                 dev->ifindex = tfile->ifindex;
2775                 dev->sysfs_groups[0] = &tun_attr_group;
2776
2777                 tun = netdev_priv(dev);
2778                 tun->dev = dev;
2779                 tun->flags = flags;
2780                 tun->txflt.count = 0;
2781                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2782
2783                 tun->align = NET_SKB_PAD;
2784                 tun->filter_attached = false;
2785                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2786                 tun->rx_batched = 0;
2787                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2788
2789                 tun->ifr = ifr;
2790                 tun->file = file;
2791
2792                 tun_net_initialize(dev);
2793
2794                 err = register_netdevice(tun->dev);
2795                 if (err < 0) {
2796                         free_netdev(dev);
2797                         return err;
2798                 }
2799                 /* free_netdev() won't check refcnt, to avoid race
2800                  * with dev_put() we need publish tun after registration.
2801                  */
2802                 rcu_assign_pointer(tfile->tun, tun);
2803         }
2804
2805         netif_carrier_on(tun->dev);
2806
2807         /* Make sure persistent devices do not get stuck in
2808          * xoff state.
2809          */
2810         if (netif_running(tun->dev))
2811                 netif_tx_wake_all_queues(tun->dev);
2812
2813         strcpy(ifr->ifr_name, tun->dev->name);
2814         return 0;
2815 }
2816
2817 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2818 {
2819         strcpy(ifr->ifr_name, tun->dev->name);
2820
2821         ifr->ifr_flags = tun_flags(tun);
2822
2823 }
2824
2825 /* This is like a cut-down ethtool ops, except done via tun fd so no
2826  * privs required. */
2827 static int set_offload(struct tun_struct *tun, unsigned long arg)
2828 {
2829         netdev_features_t features = 0;
2830
2831         if (arg & TUN_F_CSUM) {
2832                 features |= NETIF_F_HW_CSUM;
2833                 arg &= ~TUN_F_CSUM;
2834
2835                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2836                         if (arg & TUN_F_TSO_ECN) {
2837                                 features |= NETIF_F_TSO_ECN;
2838                                 arg &= ~TUN_F_TSO_ECN;
2839                         }
2840                         if (arg & TUN_F_TSO4)
2841                                 features |= NETIF_F_TSO;
2842                         if (arg & TUN_F_TSO6)
2843                                 features |= NETIF_F_TSO6;
2844                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2845                 }
2846
2847                 arg &= ~TUN_F_UFO;
2848         }
2849
2850         /* This gives the user a way to test for new features in future by
2851          * trying to set them. */
2852         if (arg)
2853                 return -EINVAL;
2854
2855         tun->set_features = features;
2856         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2857         tun->dev->wanted_features |= features;
2858         netdev_update_features(tun->dev);
2859
2860         return 0;
2861 }
2862
2863 static void tun_detach_filter(struct tun_struct *tun, int n)
2864 {
2865         int i;
2866         struct tun_file *tfile;
2867
2868         for (i = 0; i < n; i++) {
2869                 tfile = rtnl_dereference(tun->tfiles[i]);
2870                 lock_sock(tfile->socket.sk);
2871                 sk_detach_filter(tfile->socket.sk);
2872                 release_sock(tfile->socket.sk);
2873         }
2874
2875         tun->filter_attached = false;
2876 }
2877
2878 static int tun_attach_filter(struct tun_struct *tun)
2879 {
2880         int i, ret = 0;
2881         struct tun_file *tfile;
2882
2883         for (i = 0; i < tun->numqueues; i++) {
2884                 tfile = rtnl_dereference(tun->tfiles[i]);
2885                 lock_sock(tfile->socket.sk);
2886                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2887                 release_sock(tfile->socket.sk);
2888                 if (ret) {
2889                         tun_detach_filter(tun, i);
2890                         return ret;
2891                 }
2892         }
2893
2894         tun->filter_attached = true;
2895         return ret;
2896 }
2897
2898 static void tun_set_sndbuf(struct tun_struct *tun)
2899 {
2900         struct tun_file *tfile;
2901         int i;
2902
2903         for (i = 0; i < tun->numqueues; i++) {
