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 <linux/seq_file.h>
  66 #include <linux/uio.h>
  67 #include <linux/skb_array.h>
  68 #include <linux/bpf.h>
  69 #include <linux/bpf_trace.h>
  70 #include <linux/mutex.h>
  71
  72 #include <linux/uaccess.h>
  73 #include <linux/proc_fs.h>
  74
  75 static void tun_default_link_ksettings(struct net_device *dev,
  76                                        struct ethtool_link_ksettings *cmd);
  77
  78 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  79
  80 /* TUN device flags */
  81
  82 /* IFF_ATTACH_QUEUE is never stored in device flags,
  83  * overload it to mean fasync when stored there.
  84  */
  85 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  86 /* High bits in flags field are unused. */
  87 #define TUN_VNET_LE     0x80000000
  88 #define TUN_VNET_BE     0x40000000
  89
  90 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
  91                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
  92
  93 #define GOODCOPY_LEN 128
  94
  95 #define FLT_EXACT_COUNT 8
  96 struct tap_filter {
  97         unsigned int    count;    /* Number of addrs. Zero means disabled */
  98         u32             mask[2];  /* Mask of the hashed addrs */
  99         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 100 };
 101
 102 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 103  * to max number of VCPUs in guest. */
 104 #define MAX_TAP_QUEUES 256
 105 #define MAX_TAP_FLOWS  4096
 106
 107 #define TUN_FLOW_EXPIRE (3 * HZ)
 108
 109 struct tun_pcpu_stats {
 110         u64_stats_t rx_packets;
 111         u64_stats_t rx_bytes;
 112         u64_stats_t tx_packets;
 113         u64_stats_t tx_bytes;
 114         struct u64_stats_sync syncp;
 115         u32 rx_dropped;
 116         u32 tx_dropped;
 117         u32 rx_frame_errors;
 118 };
 119
 120 /* A tun_file connects an open character device to a tuntap netdevice. It
 121  * also contains all socket related structures (except sock_fprog and tap_filter)
 122  * to serve as one transmit queue for tuntap device. The sock_fprog and
 123  * tap_filter were kept in tun_struct since they were used for filtering for the
 124  * netdevice not for a specific queue (at least I didn't see the requirement for
 125  * this).
 126  *
 127  * RCU usage:
 128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 129  * other can only be read while rcu_read_lock or rtnl_lock is held.
 130  */
 131 struct tun_file {
 132         struct sock sk;
 133         struct socket socket;
 134         struct tun_struct __rcu *tun;
 135         struct fasync_struct *fasync;
 136         /* only used for fasnyc */
 137         unsigned int flags;
 138         union {
 139                 u16 queue_index;
 140                 unsigned int ifindex;
 141         };
 142         struct napi_struct napi;
 143         bool napi_enabled;
 144         bool napi_frags_enabled;
 145         struct mutex napi_mutex;        /* Protects access to the above napi */
 146         struct list_head next;
 147         struct tun_struct *detached;
 148         struct ptr_ring tx_ring;
 149         struct xdp_rxq_info xdp_rxq;
 150 };
 151
 152 struct tun_page {
 153         struct page *page;
 154         int count;
 155 };
 156
 157 struct tun_flow_entry {
 158         struct hlist_node hash_link;
 159         struct rcu_head rcu;
 160         struct tun_struct *tun;
 161
 162         u32 rxhash;
 163         u32 rps_rxhash;
 164         int queue_index;
 165         unsigned long updated ____cacheline_aligned_in_smp;
 166 };
 167
 168 #define TUN_NUM_FLOW_ENTRIES 1024
 169 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 170
 171 struct tun_prog {
 172         struct rcu_head rcu;
 173         struct bpf_prog *prog;
 174 };
 175
 176 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 177  * device, socket filter, sndbuf and vnet header size were restore when the
 178  * file were attached to a persist device.
 179  */
 180 struct tun_struct {
 181         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 182         unsigned int            numqueues;
 183         unsigned int            flags;
 184         kuid_t                  owner;
 185         kgid_t                  group;
 186
 187         struct net_device       *dev;
 188         netdev_features_t       set_features;
 189 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 190                           NETIF_F_TSO6)
 191
 192         int                     align;
 193         int                     vnet_hdr_sz;
 194         int                     sndbuf;
 195         struct tap_filter       txflt;
 196         struct sock_fprog       fprog;
 197         /* protected by rtnl lock */
 198         bool                    filter_attached;
 199         u32                     msg_enable;
 200         spinlock_t lock;
 201         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202         struct timer_list flow_gc_timer;
 203         unsigned long ageing_time;
 204         unsigned int numdisabled;
 205         struct list_head disabled;
 206         void *security;
 207         u32 flow_count;
 208         u32 rx_batched;
 209         struct tun_pcpu_stats __percpu *pcpu_stats;
 210         struct bpf_prog __rcu *xdp_prog;
 211         struct tun_prog __rcu *steering_prog;
 212         struct tun_prog __rcu *filter_prog;
 213         struct ethtool_link_ksettings link_ksettings;
 214 };
 215
 216 struct veth {
 217         __be16 h_vlan_proto;
 218         __be16 h_vlan_TCI;
 219 };
 220
 221 bool tun_is_xdp_frame(void *ptr)
 222 {
 223         return (unsigned long)ptr & TUN_XDP_FLAG;
 224 }
 225 EXPORT_SYMBOL(tun_is_xdp_frame);
 226
 227 void *tun_xdp_to_ptr(void *ptr)
 228 {
 229         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 230 }
 231 EXPORT_SYMBOL(tun_xdp_to_ptr);
 232
 233 void *tun_ptr_to_xdp(void *ptr)
 234 {
 235         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 236 }
 237 EXPORT_SYMBOL(tun_ptr_to_xdp);
 238
 239 static int tun_napi_receive(struct napi_struct *napi, int budget)
 240 {
 241         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 242         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 243         struct sk_buff_head process_queue;
 244         struct sk_buff *skb;
 245         int received = 0;
 246
 247         __skb_queue_head_init(&process_queue);
 248
 249         spin_lock(&queue->lock);
 250         skb_queue_splice_tail_init(queue, &process_queue);
 251         spin_unlock(&queue->lock);
 252
 253         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 254                 napi_gro_receive(napi, skb);
 255                 ++received;
 256         }
 257
 258         if (!skb_queue_empty(&process_queue)) {
 259                 spin_lock(&queue->lock);
 260                 skb_queue_splice(&process_queue, queue);
 261                 spin_unlock(&queue->lock);
 262         }
 263
 264         return received;
 265 }
 266
 267 static int tun_napi_poll(struct napi_struct *napi, int budget)
 268 {
 269         unsigned int received;
 270
 271         received = tun_napi_receive(napi, budget);
 272
 273         if (received < budget)
 274                 napi_complete_done(napi, received);
 275
 276         return received;
 277 }
 278
 279 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 280                           bool napi_en, bool napi_frags)
 281 {
 282         tfile->napi_enabled = napi_en;
 283         tfile->napi_frags_enabled = napi_en && napi_frags;
 284         if (napi_en) {
 285                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 286                                   NAPI_POLL_WEIGHT);
 287                 napi_enable(&tfile->napi);
 288         }
 289 }
 290
 291 static void tun_napi_disable(struct tun_file *tfile)
 292 {
 293         if (tfile->napi_enabled)
 294                 napi_disable(&tfile->napi);
 295 }
 296
 297 static void tun_napi_del(struct tun_file *tfile)
 298 {
 299         if (tfile->napi_enabled)
 300                 netif_napi_del(&tfile->napi);
 301 }
 302
 303 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 304 {
 305         return tfile->napi_frags_enabled;
 306 }
 307
 308 #ifdef CONFIG_TUN_VNET_CROSS_LE
 309 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 310 {
 311         return tun->flags & TUN_VNET_BE ? false :
 312                 virtio_legacy_is_little_endian();
 313 }
 314
 315 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 316 {
 317         int be = !!(tun->flags & TUN_VNET_BE);
 318
 319         if (put_user(be, argp))
 320                 return -EFAULT;
 321
 322         return 0;
 323 }
 324
 325 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 326 {
 327         int be;
 328
 329         if (get_user(be, argp))
 330                 return -EFAULT;
 331
 332         if (be)
 333                 tun->flags |= TUN_VNET_BE;
 334         else
 335                 tun->flags &= ~TUN_VNET_BE;
 336
 337         return 0;
 338 }
 339 #else
 340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 341 {
 342         return virtio_legacy_is_little_endian();
 343 }
 344
 345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 346 {
 347         return -EINVAL;
 348 }
 349
 350 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 351 {
 352         return -EINVAL;
 353 }
 354 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 355
 356 static inline bool tun_is_little_endian(struct tun_struct *tun)
 357 {
 358         return tun->flags & TUN_VNET_LE ||
 359                 tun_legacy_is_little_endian(tun);
 360 }
 361
 362 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 363 {
 364         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 365 }
 366
 367 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 368 {
 369         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 370 }
 371
 372 static inline u32 tun_hashfn(u32 rxhash)
 373 {
 374         return rxhash & TUN_MASK_FLOW_ENTRIES;
 375 }
 376
 377 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 378 {
 379         struct tun_flow_entry *e;
 380
 381         hlist_for_each_entry_rcu(e, head, hash_link) {
 382                 if (e->rxhash == rxhash)
 383                         return e;
 384         }
 385         return NULL;
 386 }
 387
 388 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 389                                               struct hlist_head *head,
 390                                               u32 rxhash, u16 queue_index)
 391 {
 392         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 393
 394         if (e) {
 395                 netif_info(tun, tx_queued, tun->dev,
 396                            "create flow: hash %u index %u\n",
 397                            rxhash, queue_index);
 398                 e->updated = jiffies;
 399                 e->rxhash = rxhash;
 400                 e->rps_rxhash = 0;
 401                 e->queue_index = queue_index;
 402                 e->tun = tun;
 403                 hlist_add_head_rcu(&e->hash_link, head);
 404                 ++tun->flow_count;
 405         }
 406         return e;
 407 }
 408
 409 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 410 {
 411         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 412                    e->rxhash, e->queue_index);
 413         hlist_del_rcu(&e->hash_link);
 414         kfree_rcu(e, rcu);
 415         --tun->flow_count;
 416 }
 417
 418 static void tun_flow_flush(struct tun_struct *tun)
 419 {
 420         int i;
 421
 422         spin_lock_bh(&tun->lock);
 423         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 424                 struct tun_flow_entry *e;
 425                 struct hlist_node *n;
 426
 427                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 428                         tun_flow_delete(tun, e);
 429         }
 430         spin_unlock_bh(&tun->lock);
 431 }
 432
 433 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 434 {
 435         int i;
 436
 437         spin_lock_bh(&tun->lock);
 438         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 439                 struct tun_flow_entry *e;
 440                 struct hlist_node *n;
 441
 442                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 443                         if (e->queue_index == queue_index)
 444                                 tun_flow_delete(tun, e);
 445                 }
 446         }
 447         spin_unlock_bh(&tun->lock);
 448 }
 449
 450 static void tun_flow_cleanup(struct timer_list *t)
 451 {
 452         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 453         unsigned long delay = tun->ageing_time;
 454         unsigned long next_timer = jiffies + delay;
 455         unsigned long count = 0;
 456         int i;
 457
 458         spin_lock(&tun->lock);
 459         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 460                 struct tun_flow_entry *e;
 461                 struct hlist_node *n;
 462
 463                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 464                         unsigned long this_timer;
 465
 466                         this_timer = e->updated + delay;
 467                         if (time_before_eq(this_timer, jiffies)) {
 468                                 tun_flow_delete(tun, e);
 469                                 continue;
 470                         }
 471                         count++;
 472                         if (time_before(this_timer, next_timer))
 473                                 next_timer = this_timer;
 474                 }
 475         }
 476
 477         if (count)
 478                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 479         spin_unlock(&tun->lock);
 480 }
 481
 482 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 483                             struct tun_file *tfile)
 484 {
 485         struct hlist_head *head;
 486         struct tun_flow_entry *e;
 487         unsigned long delay = tun->ageing_time;
 488         u16 queue_index = tfile->queue_index;
 489
 490         head = &tun->flows[tun_hashfn(rxhash)];
 491
 492         rcu_read_lock();
 493
 494         e = tun_flow_find(head, rxhash);
 495         if (likely(e)) {
 496                 /* TODO: keep queueing to old queue until it's empty? */
 497                 if (READ_ONCE(e->queue_index) != queue_index)
 498                         WRITE_ONCE(e->queue_index, queue_index);
 499                 if (e->updated != jiffies)
 500                         e->updated = jiffies;
 501                 sock_rps_record_flow_hash(e->rps_rxhash);
 502         } else {
 503                 spin_lock_bh(&tun->lock);
 504                 if (!tun_flow_find(head, rxhash) &&
 505                     tun->flow_count < MAX_TAP_FLOWS)
 506                         tun_flow_create(tun, head, rxhash, queue_index);
 507
 508                 if (!timer_pending(&tun->flow_gc_timer))
 509                         mod_timer(&tun->flow_gc_timer,
 510                                   round_jiffies_up(jiffies + delay));
 511                 spin_unlock_bh(&tun->lock);
 512         }
 513
 514         rcu_read_unlock();
 515 }
 516
 517 /* Save the hash received in the stack receive path and update the
 518  * flow_hash table accordingly.
