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 = xdp_convert_buff_to_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                 xdp.frame_sz = buflen;
1675
1676                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1677                 if (act == XDP_REDIRECT || act == XDP_TX) {
1678                         get_page(alloc_frag->page);
1679                         alloc_frag->offset += buflen;
1680                 }
1681                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1682                 if (err < 0) {
1683                         if (act == XDP_REDIRECT || act == XDP_TX)
1684                                 put_page(alloc_frag->page);
1685                         goto out;
1686                 }
1687
1688                 if (err == XDP_REDIRECT)
1689                         xdp_do_flush();
1690                 if (err != XDP_PASS)
1691                         goto out;
1692
1693                 pad = xdp.data - xdp.data_hard_start;
1694                 len = xdp.data_end - xdp.data;
1695         }
1696         rcu_read_unlock();
1697         local_bh_enable();
1698
1699         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1700
1701 out:
1702         rcu_read_unlock();
1703         local_bh_enable();
1704         return NULL;
1705 }
1706
1707 /* Get packet from user space buffer */
1708 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1709                             void *msg_control, struct iov_iter *from,
1710                             int noblock, bool more)
1711 {
1712         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1713         struct sk_buff *skb;
1714         size_t total_len = iov_iter_count(from);
1715         size_t len = total_len, align = tun->align, linear;
1716         struct virtio_net_hdr gso = { 0 };
1717         struct tun_pcpu_stats *stats;
1718         int good_linear;
1719         int copylen;
1720         bool zerocopy = false;
1721         int err;
1722         u32 rxhash = 0;
1723         int skb_xdp = 1;
1724         bool frags = tun_napi_frags_enabled(tfile);
1725
1726         if (!(tun->flags & IFF_NO_PI)) {
1727                 if (len < sizeof(pi))
1728                         return -EINVAL;
1729                 len -= sizeof(pi);
1730
1731                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1732                         return -EFAULT;
1733         }
1734
1735         if (tun->flags & IFF_VNET_HDR) {
1736                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1737
1738                 if (len < vnet_hdr_sz)
1739                         return -EINVAL;
1740                 len -= vnet_hdr_sz;
1741
1742                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1743                         return -EFAULT;
1744
1745                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1746                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1747                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1748
1749                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1750                         return -EINVAL;
1751                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1752         }
1753
1754         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1755                 align += NET_IP_ALIGN;
1756                 if (unlikely(len < ETH_HLEN ||
1757                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1758                         return -EINVAL;
1759         }
1760
1761         good_linear = SKB_MAX_HEAD(align);
1762
1763         if (msg_control) {
1764                 struct iov_iter i = *from;
1765
1766                 /* There are 256 bytes to be copied in skb, so there is
1767                  * enough room for skb expand head in case it is used.
1768                  * The rest of the buffer is mapped from userspace.
1769                  */
1770                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1771                 if (copylen > good_linear)
1772                         copylen = good_linear;
1773                 linear = copylen;
1774                 iov_iter_advance(&i, copylen);
1775                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1776                         zerocopy = true;
1777         }
1778
1779         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1780                 /* For the packet that is not easy to be processed
1781                  * (e.g gso or jumbo packet), we will do it at after
1782                  * skb was created with generic XDP routine.
1783                  */
1784                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1785                 if (IS_ERR(skb)) {
1786                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1787                         return PTR_ERR(skb);
1788                 }
1789                 if (!skb)
1790                         return total_len;
1791         } else {
1792                 if (!zerocopy) {
1793                         copylen = len;
1794                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1795                                 linear = good_linear;
1796                         else
1797                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1798                 }
1799
1800                 if (frags) {
1801                         mutex_lock(&tfile->napi_mutex);
1802                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1803                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1804                          * If zerocopy is enabled, then this layout will be
1805                          * overwritten by zerocopy_sg_from_iter().
1806                          */
1807                         zerocopy = false;
1808                 } else {
1809                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1810                                             noblock);
1811                 }
1812
1813                 if (IS_ERR(skb)) {
1814                         if (PTR_ERR(skb) != -EAGAIN)
1815                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1816                         if (frags)
1817                                 mutex_unlock(&tfile->napi_mutex);
1818                         return PTR_ERR(skb);
1819                 }
1820
1821                 if (zerocopy)
1822                         err = zerocopy_sg_from_iter(skb, from);
1823                 else
1824                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1825
1826                 if (err) {
1827                         err = -EFAULT;
1828 drop:
1829                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1830                         kfree_skb(skb);
1831                         if (frags) {
1832                                 tfile->napi.skb = NULL;
1833                                 mutex_unlock(&tfile->napi_mutex);
1834                         }
1835
1836                         return err;
1837                 }
1838         }
1839
1840         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1841                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1842                 kfree_skb(skb);
1843                 if (frags) {
1844                         tfile->napi.skb = NULL;
1845                         mutex_unlock(&tfile->napi_mutex);
1846                 }
1847
1848                 return -EINVAL;
1849         }
1850
1851         switch (tun->flags & TUN_TYPE_MASK) {
1852         case IFF_TUN:
1853                 if (tun->flags & IFF_NO_PI) {
1854                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1855
1856                         switch (ip_version) {
1857                         case 4:
1858                                 pi.proto = htons(ETH_P_IP);
1859                                 break;
1860                         case 6:
1861                                 pi.proto = htons(ETH_P_IPV6);
1862                                 break;
1863                         default:
1864                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1865                                 kfree_skb(skb);
1866                                 return -EINVAL;
1867                         }
1868                 }
1869
1870                 skb_reset_mac_header(skb);
1871                 skb->protocol = pi.proto;
1872                 skb->dev = tun->dev;
1873                 break;
1874         case IFF_TAP:
1875                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1876                         err = -ENOMEM;
1877                         goto drop;
1878                 }
1879                 skb->protocol = eth_type_trans(skb, tun->dev);
1880                 break;
1881         }
1882
1883         /* copy skb_ubuf_info for callback when skb has no error */
1884         if (zerocopy) {
1885                 skb_shinfo(skb)->destructor_arg = msg_control;
1886                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1887                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1888         } else if (msg_control) {
1889                 struct ubuf_info *uarg = msg_control;
1890                 uarg->callback(uarg, false);
1891         }
1892
1893         skb_reset_network_header(skb);
1894         skb_probe_transport_header(skb);
1895         skb_record_rx_queue(skb, tfile->queue_index);
1896
1897         if (skb_xdp) {
1898                 struct bpf_prog *xdp_prog;
1899                 int ret;
1900
1901                 local_bh_disable();
1902                 rcu_read_lock();
1903                 xdp_prog = rcu_dereference(tun->xdp_prog);
1904                 if (xdp_prog) {
1905                         ret = do_xdp_generic(xdp_prog, skb);
1906                         if (ret != XDP_PASS) {
1907                                 rcu_read_unlock();
1908                                 local_bh_enable();
1909                                 if (frags) {
1910                                         tfile->napi.skb = NULL;
1911                                         mutex_unlock(&tfile->napi_mutex);
1912                                 }
1913                                 return total_len;
1914                         }
1915                 }
1916                 rcu_read_unlock();
1917                 local_bh_enable();
1918         }
1919
1920         /* Compute the costly rx hash only if needed for flow updates.
1921          * We may get a very small possibility of OOO during switching, not
1922          * worth to optimize.
