drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <underley@underley.eu.org>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72
  73 #include <linux/uaccess.h>
  74 #include <linux/proc_fs.h>
  75
  76 static void tun_default_link_ksettings(struct net_device *dev,
  77                                        struct ethtool_link_ksettings *cmd);
  78
  79 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  80
  81 /* TUN device flags */
  82
  83 /* IFF_ATTACH_QUEUE is never stored in device flags,
  84  * overload it to mean fasync when stored there.
  85  */
  86 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  87 /* High bits in flags field are unused. */
  88 #define TUN_VNET_LE     0x80000000
  89 #define TUN_VNET_BE     0x40000000
  90
  91 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
  92                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
  93
  94 #define GOODCOPY_LEN 128
  95
  96 #define FLT_EXACT_COUNT 8
  97 struct tap_filter {
  98         unsigned int    count;    /* Number of addrs. Zero means disabled */
  99         u32             mask[2];  /* Mask of the hashed addrs */
 100         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 101 };
 102
 103 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 104  * to max number of VCPUs in guest. */
 105 #define MAX_TAP_QUEUES 256
 106 #define MAX_TAP_FLOWS  4096
 107
 108 #define TUN_FLOW_EXPIRE (3 * HZ)
 109
 110 struct tun_pcpu_stats {
 111         u64_stats_t rx_packets;
 112         u64_stats_t rx_bytes;
 113         u64_stats_t tx_packets;
 114         u64_stats_t tx_bytes;
 115         struct u64_stats_sync syncp;
 116         u32 rx_dropped;
 117         u32 tx_dropped;
 118         u32 rx_frame_errors;
 119 };
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related structures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct tun_struct __rcu *tun;
 136         struct fasync_struct *fasync;
 137         /* only used for fasnyc */
 138         unsigned int flags;
 139         union {
 140                 u16 queue_index;
 141                 unsigned int ifindex;
 142         };
 143         struct napi_struct napi;
 144         bool napi_enabled;
 145         bool napi_frags_enabled;
 146         struct mutex napi_mutex;        /* Protects access to the above napi */
 147         struct list_head next;
 148         struct tun_struct *detached;
 149         struct ptr_ring tx_ring;
 150         struct xdp_rxq_info xdp_rxq;
 151 };
 152
 153 struct tun_page {
 154         struct page *page;
 155         int count;
 156 };
 157
 158 struct tun_flow_entry {
 159         struct hlist_node hash_link;
 160         struct rcu_head rcu;
 161         struct tun_struct *tun;
 162
 163         u32 rxhash;
 164         u32 rps_rxhash;
 165         int queue_index;
 166         unsigned long updated ____cacheline_aligned_in_smp;
 167 };
 168
 169 #define TUN_NUM_FLOW_ENTRIES 1024
 170 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 171
 172 struct tun_prog {
 173         struct rcu_head rcu;
 174         struct bpf_prog *prog;
 175 };
 176
 177 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 178  * device, socket filter, sndbuf and vnet header size were restore when the
 179  * file were attached to a persist device.