2904                 tfile = rtnl_dereference(tun->tfiles[i]);
2905                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2906         }
2907 }
2908
2909 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2910 {
2911         struct tun_file *tfile = file->private_data;
2912         struct tun_struct *tun;
2913         int ret = 0;
2914
2915         rtnl_lock();
2916
2917         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2918                 tun = tfile->detached;
2919                 if (!tun) {
2920                         ret = -EINVAL;
2921                         goto unlock;
2922                 }
2923                 ret = security_tun_dev_attach_queue(tun->security);
2924                 if (ret < 0)
2925                         goto unlock;
2926                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2927                                  tun->flags & IFF_NAPI_FRAGS, true);
2928         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2929                 tun = rtnl_dereference(tfile->tun);
2930                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2931                         ret = -EINVAL;
2932                 else
2933                         __tun_detach(tfile, false);
2934         } else
2935                 ret = -EINVAL;
2936
2937         if (ret >= 0)
2938                 netdev_state_change(tun->dev);
2939
2940 unlock:
2941         rtnl_unlock();
2942         return ret;
2943 }
2944
2945 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2946                         void __user *data)
2947 {
2948         struct bpf_prog *prog;
2949         int fd;
2950
2951         if (copy_from_user(&fd, data, sizeof(fd)))
2952                 return -EFAULT;
2953
2954         if (fd == -1) {
2955                 prog = NULL;
2956         } else {
2957                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2958                 if (IS_ERR(prog))
2959                         return PTR_ERR(prog);
2960         }
2961
2962         return __tun_set_ebpf(tun, prog_p, prog);
2963 }
2964
2965 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2966 static unsigned char tun_get_addr_len(unsigned short type)
2967 {
2968         switch (type) {
2969         case ARPHRD_IP6GRE:
2970         case ARPHRD_TUNNEL6:
2971                 return sizeof(struct in6_addr);
2972         case ARPHRD_IPGRE:
2973         case ARPHRD_TUNNEL:
2974         case ARPHRD_SIT:
2975                 return 4;
2976         case ARPHRD_ETHER:
2977                 return ETH_ALEN;
2978         case ARPHRD_IEEE802154:
2979         case ARPHRD_IEEE802154_MONITOR:
2980                 return IEEE802154_EXTENDED_ADDR_LEN;
2981         case ARPHRD_PHONET_PIPE:
2982         case ARPHRD_PPP:
2983         case ARPHRD_NONE:
2984                 return 0;
2985         case ARPHRD_6LOWPAN:
2986                 return EUI64_ADDR_LEN;
2987         case ARPHRD_FDDI:
2988                 return FDDI_K_ALEN;
2989         case ARPHRD_HIPPI:
2990                 return HIPPI_ALEN;
2991         case ARPHRD_IEEE802:
2992                 return FC_ALEN;
2993         case ARPHRD_ROSE:
2994                 return ROSE_ADDR_LEN;
2995         case ARPHRD_NETROM:
2996                 return AX25_ADDR_LEN;
2997         case ARPHRD_LOCALTLK:
2998                 return LTALK_ALEN;
2999         default:
3000                 return 0;
3001         }
3002 }
3003
3004 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3005                             unsigned long arg, int ifreq_len)
3006 {
3007         struct tun_file *tfile = file->private_data;
3008         struct net *net = sock_net(&tfile->sk);
3009         struct tun_struct *tun;
3010         void __user* argp = (void __user*)arg;
3011         unsigned int ifindex, carrier;
3012         struct ifreq ifr;
3013         kuid_t owner;
3014         kgid_t group;
3015         int sndbuf;
3016         int vnet_hdr_sz;
3017         int le;
3018         int ret;
3019         bool do_notify = false;
3020
3021         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3022             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3023                 if (copy_from_user(&ifr, argp, ifreq_len))
3024                         return -EFAULT;
3025         } else {
3026                 memset(&ifr, 0, sizeof(ifr));
3027         }
3028         if (cmd == TUNGETFEATURES) {
3029                 /* Currently this just means: "what IFF flags are valid?".
3030                  * This is needed because we never checked for invalid flags on
3031                  * TUNSETIFF.