 519  */
 520 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 521 {
 522         if (unlikely(e->rps_rxhash != hash))
 523                 e->rps_rxhash = hash;
 524 }
 525
 526 /* We try to identify a flow through its rxhash. The reason that
 527  * we do not check rxq no. is because some cards(e.g 82599), chooses
 528  * the rxq based on the txq where the last packet of the flow comes. As
 529  * the userspace application move between processors, we may get a
 530  * different rxq no. here.
 531  */
 532 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 533 {
 534         struct tun_flow_entry *e;
 535         u32 txq = 0;
 536         u32 numqueues = 0;
 537
 538         numqueues = READ_ONCE(tun->numqueues);
 539
 540         txq = __skb_get_hash_symmetric(skb);
 541         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 542         if (e) {
 543                 tun_flow_save_rps_rxhash(e, txq);
 544                 txq = e->queue_index;
 545         } else {
 546                 /* use multiply and shift instead of expensive divide */
 547                 txq = ((u64)txq * numqueues) >> 32;
 548         }
 549
 550         return txq;
 551 }
 552
 553 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 554 {
 555         struct tun_prog *prog;
 556         u32 numqueues;
 557         u16 ret = 0;
 558
 559         numqueues = READ_ONCE(tun->numqueues);
 560         if (!numqueues)
 561                 return 0;
 562
 563         prog = rcu_dereference(tun->steering_prog);
 564         if (prog)
 565                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 566
 567         return ret % numqueues;
 568 }
 569
 570 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 571                             struct net_device *sb_dev)
 572 {
 573         struct tun_struct *tun = netdev_priv(dev);
 574         u16 ret;
 575
 576         rcu_read_lock();
 577         if (rcu_dereference(tun->steering_prog))
 578                 ret = tun_ebpf_select_queue(tun, skb);
 579         else
 580                 ret = tun_automq_select_queue(tun, skb);
 581         rcu_read_unlock();
 582
 583         return ret;
 584 }
 585
 586 static inline bool tun_not_capable(struct tun_struct *tun)
 587 {
 588         const struct cred *cred = current_cred();
 589         struct net *net = dev_net(tun->dev);
 590
 591         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 592                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 593                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 594 }
 595
 596 static void tun_set_real_num_queues(struct tun_struct *tun)
 597 {
 598         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 599         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 600 }
 601
 602 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 603 {
 604         tfile->detached = tun;
 605         list_add_tail(&tfile->next, &tun->disabled);
 606         ++tun->numdisabled;
 607 }
 608
 609 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 610 {
 611         struct tun_struct *tun = tfile->detached;
 612
 613         tfile->detached = NULL;
 614         list_del_init(&tfile->next);
 615         --tun->numdisabled;
 616         return tun;
 617 }
 618
 619 void tun_ptr_free(void *ptr)
 620 {
 621         if (!ptr)
 622                 return;
 623         if (tun_is_xdp_frame(ptr)) {
 624                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 625
 626                 xdp_return_frame(xdpf);
 627         } else {
 628                 __skb_array_destroy_skb(ptr);
 629         }
 630 }
 631 EXPORT_SYMBOL_GPL(tun_ptr_free);
 632
 633 static void tun_queue_purge(struct tun_file *tfile)
 634 {
 635         void *ptr;
 636
 637         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 638                 tun_ptr_free(ptr);
 639
 640         skb_queue_purge(&tfile->sk.sk_write_queue);
 641         skb_queue_purge(&tfile->sk.sk_error_queue);
 642 }
 643
 644 static void __tun_detach(struct tun_file *tfile, bool clean)
 645 {
 646         struct tun_file *ntfile;
 647         struct tun_struct *tun;
 648
 649         tun = rtnl_dereference(tfile->tun);
 650
 651         if (tun && clean) {
 652                 tun_napi_disable(tfile);
 653                 tun_napi_del(tfile);
 654         }
 655
 656         if (tun && !tfile->detached) {
 657                 u16 index = tfile->queue_index;
 658                 BUG_ON(index >= tun->numqueues);
 659
 660                 rcu_assign_pointer(tun->tfiles[index],
 661                                    tun->tfiles[tun->numqueues - 1]);
 662                 ntfile = rtnl_dereference(tun->tfiles[index]);
 663                 ntfile->queue_index = index;
 664                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 665                                    NULL);
 666
 667                 --tun->numqueues;
 668                 if (clean) {
 669                         RCU_INIT_POINTER(tfile->tun, NULL);
 670                         sock_put(&tfile->sk);
 671                 } else
 672                         tun_disable_queue(tun, tfile);
 673
 674                 synchronize_net();
 675                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 676                 /* Drop read queue */
 677                 tun_queue_purge(tfile);
 678                 tun_set_real_num_queues(tun);
 679         } else if (tfile->detached && clean) {
 680                 tun = tun_enable_queue(tfile);
 681                 sock_put(&tfile->sk);
 682         }
 683
 684         if (clean) {
 685                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 686                         netif_carrier_off(tun->dev);
 687
 688                         if (!(tun->flags & IFF_PERSIST) &&
 689                             tun->dev->reg_state == NETREG_REGISTERED)
 690                                 unregister_netdevice(tun->dev);
 691                 }
 692                 if (tun)
 693                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 694                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 695                 sock_put(&tfile->sk);
 696         }
 697 }
 698
 699 static void tun_detach(struct tun_file *tfile, bool clean)
 700 {
 701         struct tun_struct *tun;
 702         struct net_device *dev;
 703
 704         rtnl_lock();
 705         tun = rtnl_dereference(tfile->tun);
 706         dev = tun ? tun->dev : NULL;
 707         __tun_detach(tfile, clean);
 708         if (dev)
 709                 netdev_state_change(dev);
 710         rtnl_unlock();
 711 }
 712
 713 static void tun_detach_all(struct net_device *dev)
 714 {
 715         struct tun_struct *tun = netdev_priv(dev);
 716         struct tun_file *tfile, *tmp;
 717         int i, n = tun->numqueues;
 718
 719         for (i = 0; i < n; i++) {
 720                 tfile = rtnl_dereference(tun->tfiles[i]);
 721                 BUG_ON(!tfile);
 722                 tun_napi_disable(tfile);
 723                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 724                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 725                 RCU_INIT_POINTER(tfile->tun, NULL);
 726                 --tun->numqueues;
 727         }
 728         list_for_each_entry(tfile, &tun->disabled, next) {
 729                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 730                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 731                 RCU_INIT_POINTER(tfile->tun, NULL);
 732         }
 733         BUG_ON(tun->numqueues != 0);
 734
 735         synchronize_net();
 736         for (i = 0; i < n; i++) {
 737                 tfile = rtnl_dereference(tun->tfiles[i]);
 738                 tun_napi_del(tfile);
 739                 /* Drop read queue */
 740                 tun_queue_purge(tfile);
 741                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 742                 sock_put(&tfile->sk);
 743         }
 744         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 745                 tun_enable_queue(tfile);
 746                 tun_queue_purge(tfile);
 747                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 748                 sock_put(&tfile->sk);
 749         }
 750         BUG_ON(tun->numdisabled != 0);
 751
 752         if (tun->flags & IFF_PERSIST)
 753                 module_put(THIS_MODULE);
 754 }
 755
 756 static int tun_attach(struct tun_struct *tun, struct file *file,
 757                       bool skip_filter, bool napi, bool napi_frags,
 758                       bool publish_tun)
 759 {
 760         struct tun_file *tfile = file->private_data;
 761         struct net_device *dev = tun->dev;
 762         int err;
 763
 764         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 765         if (err < 0)
 766                 goto out;
 767
 768         err = -EINVAL;
 769         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 770                 goto out;
 771
 772         err = -EBUSY;
 773         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 774                 goto out;
 775
 776         err = -E2BIG;
 777         if (!tfile->detached &&
 778             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 779                 goto out;
 780
 781         err = 0;
 782
 783         /* Re-attach the filter to persist device */
 784         if (!skip_filter && (tun->filter_attached == true)) {
 785                 lock_sock(tfile->socket.sk);
 786                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 787                 release_sock(tfile->socket.sk);
 788                 if (!err)
 789                         goto out;
 790         }
 791
 792         if (!tfile->detached &&
 793             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 794                             GFP_KERNEL, tun_ptr_free)) {
 795                 err = -ENOMEM;
 796                 goto out;
 797         }
 798
 799         tfile->queue_index = tun->numqueues;
 800         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 801
 802         if (tfile->detached) {
 803                 /* Re-attach detached tfile, updating XDP queue_index */
 804                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 805
 806                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 807                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 808         } else {
 809                 /* Setup XDP RX-queue info, for new tfile getting attached */
 810                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 811                                        tun->dev, tfile->queue_index);
 812                 if (err < 0)
 813                         goto out;
 814                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 815                                                  MEM_TYPE_PAGE_SHARED, NULL);
 816                 if (err < 0) {
 817                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 818                         goto out;
 819                 }
 820                 err = 0;
 821         }
 822
 823         if (tfile->detached) {
 824                 tun_enable_queue(tfile);
 825         } else {
 826                 sock_hold(&tfile->sk);
 827                 tun_napi_init(tun, tfile, napi, napi_frags);
 828         }
 829
 830         if (rtnl_dereference(tun->xdp_prog))
 831                 sock_set_flag(&tfile->sk, SOCK_XDP);
 832
 833         /* device is allowed to go away first, so no need to hold extra
 834          * refcnt.
 835          */
 836
 837         /* Publish tfile->tun and tun->tfiles only after we've fully
 838          * initialized tfile; otherwise we risk using half-initialized
 839          * object.
 840          */
 841         if (publish_tun)
 842                 rcu_assign_pointer(tfile->tun, tun);
 843         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 844         tun->numqueues++;
 845         tun_set_real_num_queues(tun);
 846 out:
 847         return err;
 848 }
 849
 850 static struct tun_struct *tun_get(struct tun_file *tfile)
 851 {
 852         struct tun_struct *tun;
 853
 854         rcu_read_lock();
 855         tun = rcu_dereference(tfile->tun);
 856         if (tun)
 857                 dev_hold(tun->dev);
 858         rcu_read_unlock();
 859
 860         return tun;
 861 }
 862
 863 static void tun_put(struct tun_struct *tun)
 864 {
 865         dev_put(tun->dev);
 866 }
 867
 868 /* TAP filtering */
 869 static void addr_hash_set(u32 *mask, const u8 *addr)
 870 {
 871         int n = ether_crc(ETH_ALEN, addr) >> 26;
 872         mask[n >> 5] |= (1 << (n & 31));
 873 }
 874
 875 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 876 {
 877         int n = ether_crc(ETH_ALEN, addr) >> 26;
 878         return mask[n >> 5] & (1 << (n & 31));
 879 }
 880
 881 static int update_filter(struct tap_filter *filter, void __user *arg)
 882 {
 883         struct { u8 u[ETH_ALEN]; } *addr;
 884         struct tun_filter uf;
 885         int err, alen, n, nexact;
 886
 887         if (copy_from_user(&uf, arg, sizeof(uf)))
 888                 return -EFAULT;
 889
 890         if (!uf.count) {
 891                 /* Disabled */
 892                 filter->count = 0;
 893                 return 0;
 894         }
 895
 896         alen = ETH_ALEN * uf.count;
 897         addr = memdup_user(arg + sizeof(uf), alen);
 898         if (IS_ERR(addr))
 899                 return PTR_ERR(addr);
 900
 901         /* The filter is updated without holding any locks. Which is
 902          * perfectly safe. We disable it first and in the worst
 903          * case we'll accept a few undesired packets. */
 904         filter->count = 0;
 905         wmb();
 906
 907         /* Use first set of addresses as an exact filter */
 908         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 909                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 910
 911         nexact = n;
 912
 913         /* Remaining multicast addresses are hashed,
 914          * unicast will leave the filter disabled. */
 915         memset(filter->mask, 0, sizeof(filter->mask));
 916         for (; n < uf.count; n++) {
 917                 if (!is_multicast_ether_addr(addr[n].u)) {
 918                         err = 0; /* no filter */
 919                         goto free_addr;
 920                 }
 921                 addr_hash_set(filter->mask, addr[n].u);
 922         }
 923
 924         /* For ALLMULTI just set the mask to all ones.
 925          * This overrides the mask populated above. */
 926         if ((uf.flags & TUN_FLT_ALLMULTI))
 927                 memset(filter->mask, ~0, sizeof(filter->mask));
 928
 929         /* Now enable the filter */
 930         wmb();
 931         filter->count = nexact;
 932
 933         /* Return the number of exact filters */
 934         err = nexact;
 935 free_addr:
 936         kfree(addr);
 937         return err;
 938 }
 939
 940 /* Returns: 0 - drop, !=0 - accept */
 941 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 942 {
 943         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 944          * at this point. */
 945         struct ethhdr *eh = (struct ethhdr *) skb->data;
 946         int i;
 947
 948         /* Exact match */
 949         for (i = 0; i < filter->count; i++)
 950                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 951                         return 1;
 952
 953         /* Inexact match (multicast only) */
 954         if (is_multicast_ether_addr(eh->h_dest))
 955                 return addr_hash_test(filter->mask, eh->h_dest);
 956
 957         return 0;
 958 }
 959
 960 /*
 961  * Checks whether the packet is accepted or not.