1923          */
1924         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1925             !tfile->detached)
1926                 rxhash = __skb_get_hash_symmetric(skb);
1927
1928         rcu_read_lock();
1929         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1930                 err = -EIO;
1931                 rcu_read_unlock();
1932                 goto drop;
1933         }
1934
1935         if (frags) {
1936                 u32 headlen;
1937
1938                 /* Exercise flow dissector code path. */
1939                 skb_push(skb, ETH_HLEN);
1940                 headlen = eth_get_headlen(tun->dev, skb->data,
1941                                           skb_headlen(skb));
1942
1943                 if (unlikely(headlen > skb_headlen(skb))) {
1944                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1945                         napi_free_frags(&tfile->napi);
1946                         rcu_read_unlock();
1947                         mutex_unlock(&tfile->napi_mutex);
1948                         WARN_ON(1);
1949                         return -ENOMEM;
1950                 }
1951
1952                 local_bh_disable();
1953                 napi_gro_frags(&tfile->napi);
1954                 local_bh_enable();
1955                 mutex_unlock(&tfile->napi_mutex);
1956         } else if (tfile->napi_enabled) {
1957                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1958                 int queue_len;
1959
1960                 spin_lock_bh(&queue->lock);
1961                 __skb_queue_tail(queue, skb);
1962                 queue_len = skb_queue_len(queue);
1963                 spin_unlock(&queue->lock);
1964
1965                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1966                         napi_schedule(&tfile->napi);
1967
1968                 local_bh_enable();
1969         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1970                 tun_rx_batched(tun, tfile, skb, more);
1971         } else {
1972                 netif_rx_ni(skb);
1973         }
1974         rcu_read_unlock();
1975
1976         stats = get_cpu_ptr(tun->pcpu_stats);
1977         u64_stats_update_begin(&stats->syncp);
1978         u64_stats_inc(&stats->rx_packets);
1979         u64_stats_add(&stats->rx_bytes, len);
1980         u64_stats_update_end(&stats->syncp);
1981         put_cpu_ptr(stats);
1982
1983         if (rxhash)
1984                 tun_flow_update(tun, rxhash, tfile);
1985
1986         return total_len;
1987 }
1988
1989 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1990 {
1991         struct file *file = iocb->ki_filp;
1992         struct tun_file *tfile = file->private_data;
1993         struct tun_struct *tun = tun_get(tfile);
1994         ssize_t result;
1995
1996         if (!tun)
1997                 return -EBADFD;
1998
1999         result = tun_get_user(tun, tfile, NULL, from,
2000                               file->f_flags & O_NONBLOCK, false);
2001
2002         tun_put(tun);
2003         return result;
2004 }
2005
2006 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2007                                 struct tun_file *tfile,
2008                                 struct xdp_frame *xdp_frame,
2009                                 struct iov_iter *iter)
2010 {
2011         int vnet_hdr_sz = 0;
2012         size_t size = xdp_frame->len;
2013         struct tun_pcpu_stats *stats;
2014         size_t ret;
2015
2016         if (tun->flags & IFF_VNET_HDR) {
2017                 struct virtio_net_hdr gso = { 0 };
2018
2019                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2020                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2021                         return -EINVAL;
2022                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2023                              sizeof(gso)))
2024                         return -EFAULT;
2025                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2026         }
2027
2028         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2029
2030         stats = get_cpu_ptr(tun->pcpu_stats);
2031         u64_stats_update_begin(&stats->syncp);
2032         u64_stats_inc(&stats->tx_packets);
2033         u64_stats_add(&stats->tx_bytes, ret);
2034         u64_stats_update_end(&stats->syncp);
2035         put_cpu_ptr(tun->pcpu_stats);
2036
2037         return ret;
2038 }
2039
2040 /* Put packet to the user space buffer */
2041 static ssize_t tun_put_user(struct tun_struct *tun,
2042                             struct tun_file *tfile,
2043                             struct sk_buff *skb,
2044                             struct iov_iter *iter)
2045 {
2046         struct tun_pi pi = { 0, skb->protocol };
2047         struct tun_pcpu_stats *stats;
2048         ssize_t total;
2049         int vlan_offset = 0;
2050         int vlan_hlen = 0;
2051         int vnet_hdr_sz = 0;
2052
2053         if (skb_vlan_tag_present(skb))
2054                 vlan_hlen = VLAN_HLEN;
2055
2056         if (tun->flags & IFF_VNET_HDR)
2057                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2058
2059         total = skb->len + vlan_hlen + vnet_hdr_sz;
2060
2061         if (!(tun->flags & IFF_NO_PI)) {
2062                 if (iov_iter_count(iter) < sizeof(pi))
2063                         return -EINVAL;
2064
2065                 total += sizeof(pi);
2066                 if (iov_iter_count(iter) < total) {
2067                         /* Packet will be striped */
2068                         pi.flags |= TUN_PKT_STRIP;
2069                 }
2070
2071                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2072                         return -EFAULT;
2073         }
2074
2075         if (vnet_hdr_sz) {
2076                 struct virtio_net_hdr gso;
2077
2078                 if (iov_iter_count(iter) < vnet_hdr_sz)
2079                         return -EINVAL;
2080
2081                 if (virtio_net_hdr_from_skb(skb, &gso,
2082                                             tun_is_little_endian(tun), true,
2083                                             vlan_hlen)) {
2084                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2085                         pr_err("unexpected GSO type: "
2086                                "0x%x, gso_size %d, hdr_len %d\n",
2087                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2088                                tun16_to_cpu(tun, gso.hdr_len));
2089                         print_hex_dump(KERN_ERR, "tun: ",
2090                                        DUMP_PREFIX_NONE,
2091                                        16, 1, skb->head,
2092                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2093                         WARN_ON_ONCE(1);
2094                         return -EINVAL;
2095                 }
2096
2097                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2098                         return -EFAULT;
2099
2100                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2101         }
2102
2103         if (vlan_hlen) {
2104                 int ret;
2105                 struct veth veth;
2106
2107                 veth.h_vlan_proto = skb->vlan_proto;
2108                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2109
2110                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2111
2112                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2113                 if (ret || !iov_iter_count(iter))
2114                         goto done;
2115
2116                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2117                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2118                         goto done;
2119         }
2120
2121         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2122
2123 done:
2124         /* caller is in process context, */
2125         stats = get_cpu_ptr(tun->pcpu_stats);
2126         u64_stats_update_begin(&stats->syncp);
2127         u64_stats_inc(&stats->tx_packets);
2128         u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2129         u64_stats_update_end(&stats->syncp);
2130         put_cpu_ptr(tun->pcpu_stats);
2131
2132         return total;
2133 }
2134
2135 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2136 {
2137         DECLARE_WAITQUEUE(wait, current);
2138         void *ptr = NULL;
2139         int error = 0;
2140
2141         ptr = ptr_ring_consume(&tfile->tx_ring);
2142         if (ptr)
2143                 goto out;
2144         if (noblock) {
2145                 error = -EAGAIN;
2146                 goto out;
2147         }
2148
2149         add_wait_queue(&tfile->socket.wq.wait, &wait);
2150
2151         while (1) {
2152                 set_current_state(TASK_INTERRUPTIBLE);
2153                 ptr = ptr_ring_consume(&tfile->tx_ring);
2154                 if (ptr)
2155                         break;
2156                 if (signal_pending(current)) {
2157                         error = -ERESTARTSYS;
2158                         break;
2159                 }
2160                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2161                         error = -EFAULT;
2162                         break;
2163                 }
2164
2165                 schedule();
2166         }
2167
2168         __set_current_state(TASK_RUNNING);
2169         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2170
2171 out:
2172         *err = error;
2173         return ptr;
2174 }
2175
2176 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2177                            struct iov_iter *to,
2178                            int noblock, void *ptr)
2179 {
2180         ssize_t ret;
2181         int err;
2182
2183         if (!iov_iter_count(to)) {