3032                  */
3033                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3034                                 (unsigned int __user*)argp);
3035         } else if (cmd == TUNSETQUEUE) {
3036                 return tun_set_queue(file, &ifr);
3037         } else if (cmd == SIOCGSKNS) {
3038                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3039                         return -EPERM;
3040                 return open_related_ns(&net->ns, get_net_ns);
3041         }
3042
3043         rtnl_lock();
3044
3045         tun = tun_get(tfile);
3046         if (cmd == TUNSETIFF) {
3047                 ret = -EEXIST;
3048                 if (tun)
3049                         goto unlock;
3050
3051                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3052
3053                 ret = tun_set_iff(net, file, &ifr);
3054
3055                 if (ret)
3056                         goto unlock;
3057
3058                 if (copy_to_user(argp, &ifr, ifreq_len))
3059                         ret = -EFAULT;
3060                 goto unlock;
3061         }
3062         if (cmd == TUNSETIFINDEX) {
3063                 ret = -EPERM;
3064                 if (tun)
3065                         goto unlock;
3066
3067                 ret = -EFAULT;
3068                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3069                         goto unlock;
3070
3071                 ret = 0;
3072                 tfile->ifindex = ifindex;
3073                 goto unlock;
3074         }
3075
3076         ret = -EBADFD;
3077         if (!tun)
3078                 goto unlock;
3079
3080         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3081
3082         net = dev_net(tun->dev);
3083         ret = 0;
3084         switch (cmd) {
3085         case TUNGETIFF:
3086                 tun_get_iff(tun, &ifr);
3087
3088                 if (tfile->detached)
3089                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3090                 if (!tfile->socket.sk->sk_filter)
3091                         ifr.ifr_flags |= IFF_NOFILTER;
3092
3093                 if (copy_to_user(argp, &ifr, ifreq_len))
3094                         ret = -EFAULT;
3095                 break;
3096
3097         case TUNSETNOCSUM:
3098                 /* Disable/Enable checksum */
3099
3100                 /* [unimplemented] */
3101                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3102                            arg ? "disabled" : "enabled");
3103                 break;
3104
3105         case TUNSETPERSIST:
3106                 /* Disable/Enable persist mode. Keep an extra reference to the
3107                  * module to prevent the module being unprobed.
3108                  */
3109                 if (arg && !(tun->flags & IFF_PERSIST)) {
3110                         tun->flags |= IFF_PERSIST;
3111                         __module_get(THIS_MODULE);
3112                         do_notify = true;
3113                 }
3114                 if (!arg && (tun->flags & IFF_PERSIST)) {
3115                         tun->flags &= ~IFF_PERSIST;
3116                         module_put(THIS_MODULE);
3117                         do_notify = true;
3118                 }
3119
3120                 netif_info(tun, drv, tun->dev, "persist %s\n",
3121                            arg ? "enabled" : "disabled");
3122                 break;
3123
3124         case TUNSETOWNER:
3125                 /* Set owner of the device */
3126                 owner = make_kuid(current_user_ns(), arg);
3127                 if (!uid_valid(owner)) {
3128                         ret = -EINVAL;
3129                         break;
3130                 }
3131                 tun->owner = owner;
3132                 do_notify = true;
3133                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3134                            from_kuid(&init_user_ns, tun->owner));
3135                 break;
3136
3137         case TUNSETGROUP:
3138                 /* Set group of the device */
3139                 group = make_kgid(current_user_ns(), arg);
3140                 if (!gid_valid(group)) {
3141                         ret = -EINVAL;
3142                         break;
3143                 }
3144                 tun->group = group;
3145                 do_notify = true;
3146                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3147                            from_kgid(&init_user_ns, tun->group));
3148                 break;
3149
3150         case TUNSETLINK:
3151                 /* Only allow setting the type when the interface is down */
3152                 if (tun->dev->flags & IFF_UP) {
3153                         netif_info(tun, drv, tun->dev,
3154                                    "Linktype set failed because interface is up\n");
3155                         ret = -EBUSY;
3156                 } else {
3157                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3158                                                        tun->dev);
3159                         ret = notifier_to_errno(ret);
3160                         if (ret) {
3161                                 netif_info(tun, drv, tun->dev,
3162                                            "Refused to change device type\n");
3163                                 break;
3164                         }
3165                         tun->dev->type = (int) arg;
3166                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3167                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3168                                    tun->dev->type);
3169                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3170                                                  tun->dev);
3171                 }
3172                 break;
3173
3174         case TUNSETDEBUG:
3175                 tun->msg_enable = (u32)arg;
3176                 break;
3177
3178         case TUNSETOFFLOAD:
3179                 ret = set_offload(tun, arg);
3180                 break;
3181
3182         case TUNSETTXFILTER:
3183                 /* Can be set only for TAPs */
3184                 ret = -EINVAL;
3185                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3186                         break;