 962  * Returns: 0 - drop, !=0 - accept
 963  */
 964 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 965 {
 966         if (!filter->count)
 967                 return 1;
 968
 969         return run_filter(filter, skb);
 970 }
 971
 972 /* Network device part of the driver */
 973
 974 static const struct ethtool_ops tun_ethtool_ops;
 975
 976 /* Net device detach from fd. */
 977 static void tun_net_uninit(struct net_device *dev)
 978 {
 979         tun_detach_all(dev);
 980 }
 981
 982 /* Net device open. */
 983 static int tun_net_open(struct net_device *dev)
 984 {
 985         netif_tx_start_all_queues(dev);
 986
 987         return 0;
 988 }
 989
 990 /* Net device close. */
 991 static int tun_net_close(struct net_device *dev)
 992 {
 993         netif_tx_stop_all_queues(dev);
 994         return 0;
 995 }
 996
 997 /* Net device start xmit */
 998 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
 999 {
1000 #ifdef CONFIG_RPS
1001         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1002                 /* Select queue was not called for the skbuff, so we extract the
1003                  * RPS hash and save it into the flow_table here.
1004                  */
1005                 struct tun_flow_entry *e;
1006                 __u32 rxhash;
1007
1008                 rxhash = __skb_get_hash_symmetric(skb);
1009                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1010                 if (e)
1011                         tun_flow_save_rps_rxhash(e, rxhash);
1012         }
1013 #endif
1014 }
1015
1016 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1017                                     struct sk_buff *skb,
1018                                     int len)
1019 {
1020         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1021
1022         if (prog)
1023                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1024
1025         return len;
1026 }
1027
1028 /* Net device start xmit */
1029 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1030 {
1031         struct tun_struct *tun = netdev_priv(dev);
1032         int txq = skb->queue_mapping;
1033         struct tun_file *tfile;
1034         int len = skb->len;
1035
1036         rcu_read_lock();
1037         tfile = rcu_dereference(tun->tfiles[txq]);
1038
1039         /* Drop packet if interface is not attached */
1040         if (!tfile)
1041                 goto drop;
1042
1043         if (!rcu_dereference(tun->steering_prog))
1044                 tun_automq_xmit(tun, skb);
1045
1046         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1047
1048         /* Drop if the filter does not like it.
1049          * This is a noop if the filter is disabled.
1050          * Filter can be enabled only for the TAP devices. */
1051         if (!check_filter(&tun->txflt, skb))
1052                 goto drop;
1053
1054         if (tfile->socket.sk->sk_filter &&
1055             sk_filter(tfile->socket.sk, skb))
1056                 goto drop;
1057
1058         len = run_ebpf_filter(tun, skb, len);
1059         if (len == 0 || pskb_trim(skb, len))
1060                 goto drop;
1061
1062         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1063                 goto drop;
1064
1065         skb_tx_timestamp(skb);
1066
1067         /* Orphan the skb - required as we might hang on to it
1068          * for indefinite time.
1069          */
1070         skb_orphan(skb);
1071
1072         nf_reset_ct(skb);
1073
1074         if (ptr_ring_produce(&tfile->tx_ring, skb))
1075                 goto drop;
1076
1077         /* Notify and wake up reader process */
1078         if (tfile->flags & TUN_FASYNC)
1079                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1080         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1081
1082         rcu_read_unlock();
1083         return NETDEV_TX_OK;
1084
1085 drop:
1086         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1087         skb_tx_error(skb);
1088         kfree_skb(skb);
1089         rcu_read_unlock();
1090         return NET_XMIT_DROP;
1091 }
1092
1093 static void tun_net_mclist(struct net_device *dev)
1094 {
1095         /*
1096          * This callback is supposed to deal with mc filter in
1097          * _rx_ path and has nothing to do with the _tx_ path.
1098          * In rx path we always accept everything userspace gives us.
1099          */
1100 }
1101
1102 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1103         netdev_features_t features)
1104 {
1105         struct tun_struct *tun = netdev_priv(dev);
1106
1107         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1108 }
1109
1110 static void tun_set_headroom(struct net_device *dev, int new_hr)
1111 {
1112         struct tun_struct *tun = netdev_priv(dev);
1113
1114         if (new_hr < NET_SKB_PAD)
1115                 new_hr = NET_SKB_PAD;
1116
1117         tun->align = new_hr;
1118 }
1119
1120 static void
1121 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1122 {
1123         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1124         struct tun_struct *tun = netdev_priv(dev);
1125         struct tun_pcpu_stats *p;
1126         int i;
1127
1128         for_each_possible_cpu(i) {
1129                 u64 rxpackets, rxbytes, txpackets, txbytes;
1130                 unsigned int start;
1131
1132                 p = per_cpu_ptr(tun->pcpu_stats, i);
1133                 do {
1134                         start = u64_stats_fetch_begin(&p->syncp);
1135                         rxpackets       = u64_stats_read(&p->rx_packets);
1136                         rxbytes         = u64_stats_read(&p->rx_bytes);
1137                         txpackets       = u64_stats_read(&p->tx_packets);
1138                         txbytes         = u64_stats_read(&p->tx_bytes);
1139                 } while (u64_stats_fetch_retry(&p->syncp, start));
1140
1141                 stats->rx_packets       += rxpackets;
1142                 stats->rx_bytes         += rxbytes;
1143                 stats->tx_packets       += txpackets;
1144                 stats->tx_bytes         += txbytes;
1145
1146                 /* u32 counters */
1147                 rx_dropped      += p->rx_dropped;
1148                 rx_frame_errors += p->rx_frame_errors;
1149                 tx_dropped      += p->tx_dropped;
1150         }
1151         stats->rx_dropped  = rx_dropped;
1152         stats->rx_frame_errors = rx_frame_errors;
1153         stats->tx_dropped = tx_dropped;
1154 }
1155
1156 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1157                        struct netlink_ext_ack *extack)
1158 {
1159         struct tun_struct *tun = netdev_priv(dev);
1160         struct tun_file *tfile;
1161         struct bpf_prog *old_prog;
1162         int i;
1163
1164         old_prog = rtnl_dereference(tun->xdp_prog);
1165         rcu_assign_pointer(tun->xdp_prog, prog);
1166         if (old_prog)
1167                 bpf_prog_put(old_prog);
1168
1169         for (i = 0; i < tun->numqueues; i++) {
1170                 tfile = rtnl_dereference(tun->tfiles[i]);
1171                 if (prog)
1172                         sock_set_flag(&tfile->sk, SOCK_XDP);
1173                 else
1174                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1175         }
1176         list_for_each_entry(tfile, &tun->disabled, next) {
1177                 if (prog)
1178                         sock_set_flag(&tfile->sk, SOCK_XDP);
1179                 else
1180                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1181         }
1182
1183         return 0;
1184 }
1185
1186 static u32 tun_xdp_query(struct net_device *dev)
1187 {
1188         struct tun_struct *tun = netdev_priv(dev);
1189         const struct bpf_prog *xdp_prog;
1190
1191         xdp_prog = rtnl_dereference(tun->xdp_prog);
1192         if (xdp_prog)
1193                 return xdp_prog->aux->id;
1194
1195         return 0;
1196 }
1197
1198 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1199 {
1200         switch (xdp->command) {
1201         case XDP_SETUP_PROG:
1202                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1203         case XDP_QUERY_PROG:
1204                 xdp->prog_id = tun_xdp_query(dev);
1205                 return 0;
1206         default:
1207                 return -EINVAL;
1208         }
1209 }
1210
1211 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1212 {
1213         if (new_carrier) {
1214                 struct tun_struct *tun = netdev_priv(dev);
1215
1216                 if (!tun->numqueues)
1217                         return -EPERM;
1218
1219                 netif_carrier_on(dev);
1220         } else {
1221                 netif_carrier_off(dev);
1222         }
1223         return 0;
1224 }
1225
1226 static const struct net_device_ops tun_netdev_ops = {
1227         .ndo_uninit             = tun_net_uninit,
1228         .ndo_open               = tun_net_open,
1229         .ndo_stop               = tun_net_close,
1230         .ndo_start_xmit         = tun_net_xmit,
1231         .ndo_fix_features       = tun_net_fix_features,
1232         .ndo_select_queue       = tun_select_queue,
1233         .ndo_set_rx_headroom    = tun_set_headroom,
1234         .ndo_get_stats64        = tun_net_get_stats64,
1235         .ndo_change_carrier     = tun_net_change_carrier,
1236 };
1237
1238 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1239 {
1240         /* Notify and wake up reader process */
1241         if (tfile->flags & TUN_FASYNC)
1242                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1243         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1244 }
1245
1246 static int tun_xdp_xmit(struct net_device *dev, int n,
1247                         struct xdp_frame **frames, u32 flags)
1248 {
1249         struct tun_struct *tun = netdev_priv(dev);
1250         struct tun_file *tfile;
1251         u32 numqueues;
1252         int drops = 0;
1253         int cnt = n;
1254         int i;
1255
1256         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1257                 return -EINVAL;
1258
1259         rcu_read_lock();
1260
1261 resample:
1262         numqueues = READ_ONCE(tun->numqueues);
1263         if (!numqueues) {
1264                 rcu_read_unlock();
1265                 return -ENXIO; /* Caller will free/return all frames */
1266         }
1267
1268         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1269                                             numqueues]);
1270         if (unlikely(!tfile))
1271                 goto resample;
1272
1273         spin_lock(&tfile->tx_ring.producer_lock);
1274         for (i = 0; i < n; i++) {
1275                 struct xdp_frame *xdp = frames[i];
1276                 /* Encode the XDP flag into lowest bit for consumer to differ
1277                  * XDP buffer from sk_buff.
1278                  */
1279                 void *frame = tun_xdp_to_ptr(xdp);
1280
1281                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1282                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1283                         xdp_return_frame_rx_napi(xdp);
1284                         drops++;
1285                 }
1286         }
1287         spin_unlock(&tfile->tx_ring.producer_lock);
1288
1289         if (flags & XDP_XMIT_FLUSH)
1290                 __tun_xdp_flush_tfile(tfile);
1291
1292         rcu_read_unlock();
1293         return cnt - drops;
1294 }
1295
1296 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1297 {
1298         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1299
1300         if (unlikely(!frame))
1301                 return -EOVERFLOW;
1302
1303         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1304 }
1305
1306 static const struct net_device_ops tap_netdev_ops = {
1307         .ndo_uninit             = tun_net_uninit,
1308         .ndo_open               = tun_net_open,
1309         .ndo_stop               = tun_net_close,
1310         .ndo_start_xmit         = tun_net_xmit,
1311         .ndo_fix_features       = tun_net_fix_features,
1312         .ndo_set_rx_mode        = tun_net_mclist,
1313         .ndo_set_mac_address    = eth_mac_addr,
1314         .ndo_validate_addr      = eth_validate_addr,
1315         .ndo_select_queue       = tun_select_queue,
1316         .ndo_features_check     = passthru_features_check,
1317         .ndo_set_rx_headroom    = tun_set_headroom,
1318         .ndo_get_stats64        = tun_net_get_stats64,
1319         .ndo_bpf                = tun_xdp,
1320         .ndo_xdp_xmit           = tun_xdp_xmit,
1321         .ndo_change_carrier     = tun_net_change_carrier,
1322 };
1323
1324 static void tun_flow_init(struct tun_struct *tun)
1325 {
1326         int i;
1327
1328         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1329                 INIT_HLIST_HEAD(&tun->flows[i]);
1330
1331         tun->ageing_time = TUN_FLOW_EXPIRE;
1332         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1333         mod_timer(&tun->flow_gc_timer,
1334                   round_jiffies_up(jiffies + tun->ageing_time));
1335 }
1336
1337 static void tun_flow_uninit(struct tun_struct *tun)
1338 {
1339         del_timer_sync(&tun->flow_gc_timer);
1340         tun_flow_flush(tun);
1341 }
1342
1343 #define MIN_MTU 68
1344 #define MAX_MTU 65535
1345
1346 /* Initialize net device. */
1347 static void tun_net_init(struct net_device *dev)
1348 {
1349         struct tun_struct *tun = netdev_priv(dev);
1350
1351         switch (tun->flags & TUN_TYPE_MASK) {
1352         case IFF_TUN:
1353                 dev->netdev_ops = &tun_netdev_ops;
1354
1355                 /* Point-to-Point TUN Device */
1356                 dev->hard_header_len = 0;
1357                 dev->addr_len = 0;
1358                 dev->mtu = 1500;
1359
1360                 /* Zero header length */
1361                 dev->type = ARPHRD_NONE;
1362                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1363                 break;
1364
1365         case IFF_TAP:
1366                 dev->netdev_ops = &tap_netdev_ops;
1367                 /* Ethernet TAP Device */
1368                 ether_setup(dev);
1369                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1370                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1371
1372                 eth_hw_addr_random(dev);
1373
1374                 break;
1375         }
1376
1377         dev->min_mtu = MIN_MTU;
1378         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1379 }
1380
1381 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1382 {
1383         struct sock *sk = tfile->socket.sk;
1384
1385         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1386 }
1387
1388 /* Character device part */
1389
1390 /* Poll */
1391 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1392 {
1393         struct tun_file *tfile = file->private_data;
1394         struct tun_struct *tun = tun_get(tfile);
1395         struct sock *sk;
1396         __poll_t mask = 0;
1397
1398         if (!tun)
1399                 return EPOLLERR;
1400
1401         sk = tfile->socket.sk;
1402
1403         poll_wait(file, sk_sleep(sk), wait);
1404
1405         if (!ptr_ring_empty(&tfile->tx_ring))
1406                 mask |= EPOLLIN | EPOLLRDNORM;
1407
1408         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1409          * guarantee EPOLLOUT to be raised by either here or
1410          * tun_sock_write_space(). Then process could get notification
1411          * after it writes to a down device and meets -EIO.