2184                 tun_ptr_free(ptr);
2185                 return 0;
2186         }
2187
2188         if (!ptr) {
2189                 /* Read frames from ring */
2190                 ptr = tun_ring_recv(tfile, noblock, &err);
2191                 if (!ptr)
2192                         return err;
2193         }
2194
2195         if (tun_is_xdp_frame(ptr)) {
2196                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2197
2198                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2199                 xdp_return_frame(xdpf);
2200         } else {
2201                 struct sk_buff *skb = ptr;
2202
2203                 ret = tun_put_user(tun, tfile, skb, to);
2204                 if (unlikely(ret < 0))
2205                         kfree_skb(skb);
2206                 else
2207                         consume_skb(skb);
2208         }
2209
2210         return ret;
2211 }
2212
2213 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2214 {
2215         struct file *file = iocb->ki_filp;
2216         struct tun_file *tfile = file->private_data;
2217         struct tun_struct *tun = tun_get(tfile);
2218         ssize_t len = iov_iter_count(to), ret;
2219
2220         if (!tun)
2221                 return -EBADFD;
2222         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2223         ret = min_t(ssize_t, ret, len);
2224         if (ret > 0)
2225                 iocb->ki_pos = ret;
2226         tun_put(tun);
2227         return ret;
2228 }
2229
2230 static void tun_prog_free(struct rcu_head *rcu)
2231 {
2232         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2233
2234         bpf_prog_destroy(prog->prog);
2235         kfree(prog);
2236 }
2237
2238 static int __tun_set_ebpf(struct tun_struct *tun,
2239                           struct tun_prog __rcu **prog_p,
2240                           struct bpf_prog *prog)
2241 {
2242         struct tun_prog *old, *new = NULL;
2243
2244         if (prog) {
2245                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2246                 if (!new)
2247                         return -ENOMEM;
2248                 new->prog = prog;
2249         }
2250
2251         spin_lock_bh(&tun->lock);
2252         old = rcu_dereference_protected(*prog_p,
2253                                         lockdep_is_held(&tun->lock));
2254         rcu_assign_pointer(*prog_p, new);
2255         spin_unlock_bh(&tun->lock);
2256
2257         if (old)
2258                 call_rcu(&old->rcu, tun_prog_free);
2259
2260         return 0;
2261 }
2262
2263 static void tun_free_netdev(struct net_device *dev)
2264 {
2265         struct tun_struct *tun = netdev_priv(dev);
2266
2267         BUG_ON(!(list_empty(&tun->disabled)));
2268
2269         free_percpu(tun->pcpu_stats);
2270         /* We clear pcpu_stats so that tun_set_iff() can tell if
2271          * tun_free_netdev() has been called from register_netdevice().
2272          */
2273         tun->pcpu_stats = NULL;
2274
2275         tun_flow_uninit(tun);
2276         security_tun_dev_free_security(tun->security);
2277         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2278         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2279 }
2280
2281 static void tun_setup(struct net_device *dev)
2282 {
2283         struct tun_struct *tun = netdev_priv(dev);
2284
2285         tun->owner = INVALID_UID;
2286         tun->group = INVALID_GID;
2287         tun_default_link_ksettings(dev, &tun->link_ksettings);
2288
2289         dev->ethtool_ops = &tun_ethtool_ops;
2290         dev->needs_free_netdev = true;
2291         dev->priv_destructor = tun_free_netdev;
2292         /* We prefer our own queue length */
2293         dev->tx_queue_len = TUN_READQ_SIZE;
2294 }
2295
2296 /* Trivial set of netlink ops to allow deleting tun or tap
2297  * device with netlink.
2298  */
2299 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2300                         struct netlink_ext_ack *extack)
2301 {
2302         NL_SET_ERR_MSG(extack,
2303                        "tun/tap creation via rtnetlink is not supported.");
2304         return -EOPNOTSUPP;
2305 }
2306
2307 static size_t tun_get_size(const struct net_device *dev)
2308 {
2309         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2310         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2311
2312         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2313                nla_total_size(sizeof(gid_t)) + /* GROUP */
2314                nla_total_size(sizeof(u8)) + /* TYPE */
2315                nla_total_size(sizeof(u8)) + /* PI */
2316                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2317                nla_total_size(sizeof(u8)) + /* PERSIST */
2318                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2319                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2320                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2321                0;
2322 }
2323
2324 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2325 {
2326         struct tun_struct *tun = netdev_priv(dev);
2327
2328         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2329                 goto nla_put_failure;
2330         if (uid_valid(tun->owner) &&
2331             nla_put_u32(skb, IFLA_TUN_OWNER,
2332                         from_kuid_munged(current_user_ns(), tun->owner)))
2333                 goto nla_put_failure;
2334         if (gid_valid(tun->group) &&
2335             nla_put_u32(skb, IFLA_TUN_GROUP,
2336                         from_kgid_munged(current_user_ns(), tun->group)))
2337                 goto nla_put_failure;
2338         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2339                 goto nla_put_failure;
2340         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2341                 goto nla_put_failure;
2342         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2343                 goto nla_put_failure;
2344         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2345                        !!(tun->flags & IFF_MULTI_QUEUE)))
2346                 goto nla_put_failure;
2347         if (tun->flags & IFF_MULTI_QUEUE) {
2348                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2349                         goto nla_put_failure;
2350                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2351                                 tun->numdisabled))
2352                         goto nla_put_failure;
2353         }
2354
2355         return 0;
2356
2357 nla_put_failure:
2358         return -EMSGSIZE;
2359 }
2360
2361 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2362         .kind           = DRV_NAME,
2363         .priv_size      = sizeof(struct tun_struct),
2364         .setup          = tun_setup,
2365         .validate       = tun_validate,
2366         .get_size       = tun_get_size,
2367         .fill_info      = tun_fill_info,
2368 };
2369
2370 static void tun_sock_write_space(struct sock *sk)
2371 {
2372         struct tun_file *tfile;
2373         wait_queue_head_t *wqueue;
2374
2375         if (!sock_writeable(sk))
2376                 return;
2377
2378         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2379                 return;
2380
2381         wqueue = sk_sleep(sk);
2382         if (wqueue && waitqueue_active(wqueue))
2383                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2384                                                 EPOLLWRNORM | EPOLLWRBAND);
2385
2386         tfile = container_of(sk, struct tun_file, sk);
2387         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2388 }
2389
2390 static void tun_put_page(struct tun_page *tpage)
2391 {
2392         if (tpage->page)
2393                 __page_frag_cache_drain(tpage->page, tpage->count);
2394 }
2395
2396 static int tun_xdp_one(struct tun_struct *tun,
2397                        struct tun_file *tfile,
2398                        struct xdp_buff *xdp, int *flush,
2399                        struct tun_page *tpage)
2400 {
2401         unsigned int datasize = xdp->data_end - xdp->data;
2402         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2403         struct virtio_net_hdr *gso = &hdr->gso;
2404         struct tun_pcpu_stats *stats;
2405         struct bpf_prog *xdp_prog;
2406         struct sk_buff *skb = NULL;
2407         u32 rxhash = 0, act;
2408         int buflen = hdr->buflen;
2409         int err = 0;
2410         bool skb_xdp = false;
2411         struct page *page;
2412
2413         xdp_prog = rcu_dereference(tun->xdp_prog);
2414         if (xdp_prog) {
2415                 if (gso->gso_type) {
2416                         skb_xdp = true;
2417                         goto build;
2418                 }
2419                 xdp_set_data_meta_invalid(xdp);
2420                 xdp->rxq = &tfile->xdp_rxq;
2421                 xdp->frame_sz = buflen;
2422
2423                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2424                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2425                 if (err < 0) {
2426                         put_page(virt_to_head_page(xdp->data));
2427                         return err;
2428                 }
2429
2430                 switch (err) {
2431                 case XDP_REDIRECT:
2432                         *flush = true;
2433                         /* fall through */
2434                 case XDP_TX:
2435                         return 0;
2436                 case XDP_PASS:
2437                         break;
2438                 default:
2439                         page = virt_to_head_page(xdp->data);
2440                         if (tpage->page == page) {
2441                                 ++tpage->count;
2442                         } else {