3187                 ret = update_filter(&tun->txflt, (void __user *)arg);
3188                 break;
3189
3190         case SIOCGIFHWADDR:
3191                 /* Get hw address */
3192                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3193                 if (copy_to_user(argp, &ifr, ifreq_len))
3194                         ret = -EFAULT;
3195                 break;
3196
3197         case SIOCSIFHWADDR:
3198                 /* Set hw address */
3199                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3200                 break;
3201
3202         case TUNGETSNDBUF:
3203                 sndbuf = tfile->socket.sk->sk_sndbuf;
3204                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3205                         ret = -EFAULT;
3206                 break;
3207
3208         case TUNSETSNDBUF:
3209                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3210                         ret = -EFAULT;
3211                         break;
3212                 }
3213                 if (sndbuf <= 0) {
3214                         ret = -EINVAL;
3215                         break;
3216                 }
3217
3218                 tun->sndbuf = sndbuf;
3219                 tun_set_sndbuf(tun);
3220                 break;
3221
3222         case TUNGETVNETHDRSZ:
3223                 vnet_hdr_sz = tun->vnet_hdr_sz;
3224                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3225                         ret = -EFAULT;
3226                 break;
3227
3228         case TUNSETVNETHDRSZ:
3229                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3230                         ret = -EFAULT;
3231                         break;
3232                 }
3233                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3234                         ret = -EINVAL;
3235                         break;
3236                 }
3237
3238                 tun->vnet_hdr_sz = vnet_hdr_sz;
3239                 break;
3240
3241         case TUNGETVNETLE:
3242                 le = !!(tun->flags & TUN_VNET_LE);
3243                 if (put_user(le, (int __user *)argp))
3244                         ret = -EFAULT;
3245                 break;
3246
3247         case TUNSETVNETLE:
3248                 if (get_user(le, (int __user *)argp)) {
3249                         ret = -EFAULT;
3250                         break;
3251                 }
3252                 if (le)
3253                         tun->flags |= TUN_VNET_LE;
3254                 else
3255                         tun->flags &= ~TUN_VNET_LE;
3256                 break;
3257
3258         case TUNGETVNETBE:
3259                 ret = tun_get_vnet_be(tun, argp);
3260                 break;
3261
3262         case TUNSETVNETBE:
3263                 ret = tun_set_vnet_be(tun, argp);
3264                 break;
3265
3266         case TUNATTACHFILTER:
3267                 /* Can be set only for TAPs */
3268                 ret = -EINVAL;
3269                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3270                         break;
3271                 ret = -EFAULT;
3272                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3273                         break;
3274
3275                 ret = tun_attach_filter(tun);
3276                 break;
3277
3278         case TUNDETACHFILTER:
3279                 /* Can be set only for TAPs */
3280                 ret = -EINVAL;
3281                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3282                         break;
3283                 ret = 0;
3284                 tun_detach_filter(tun, tun->numqueues);
3285                 break;
3286
3287         case TUNGETFILTER:
3288                 ret = -EINVAL;
3289                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3290                         break;
3291                 ret = -EFAULT;
3292                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3293                         break;
3294                 ret = 0;
3295                 break;
3296
3297         case TUNSETSTEERINGEBPF:
3298                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3299                 break;
3300
3301         case TUNSETFILTEREBPF:
3302                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3303                 break;
3304
3305         case TUNSETCARRIER:
3306                 ret = -EFAULT;
3307                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3308                         goto unlock;
3309
3310                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3311                 break;
3312
3313         case TUNGETDEVNETNS:
3314                 ret = -EPERM;
3315                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3316                         goto unlock;
3317                 ret = open_related_ns(&net->ns, get_net_ns);
3318                 break;
3319
3320         default:
3321                 ret = -EINVAL;
3322                 break;
3323         }
3324
3325         if (do_notify)
3326                 netdev_state_change(tun->dev);
3327
3328 unlock:
3329         rtnl_unlock();
3330         if (tun)
3331                 tun_put(tun);
3332         return ret;
3333 }
3334
3335 static long tun_chr_ioctl(struct file *file,
3336                           unsigned int cmd, unsigned long arg)
3337 {
3338         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3339 }
3340
3341 #ifdef CONFIG_COMPAT
3342 static long tun_chr_compat_ioctl(struct file *file,
3343                          unsigned int cmd, unsigned long arg)
3344 {
3345         switch (cmd) {
3346         case TUNSETIFF:
3347         case TUNGETIFF:
3348         case TUNSETTXFILTER:
3349         case TUNGETSNDBUF:
3350         case TUNSETSNDBUF:
3351         case SIOCGIFHWADDR:
3352         case SIOCSIFHWADDR:
3353                 arg = (unsigned long)compat_ptr(arg);
3354                 break;
3355         default:
3356                 arg = (compat_ulong_t)arg;
3357                 break;
3358         }
3359
3360         /*
3361          * compat_ifreq is shorter than ifreq, so we must not access beyond
3362          * the end of that structure. All fields that are used in this
3363          * driver are compatible though, we don't need to convert the
3364          * contents.