1412          */
1413         if (tun_sock_writeable(tun, tfile) ||
1414             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1415              tun_sock_writeable(tun, tfile)))
1416                 mask |= EPOLLOUT | EPOLLWRNORM;
1417
1418         if (tun->dev->reg_state != NETREG_REGISTERED)
1419                 mask = EPOLLERR;
1420
1421         tun_put(tun);
1422         return mask;
1423 }
1424
1425 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1426                                             size_t len,
1427                                             const struct iov_iter *it)
1428 {
1429         struct sk_buff *skb;
1430         size_t linear;
1431         int err;
1432         int i;
1433
1434         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1435                 return ERR_PTR(-ENOMEM);
1436
1437         local_bh_disable();
1438         skb = napi_get_frags(&tfile->napi);
1439         local_bh_enable();
1440         if (!skb)
1441                 return ERR_PTR(-ENOMEM);
1442
1443         linear = iov_iter_single_seg_count(it);
1444         err = __skb_grow(skb, linear);
1445         if (err)
1446                 goto free;
1447
1448         skb->len = len;
1449         skb->data_len = len - linear;
1450         skb->truesize += skb->data_len;
1451
1452         for (i = 1; i < it->nr_segs; i++) {
1453                 size_t fragsz = it->iov[i].iov_len;
1454                 struct page *page;
1455                 void *frag;
1456
1457                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1458                         err = -EINVAL;
1459                         goto free;
1460                 }
1461                 frag = netdev_alloc_frag(fragsz);
1462                 if (!frag) {
1463                         err = -ENOMEM;
1464                         goto free;
1465                 }
1466                 page = virt_to_head_page(frag);
1467                 skb_fill_page_desc(skb, i - 1, page,
1468                                    frag - page_address(page), fragsz);
1469         }
1470
1471         return skb;
1472 free:
1473         /* frees skb and all frags allocated with napi_alloc_frag() */
1474         napi_free_frags(&tfile->napi);
1475         return ERR_PTR(err);
1476 }
1477
1478 /* prepad is the amount to reserve at front.  len is length after that.
1479  * linear is a hint as to how much to copy (usually headers). */
1480 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1481                                      size_t prepad, size_t len,
1482                                      size_t linear, int noblock)
1483 {
1484         struct sock *sk = tfile->socket.sk;
1485         struct sk_buff *skb;
1486         int err;
1487
1488         /* Under a page?  Don't bother with paged skb. */
1489         if (prepad + len < PAGE_SIZE || !linear)
1490                 linear = len;
1491
1492         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1493                                    &err, 0);
1494         if (!skb)
1495                 return ERR_PTR(err);
1496
1497         skb_reserve(skb, prepad);
1498         skb_put(skb, linear);
1499         skb->data_len = len - linear;
1500         skb->len += len - linear;
1501
1502         return skb;
1503 }
1504
1505 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1506                            struct sk_buff *skb, int more)
1507 {
1508         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1509         struct sk_buff_head process_queue;
1510         u32 rx_batched = tun->rx_batched;
1511         bool rcv = false;
1512
1513         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1514                 local_bh_disable();
1515                 skb_record_rx_queue(skb, tfile->queue_index);
1516                 netif_receive_skb(skb);
1517                 local_bh_enable();
1518                 return;
1519         }
1520
1521         spin_lock(&queue->lock);
1522         if (!more || skb_queue_len(queue) == rx_batched) {
1523                 __skb_queue_head_init(&process_queue);
1524                 skb_queue_splice_tail_init(queue, &process_queue);
1525                 rcv = true;
1526         } else {
1527                 __skb_queue_tail(queue, skb);
1528         }
1529         spin_unlock(&queue->lock);
1530
1531         if (rcv) {
1532                 struct sk_buff *nskb;
1533
1534                 local_bh_disable();
1535                 while ((nskb = __skb_dequeue(&process_queue))) {
1536                         skb_record_rx_queue(nskb, tfile->queue_index);
1537                         netif_receive_skb(nskb);
1538                 }
1539                 skb_record_rx_queue(skb, tfile->queue_index);
1540                 netif_receive_skb(skb);
1541                 local_bh_enable();
1542         }
1543 }
1544
1545 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1546                               int len, int noblock, bool zerocopy)
1547 {
1548         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1549                 return false;
1550
1551         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1552                 return false;
1553
1554         if (!noblock)
1555                 return false;
1556
1557         if (zerocopy)
1558                 return false;
1559
1560         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1561             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1562                 return false;
1563
1564         return true;
1565 }
1566
1567 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1568                                        struct page_frag *alloc_frag, char *buf,
1569                                        int buflen, int len, int pad)
1570 {
1571         struct sk_buff *skb = build_skb(buf, buflen);
1572
1573         if (!skb)
1574                 return ERR_PTR(-ENOMEM);
1575
1576         skb_reserve(skb, pad);
1577         skb_put(skb, len);
1578         skb_set_owner_w(skb, tfile->socket.sk);
1579
1580         get_page(alloc_frag->page);
1581         alloc_frag->offset += buflen;
1582
1583         return skb;
1584 }
1585
1586 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1587                        struct xdp_buff *xdp, u32 act)
1588 {
1589         int err;
1590
1591         switch (act) {
1592         case XDP_REDIRECT:
1593                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1594                 if (err)
1595                         return err;
1596                 break;
1597         case XDP_TX:
1598                 err = tun_xdp_tx(tun->dev, xdp);
1599                 if (err < 0)
1600                         return err;
1601                 break;
1602         case XDP_PASS:
1603                 break;
1604         default:
1605                 bpf_warn_invalid_xdp_action(act);
1606                 /* fall through */
1607         case XDP_ABORTED:
1608                 trace_xdp_exception(tun->dev, xdp_prog, act);
1609                 /* fall through */
1610         case XDP_DROP:
1611                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1612                 break;
1613         }
1614
1615         return act;
1616 }
1617
1618 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1619                                      struct tun_file *tfile,
1620                                      struct iov_iter *from,
1621                                      struct virtio_net_hdr *hdr,
1622                                      int len, int *skb_xdp)
1623 {
1624         struct page_frag *alloc_frag = &current->task_frag;
1625         struct bpf_prog *xdp_prog;
1626         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1627         char *buf;
1628         size_t copied;
1629         int pad = TUN_RX_PAD;
1630         int err = 0;
1631
1632         rcu_read_lock();
1633         xdp_prog = rcu_dereference(tun->xdp_prog);
1634         if (xdp_prog)
1635                 pad += XDP_PACKET_HEADROOM;
1636         buflen += SKB_DATA_ALIGN(len + pad);
1637         rcu_read_unlock();
1638
1639         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1640         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1641                 return ERR_PTR(-ENOMEM);
1642
1643         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1644         copied = copy_page_from_iter(alloc_frag->page,
1645                                      alloc_frag->offset + pad,
1646                                      len, from);
1647         if (copied != len)
1648                 return ERR_PTR(-EFAULT);
1649
1650         /* There's a small window that XDP may be set after the check
1651          * of xdp_prog above, this should be rare and for simplicity
1652          * we do XDP on skb in case the headroom is not enough.
1653          */
1654         if (hdr->gso_type || !xdp_prog) {
1655                 *skb_xdp = 1;
1656                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1657                                        pad);
1658         }
1659
1660         *skb_xdp = 0;
1661
1662         local_bh_disable();
1663         rcu_read_lock();
1664         xdp_prog = rcu_dereference(tun->xdp_prog);
1665         if (xdp_prog) {
1666                 struct xdp_buff xdp;
1667                 u32 act;
1668
1669                 xdp.data_hard_start = buf;
1670                 xdp.data = buf + pad;
1671                 xdp_set_data_meta_invalid(&xdp);
1672                 xdp.data_end = xdp.data + len;
1673                 xdp.rxq = &tfile->xdp_rxq;
1674
1675                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1676                 if (act == XDP_REDIRECT || act == XDP_TX) {
1677                         get_page(alloc_frag->page);
1678                         alloc_frag->offset += buflen;
1679                 }
1680                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1681                 if (err < 0) {
1682                         if (act == XDP_REDIRECT || act == XDP_TX)
1683                                 put_page(alloc_frag->page);
1684                         goto out;
1685                 }
1686
1687                 if (err == XDP_REDIRECT)
1688                         xdp_do_flush();
1689                 if (err != XDP_PASS)
1690                         goto out;
1691
1692                 pad = xdp.data - xdp.data_hard_start;
1693                 len = xdp.data_end - xdp.data;
1694         }
1695         rcu_read_unlock();
1696         local_bh_enable();
1697
1698         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1699
1700 out:
1701         rcu_read_unlock();
1702         local_bh_enable();
1703         return NULL;
1704 }
1705
1706 /* Get packet from user space buffer */
1707 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1708                             void *msg_control, struct iov_iter *from,
1709                             int noblock, bool more)
1710 {
1711         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1712         struct sk_buff *skb;
1713         size_t total_len = iov_iter_count(from);
1714         size_t len = total_len, align = tun->align, linear;
1715         struct virtio_net_hdr gso = { 0 };
1716         struct tun_pcpu_stats *stats;
1717         int good_linear;
1718         int copylen;
1719         bool zerocopy = false;
1720         int err;
1721         u32 rxhash = 0;
1722         int skb_xdp = 1;
1723         bool frags = tun_napi_frags_enabled(tfile);
1724
1725         if (!(tun->flags & IFF_NO_PI)) {
1726                 if (len < sizeof(pi))
1727                         return -EINVAL;
1728                 len -= sizeof(pi);
1729
1730                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1731                         return -EFAULT;
1732         }
1733
1734         if (tun->flags & IFF_VNET_HDR) {
1735                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1736
1737                 if (len < vnet_hdr_sz)
1738                         return -EINVAL;
1739                 len -= vnet_hdr_sz;
1740
1741                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1742                         return -EFAULT;
1743
1744                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1745                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1746                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1747
1748                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1749                         return -EINVAL;
1750                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1751         }
1752
1753         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1754                 align += NET_IP_ALIGN;
1755                 if (unlikely(len < ETH_HLEN ||
1756                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1757                         return -EINVAL;
1758         }
1759
1760         good_linear = SKB_MAX_HEAD(align);
1761
1762         if (msg_control) {
1763                 struct iov_iter i = *from;
1764
1765                 /* There are 256 bytes to be copied in skb, so there is
1766                  * enough room for skb expand head in case it is used.
1767                  * The rest of the buffer is mapped from userspace.
1768                  */
1769                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1770                 if (copylen > good_linear)
1771                         copylen = good_linear;
1772                 linear = copylen;
1773                 iov_iter_advance(&i, copylen);
1774                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1775                         zerocopy = true;
1776         }
1777
1778         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1779                 /* For the packet that is not easy to be processed
1780                  * (e.g gso or jumbo packet), we will do it at after
1781                  * skb was created with generic XDP routine.
1782                  */
1783                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1784                 if (IS_ERR(skb)) {
1785                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1786                         return PTR_ERR(skb);
1787                 }
1788                 if (!skb)
1789                         return total_len;
1790         } else {
1791                 if (!zerocopy) {
1792                         copylen = len;
1793                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1794                                 linear = good_linear;
1795                         else
1796                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1797                 }
1798
1799                 if (frags) {
1800                         mutex_lock(&tfile->napi_mutex);
1801                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1802                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1803                          * If zerocopy is enabled, then this layout will be
1804                          * overwritten by zerocopy_sg_from_iter().
1805                          */
1806                         zerocopy = false;
1807                 } else {
1808                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1809                                             noblock);
1810                 }
1811
1812                 if (IS_ERR(skb)) {
1813                         if (PTR_ERR(skb) != -EAGAIN)
1814                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1815                         if (frags)
1816                                 mutex_unlock(&tfile->napi_mutex);
1817                         return PTR_ERR(skb);
1818                 }
1819
1820                 if (zerocopy)
1821                         err = zerocopy_sg_from_iter(skb, from);
1822                 else
1823                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1824
1825                 if (err) {
1826                         err = -EFAULT;
1827 drop:
1828                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1829                         kfree_skb(skb);
1830                         if (frags) {
1831                                 tfile->napi.skb = NULL;
1832                                 mutex_unlock(&tfile->napi_mutex);
1833                         }
1834
1835                         return err;
1836                 }
1837         }
1838
1839         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1840                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1841                 kfree_skb(skb);
1842                 if (frags) {
1843                         tfile->napi.skb = NULL;
1844                         mutex_unlock(&tfile->napi_mutex);
1845                 }
1846
1847                 return -EINVAL;
1848         }
1849
1850         switch (tun->flags & TUN_TYPE_MASK) {
1851         case IFF_TUN:
1852                 if (tun->flags & IFF_NO_PI) {
1853                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1854
1855                         switch (ip_version) {
1856                         case 4:
1857                                 pi.proto = htons(ETH_P_IP);
1858                                 break;
1859                         case 6:
1860                                 pi.proto = htons(ETH_P_IPV6);
1861                                 break;
1862                         default:
1863                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1864                                 kfree_skb(skb);
1865                                 return -EINVAL;
1866                         }
1867                 }
1868
1869                 skb_reset_mac_header(skb);
1870                 skb->protocol = pi.proto;
1871                 skb->dev = tun->dev;
1872                 break;
1873         case IFF_TAP:
1874                 if (!frags)
1875                         skb->protocol = eth_type_trans(skb, tun->dev);
1876                 break;
1877         }
1878
1879         /* copy skb_ubuf_info for callback when skb has no error */
1880         if (zerocopy) {
1881                 skb_shinfo(skb)->destructor_arg = msg_control;
1882                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1883                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1884         } else if (msg_control) {
1885                 struct ubuf_info *uarg = msg_control;
1886                 uarg->callback(uarg, false);
1887         }
1888
1889         skb_reset_network_header(skb);
1890         skb_probe_transport_header(skb);
1891         skb_record_rx_queue(skb, tfile->queue_index);
1892
1893         if (skb_xdp) {
1894                 struct bpf_prog *xdp_prog;
1895                 int ret;
1896
1897                 local_bh_disable();
1898                 rcu_read_lock();
1899                 xdp_prog = rcu_dereference(tun->xdp_prog);
1900                 if (xdp_prog) {
1901                         ret = do_xdp_generic(xdp_prog, skb);
1902                         if (ret != XDP_PASS) {
1903                                 rcu_read_unlock();
1904                                 local_bh_enable();
1905                                 if (frags) {
1906                                         tfile->napi.skb = NULL;
1907                                         mutex_unlock(&tfile->napi_mutex);
1908                                 }
1909                                 return total_len;
1910                         }
1911                 }
1912                 rcu_read_unlock();
1913                 local_bh_enable();
1914         }
1915
1916         /* Compute the costly rx hash only if needed for flow updates.