2443                                 tun_put_page(tpage);
2444                                 tpage->page = page;
2445                                 tpage->count = 1;
2446                         }
2447                         return 0;
2448                 }
2449         }
2450
2451 build:
2452         skb = build_skb(xdp->data_hard_start, buflen);
2453         if (!skb) {
2454                 err = -ENOMEM;
2455                 goto out;
2456         }
2457
2458         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2459         skb_put(skb, xdp->data_end - xdp->data);
2460
2461         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2462                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2463                 kfree_skb(skb);
2464                 err = -EINVAL;
2465                 goto out;
2466         }
2467
2468         skb->protocol = eth_type_trans(skb, tun->dev);
2469         skb_reset_network_header(skb);
2470         skb_probe_transport_header(skb);
2471         skb_record_rx_queue(skb, tfile->queue_index);
2472
2473         if (skb_xdp) {
2474                 err = do_xdp_generic(xdp_prog, skb);
2475                 if (err != XDP_PASS)
2476                         goto out;
2477         }
2478
2479         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2480             !tfile->detached)
2481                 rxhash = __skb_get_hash_symmetric(skb);
2482
2483         netif_receive_skb(skb);
2484
2485         /* No need for get_cpu_ptr() here since this function is
2486          * always called with bh disabled
2487          */
2488         stats = this_cpu_ptr(tun->pcpu_stats);
2489         u64_stats_update_begin(&stats->syncp);
2490         u64_stats_inc(&stats->rx_packets);
2491         u64_stats_add(&stats->rx_bytes, datasize);
2492         u64_stats_update_end(&stats->syncp);
2493
2494         if (rxhash)
2495                 tun_flow_update(tun, rxhash, tfile);
2496
2497 out:
2498         return err;
2499 }
2500
2501 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2502 {
2503         int ret, i;
2504         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2505         struct tun_struct *tun = tun_get(tfile);
2506         struct tun_msg_ctl *ctl = m->msg_control;
2507         struct xdp_buff *xdp;
2508
2509         if (!tun)
2510                 return -EBADFD;
2511
2512         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2513                 struct tun_page tpage;
2514                 int n = ctl->num;
2515                 int flush = 0;
2516
2517                 memset(&tpage, 0, sizeof(tpage));
2518
2519                 local_bh_disable();
2520                 rcu_read_lock();
2521
2522                 for (i = 0; i < n; i++) {
2523                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2524                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2525                 }
2526
2527                 if (flush)
2528                         xdp_do_flush();
2529
2530                 rcu_read_unlock();
2531                 local_bh_enable();
2532
2533                 tun_put_page(&tpage);
2534
2535                 ret = total_len;
2536                 goto out;
2537         }
2538
2539         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2540                            m->msg_flags & MSG_DONTWAIT,
2541                            m->msg_flags & MSG_MORE);
2542 out:
2543         tun_put(tun);
2544         return ret;
2545 }
2546
2547 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2548                        int flags)
2549 {
2550         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2551         struct tun_struct *tun = tun_get(tfile);
2552         void *ptr = m->msg_control;
2553         int ret;
2554
2555         if (!tun) {
2556                 ret = -EBADFD;
2557                 goto out_free;
2558         }
2559
2560         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2561                 ret = -EINVAL;
2562                 goto out_put_tun;
2563         }
2564         if (flags & MSG_ERRQUEUE) {
2565                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2566                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2567                 goto out;
2568         }
2569         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2570         if (ret > (ssize_t)total_len) {
2571                 m->msg_flags |= MSG_TRUNC;
2572                 ret = flags & MSG_TRUNC ? ret : total_len;
2573         }
2574 out:
2575         tun_put(tun);
2576         return ret;
2577
2578 out_put_tun:
2579         tun_put(tun);
2580 out_free:
2581         tun_ptr_free(ptr);
2582         return ret;
2583 }
2584
2585 static int tun_ptr_peek_len(void *ptr)
2586 {
2587         if (likely(ptr)) {
2588                 if (tun_is_xdp_frame(ptr)) {
2589                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2590
2591                         return xdpf->len;
2592                 }
2593                 return __skb_array_len_with_tag(ptr);
2594         } else {
2595                 return 0;
2596         }
2597 }
2598
2599 static int tun_peek_len(struct socket *sock)
2600 {
2601         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2602         struct tun_struct *tun;
2603         int ret = 0;
2604
2605         tun = tun_get(tfile);
2606         if (!tun)
2607                 return 0;
2608
2609         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2610         tun_put(tun);
2611
2612         return ret;
2613 }
2614
2615 /* Ops structure to mimic raw sockets with tun */
2616 static const struct proto_ops tun_socket_ops = {
2617         .peek_len = tun_peek_len,
2618         .sendmsg = tun_sendmsg,
2619         .recvmsg = tun_recvmsg,
2620 };
2621
2622 static struct proto tun_proto = {
2623         .name           = "tun",
2624         .owner          = THIS_MODULE,
2625         .obj_size       = sizeof(struct tun_file),
2626 };
2627
2628 static int tun_flags(struct tun_struct *tun)
2629 {
2630         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2631 }
2632
2633 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2634                               char *buf)
2635 {
2636         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2637         return sprintf(buf, "0x%x\n", tun_flags(tun));
2638 }
2639
2640 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2641                               char *buf)
2642 {
2643         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2644         return uid_valid(tun->owner)?
2645                 sprintf(buf, "%u\n",
2646                         from_kuid_munged(current_user_ns(), tun->owner)):
2647                 sprintf(buf, "-1\n");
2648 }
2649
2650 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2651                               char *buf)
2652 {
2653         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2654         return gid_valid(tun->group) ?
2655                 sprintf(buf, "%u\n",
2656                         from_kgid_munged(current_user_ns(), tun->group)):
2657                 sprintf(buf, "-1\n");
2658 }
2659
2660 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2661 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2662 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2663
2664 static struct attribute *tun_dev_attrs[] = {
2665         &dev_attr_tun_flags.attr,
2666         &dev_attr_owner.attr,
2667         &dev_attr_group.attr,
2668         NULL
2669 };
2670
2671 static const struct attribute_group tun_attr_group = {
2672         .attrs = tun_dev_attrs
2673 };
2674
2675 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2676 {
2677         struct tun_struct *tun;
2678         struct tun_file *tfile = file->private_data;
2679         struct net_device *dev;
2680         int err;
2681
2682         if (tfile->detached)
2683                 return -EINVAL;
2684
2685         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2686                 if (!capable(CAP_NET_ADMIN))
2687                         return -EPERM;
2688
2689                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2690                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2691                         return -EINVAL;
2692         }
2693
2694         dev = __dev_get_by_name(net, ifr->ifr_name);
2695         if (dev) {
2696                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2697                         return -EBUSY;
2698                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2699                         tun = netdev_priv(dev);
2700                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2701                         tun = netdev_priv(dev);
2702                 else
2703                         return -EINVAL;
2704
2705                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2706                     !!(tun->flags & IFF_MULTI_QUEUE))
2707                         return -EINVAL;
2708
2709                 if (tun_not_capable(tun))
2710                         return -EPERM;
2711                 err = security_tun_dev_open(tun->security);
2712                 if (err < 0)
2713                         return err;
2714
2715                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2716                                  ifr->ifr_flags & IFF_NAPI,
2717                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2718                 if (err < 0)
2719                         return err;
2720
2721                 if (tun->flags & IFF_MULTI_QUEUE &&
2722                     (tun->numqueues + tun->numdisabled > 1)) {
2723                         /* One or more queue has already been attached, no need
2724                          * to initialize the device again.