3365          */
3366         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3367 }
3368 #endif /* CONFIG_COMPAT */
3369
3370 static int tun_chr_fasync(int fd, struct file *file, int on)
3371 {
3372         struct tun_file *tfile = file->private_data;
3373         int ret;
3374
3375         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3376                 goto out;
3377
3378         if (on) {
3379                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3380                 tfile->flags |= TUN_FASYNC;
3381         } else
3382                 tfile->flags &= ~TUN_FASYNC;
3383         ret = 0;
3384 out:
3385         return ret;
3386 }
3387
3388 static int tun_chr_open(struct inode *inode, struct file * file)
3389 {
3390         struct net *net = current->nsproxy->net_ns;
3391         struct tun_file *tfile;
3392
3393         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3394                                             &tun_proto, 0);
3395         if (!tfile)
3396                 return -ENOMEM;
3397         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3398                 sk_free(&tfile->sk);
3399                 return -ENOMEM;
3400         }
3401
3402         mutex_init(&tfile->napi_mutex);
3403         RCU_INIT_POINTER(tfile->tun, NULL);
3404         tfile->flags = 0;
3405         tfile->ifindex = 0;
3406
3407         init_waitqueue_head(&tfile->socket.wq.wait);
3408
3409         tfile->socket.file = file;
3410         tfile->socket.ops = &tun_socket_ops;
3411
3412         sock_init_data(&tfile->socket, &tfile->sk);
3413
3414         tfile->sk.sk_write_space = tun_sock_write_space;
3415         tfile->sk.sk_sndbuf = INT_MAX;
3416
3417         file->private_data = tfile;
3418         INIT_LIST_HEAD(&tfile->next);
3419
3420         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3421
3422         return 0;
3423 }
3424
3425 static int tun_chr_close(struct inode *inode, struct file *file)
3426 {
3427         struct tun_file *tfile = file->private_data;
3428
3429         tun_detach(tfile, true);
3430
3431         return 0;
3432 }
3433
3434 #ifdef CONFIG_PROC_FS
3435 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3436 {
3437         struct tun_file *tfile = file->private_data;
3438         struct tun_struct *tun;
3439         struct ifreq ifr;
3440
3441         memset(&ifr, 0, sizeof(ifr));
3442
3443         rtnl_lock();
3444         tun = tun_get(tfile);
3445         if (tun)
3446                 tun_get_iff(tun, &ifr);
3447         rtnl_unlock();
3448
3449         if (tun)
3450                 tun_put(tun);
3451
3452         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3453 }
3454 #endif
3455
3456 static const struct file_operations tun_fops = {
3457         .owner  = THIS_MODULE,
3458         .llseek = no_llseek,
3459         .read_iter  = tun_chr_read_iter,
3460         .write_iter = tun_chr_write_iter,
3461         .poll   = tun_chr_poll,
3462         .unlocked_ioctl = tun_chr_ioctl,
3463 #ifdef CONFIG_COMPAT
3464         .compat_ioctl = tun_chr_compat_ioctl,
3465 #endif
3466         .open   = tun_chr_open,
3467         .release = tun_chr_close,
3468         .fasync = tun_chr_fasync,
3469 #ifdef CONFIG_PROC_FS
3470         .show_fdinfo = tun_chr_show_fdinfo,
3471 #endif
3472 };
3473
3474 static struct miscdevice tun_miscdev = {
3475         .minor = TUN_MINOR,
3476         .name = "tun",
3477         .nodename = "net/tun",
3478         .fops = &tun_fops,
3479 };
3480
3481 /* ethtool interface */
3482
3483 static void tun_default_link_ksettings(struct net_device *dev,
3484                                        struct ethtool_link_ksettings *cmd)
3485 {
3486         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3487         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3488         cmd->base.speed         = SPEED_10;
3489         cmd->base.duplex        = DUPLEX_FULL;
3490         cmd->base.port          = PORT_TP;
3491         cmd->base.phy_address   = 0;