1917          * We may get a very small possibility of OOO during switching, not
1918          * worth to optimize.
1919          */
1920         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1921             !tfile->detached)
1922                 rxhash = __skb_get_hash_symmetric(skb);
1923
1924         rcu_read_lock();
1925         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1926                 err = -EIO;
1927                 rcu_read_unlock();
1928                 goto drop;
1929         }
1930
1931         if (frags) {
1932                 /* Exercise flow dissector code path. */
1933                 u32 headlen = eth_get_headlen(tun->dev, skb->data,
1934                                               skb_headlen(skb));
1935
1936                 if (unlikely(headlen > skb_headlen(skb))) {
1937                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1938                         napi_free_frags(&tfile->napi);
1939                         rcu_read_unlock();
1940                         mutex_unlock(&tfile->napi_mutex);
1941                         WARN_ON(1);
1942                         return -ENOMEM;
1943                 }
1944
1945                 local_bh_disable();
1946                 napi_gro_frags(&tfile->napi);
1947                 local_bh_enable();
1948                 mutex_unlock(&tfile->napi_mutex);
1949         } else if (tfile->napi_enabled) {
1950                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1951                 int queue_len;
1952
1953                 spin_lock_bh(&queue->lock);
1954                 __skb_queue_tail(queue, skb);
1955                 queue_len = skb_queue_len(queue);
1956                 spin_unlock(&queue->lock);
1957
1958                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1959                         napi_schedule(&tfile->napi);
1960
1961                 local_bh_enable();
1962         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1963                 tun_rx_batched(tun, tfile, skb, more);
1964         } else {
1965                 netif_rx_ni(skb);
1966         }
1967         rcu_read_unlock();
1968
1969         stats = get_cpu_ptr(tun->pcpu_stats);
1970         u64_stats_update_begin(&stats->syncp);
1971         u64_stats_inc(&stats->rx_packets);
1972         u64_stats_add(&stats->rx_bytes, len);
1973         u64_stats_update_end(&stats->syncp);
1974         put_cpu_ptr(stats);
1975
1976         if (rxhash)
1977                 tun_flow_update(tun, rxhash, tfile);
1978
1979         return total_len;
1980 }
1981
1982 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1983 {
1984         struct file *file = iocb->ki_filp;
1985         struct tun_file *tfile = file->private_data;
1986         struct tun_struct *tun = tun_get(tfile);
1987         ssize_t result;
1988
1989         if (!tun)
1990                 return -EBADFD;
1991
1992         result = tun_get_user(tun, tfile, NULL, from,
1993                               file->f_flags & O_NONBLOCK, false);
1994
1995         tun_put(tun);
1996         return result;
1997 }
1998
1999 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2000                                 struct tun_file *tfile,
2001                                 struct xdp_frame *xdp_frame,
2002                                 struct iov_iter *iter)
2003 {
2004         int vnet_hdr_sz = 0;
2005         size_t size = xdp_frame->len;
2006         struct tun_pcpu_stats *stats;
2007         size_t ret;
2008
2009         if (tun->flags & IFF_VNET_HDR) {
2010                 struct virtio_net_hdr gso = { 0 };
2011
2012                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2013                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2014                         return -EINVAL;
2015                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2016                              sizeof(gso)))
2017                         return -EFAULT;
2018                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2019         }
2020
2021         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2022
2023         stats = get_cpu_ptr(tun->pcpu_stats);
2024         u64_stats_update_begin(&stats->syncp);
2025         u64_stats_inc(&stats->tx_packets);
2026         u64_stats_add(&stats->tx_bytes, ret);
2027         u64_stats_update_end(&stats->syncp);
2028         put_cpu_ptr(tun->pcpu_stats);
2029
2030         return ret;
2031 }
2032
2033 /* Put packet to the user space buffer */
2034 static ssize_t tun_put_user(struct tun_struct *tun,
2035                             struct tun_file *tfile,
2036                             struct sk_buff *skb,
2037                             struct iov_iter *iter)
2038 {
2039         struct tun_pi pi = { 0, skb->protocol };
2040         struct tun_pcpu_stats *stats;
2041         ssize_t total;
2042         int vlan_offset = 0;
2043         int vlan_hlen = 0;
2044         int vnet_hdr_sz = 0;
2045
2046         if (skb_vlan_tag_present(skb))
2047                 vlan_hlen = VLAN_HLEN;
2048
2049         if (tun->flags & IFF_VNET_HDR)
2050                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2051
2052         total = skb->len + vlan_hlen + vnet_hdr_sz;
2053
2054         if (!(tun->flags & IFF_NO_PI)) {
2055                 if (iov_iter_count(iter) < sizeof(pi))
2056                         return -EINVAL;
2057
2058                 total += sizeof(pi);
2059                 if (iov_iter_count(iter) < total) {
2060                         /* Packet will be striped */
2061                         pi.flags |= TUN_PKT_STRIP;
2062                 }
2063
2064                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2065                         return -EFAULT;
2066         }
2067
2068         if (vnet_hdr_sz) {
2069                 struct virtio_net_hdr gso;
2070
2071                 if (iov_iter_count(iter) < vnet_hdr_sz)
2072                         return -EINVAL;
2073
2074                 if (virtio_net_hdr_from_skb(skb, &gso,
2075                                             tun_is_little_endian(tun), true,
2076                                             vlan_hlen)) {
2077                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2078                         pr_err("unexpected GSO type: "
2079                                "0x%x, gso_size %d, hdr_len %d\n",
2080                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2081                                tun16_to_cpu(tun, gso.hdr_len));
2082                         print_hex_dump(KERN_ERR, "tun: ",
2083                                        DUMP_PREFIX_NONE,
2084                                        16, 1, skb->head,
2085                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2086                         WARN_ON_ONCE(1);
2087                         return -EINVAL;
2088                 }
2089
2090                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2091                         return -EFAULT;
2092
2093                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2094         }
2095
2096         if (vlan_hlen) {
2097                 int ret;
2098                 struct veth veth;
2099
2100                 veth.h_vlan_proto = skb->vlan_proto;
2101                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2102
2103                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2104
2105                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2106                 if (ret || !iov_iter_count(iter))
2107                         goto done;
2108
2109                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2110                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2111                         goto done;
2112         }
2113
2114         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2115
2116 done:
2117         /* caller is in process context, */
2118         stats = get_cpu_ptr(tun->pcpu_stats);
2119         u64_stats_update_begin(&stats->syncp);
2120         u64_stats_inc(&stats->tx_packets);
2121         u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2122         u64_stats_update_end(&stats->syncp);
2123         put_cpu_ptr(tun->pcpu_stats);
2124
2125         return total;
2126 }
2127
2128 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2129 {
2130         DECLARE_WAITQUEUE(wait, current);
2131         void *ptr = NULL;
2132         int error = 0;
2133
2134         ptr = ptr_ring_consume(&tfile->tx_ring);
2135         if (ptr)
2136                 goto out;
2137         if (noblock) {
2138                 error = -EAGAIN;
2139                 goto out;
2140         }
2141
2142         add_wait_queue(&tfile->socket.wq.wait, &wait);
2143
2144         while (1) {
2145                 set_current_state(TASK_INTERRUPTIBLE);
2146                 ptr = ptr_ring_consume(&tfile->tx_ring);
2147                 if (ptr)
2148                         break;
2149                 if (signal_pending(current)) {
2150                         error = -ERESTARTSYS;
2151                         break;
2152                 }
2153                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2154                         error = -EFAULT;
2155                         break;
2156                 }
2157
2158                 schedule();
2159         }
2160
2161         __set_current_state(TASK_RUNNING);
2162         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2163
2164 out:
2165         *err = error;
2166         return ptr;
2167 }
2168
2169 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2170                            struct iov_iter *to,
2171                            int noblock, void *ptr)
2172 {
2173         ssize_t ret;
2174         int err;
2175
2176         if (!iov_iter_count(to)) {
2177                 tun_ptr_free(ptr);
2178                 return 0;
2179         }
2180
2181         if (!ptr) {
2182                 /* Read frames from ring */
2183                 ptr = tun_ring_recv(tfile, noblock, &err);
2184                 if (!ptr)
2185                         return err;
2186         }
2187
2188         if (tun_is_xdp_frame(ptr)) {
2189                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2190
2191                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2192                 xdp_return_frame(xdpf);
2193         } else {
2194                 struct sk_buff *skb = ptr;
2195
2196                 ret = tun_put_user(tun, tfile, skb, to);
2197                 if (unlikely(ret < 0))
2198                         kfree_skb(skb);
2199                 else
2200                         consume_skb(skb);
2201         }
2202
2203         return ret;
2204 }
2205
2206 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2207 {
2208         struct file *file = iocb->ki_filp;
2209         struct tun_file *tfile = file->private_data;
2210         struct tun_struct *tun = tun_get(tfile);
2211         ssize_t len = iov_iter_count(to), ret;
2212
2213         if (!tun)
2214                 return -EBADFD;
2215         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2216         ret = min_t(ssize_t, ret, len);
2217         if (ret > 0)
2218                 iocb->ki_pos = ret;
2219         tun_put(tun);
2220         return ret;
2221 }
2222
2223 static void tun_prog_free(struct rcu_head *rcu)
2224 {
2225         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2226
2227         bpf_prog_destroy(prog->prog);
2228         kfree(prog);
2229 }
2230
2231 static int __tun_set_ebpf(struct tun_struct *tun,
2232                           struct tun_prog __rcu **prog_p,
2233                           struct bpf_prog *prog)
2234 {
2235         struct tun_prog *old, *new = NULL;
2236
2237         if (prog) {
2238                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2239                 if (!new)
2240                         return -ENOMEM;
2241                 new->prog = prog;
2242         }
2243
2244         spin_lock_bh(&tun->lock);
2245         old = rcu_dereference_protected(*prog_p,
2246                                         lockdep_is_held(&tun->lock));
2247         rcu_assign_pointer(*prog_p, new);
2248         spin_unlock_bh(&tun->lock);
2249
2250         if (old)
2251                 call_rcu(&old->rcu, tun_prog_free);
2252
2253         return 0;
2254 }
2255
2256 static void tun_free_netdev(struct net_device *dev)
2257 {
2258         struct tun_struct *tun = netdev_priv(dev);
2259
2260         BUG_ON(!(list_empty(&tun->disabled)));
2261
2262         free_percpu(tun->pcpu_stats);
2263         /* We clear pcpu_stats so that tun_set_iff() can tell if
2264          * tun_free_netdev() has been called from register_netdevice().
2265          */
2266         tun->pcpu_stats = NULL;
2267
2268         tun_flow_uninit(tun);
2269         security_tun_dev_free_security(tun->security);
2270         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2271         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2272 }
2273
2274 static void tun_setup(struct net_device *dev)
2275 {
2276         struct tun_struct *tun = netdev_priv(dev);
2277
2278         tun->owner = INVALID_UID;
2279         tun->group = INVALID_GID;
2280         tun_default_link_ksettings(dev, &tun->link_ksettings);
2281
2282         dev->ethtool_ops = &tun_ethtool_ops;
2283         dev->needs_free_netdev = true;
2284         dev->priv_destructor = tun_free_netdev;
2285         /* We prefer our own queue length */
2286         dev->tx_queue_len = TUN_READQ_SIZE;
2287 }
2288
2289 /* Trivial set of netlink ops to allow deleting tun or tap
2290  * device with netlink.