2725                          */
2726                         netdev_state_change(dev);
2727                         return 0;
2728                 }
2729
2730                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2731                               (ifr->ifr_flags & TUN_FEATURES);
2732
2733                 netdev_state_change(dev);
2734         } else {
2735                 char *name;
2736                 unsigned long flags = 0;
2737                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2738                              MAX_TAP_QUEUES : 1;
2739
2740                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2741                         return -EPERM;
2742                 err = security_tun_dev_create();
2743                 if (err < 0)
2744                         return err;
2745
2746                 /* Set dev type */
2747                 if (ifr->ifr_flags & IFF_TUN) {
2748                         /* TUN device */
2749                         flags |= IFF_TUN;
2750                         name = "tun%d";
2751                 } else if (ifr->ifr_flags & IFF_TAP) {
2752                         /* TAP device */
2753                         flags |= IFF_TAP;
2754                         name = "tap%d";
2755                 } else
2756                         return -EINVAL;
2757
2758                 if (*ifr->ifr_name)
2759                         name = ifr->ifr_name;
2760
2761                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2762                                        NET_NAME_UNKNOWN, tun_setup, queues,
2763                                        queues);
2764
2765                 if (!dev)
2766                         return -ENOMEM;
2767
2768                 dev_net_set(dev, net);
2769                 dev->rtnl_link_ops = &tun_link_ops;
2770                 dev->ifindex = tfile->ifindex;
2771                 dev->sysfs_groups[0] = &tun_attr_group;
2772
2773                 tun = netdev_priv(dev);
2774                 tun->dev = dev;
2775                 tun->flags = flags;
2776                 tun->txflt.count = 0;
2777                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2778
2779                 tun->align = NET_SKB_PAD;
2780                 tun->filter_attached = false;
2781                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2782                 tun->rx_batched = 0;
2783                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2784
2785                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2786                 if (!tun->pcpu_stats) {
2787                         err = -ENOMEM;
2788                         goto err_free_dev;
2789                 }
2790
2791                 spin_lock_init(&tun->lock);
2792
2793                 err = security_tun_dev_alloc_security(&tun->security);
2794                 if (err < 0)
2795                         goto err_free_stat;
2796
2797                 tun_net_init(dev);
2798                 tun_flow_init(tun);
2799
2800                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2801                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2802                                    NETIF_F_HW_VLAN_STAG_TX;
2803                 dev->features = dev->hw_features | NETIF_F_LLTX;
2804                 dev->vlan_features = dev->features &
2805                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2806                                        NETIF_F_HW_VLAN_STAG_TX);
2807
2808                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2809                               (ifr->ifr_flags & TUN_FEATURES);
2810
2811                 INIT_LIST_HEAD(&tun->disabled);
2812                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2813                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2814                 if (err < 0)
2815                         goto err_free_flow;
2816
2817                 err = register_netdevice(tun->dev);
2818                 if (err < 0)
2819                         goto err_detach;
2820                 /* free_netdev() won't check refcnt, to aovid race
2821                  * with dev_put() we need publish tun after registration.
2822                  */
2823                 rcu_assign_pointer(tfile->tun, tun);
2824         }
2825
2826         netif_carrier_on(tun->dev);
2827
2828         /* Make sure persistent devices do not get stuck in
2829          * xoff state.
2830          */
2831         if (netif_running(tun->dev))
2832                 netif_tx_wake_all_queues(tun->dev);
2833
2834         strcpy(ifr->ifr_name, tun->dev->name);
2835         return 0;
2836
2837 err_detach:
2838         tun_detach_all(dev);
2839         /* We are here because register_netdevice() has failed.
2840          * If register_netdevice() already called tun_free_netdev()
2841          * while dealing with the error, tun->pcpu_stats has been cleared.
2842          */
2843         if (!tun->pcpu_stats)
2844                 goto err_free_dev;
2845
2846 err_free_flow:
2847         tun_flow_uninit(tun);
2848         security_tun_dev_free_security(tun->security);
2849 err_free_stat:
2850         free_percpu(tun->pcpu_stats);
2851 err_free_dev:
2852         free_netdev(dev);
2853         return err;
2854 }
2855
2856 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2857 {
2858         strcpy(ifr->ifr_name, tun->dev->name);
2859
2860         ifr->ifr_flags = tun_flags(tun);
2861
2862 }
2863
2864 /* This is like a cut-down ethtool ops, except done via tun fd so no
2865  * privs required. */
2866 static int set_offload(struct tun_struct *tun, unsigned long arg)
2867 {
2868         netdev_features_t features = 0;
2869
2870         if (arg & TUN_F_CSUM) {
2871                 features |= NETIF_F_HW_CSUM;
2872                 arg &= ~TUN_F_CSUM;
2873
2874                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2875                         if (arg & TUN_F_TSO_ECN) {
2876                                 features |= NETIF_F_TSO_ECN;
2877                                 arg &= ~TUN_F_TSO_ECN;
2878                         }
2879                         if (arg & TUN_F_TSO4)
2880                                 features |= NETIF_F_TSO;
2881                         if (arg & TUN_F_TSO6)
2882                                 features |= NETIF_F_TSO6;
2883                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2884                 }
2885
2886                 arg &= ~TUN_F_UFO;
2887         }
2888
2889         /* This gives the user a way to test for new features in future by
2890          * trying to set them. */
2891         if (arg)
2892                 return -EINVAL;
2893
2894         tun->set_features = features;
2895         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2896         tun->dev->wanted_features |= features;
2897         netdev_update_features(tun->dev);
2898
2899         return 0;
2900 }
2901
2902 static void tun_detach_filter(struct tun_struct *tun, int n)
2903 {
2904         int i;
2905         struct tun_file *tfile;
2906
2907         for (i = 0; i < n; i++) {
2908                 tfile = rtnl_dereference(tun->tfiles[i]);
2909                 lock_sock(tfile->socket.sk);
2910                 sk_detach_filter(tfile->socket.sk);
2911                 release_sock(tfile->socket.sk);
2912         }
2913
2914         tun->filter_attached = false;
2915 }
2916
2917 static int tun_attach_filter(struct tun_struct *tun)
2918 {
2919         int i, ret = 0;
2920         struct tun_file *tfile;
2921
2922         for (i = 0; i < tun->numqueues; i++) {
2923                 tfile = rtnl_dereference(tun->tfiles[i]);
2924                 lock_sock(tfile->socket.sk);
2925                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2926                 release_sock(tfile->socket.sk);
2927                 if (ret) {
2928                         tun_detach_filter(tun, i);
2929                         return ret;
2930                 }
2931         }
2932
2933         tun->filter_attached = true;
2934         return ret;
2935 }
2936
2937 static void tun_set_sndbuf(struct tun_struct *tun)
2938 {
2939         struct tun_file *tfile;
2940         int i;
2941
2942         for (i = 0; i < tun->numqueues; i++) {
2943                 tfile = rtnl_dereference(tun->tfiles[i]);
2944                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2945         }
2946 }
2947
2948 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2949 {
2950         struct tun_file *tfile = file->private_data;
2951         struct tun_struct *tun;
2952         int ret = 0;
2953
2954         rtnl_lock();
2955
2956         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2957                 tun = tfile->detached;
2958                 if (!tun) {
2959                         ret = -EINVAL;
2960                         goto unlock;
2961                 }
2962                 ret = security_tun_dev_attach_queue(tun->security);
2963                 if (ret < 0)
2964                         goto unlock;
2965                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2966                                  tun->flags & IFF_NAPI_FRAGS, true);
2967         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2968                 tun = rtnl_dereference(tfile->tun);
2969                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2970                         ret = -EINVAL;
2971                 else
2972                         __tun_detach(tfile, false);
2973         } else
2974                 ret = -EINVAL;
2975
2976         if (ret >= 0)
2977                 netdev_state_change(tun->dev);
2978
2979 unlock:
2980         rtnl_unlock();
2981         return ret;
2982 }
2983
2984 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2985                         void __user *data)
2986 {
2987         struct bpf_prog *prog;
2988         int fd;
2989
2990         if (copy_from_user(&fd, data, sizeof(fd)))
2991                 return -EFAULT;
2992
2993         if (fd == -1) {
2994                 prog = NULL;
2995         } else {
2996                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2997                 if (IS_ERR(prog))
2998                         return PTR_ERR(prog);
2999         }
3000
3001         return __tun_set_ebpf(tun, prog_p, prog);
3002 }
3003
3004 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3005                             unsigned long arg, int ifreq_len)
3006 {
3007         struct tun_file *tfile = file->private_data;
3008         struct net *net = sock_net(&tfile->sk);
3009         struct tun_struct *tun;
3010         void __user* argp = (void __user*)arg;
3011         unsigned int ifindex, carrier;
3012         struct ifreq ifr;
3013         kuid_t owner;
3014         kgid_t group;
3015         int sndbuf;
3016         int vnet_hdr_sz;
3017         int le;
3018         int ret;
3019         bool do_notify = false;
3020
3021         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3022             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3023                 if (copy_from_user(&ifr, argp, ifreq_len))
3024                         return -EFAULT;
3025         } else {
3026                 memset(&ifr, 0, sizeof(ifr));
3027         }
3028         if (cmd == TUNGETFEATURES) {
3029                 /* Currently this just means: "what IFF flags are valid?".