3492         cmd->base.autoneg       = AUTONEG_DISABLE;
3493 }
3494
3495 static int tun_get_link_ksettings(struct net_device *dev,
3496                                   struct ethtool_link_ksettings *cmd)
3497 {
3498         struct tun_struct *tun = netdev_priv(dev);
3499
3500         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3501         return 0;
3502 }
3503
3504 static int tun_set_link_ksettings(struct net_device *dev,
3505                                   const struct ethtool_link_ksettings *cmd)
3506 {
3507         struct tun_struct *tun = netdev_priv(dev);
3508
3509         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3510         return 0;
3511 }
3512
3513 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3514 {
3515         struct tun_struct *tun = netdev_priv(dev);
3516
3517         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3518         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3519
3520         switch (tun->flags & TUN_TYPE_MASK) {
3521         case IFF_TUN:
3522                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3523                 break;
3524         case IFF_TAP:
3525                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3526                 break;
3527         }
3528 }
3529
3530 static u32 tun_get_msglevel(struct net_device *dev)
3531 {
3532         struct tun_struct *tun = netdev_priv(dev);
3533
3534         return tun->msg_enable;
3535 }
3536
3537 static void tun_set_msglevel(struct net_device *dev, u32 value)
3538 {
3539         struct tun_struct *tun = netdev_priv(dev);
3540
3541         tun->msg_enable = value;
3542 }
3543
3544 static int tun_get_coalesce(struct net_device *dev,
3545                             struct ethtool_coalesce *ec,
3546                             struct kernel_ethtool_coalesce *kernel_coal,
3547                             struct netlink_ext_ack *extack)
3548 {
3549         struct tun_struct *tun = netdev_priv(dev);
3550
3551         ec->rx_max_coalesced_frames = tun->rx_batched;
3552
3553         return 0;
3554 }
3555
3556 static int tun_set_coalesce(struct net_device *dev,
3557                             struct ethtool_coalesce *ec,
3558                             struct kernel_ethtool_coalesce *kernel_coal,
3559                             struct netlink_ext_ack *extack)
3560 {
3561         struct tun_struct *tun = netdev_priv(dev);
3562
3563         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3564                 tun->rx_batched = NAPI_POLL_WEIGHT;
3565         else
3566                 tun->rx_batched = ec->rx_max_coalesced_frames;
3567
3568         return 0;
3569 }
3570
3571 static const struct ethtool_ops tun_ethtool_ops = {
3572         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3573         .get_drvinfo    = tun_get_drvinfo,
3574         .get_msglevel   = tun_get_msglevel,
3575         .set_msglevel   = tun_set_msglevel,
3576         .get_link       = ethtool_op_get_link,
3577         .get_ts_info    = ethtool_op_get_ts_info,
3578         .get_coalesce   = tun_get_coalesce,
3579         .set_coalesce   = tun_set_coalesce,
3580         .get_link_ksettings = tun_get_link_ksettings,
3581         .set_link_ksettings = tun_set_link_ksettings,
3582 };
3583
3584 static int tun_queue_resize(struct tun_struct *tun)
3585 {
3586         struct net_device *dev = tun->dev;
3587         struct tun_file *tfile;
3588         struct ptr_ring **rings;
3589         int n = tun->numqueues + tun->numdisabled;
3590         int ret, i;
3591
3592         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3593         if (!rings)
3594                 return -ENOMEM;
3595
3596         for (i = 0; i < tun->numqueues; i++) {
3597                 tfile = rtnl_dereference(tun->tfiles[i]);
3598                 rings[i] = &tfile->tx_ring;
3599         }
3600         list_for_each_entry(tfile, &tun->disabled, next)
3601                 rings[i++] = &tfile->tx_ring;
3602