2291  */
2292 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2293                         struct netlink_ext_ack *extack)
2294 {
2295         NL_SET_ERR_MSG(extack,
2296                        "tun/tap creation via rtnetlink is not supported.");
2297         return -EOPNOTSUPP;
2298 }
2299
2300 static size_t tun_get_size(const struct net_device *dev)
2301 {
2302         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2303         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2304
2305         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2306                nla_total_size(sizeof(gid_t)) + /* GROUP */
2307                nla_total_size(sizeof(u8)) + /* TYPE */
2308                nla_total_size(sizeof(u8)) + /* PI */
2309                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2310                nla_total_size(sizeof(u8)) + /* PERSIST */
2311                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2312                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2313                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2314                0;
2315 }
2316
2317 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2318 {
2319         struct tun_struct *tun = netdev_priv(dev);
2320
2321         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2322                 goto nla_put_failure;
2323         if (uid_valid(tun->owner) &&
2324             nla_put_u32(skb, IFLA_TUN_OWNER,
2325                         from_kuid_munged(current_user_ns(), tun->owner)))
2326                 goto nla_put_failure;
2327         if (gid_valid(tun->group) &&
2328             nla_put_u32(skb, IFLA_TUN_GROUP,
2329                         from_kgid_munged(current_user_ns(), tun->group)))
2330                 goto nla_put_failure;
2331         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2332                 goto nla_put_failure;
2333         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2334                 goto nla_put_failure;
2335         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2336                 goto nla_put_failure;
2337         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2338                        !!(tun->flags & IFF_MULTI_QUEUE)))
2339                 goto nla_put_failure;
2340         if (tun->flags & IFF_MULTI_QUEUE) {
2341                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2342                         goto nla_put_failure;
2343                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2344                                 tun->numdisabled))
2345                         goto nla_put_failure;
2346         }
2347
2348         return 0;
2349
2350 nla_put_failure:
2351         return -EMSGSIZE;
2352 }
2353
2354 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2355         .kind           = DRV_NAME,
2356         .priv_size      = sizeof(struct tun_struct),
2357         .setup          = tun_setup,
2358         .validate       = tun_validate,
2359         .get_size       = tun_get_size,
2360         .fill_info      = tun_fill_info,
2361 };
2362
2363 static void tun_sock_write_space(struct sock *sk)
2364 {
2365         struct tun_file *tfile;
2366         wait_queue_head_t *wqueue;
2367
2368         if (!sock_writeable(sk))
2369                 return;
2370
2371         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2372                 return;
2373
2374         wqueue = sk_sleep(sk);
2375         if (wqueue && waitqueue_active(wqueue))
2376                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2377                                                 EPOLLWRNORM | EPOLLWRBAND);
2378
2379         tfile = container_of(sk, struct tun_file, sk);
2380         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2381 }
2382
2383 static void tun_put_page(struct tun_page *tpage)
2384 {
2385         if (tpage->page)
2386                 __page_frag_cache_drain(tpage->page, tpage->count);
2387 }
2388
2389 static int tun_xdp_one(struct tun_struct *tun,
2390                        struct tun_file *tfile,
2391                        struct xdp_buff *xdp, int *flush,
2392                        struct tun_page *tpage)
2393 {
2394         unsigned int datasize = xdp->data_end - xdp->data;
2395         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2396         struct virtio_net_hdr *gso = &hdr->gso;
2397         struct tun_pcpu_stats *stats;
2398         struct bpf_prog *xdp_prog;
2399         struct sk_buff *skb = NULL;
2400         u32 rxhash = 0, act;
2401         int buflen = hdr->buflen;
2402         int err = 0;
2403         bool skb_xdp = false;
2404         struct page *page;
2405
2406         xdp_prog = rcu_dereference(tun->xdp_prog);
2407         if (xdp_prog) {
2408                 if (gso->gso_type) {
2409                         skb_xdp = true;
2410                         goto build;
2411                 }
2412                 xdp_set_data_meta_invalid(xdp);
2413                 xdp->rxq = &tfile->xdp_rxq;
2414
2415                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2416                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2417                 if (err < 0) {
2418                         put_page(virt_to_head_page(xdp->data));
2419                         return err;
2420                 }
2421
2422                 switch (err) {
2423                 case XDP_REDIRECT:
2424                         *flush = true;
2425                         /* fall through */
2426                 case XDP_TX:
2427                         return 0;
2428                 case XDP_PASS:
2429                         break;
2430                 default:
2431                         page = virt_to_head_page(xdp->data);
2432                         if (tpage->page == page) {
2433                                 ++tpage->count;
2434                         } else {
2435                                 tun_put_page(tpage);
2436                                 tpage->page = page;
2437                                 tpage->count = 1;
2438                         }
2439                         return 0;
2440                 }
2441         }
2442
2443 build:
2444         skb = build_skb(xdp->data_hard_start, buflen);
2445         if (!skb) {
2446                 err = -ENOMEM;
2447                 goto out;
2448         }
2449
2450         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2451         skb_put(skb, xdp->data_end - xdp->data);
2452
2453         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2454                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2455                 kfree_skb(skb);
2456                 err = -EINVAL;
2457                 goto out;
2458         }
2459
2460         skb->protocol = eth_type_trans(skb, tun->dev);
2461         skb_reset_network_header(skb);
2462         skb_probe_transport_header(skb);
2463         skb_record_rx_queue(skb, tfile->queue_index);
2464
2465         if (skb_xdp) {
2466                 err = do_xdp_generic(xdp_prog, skb);
2467                 if (err != XDP_PASS)
2468                         goto out;
2469         }
2470
2471         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2472             !tfile->detached)
2473                 rxhash = __skb_get_hash_symmetric(skb);
2474
2475         netif_receive_skb(skb);
2476
2477         /* No need for get_cpu_ptr() here since this function is
2478          * always called with bh disabled
2479          */
2480         stats = this_cpu_ptr(tun->pcpu_stats);
2481         u64_stats_update_begin(&stats->syncp);
2482         u64_stats_inc(&stats->rx_packets);
2483         u64_stats_add(&stats->rx_bytes, datasize);
2484         u64_stats_update_end(&stats->syncp);
2485
2486         if (rxhash)
2487                 tun_flow_update(tun, rxhash, tfile);
2488
2489 out:
2490         return err;
2491 }
2492
2493 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2494 {
2495         int ret, i;
2496         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2497         struct tun_struct *tun = tun_get(tfile);
2498         struct tun_msg_ctl *ctl = m->msg_control;
2499         struct xdp_buff *xdp;
2500
2501         if (!tun)
2502                 return -EBADFD;
2503
2504         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2505                 struct tun_page tpage;
2506                 int n = ctl->num;
2507                 int flush = 0;
2508
2509                 memset(&tpage, 0, sizeof(tpage));
2510
2511                 local_bh_disable();
2512                 rcu_read_lock();
2513
2514                 for (i = 0; i < n; i++) {
2515                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2516                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2517                 }
2518
2519                 if (flush)
2520                         xdp_do_flush();
2521
2522                 rcu_read_unlock();
2523                 local_bh_enable();
2524
2525                 tun_put_page(&tpage);
2526
2527                 ret = total_len;
2528                 goto out;
2529         }
2530
2531         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2532                            m->msg_flags & MSG_DONTWAIT,
2533                            m->msg_flags & MSG_MORE);
2534 out:
2535         tun_put(tun);
2536         return ret;
2537 }
2538
2539 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2540                        int flags)
2541 {
2542         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2543         struct tun_struct *tun = tun_get(tfile);
2544         void *ptr = m->msg_control;
2545         int ret;
2546
2547         if (!tun) {
2548                 ret = -EBADFD;
2549                 goto out_free;
2550         }
2551
2552         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2553                 ret = -EINVAL;
2554                 goto out_put_tun;
2555         }
2556         if (flags & MSG_ERRQUEUE) {
2557                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2558                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2559                 goto out;
2560         }
2561         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2562         if (ret > (ssize_t)total_len) {
2563                 m->msg_flags |= MSG_TRUNC;
2564                 ret = flags & MSG_TRUNC ? ret : total_len;
2565         }
2566 out:
2567         tun_put(tun);
2568         return ret;
2569
2570 out_put_tun:
2571         tun_put(tun);
2572 out_free:
2573         tun_ptr_free(ptr);
2574         return ret;
2575 }
2576
2577 static int tun_ptr_peek_len(void *ptr)
2578 {
2579         if (likely(ptr)) {
2580                 if (tun_is_xdp_frame(ptr)) {
2581                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2582
2583                         return xdpf->len;
2584                 }
2585                 return __skb_array_len_with_tag(ptr);
2586         } else {
2587                 return 0;
2588         }
2589 }
2590
2591 static int tun_peek_len(struct socket *sock)
2592 {
2593         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2594         struct tun_struct *tun;
2595         int ret = 0;
2596
2597         tun = tun_get(tfile);
2598         if (!tun)
2599                 return 0;
2600
2601         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2602         tun_put(tun);
2603
2604         return ret;
2605 }
2606
2607 /* Ops structure to mimic raw sockets with tun */
2608 static const struct proto_ops tun_socket_ops = {
2609         .peek_len = tun_peek_len,
2610         .sendmsg = tun_sendmsg,
2611         .recvmsg = tun_recvmsg,
2612 };
2613
2614 static struct proto tun_proto = {
2615         .name           = "tun",
2616         .owner          = THIS_MODULE,
2617         .obj_size       = sizeof(struct tun_file),
2618 };
2619
2620 static int tun_flags(struct tun_struct *tun)
2621 {
2622         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2623 }
2624
2625 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2626                               char *buf)
2627 {
2628         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2629         return sprintf(buf, "0x%x\n", tun_flags(tun));
2630 }
2631
2632 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2633                               char *buf)
2634 {
2635         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2636         return uid_valid(tun->owner)?
2637                 sprintf(buf, "%u\n",
2638                         from_kuid_munged(current_user_ns(), tun->owner)):
2639                 sprintf(buf, "-1\n");
2640 }
2641
2642 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2643                               char *buf)
2644 {
2645         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2646         return gid_valid(tun->group) ?
2647                 sprintf(buf, "%u\n",
2648                         from_kgid_munged(current_user_ns(), tun->group)):
2649                 sprintf(buf, "-1\n");
2650 }
2651
2652 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2653 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2654 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2655
2656 static struct attribute *tun_dev_attrs[] = {
2657         &dev_attr_tun_flags.attr,
2658         &dev_attr_owner.attr,
2659         &dev_attr_group.attr,
2660         NULL
2661 };
2662
2663 static const struct attribute_group tun_attr_group = {
2664         .attrs = tun_dev_attrs
2665 };
2666
2667 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2668 {
2669         struct tun_struct *tun;
2670         struct tun_file *tfile = file->private_data;
2671         struct net_device *dev;
2672         int err;
2673
2674         if (tfile->detached)
2675                 return -EINVAL;
2676
2677         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2678                 if (!capable(CAP_NET_ADMIN))
2679                         return -EPERM;
2680
2681                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2682                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2683                         return -EINVAL;
2684         }
2685
2686         dev = __dev_get_by_name(net, ifr->ifr_name);
2687         if (dev) {
2688                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2689                         return -EBUSY;
2690                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2691                         tun = netdev_priv(dev);
2692                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2693                         tun = netdev_priv(dev);
2694                 else
2695                         return -EINVAL;
2696
2697                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2698                     !!(tun->flags & IFF_MULTI_QUEUE))
2699                         return -EINVAL;
2700
2701                 if (tun_not_capable(tun))
2702                         return -EPERM;
2703                 err = security_tun_dev_open(tun->security);
2704                 if (err < 0)
2705                         return err;
2706
2707                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2708                                  ifr->ifr_flags & IFF_NAPI,
2709                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2710                 if (err < 0)
2711                         return err;
2712
2713                 if (tun->flags & IFF_MULTI_QUEUE &&
2714                     (tun->numqueues + tun->numdisabled > 1)) {
2715                         /* One or more queue has already been attached, no need
2716                          * to initialize the device again.
2717                          */
2718                         netdev_state_change(dev);
2719                         return 0;
2720                 }
2721
2722                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2723                               (ifr->ifr_flags & TUN_FEATURES);
2724
2725                 netdev_state_change(dev);
2726         } else {
2727                 char *name;
2728                 unsigned long flags = 0;
2729                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2730                              MAX_TAP_QUEUES : 1;
2731
2732                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2733                         return -EPERM;
2734                 err = security_tun_dev_create();
2735                 if (err < 0)
2736                         return err;
2737
2738                 /* Set dev type */
2739                 if (ifr->ifr_flags & IFF_TUN) {
2740                         /* TUN device */
2741                         flags |= IFF_TUN;
2742                         name = "tun%d";
2743                 } else if (ifr->ifr_flags & IFF_TAP) {
2744                         /* TAP device */
2745                         flags |= IFF_TAP;
2746                         name = "tap%d";
2747                 } else
2748                         return -EINVAL;
2749
2750                 if (*ifr->ifr_name)
2751                         name = ifr->ifr_name;
2752
2753                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2754                                        NET_NAME_UNKNOWN, tun_setup, queues,
2755                                        queues);
2756
2757                 if (!dev)
2758                         return -ENOMEM;
2759
2760                 dev_net_set(dev, net);
2761                 dev->rtnl_link_ops = &tun_link_ops;
2762                 dev->ifindex = tfile->ifindex;
2763                 dev->sysfs_groups[0] = &tun_attr_group;
2764
2765                 tun = netdev_priv(dev);
2766                 tun->dev = dev;
2767                 tun->flags = flags;
2768                 tun->txflt.count = 0;
2769                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2770
2771                 tun->align = NET_SKB_PAD;
2772                 tun->filter_attached = false;
2773                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2774                 tun->rx_batched = 0;
2775                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2776
2777                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2778                 if (!tun->pcpu_stats) {
2779                         err = -ENOMEM;
2780                         goto err_free_dev;
2781                 }
2782
2783                 spin_lock_init(&tun->lock);
2784
2785                 err = security_tun_dev_alloc_security(&tun->security);
2786                 if (err < 0)
2787                         goto err_free_stat;
2788
2789                 tun_net_init(dev);
2790                 tun_flow_init(tun);
2791
2792                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2793                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2794                                    NETIF_F_HW_VLAN_STAG_TX;
2795                 dev->features = dev->hw_features | NETIF_F_LLTX;
2796                 dev->vlan_features = dev->features &
2797                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2798                                        NETIF_F_HW_VLAN_STAG_TX);
2799
2800                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2801                               (ifr->ifr_flags & TUN_FEATURES);
2802
2803                 INIT_LIST_HEAD(&tun->disabled);
2804                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2805                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2806                 if (err < 0)
2807                         goto err_free_flow;
2808
2809                 err = register_netdevice(tun->dev);
2810                 if (err < 0)
2811                         goto err_detach;
2812                 /* free_netdev() won't check refcnt, to aovid race
2813                  * with dev_put() we need publish tun after registration.