3030                  * This is needed because we never checked for invalid flags on
3031                  * TUNSETIFF.
3032                  */
3033                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3034                                 (unsigned int __user*)argp);
3035         } else if (cmd == TUNSETQUEUE) {
3036                 return tun_set_queue(file, &ifr);
3037         } else if (cmd == SIOCGSKNS) {
3038                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3039                         return -EPERM;
3040                 return open_related_ns(&net->ns, get_net_ns);
3041         }
3042
3043         ret = 0;
3044         rtnl_lock();
3045
3046         tun = tun_get(tfile);
3047         if (cmd == TUNSETIFF) {
3048                 ret = -EEXIST;
3049                 if (tun)
3050                         goto unlock;
3051
3052                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3053
3054                 ret = tun_set_iff(net, file, &ifr);
3055
3056                 if (ret)
3057                         goto unlock;
3058
3059                 if (copy_to_user(argp, &ifr, ifreq_len))
3060                         ret = -EFAULT;
3061                 goto unlock;
3062         }
3063         if (cmd == TUNSETIFINDEX) {
3064                 ret = -EPERM;
3065                 if (tun)
3066                         goto unlock;
3067
3068                 ret = -EFAULT;
3069                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3070                         goto unlock;
3071
3072                 ret = 0;
3073                 tfile->ifindex = ifindex;
3074                 goto unlock;
3075         }
3076
3077         ret = -EBADFD;
3078         if (!tun)
3079                 goto unlock;
3080
3081         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3082
3083         net = dev_net(tun->dev);
3084         ret = 0;
3085         switch (cmd) {
3086         case TUNGETIFF:
3087                 tun_get_iff(tun, &ifr);
3088
3089                 if (tfile->detached)
3090                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3091                 if (!tfile->socket.sk->sk_filter)
3092                         ifr.ifr_flags |= IFF_NOFILTER;
3093
3094                 if (copy_to_user(argp, &ifr, ifreq_len))
3095                         ret = -EFAULT;
3096                 break;
3097
3098         case TUNSETNOCSUM:
3099                 /* Disable/Enable checksum */
3100
3101                 /* [unimplemented] */
3102                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3103                            arg ? "disabled" : "enabled");
3104                 break;
3105
3106         case TUNSETPERSIST:
3107                 /* Disable/Enable persist mode. Keep an extra reference to the
3108                  * module to prevent the module being unprobed.
3109                  */
3110                 if (arg && !(tun->flags & IFF_PERSIST)) {
3111                         tun->flags |= IFF_PERSIST;
3112                         __module_get(THIS_MODULE);
3113                         do_notify = true;
3114                 }
3115                 if (!arg && (tun->flags & IFF_PERSIST)) {
3116                         tun->flags &= ~IFF_PERSIST;
3117                         module_put(THIS_MODULE);
3118                         do_notify = true;
3119                 }
3120
3121                 netif_info(tun, drv, tun->dev, "persist %s\n",
3122                            arg ? "enabled" : "disabled");
3123                 break;
3124
3125         case TUNSETOWNER:
3126                 /* Set owner of the device */
3127                 owner = make_kuid(current_user_ns(), arg);
3128                 if (!uid_valid(owner)) {
3129                         ret = -EINVAL;
3130                         break;
3131                 }
3132                 tun->owner = owner;
3133                 do_notify = true;
3134                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3135                            from_kuid(&init_user_ns, tun->owner));
3136                 break;
3137
3138         case TUNSETGROUP:
3139                 /* Set group of the device */
3140                 group = make_kgid(current_user_ns(), arg);
3141                 if (!gid_valid(group)) {
3142                         ret = -EINVAL;
3143                         break;
3144                 }
3145                 tun->group = group;
3146                 do_notify = true;
3147                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3148                            from_kgid(&init_user_ns, tun->group));
3149                 break;
3150
3151         case TUNSETLINK:
3152                 /* Only allow setting the type when the interface is down */
3153                 if (tun->dev->flags & IFF_UP) {
3154                         netif_info(tun, drv, tun->dev,
3155                                    "Linktype set failed because interface is up\n");
3156                         ret = -EBUSY;
3157                 } else {
3158                         tun->dev->type = (int) arg;
3159                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3160                                    tun->dev->type);
3161                         ret = 0;
3162                 }
3163                 break;
3164
3165         case TUNSETDEBUG:
3166                 tun->msg_enable = (u32)arg;
3167                 break;
3168
3169         case TUNSETOFFLOAD:
3170                 ret = set_offload(tun, arg);
3171                 break;
3172
3173         case TUNSETTXFILTER:
3174                 /* Can be set only for TAPs */
3175                 ret = -EINVAL;
3176                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3177                         break;
3178                 ret = update_filter(&tun->txflt, (void __user *)arg);
3179                 break;
3180
3181         case SIOCGIFHWADDR:
3182                 /* Get hw address */
3183                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3184                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3185                 if (copy_to_user(argp, &ifr, ifreq_len))
3186                         ret = -EFAULT;
3187                 break;
3188
3189         case SIOCSIFHWADDR:
3190                 /* Set hw address */
3191                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3192                 break;
3193
3194         case TUNGETSNDBUF:
3195                 sndbuf = tfile->socket.sk->sk_sndbuf;
3196                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3197                         ret = -EFAULT;
3198                 break;
3199
3200         case TUNSETSNDBUF:
3201                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3202                         ret = -EFAULT;
3203                         break;
3204                 }
3205                 if (sndbuf <= 0) {
3206                         ret = -EINVAL;
3207                         break;
3208                 }
3209
3210                 tun->sndbuf = sndbuf;
3211                 tun_set_sndbuf(tun);
3212                 break;
3213
3214         case TUNGETVNETHDRSZ:
3215                 vnet_hdr_sz = tun->vnet_hdr_sz;
3216                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3217                         ret = -EFAULT;
3218                 break;
3219
3220         case TUNSETVNETHDRSZ:
3221                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3222                         ret = -EFAULT;
3223                         break;
3224                 }
3225                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3226                         ret = -EINVAL;
3227                         break;
3228                 }
3229
3230                 tun->vnet_hdr_sz = vnet_hdr_sz;
3231                 break;
3232
3233         case TUNGETVNETLE:
3234                 le = !!(tun->flags & TUN_VNET_LE);