3603         ret = ptr_ring_resize_multiple(rings, n,
3604                                        dev->tx_queue_len, GFP_KERNEL,
3605                                        tun_ptr_free);
3606
3607         kfree(rings);
3608         return ret;
3609 }
3610
3611 static int tun_device_event(struct notifier_block *unused,
3612                             unsigned long event, void *ptr)
3613 {
3614         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3615         struct tun_struct *tun = netdev_priv(dev);
3616         int i;
3617
3618         if (dev->rtnl_link_ops != &tun_link_ops)
3619                 return NOTIFY_DONE;
3620
3621         switch (event) {
3622         case NETDEV_CHANGE_TX_QUEUE_LEN:
3623                 if (tun_queue_resize(tun))
3624                         return NOTIFY_BAD;
3625                 break;
3626         case NETDEV_UP:
3627                 for (i = 0; i < tun->numqueues; i++) {
3628                         struct tun_file *tfile;
3629
3630                         tfile = rtnl_dereference(tun->tfiles[i]);
3631                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3632                 }
3633                 break;
3634         default:
3635                 break;
3636         }
3637
3638         return NOTIFY_DONE;
3639 }
3640
3641 static struct notifier_block tun_notifier_block __read_mostly = {
3642         .notifier_call  = tun_device_event,
3643 };
3644
3645 static int __init tun_init(void)
3646 {
3647         int ret = 0;
3648
3649         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3650
3651         ret = rtnl_link_register(&tun_link_ops);
3652         if (ret) {
3653                 pr_err("Can't register link_ops\n");
3654                 goto err_linkops;
3655         }
3656
3657         ret = misc_register(&tun_miscdev);
3658         if (ret) {
3659                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3660                 goto err_misc;
3661         }
3662
3663         ret = register_netdevice_notifier(&tun_notifier_block);
3664         if (ret) {
3665                 pr_err("Can't register netdevice notifier\n");
3666                 goto err_notifier;
3667         }
3668
3669         return  0;
3670
3671 err_notifier:
3672         misc_deregister(&tun_miscdev);
3673 err_misc:
3674         rtnl_link_unregister(&tun_link_ops);
3675 err_linkops:
3676         return ret;
3677 }
3678
3679 static void tun_cleanup(void)
3680 {
3681         misc_deregister(&tun_miscdev);
3682         rtnl_link_unregister(&tun_link_ops);
3683         unregister_netdevice_notifier(&tun_notifier_block);
3684 }
3685
3686 /* Get an underlying socket object from tun file.  Returns error unless file is
3687  * attached to a device.  The returned object works like a packet socket, it
3688  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3689  * holding a reference to the file for as long as the socket is in use. */
3690 struct socket *tun_get_socket(struct file *file)
3691 {
3692         struct tun_file *tfile;
3693         if (file->f_op != &tun_fops)
3694                 return ERR_PTR(-EINVAL);
3695         tfile = file->private_data;
3696         if (!tfile)
3697                 return ERR_PTR(-EBADFD);
3698         return &tfile->socket;
3699 }
3700 EXPORT_SYMBOL_GPL(tun_get_socket);
3701
3702 struct ptr_ring *tun_get_tx_ring(struct file *file)
3703 {
3704         struct tun_file *tfile;
3705
3706         if (file->f_op != &tun_fops)
3707                 return ERR_PTR(-EINVAL);
3708         tfile = file->private_data;
3709         if (!tfile)
3710                 return ERR_PTR(-EBADFD);
3711         return &tfile->tx_ring;
3712 }
3713 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3714
3715 module_init(tun_init);
3716 module_exit(tun_cleanup);
3717 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3718 MODULE_AUTHOR(DRV_COPYRIGHT);
3719 MODULE_LICENSE("GPL");
3720 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3721 MODULE_ALIAS("devname:net/tun");