2814                  */
2815                 rcu_assign_pointer(tfile->tun, tun);
2816         }
2817
2818         netif_carrier_on(tun->dev);
2819
2820         /* Make sure persistent devices do not get stuck in
2821          * xoff state.
2822          */
2823         if (netif_running(tun->dev))
2824                 netif_tx_wake_all_queues(tun->dev);
2825
2826         strcpy(ifr->ifr_name, tun->dev->name);
2827         return 0;
2828
2829 err_detach:
2830         tun_detach_all(dev);
2831         /* We are here because register_netdevice() has failed.
2832          * If register_netdevice() already called tun_free_netdev()
2833          * while dealing with the error, tun->pcpu_stats has been cleared.
2834          */
2835         if (!tun->pcpu_stats)
2836                 goto err_free_dev;
2837
2838 err_free_flow:
2839         tun_flow_uninit(tun);
2840         security_tun_dev_free_security(tun->security);
2841 err_free_stat:
2842         free_percpu(tun->pcpu_stats);
2843 err_free_dev:
2844         free_netdev(dev);
2845         return err;
2846 }
2847
2848 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2849 {
2850         strcpy(ifr->ifr_name, tun->dev->name);
2851
2852         ifr->ifr_flags = tun_flags(tun);
2853
2854 }
2855
2856 /* This is like a cut-down ethtool ops, except done via tun fd so no
2857  * privs required. */
2858 static int set_offload(struct tun_struct *tun, unsigned long arg)
2859 {
2860         netdev_features_t features = 0;
2861
2862         if (arg & TUN_F_CSUM) {
2863                 features |= NETIF_F_HW_CSUM;
2864                 arg &= ~TUN_F_CSUM;
2865
2866                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2867                         if (arg & TUN_F_TSO_ECN) {
2868                                 features |= NETIF_F_TSO_ECN;
2869                                 arg &= ~TUN_F_TSO_ECN;
2870                         }
2871                         if (arg & TUN_F_TSO4)
2872                                 features |= NETIF_F_TSO;
2873                         if (arg & TUN_F_TSO6)
2874                                 features |= NETIF_F_TSO6;
2875                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2876                 }
2877
2878                 arg &= ~TUN_F_UFO;
2879         }
2880
2881         /* This gives the user a way to test for new features in future by
2882          * trying to set them. */
2883         if (arg)
2884                 return -EINVAL;
2885
2886         tun->set_features = features;
2887         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2888         tun->dev->wanted_features |= features;
2889         netdev_update_features(tun->dev);
2890
2891         return 0;
2892 }
2893
2894 static void tun_detach_filter(struct tun_struct *tun, int n)
2895 {
2896         int i;
2897         struct tun_file *tfile;
2898
2899         for (i = 0; i < n; i++) {
2900                 tfile = rtnl_dereference(tun->tfiles[i]);
2901                 lock_sock(tfile->socket.sk);
2902                 sk_detach_filter(tfile->socket.sk);
2903                 release_sock(tfile->socket.sk);
2904         }
2905
2906         tun->filter_attached = false;
2907 }
2908
2909 static int tun_attach_filter(struct tun_struct *tun)
2910 {
2911         int i, ret = 0;
2912         struct tun_file *tfile;
2913
2914         for (i = 0; i < tun->numqueues; i++) {
2915                 tfile = rtnl_dereference(tun->tfiles[i]);
2916                 lock_sock(tfile->socket.sk);
2917                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2918                 release_sock(tfile->socket.sk);
2919                 if (ret) {
2920                         tun_detach_filter(tun, i);
2921                         return ret;
2922                 }
2923         }
2924
2925         tun->filter_attached = true;
2926         return ret;
2927 }
2928
2929 static void tun_set_sndbuf(struct tun_struct *tun)
2930 {
2931         struct tun_file *tfile;
2932         int i;
2933
2934         for (i = 0; i < tun->numqueues; i++) {
2935                 tfile = rtnl_dereference(tun->tfiles[i]);
2936                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2937         }
2938 }
2939
2940 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2941 {
2942         struct tun_file *tfile = file->private_data;
2943         struct tun_struct *tun;
2944         int ret = 0;
2945
2946         rtnl_lock();
2947
2948         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2949                 tun = tfile->detached;
2950                 if (!tun) {
2951                         ret = -EINVAL;
2952                         goto unlock;
2953                 }
2954                 ret = security_tun_dev_attach_queue(tun->security);
2955                 if (ret < 0)
2956                         goto unlock;
2957                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2958                                  tun->flags & IFF_NAPI_FRAGS, true);
2959         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2960                 tun = rtnl_dereference(tfile->tun);
2961                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2962                         ret = -EINVAL;
2963                 else
2964                         __tun_detach(tfile, false);
2965         } else
2966                 ret = -EINVAL;
2967
2968         if (ret >= 0)
2969                 netdev_state_change(tun->dev);
2970
2971 unlock:
2972         rtnl_unlock();
2973         return ret;
2974 }
2975
2976 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2977                         void __user *data)
2978 {
2979         struct bpf_prog *prog;
2980         int fd;
2981
2982         if (copy_from_user(&fd, data, sizeof(fd)))
2983                 return -EFAULT;
2984
2985         if (fd == -1) {
2986                 prog = NULL;
2987         } else {
2988                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2989                 if (IS_ERR(prog))
2990                         return PTR_ERR(prog);
2991         }
2992
2993         return __tun_set_ebpf(tun, prog_p, prog);
2994 }
2995
2996 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2997                             unsigned long arg, int ifreq_len)
2998 {
2999         struct tun_file *tfile = file->private_data;
3000         struct net *net = sock_net(&tfile->sk);
3001         struct tun_struct *tun;
3002         void __user* argp = (void __user*)arg;
3003         unsigned int ifindex, carrier;
3004         struct ifreq ifr;
3005         kuid_t owner;
3006         kgid_t group;
3007         int sndbuf;
3008         int vnet_hdr_sz;
3009         int le;
3010         int ret;
3011         bool do_notify = false;
3012
3013         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3014             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3015                 if (copy_from_user(&ifr, argp, ifreq_len))
3016                         return -EFAULT;
3017         } else {
3018                 memset(&ifr, 0, sizeof(ifr));
3019         }
3020         if (cmd == TUNGETFEATURES) {
3021                 /* Currently this just means: "what IFF flags are valid?".
3022                  * This is needed because we never checked for invalid flags on
3023                  * TUNSETIFF.
3024                  */
3025                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3026                                 (unsigned int __user*)argp);
3027         } else if (cmd == TUNSETQUEUE) {
3028                 return tun_set_queue(file, &ifr);
3029         } else if (cmd == SIOCGSKNS) {
3030                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3031                         return -EPERM;
3032                 return open_related_ns(&net->ns, get_net_ns);
3033         }
3034
3035         ret = 0;
3036         rtnl_lock();
3037
3038         tun = tun_get(tfile);
3039         if (cmd == TUNSETIFF) {
3040                 ret = -EEXIST;
3041                 if (tun)
3042                         goto unlock;
3043
3044                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3045
3046                 ret = tun_set_iff(net, file, &ifr);
3047
3048                 if (ret)
3049                         goto unlock;
3050
3051                 if (copy_to_user(argp, &ifr, ifreq_len))
3052                         ret = -EFAULT;
3053                 goto unlock;
3054         }
3055         if (cmd == TUNSETIFINDEX) {
3056                 ret = -EPERM;
3057                 if (tun)
3058                         goto unlock;
3059
3060                 ret = -EFAULT;
3061                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3062                         goto unlock;
3063
3064                 ret = 0;
3065                 tfile->ifindex = ifindex;
3066                 goto unlock;
3067         }
3068
3069         ret = -EBADFD;
3070         if (!tun)
3071                 goto unlock;
3072
3073         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3074
3075         net = dev_net(tun->dev);
3076         ret = 0;
3077         switch (cmd) {
3078         case TUNGETIFF:
3079                 tun_get_iff(tun, &ifr);
3080
3081                 if (tfile->detached)
3082                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3083                 if (!tfile->socket.sk->sk_filter)
3084                         ifr.ifr_flags |= IFF_NOFILTER;
3085
3086                 if (copy_to_user(argp, &ifr, ifreq_len))
3087                         ret = -EFAULT;
3088                 break;
3089
3090         case TUNSETNOCSUM:
3091                 /* Disable/Enable checksum */
3092
3093                 /* [unimplemented] */
3094                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3095                            arg ? "disabled" : "enabled");
3096                 break;
3097
3098         case TUNSETPERSIST:
3099                 /* Disable/Enable persist mode. Keep an extra reference to the
3100                  * module to prevent the module being unprobed.
3101                  */
3102                 if (arg && !(tun->flags & IFF_PERSIST)) {
3103                         tun->flags |= IFF_PERSIST;
3104                         __module_get(THIS_MODULE);
3105                         do_notify = true;
3106                 }
3107                 if (!arg && (tun->flags & IFF_PERSIST)) {
3108                         tun->flags &= ~IFF_PERSIST;
3109                         module_put(THIS_MODULE);
3110                         do_notify = true;
3111                 }
3112
3113                 netif_info(tun, drv, tun->dev, "persist %s\n",
3114                            arg ? "enabled" : "disabled");
3115                 break;
3116
3117         case TUNSETOWNER:
3118                 /* Set owner of the device */
3119                 owner = make_kuid(current_user_ns(), arg);
3120                 if (!uid_valid(owner)) {
3121                         ret = -EINVAL;
3122                         break;
3123                 }
3124                 tun->owner = owner;
3125                 do_notify = true;
3126                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3127                            from_kuid(&init_user_ns, tun->owner));
3128                 break;
3129
3130         case TUNSETGROUP:
3131                 /* Set group of the device */
3132                 group = make_kgid(current_user_ns(), arg);
3133                 if (!gid_valid(group)) {
3134                         ret = -EINVAL;
3135                         break;
3136                 }
3137                 tun->group = group;
3138                 do_notify = true;
3139                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3140                            from_kgid(&init_user_ns, tun->group));
3141                 break;
3142
3143         case TUNSETLINK:
3144                 /* Only allow setting the type when the interface is down */
3145                 if (tun->dev->flags & IFF_UP) {
3146                         netif_info(tun, drv, tun->dev,
3147                                    "Linktype set failed because interface is up\n");
3148                         ret = -EBUSY;
3149                 } else {
3150                         tun->dev->type = (int) arg;
3151                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3152                                    tun->dev->type);
3153                         ret = 0;
3154                 }
3155                 break;
3156
3157         case TUNSETDEBUG:
3158                 tun->msg_enable = (u32)arg;
3159                 break;
3160
3161         case TUNSETOFFLOAD:
3162                 ret = set_offload(tun, arg);
3163                 break;
3164
3165         case TUNSETTXFILTER:
3166                 /* Can be set only for TAPs */
3167                 ret = -EINVAL;
3168                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3169                         break;
3170                 ret = update_filter(&tun->txflt, (void __user *)arg);
3171                 break;
3172
3173         case SIOCGIFHWADDR:
3174                 /* Get hw address */
3175                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3176                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3177                 if (copy_to_user(argp, &ifr, ifreq_len))
3178                         ret = -EFAULT;
3179                 break;
3180
3181         case SIOCSIFHWADDR:
3182                 /* Set hw address */
3183                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3184                 break;
3185
3186         case TUNGETSNDBUF:
3187                 sndbuf = tfile->socket.sk->sk_sndbuf;
3188                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3189                         ret = -EFAULT;
3190                 break;
3191
3192         case TUNSETSNDBUF:
3193                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3194                         ret = -EFAULT;
3195                         break;
3196                 }
3197                 if (sndbuf <= 0) {
3198                         ret = -EINVAL;
3199                         break;
3200                 }
3201
3202                 tun->sndbuf = sndbuf;
3203                 tun_set_sndbuf(tun);
3204                 break;
3205
3206         case TUNGETVNETHDRSZ:
3207                 vnet_hdr_sz = tun->vnet_hdr_sz;
3208                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3209                         ret = -EFAULT;
3210                 break;
3211
3212         case TUNSETVNETHDRSZ:
3213                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3214                         ret = -EFAULT;
3215                         break;
3216                 }
3217                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3218                         ret = -EINVAL;
3219                         break;
3220                 }
3221
3222                 tun->vnet_hdr_sz = vnet_hdr_sz;
3223                 break;
3224
3225         case TUNGETVNETLE:
3226                 le = !!(tun->flags & TUN_VNET_LE);
3227                 if (put_user(le, (int __user *)argp))
3228                         ret = -EFAULT;
3229                 break;
3230
3231         case TUNSETVNETLE:
3232                 if (get_user(le, (int __user *)argp)) {
3233                         ret = -EFAULT;
3234                         break;
3235                 }
3236                 if (le)
3237                         tun->flags |= TUN_VNET_LE;
3238                 else
3239                         tun->flags &= ~TUN_VNET_LE;
3240                 break;
3241
3242         case TUNGETVNETBE:
3243                 ret = tun_get_vnet_be(tun, argp);
3244                 break;
3245
3246         case TUNSETVNETBE:
3247                 ret = tun_set_vnet_be(tun, argp);
3248                 break;
3249
3250         case TUNATTACHFILTER:
3251                 /* Can be set only for TAPs */
3252                 ret = -EINVAL;
3253                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3254                         break;
3255                 ret = -EFAULT;
3256                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3257                         break;
3258
3259                 ret = tun_attach_filter(tun);
3260                 break;
3261
3262         case TUNDETACHFILTER:
3263                 /* Can be set only for TAPs */
3264                 ret = -EINVAL;
3265                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3266                         break;
3267                 ret = 0;
3268                 tun_detach_filter(tun, tun->numqueues);
3269                 break;
3270
3271         case TUNGETFILTER:
3272                 ret = -EINVAL;
3273                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3274                         break;
3275                 ret = -EFAULT;
3276                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3277                         break;
3278                 ret = 0;
3279                 break;
3280
3281         case TUNSETSTEERINGEBPF:
3282                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3283                 break;
3284
3285         case TUNSETFILTEREBPF:
3286                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3287                 break;
3288
3289         case TUNSETCARRIER:
3290                 ret = -EFAULT;
3291                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3292                         goto unlock;
3293
3294                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3295                 break;
3296
3297         case TUNGETDEVNETNS:
3298                 ret = -EPERM;
3299                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3300                         goto unlock;
3301                 ret = open_related_ns(&net->ns, get_net_ns);
3302                 break;
3303
3304         default:
3305                 ret = -EINVAL;
3306                 break;
3307         }
3308
3309         if (do_notify)
3310                 netdev_state_change(tun->dev);
3311
3312 unlock:
3313         rtnl_unlock();
3314         if (tun)
3315                 tun_put(tun);
3316         return ret;
3317 }
3318
3319 static long tun_chr_ioctl(struct file *file,
3320                           unsigned int cmd, unsigned long arg)
3321 {
3322         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3323 }
3324
3325 #ifdef CONFIG_COMPAT
3326 static long tun_chr_compat_ioctl(struct file *file,
3327                          unsigned int cmd, unsigned long arg)
3328 {
3329         switch (cmd) {
3330         case TUNSETIFF:
3331         case TUNGETIFF:
3332         case TUNSETTXFILTER:
3333         case TUNGETSNDBUF:
3334         case TUNSETSNDBUF:
3335         case SIOCGIFHWADDR:
3336         case SIOCSIFHWADDR:
3337                 arg = (unsigned long)compat_ptr(arg);
3338                 break;
3339         default:
3340                 arg = (compat_ulong_t)arg;
3341                 break;
3342         }
3343
3344         /*
3345          * compat_ifreq is shorter than ifreq, so we must not access beyond
3346          * the end of that structure. All fields that are used in this
3347          * driver are compatible though, we don't need to convert the
3348          * contents.