3235                 if (put_user(le, (int __user *)argp))
3236                         ret = -EFAULT;
3237                 break;
3238
3239         case TUNSETVNETLE:
3240                 if (get_user(le, (int __user *)argp)) {
3241                         ret = -EFAULT;
3242                         break;
3243                 }
3244                 if (le)
3245                         tun->flags |= TUN_VNET_LE;
3246                 else
3247                         tun->flags &= ~TUN_VNET_LE;
3248                 break;
3249
3250         case TUNGETVNETBE:
3251                 ret = tun_get_vnet_be(tun, argp);
3252                 break;
3253
3254         case TUNSETVNETBE:
3255                 ret = tun_set_vnet_be(tun, argp);
3256                 break;
3257
3258         case TUNATTACHFILTER:
3259                 /* Can be set only for TAPs */
3260                 ret = -EINVAL;
3261                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3262                         break;
3263                 ret = -EFAULT;
3264                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3265                         break;
3266
3267                 ret = tun_attach_filter(tun);
3268                 break;
3269
3270         case TUNDETACHFILTER:
3271                 /* Can be set only for TAPs */
3272                 ret = -EINVAL;
3273                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3274                         break;
3275                 ret = 0;
3276                 tun_detach_filter(tun, tun->numqueues);
3277                 break;
3278
3279         case TUNGETFILTER:
3280                 ret = -EINVAL;
3281                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3282                         break;
3283                 ret = -EFAULT;
3284                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3285                         break;
3286                 ret = 0;
3287                 break;
3288
3289         case TUNSETSTEERINGEBPF:
3290                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3291                 break;
3292
3293         case TUNSETFILTEREBPF:
3294                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3295                 break;
3296
3297         case TUNSETCARRIER:
3298                 ret = -EFAULT;
3299                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3300                         goto unlock;
3301
3302                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3303                 break;
3304
3305         case TUNGETDEVNETNS:
3306                 ret = -EPERM;
3307                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3308                         goto unlock;
3309                 ret = open_related_ns(&net->ns, get_net_ns);
3310                 break;
3311
3312         default:
3313                 ret = -EINVAL;
3314                 break;
3315         }
3316
3317         if (do_notify)
3318                 netdev_state_change(tun->dev);
3319
3320 unlock:
3321         rtnl_unlock();
3322         if (tun)
3323                 tun_put(tun);
3324         return ret;
3325 }
3326
3327 static long tun_chr_ioctl(struct file *file,
3328                           unsigned int cmd, unsigned long arg)
3329 {
3330         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3331 }
3332
3333 #ifdef CONFIG_COMPAT
3334 static long tun_chr_compat_ioctl(struct file *file,
3335                          unsigned int cmd, unsigned long arg)
3336 {
3337         switch (cmd) {
3338         case TUNSETIFF:
3339         case TUNGETIFF:
3340         case TUNSETTXFILTER:
3341         case TUNGETSNDBUF:
3342         case TUNSETSNDBUF:
3343         case SIOCGIFHWADDR:
3344         case SIOCSIFHWADDR:
3345                 arg = (unsigned long)compat_ptr(arg);
3346                 break;
3347         default:
3348                 arg = (compat_ulong_t)arg;
3349                 break;
3350         }
3351
3352         /*
3353          * compat_ifreq is shorter than ifreq, so we must not access beyond
3354          * the end of that structure. All fields that are used in this
3355          * driver are compatible though, we don't need to convert the
3356          * contents.
3357          */
3358         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3359 }
3360 #endif /* CONFIG_COMPAT */
3361
3362 static int tun_chr_fasync(int fd, struct file *file, int on)
3363 {
3364         struct tun_file *tfile = file->private_data;
3365         int ret;
3366
3367         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3368                 goto out;
3369
3370         if (on) {
3371                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3372                 tfile->flags |= TUN_FASYNC;
3373         } else
3374                 tfile->flags &= ~TUN_FASYNC;
3375         ret = 0;
3376 out:
3377         return ret;
3378 }
3379
3380 static int tun_chr_open(struct inode *inode, struct file * file)
3381 {
3382         struct net *net = current->nsproxy->net_ns;
3383         struct tun_file *tfile;
3384
3385         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3386                                             &tun_proto, 0);
3387         if (!tfile)
3388                 return -ENOMEM;
3389         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3390                 sk_free(&tfile->sk);
3391                 return -ENOMEM;
3392         }
3393
3394         mutex_init(&tfile->napi_mutex);
3395         RCU_INIT_POINTER(tfile->tun, NULL);
3396         tfile->flags = 0;
3397         tfile->ifindex = 0;
3398
3399         init_waitqueue_head(&tfile->socket.wq.wait);
3400
3401         tfile->socket.file = file;
3402         tfile->socket.ops = &tun_socket_ops;
3403
3404         sock_init_data(&tfile->socket, &tfile->sk);
3405
3406         tfile->sk.sk_write_space = tun_sock_write_space;
3407         tfile->sk.sk_sndbuf = INT_MAX;
3408
3409         file->private_data = tfile;
3410         INIT_LIST_HEAD(&tfile->next);
3411
3412         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3413
3414         return 0;
3415 }
3416
3417 static int tun_chr_close(struct inode *inode, struct file *file)
3418 {
3419         struct tun_file *tfile = file->private_data;
3420
3421         tun_detach(tfile, true);
3422
3423         return 0;
3424 }
3425
3426 #ifdef CONFIG_PROC_FS
3427 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3428 {
3429         struct tun_file *tfile = file->private_data;
3430         struct tun_struct *tun;
3431         struct ifreq ifr;
3432
3433         memset(&ifr, 0, sizeof(ifr));
3434
3435         rtnl_lock();
3436         tun = tun_get(tfile);
3437         if (tun)
3438                 tun_get_iff(tun, &ifr);
3439         rtnl_unlock();
3440
3441         if (tun)
3442                 tun_put(tun);
3443
3444         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3445 }
3446 #endif
3447
3448 static const struct file_operations tun_fops = {
3449         .owner  = THIS_MODULE,
3450         .llseek = no_llseek,
3451         .read_iter  = tun_chr_read_iter,
3452         .write_iter = tun_chr_write_iter,
3453         .poll   = tun_chr_poll,
3454         .unlocked_ioctl = tun_chr_ioctl,
3455 #ifdef CONFIG_COMPAT
3456         .compat_ioctl = tun_chr_compat_ioctl,
3457 #endif
3458         .open   = tun_chr_open,
3459         .release = tun_chr_close,
3460         .fasync = tun_chr_fasync,
3461 #ifdef CONFIG_PROC_FS
3462         .show_fdinfo = tun_chr_show_fdinfo,
3463 #endif
3464 };
3465
3466 static struct miscdevice tun_miscdev = {
3467         .minor = TUN_MINOR,
3468         .name = "tun",
3469         .nodename = "net/tun",
3470         .fops = &tun_fops,
3471 };
3472
3473 /* ethtool interface */
3474
3475 static void tun_default_link_ksettings(struct net_device *dev,
3476                                        struct ethtool_link_ksettings *cmd)
3477 {