3349          */
3350         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3351 }
3352 #endif /* CONFIG_COMPAT */
3353
3354 static int tun_chr_fasync(int fd, struct file *file, int on)
3355 {
3356         struct tun_file *tfile = file->private_data;
3357         int ret;
3358
3359         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3360                 goto out;
3361
3362         if (on) {
3363                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3364                 tfile->flags |= TUN_FASYNC;
3365         } else
3366                 tfile->flags &= ~TUN_FASYNC;
3367         ret = 0;
3368 out:
3369         return ret;
3370 }
3371
3372 static int tun_chr_open(struct inode *inode, struct file * file)
3373 {
3374         struct net *net = current->nsproxy->net_ns;
3375         struct tun_file *tfile;
3376
3377         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3378                                             &tun_proto, 0);
3379         if (!tfile)
3380                 return -ENOMEM;
3381         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3382                 sk_free(&tfile->sk);
3383                 return -ENOMEM;
3384         }
3385
3386         mutex_init(&tfile->napi_mutex);
3387         RCU_INIT_POINTER(tfile->tun, NULL);
3388         tfile->flags = 0;
3389         tfile->ifindex = 0;
3390
3391         init_waitqueue_head(&tfile->socket.wq.wait);
3392
3393         tfile->socket.file = file;
3394         tfile->socket.ops = &tun_socket_ops;
3395
3396         sock_init_data(&tfile->socket, &tfile->sk);
3397
3398         tfile->sk.sk_write_space = tun_sock_write_space;
3399         tfile->sk.sk_sndbuf = INT_MAX;
3400
3401         file->private_data = tfile;
3402         INIT_LIST_HEAD(&tfile->next);
3403
3404         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3405
3406         return 0;
3407 }
3408
3409 static int tun_chr_close(struct inode *inode, struct file *file)
3410 {
3411         struct tun_file *tfile = file->private_data;
3412
3413         tun_detach(tfile, true);
3414
3415         return 0;
3416 }
3417
3418 #ifdef CONFIG_PROC_FS
3419 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3420 {
3421         struct tun_file *tfile = file->private_data;
3422         struct tun_struct *tun;
3423         struct ifreq ifr;
3424
3425         memset(&ifr, 0, sizeof(ifr));
3426
3427         rtnl_lock();
3428         tun = tun_get(tfile);
3429         if (tun)
3430                 tun_get_iff(tun, &ifr);
3431         rtnl_unlock();
3432
3433         if (tun)
3434                 tun_put(tun);
3435
3436         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3437 }
3438 #endif
3439
3440 static const struct file_operations tun_fops = {
3441         .owner  = THIS_MODULE,
3442         .llseek = no_llseek,
3443         .read_iter  = tun_chr_read_iter,
3444         .write_iter = tun_chr_write_iter,
3445         .poll   = tun_chr_poll,
3446         .unlocked_ioctl = tun_chr_ioctl,
3447 #ifdef CONFIG_COMPAT
3448         .compat_ioctl = tun_chr_compat_ioctl,
3449 #endif
3450         .open   = tun_chr_open,
3451         .release = tun_chr_close,
3452         .fasync = tun_chr_fasync,
3453 #ifdef CONFIG_PROC_FS
3454         .show_fdinfo = tun_chr_show_fdinfo,
3455 #endif
3456 };
3457
3458 static struct miscdevice tun_miscdev = {
3459         .minor = TUN_MINOR,
3460         .name = "tun",
3461         .nodename = "net/tun",
3462         .fops = &tun_fops,
3463 };
3464
3465 /* ethtool interface */
3466
3467 static void tun_default_link_ksettings(struct net_device *dev,
3468                                        struct ethtool_link_ksettings *cmd)
3469 {
3470         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3471         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3472         cmd->base.speed         = SPEED_10;
3473         cmd->base.duplex        = DUPLEX_FULL;
3474         cmd->base.port          = PORT_TP;
3475         cmd->base.phy_address   = 0;
3476         cmd->base.autoneg       = AUTONEG_DISABLE;
3477 }
3478
3479 static int tun_get_link_ksettings(struct net_device *dev,
3480                                   struct ethtool_link_ksettings *cmd)
3481 {
3482         struct tun_struct *tun = netdev_priv(dev);
3483
3484         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3485         return 0;
3486 }
3487
3488 static int tun_set_link_ksettings(struct net_device *dev,
3489                                   const struct ethtool_link_ksettings *cmd)
3490 {
3491         struct tun_struct *tun = netdev_priv(dev);
3492
3493         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3494         return 0;
3495 }
3496
3497 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3498 {
3499         struct tun_struct *tun = netdev_priv(dev);
3500
3501         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3502         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3503
3504         switch (tun->flags & TUN_TYPE_MASK) {
3505         case IFF_TUN:
3506                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3507                 break;
3508         case IFF_TAP:
3509                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3510                 break;
3511         }
3512 }
3513
3514 static u32 tun_get_msglevel(struct net_device *dev)
3515 {
3516         struct tun_struct *tun = netdev_priv(dev);
3517
3518         return tun->msg_enable;
3519 }
3520
3521 static void tun_set_msglevel(struct net_device *dev, u32 value)
3522 {
3523         struct tun_struct *tun = netdev_priv(dev);
3524
3525         tun->msg_enable = value;
3526 }
3527
3528 static int tun_get_coalesce(struct net_device *dev,
3529                             struct ethtool_coalesce *ec)
3530 {
3531         struct tun_struct *tun = netdev_priv(dev);
3532
3533         ec->rx_max_coalesced_frames = tun->rx_batched;
3534
3535         return 0;
3536 }
3537
3538 static int tun_set_coalesce(struct net_device *dev,
3539                             struct ethtool_coalesce *ec)
3540 {
3541         struct tun_struct *tun = netdev_priv(dev);
3542
3543         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3544                 tun->rx_batched = NAPI_POLL_WEIGHT;
3545         else
3546                 tun->rx_batched = ec->rx_max_coalesced_frames;
3547
3548         return 0;
3549 }
3550
3551 static const struct ethtool_ops tun_ethtool_ops = {
3552         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3553         .get_drvinfo    = tun_get_drvinfo,
3554         .get_msglevel   = tun_get_msglevel,
3555         .set_msglevel   = tun_set_msglevel,
3556         .get_link       = ethtool_op_get_link,
3557         .get_ts_info    = ethtool_op_get_ts_info,
3558         .get_coalesce   = tun_get_coalesce,
3559         .set_coalesce   = tun_set_coalesce,
3560         .get_link_ksettings = tun_get_link_ksettings,
3561         .set_link_ksettings = tun_set_link_ksettings,
3562 };
3563
3564 static int tun_queue_resize(struct tun_struct *tun)
3565 {
3566         struct net_device *dev = tun->dev;
3567         struct tun_file *tfile;
3568         struct ptr_ring **rings;
3569         int n = tun->numqueues + tun->numdisabled;
3570         int ret, i;
3571
3572         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3573         if (!rings)
3574                 return -ENOMEM;
3575
3576         for (i = 0; i < tun->numqueues; i++) {
3577                 tfile = rtnl_dereference(tun->tfiles[i]);
3578                 rings[i] = &tfile->tx_ring;
3579         }
3580         list_for_each_entry(tfile, &tun->disabled, next)
3581                 rings[i++] = &tfile->tx_ring;
3582
3583         ret = ptr_ring_resize_multiple(rings, n,
3584                                        dev->tx_queue_len, GFP_KERNEL,
3585                                        tun_ptr_free);
3586
3587         kfree(rings);
3588         return ret;
3589 }
3590
3591 static int tun_device_event(struct notifier_block *unused,
3592                             unsigned long event, void *ptr)
3593 {
3594         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3595         struct tun_struct *tun = netdev_priv(dev);
3596         int i;
3597
3598         if (dev->rtnl_link_ops != &tun_link_ops)
3599                 return NOTIFY_DONE;
3600
3601         switch (event) {
3602         case NETDEV_CHANGE_TX_QUEUE_LEN:
3603                 if (tun_queue_resize(tun))
3604                         return NOTIFY_BAD;
3605                 break;
3606         case NETDEV_UP:
3607                 for (i = 0; i < tun->numqueues; i++) {
3608                         struct tun_file *tfile;
3609
3610                         tfile = rtnl_dereference(tun->tfiles[i]);
3611                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3612                 }
3613                 break;
3614         default:
3615                 break;
3616         }
3617
3618         return NOTIFY_DONE;
3619 }
3620
3621 static struct notifier_block tun_notifier_block __read_mostly = {
3622         .notifier_call  = tun_device_event,
3623 };
3624
3625 static int __init tun_init(void)
3626 {
3627         int ret = 0;
3628
3629         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3630
3631         ret = rtnl_link_register(&tun_link_ops);
3632         if (ret) {
3633                 pr_err("Can't register link_ops\n");
3634                 goto err_linkops;
3635         }
3636
3637         ret = misc_register(&tun_miscdev);
3638         if (ret) {
3639                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3640                 goto err_misc;
3641         }
3642
3643         ret = register_netdevice_notifier(&tun_notifier_block);
3644         if (ret) {
3645                 pr_err("Can't register netdevice notifier\n");
3646                 goto err_notifier;
3647         }
3648
3649         return  0;
3650
3651 err_notifier:
3652         misc_deregister(&tun_miscdev);
3653 err_misc:
3654         rtnl_link_unregister(&tun_link_ops);
3655 err_linkops:
3656         return ret;
3657 }
3658
3659 static void tun_cleanup(void)
3660 {
3661         misc_deregister(&tun_miscdev);
3662         rtnl_link_unregister(&tun_link_ops);
3663         unregister_netdevice_notifier(&tun_notifier_block);
3664 }
3665
3666 /* Get an underlying socket object from tun file.  Returns error unless file is
3667  * attached to a device.  The returned object works like a packet socket, it
3668  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3669  * holding a reference to the file for as long as the socket is in use. */
3670 struct socket *tun_get_socket(struct file *file)
3671 {
3672         struct tun_file *tfile;
3673         if (file->f_op != &tun_fops)
3674                 return ERR_PTR(-EINVAL);
3675         tfile = file->private_data;
3676         if (!tfile)
3677                 return ERR_PTR(-EBADFD);
3678         return &tfile->socket;
3679 }
3680 EXPORT_SYMBOL_GPL(tun_get_socket);
3681
3682 struct ptr_ring *tun_get_tx_ring(struct file *file)
3683 {
3684         struct tun_file *tfile;
3685
3686         if (file->f_op != &tun_fops)
3687                 return ERR_PTR(-EINVAL);
3688         tfile = file->private_data;
3689         if (!tfile)
3690                 return ERR_PTR(-EBADFD);
3691         return &tfile->tx_ring;
3692 }
3693 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3694
3695 module_init(tun_init);
3696 module_exit(tun_cleanup);
3697 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3698 MODULE_AUTHOR(DRV_COPYRIGHT);
3699 MODULE_LICENSE("GPL");
3700 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3701 MODULE_ALIAS("devname:net/tun");