3478         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3479         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3480         cmd->base.speed         = SPEED_10;
3481         cmd->base.duplex        = DUPLEX_FULL;
3482         cmd->base.port          = PORT_TP;
3483         cmd->base.phy_address   = 0;
3484         cmd->base.autoneg       = AUTONEG_DISABLE;
3485 }
3486
3487 static int tun_get_link_ksettings(struct net_device *dev,
3488                                   struct ethtool_link_ksettings *cmd)
3489 {
3490         struct tun_struct *tun = netdev_priv(dev);
3491
3492         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3493         return 0;
3494 }
3495
3496 static int tun_set_link_ksettings(struct net_device *dev,
3497                                   const struct ethtool_link_ksettings *cmd)
3498 {
3499         struct tun_struct *tun = netdev_priv(dev);
3500
3501         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3502         return 0;
3503 }
3504
3505 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3506 {
3507         struct tun_struct *tun = netdev_priv(dev);
3508
3509         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3510         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3511
3512         switch (tun->flags & TUN_TYPE_MASK) {
3513         case IFF_TUN:
3514                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3515                 break;
3516         case IFF_TAP:
3517                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3518                 break;
3519         }
3520 }
3521
3522 static u32 tun_get_msglevel(struct net_device *dev)
3523 {
3524         struct tun_struct *tun = netdev_priv(dev);
3525
3526         return tun->msg_enable;
3527 }
3528
3529 static void tun_set_msglevel(struct net_device *dev, u32 value)
3530 {
3531         struct tun_struct *tun = netdev_priv(dev);
3532
3533         tun->msg_enable = value;
3534 }
3535
3536 static int tun_get_coalesce(struct net_device *dev,
3537                             struct ethtool_coalesce *ec)
3538 {
3539         struct tun_struct *tun = netdev_priv(dev);
3540
3541         ec->rx_max_coalesced_frames = tun->rx_batched;
3542
3543         return 0;
3544 }
3545
3546 static int tun_set_coalesce(struct net_device *dev,
3547                             struct ethtool_coalesce *ec)
3548 {
3549         struct tun_struct *tun = netdev_priv(dev);
3550
3551         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3552                 tun->rx_batched = NAPI_POLL_WEIGHT;
3553         else
3554                 tun->rx_batched = ec->rx_max_coalesced_frames;
3555
3556         return 0;
3557 }
3558
3559 static const struct ethtool_ops tun_ethtool_ops = {
3560         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3561         .get_drvinfo    = tun_get_drvinfo,
3562         .get_msglevel   = tun_get_msglevel,
3563         .set_msglevel   = tun_set_msglevel,
3564         .get_link       = ethtool_op_get_link,
3565         .get_ts_info    = ethtool_op_get_ts_info,
3566         .get_coalesce   = tun_get_coalesce,
3567         .set_coalesce   = tun_set_coalesce,
3568         .get_link_ksettings = tun_get_link_ksettings,
3569         .set_link_ksettings = tun_set_link_ksettings,
3570 };
3571
3572 static int tun_queue_resize(struct tun_struct *tun)
3573 {
3574         struct net_device *dev = tun->dev;
3575         struct tun_file *tfile;
3576         struct ptr_ring **rings;
3577         int n = tun->numqueues + tun->numdisabled;
3578         int ret, i;
3579
3580         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3581         if (!rings)
3582                 return -ENOMEM;
3583
3584         for (i = 0; i < tun->numqueues; i++) {
3585                 tfile = rtnl_dereference(tun->tfiles[i]);
3586                 rings[i] = &tfile->tx_ring;
3587         }
3588         list_for_each_entry(tfile, &tun->disabled, next)
3589                 rings[i++] = &tfile->tx_ring;
3590
3591         ret = ptr_ring_resize_multiple(rings, n,
3592                                        dev->tx_queue_len, GFP_KERNEL,
3593                                        tun_ptr_free);
3594
3595         kfree(rings);
3596         return ret;
3597 }
3598
3599 static int tun_device_event(struct notifier_block *unused,
3600                             unsigned long event, void *ptr)
3601 {
3602         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3603         struct tun_struct *tun = netdev_priv(dev);
3604         int i;
3605
3606         if (dev->rtnl_link_ops != &tun_link_ops)
3607                 return NOTIFY_DONE;
3608
3609         switch (event) {
3610         case NETDEV_CHANGE_TX_QUEUE_LEN:
3611                 if (tun_queue_resize(tun))
3612                         return NOTIFY_BAD;
3613                 break;
3614         case NETDEV_UP:
3615                 for (i = 0; i < tun->numqueues; i++) {
3616                         struct tun_file *tfile;
3617
3618                         tfile = rtnl_dereference(tun->tfiles[i]);
3619                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3620                 }
3621                 break;
3622         default:
3623                 break;
3624         }
3625
3626         return NOTIFY_DONE;
3627 }
3628
3629 static struct notifier_block tun_notifier_block __read_mostly = {
3630         .notifier_call  = tun_device_event,
3631 };
3632
3633 static int __init tun_init(void)
3634 {
3635         int ret = 0;
3636
3637         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3638
3639         ret = rtnl_link_register(&tun_link_ops);
3640         if (ret) {
3641                 pr_err("Can't register link_ops\n");
3642                 goto err_linkops;
3643         }
3644
3645         ret = misc_register(&tun_miscdev);
3646         if (ret) {
3647                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3648                 goto err_misc;
3649         }
3650
3651         ret = register_netdevice_notifier(&tun_notifier_block);
3652         if (ret) {
3653                 pr_err("Can't register netdevice notifier\n");
3654                 goto err_notifier;
3655         }
3656
3657         return  0;
3658
3659 err_notifier:
3660         misc_deregister(&tun_miscdev);
3661 err_misc:
3662         rtnl_link_unregister(&tun_link_ops);
3663 err_linkops:
3664         return ret;
3665 }
3666
3667 static void tun_cleanup(void)
3668 {
3669         misc_deregister(&tun_miscdev);
3670         rtnl_link_unregister(&tun_link_ops);
3671         unregister_netdevice_notifier(&tun_notifier_block);
3672 }
3673
3674 /* Get an underlying socket object from tun file.  Returns error unless file is
3675  * attached to a device.  The returned object works like a packet socket, it
3676  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3677  * holding a reference to the file for as long as the socket is in use. */
3678 struct socket *tun_get_socket(struct file *file)
3679 {
3680         struct tun_file *tfile;
3681         if (file->f_op != &tun_fops)
3682                 return ERR_PTR(-EINVAL);
3683         tfile = file->private_data;
3684         if (!tfile)
3685                 return ERR_PTR(-EBADFD);
3686         return &tfile->socket;
3687 }
3688 EXPORT_SYMBOL_GPL(tun_get_socket);
3689
3690 struct ptr_ring *tun_get_tx_ring(struct file *file)
3691 {
3692         struct tun_file *tfile;
3693
3694         if (file->f_op != &tun_fops)
3695                 return ERR_PTR(-EINVAL);
3696         tfile = file->private_data;
3697         if (!tfile)
3698                 return ERR_PTR(-EBADFD);
3699         return &tfile->tx_ring;
3700 }
3701 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3702
3703 module_init(tun_init);
3704 module_exit(tun_cleanup);
3705 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3706 MODULE_AUTHOR(DRV_COPYRIGHT);
3707 MODULE_LICENSE("GPL");
3708 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3709 MODULE_ALIAS("devname:net/tun");