 180  */
 181 struct tun_struct {
 182         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 183         unsigned int            numqueues;
 184         unsigned int            flags;
 185         kuid_t                  owner;
 186         kgid_t                  group;
 187
 188         struct net_device       *dev;
 189         netdev_features_t       set_features;
 190 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 191                           NETIF_F_TSO6)
 192
 193         int                     align;
 194         int                     vnet_hdr_sz;
 195         int                     sndbuf;
 196         struct tap_filter       txflt;
 197         struct sock_fprog       fprog;
 198         /* protected by rtnl lock */
 199         bool                    filter_attached;
 200         u32                     msg_enable;
 201         spinlock_t lock;
 202         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 203         struct timer_list flow_gc_timer;
 204         unsigned long ageing_time;
 205         unsigned int numdisabled;
 206         struct list_head disabled;
 207         void *security;
 208         u32 flow_count;
 209         u32 rx_batched;
 210         struct tun_pcpu_stats __percpu *pcpu_stats;
 211         struct bpf_prog __rcu *xdp_prog;
 212         struct tun_prog __rcu *steering_prog;
 213         struct tun_prog __rcu *filter_prog;
 214         struct ethtool_link_ksettings link_ksettings;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static int tun_napi_receive(struct napi_struct *napi, int budget)
 223 {
 224         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 225         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 226         struct sk_buff_head process_queue;
 227         struct sk_buff *skb;
 228         int received = 0;
 229
 230         __skb_queue_head_init(&process_queue);
 231
 232         spin_lock(&queue->lock);
 233         skb_queue_splice_tail_init(queue, &process_queue);
 234         spin_unlock(&queue->lock);
 235
 236         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 237                 napi_gro_receive(napi, skb);
 238                 ++received;
 239         }
 240
 241         if (!skb_queue_empty(&process_queue)) {
 242                 spin_lock(&queue->lock);
 243                 skb_queue_splice(&process_queue, queue);
 244                 spin_unlock(&queue->lock);
 245         }
 246
 247         return received;
 248 }
 249
 250 static int tun_napi_poll(struct napi_struct *napi, int budget)
 251 {
 252         unsigned int received;
 253
 254         received = tun_napi_receive(napi, budget);
 255
 256         if (received < budget)
 257                 napi_complete_done(napi, received);
 258
 259         return received;
 260 }
 261
 262 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 263                           bool napi_en, bool napi_frags)
 264 {
 265         tfile->napi_enabled = napi_en;
 266         tfile->napi_frags_enabled = napi_en && napi_frags;
 267         if (napi_en) {
 268                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 269                                   NAPI_POLL_WEIGHT);
 270                 napi_enable(&tfile->napi);
 271         }
 272 }
 273
 274 static void tun_napi_disable(struct tun_file *tfile)
 275 {
 276         if (tfile->napi_enabled)
 277                 napi_disable(&tfile->napi);
 278 }
 279
 280 static void tun_napi_del(struct tun_file *tfile)
 281 {
 282         if (tfile->napi_enabled)
 283                 netif_napi_del(&tfile->napi);
 284 }
 285
 286 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 287 {
 288         return tfile->napi_frags_enabled;
 289 }
 290
 291 #ifdef CONFIG_TUN_VNET_CROSS_LE
 292 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 293 {
 294         return tun->flags & TUN_VNET_BE ? false :
 295                 virtio_legacy_is_little_endian();
 296 }
 297
 298 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 299 {
 300         int be = !!(tun->flags & TUN_VNET_BE);
 301
 302         if (put_user(be, argp))
 303                 return -EFAULT;
 304
 305         return 0;
 306 }
 307
 308 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 309 {
 310         int be;
 311
 312         if (get_user(be, argp))
 313                 return -EFAULT;
 314
 315         if (be)
 316                 tun->flags |= TUN_VNET_BE;
 317         else
 318                 tun->flags &= ~TUN_VNET_BE;
 319
 320         return 0;
 321 }
 322 #else
 323 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 324 {
 325         return virtio_legacy_is_little_endian();
 326 }
 327
 328 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 329 {
 330         return -EINVAL;
 331 }
 332
 333 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 334 {
 335         return -EINVAL;
 336 }
 337 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 338
 339 static inline bool tun_is_little_endian(struct tun_struct *tun)
 340 {
 341         return tun->flags & TUN_VNET_LE ||
 342                 tun_legacy_is_little_endian(tun);
 343 }
 344
 345 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 346 {
 347         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 348 }
 349
 350 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 351 {
 352         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 353 }
 354
 355 static inline u32 tun_hashfn(u32 rxhash)
 356 {
 357         return rxhash & TUN_MASK_FLOW_ENTRIES;
 358 }
 359
 360 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 361 {
 362         struct tun_flow_entry *e;
 363
 364         hlist_for_each_entry_rcu(e, head, hash_link) {
 365                 if (e->rxhash == rxhash)
 366                         return e;
 367         }
 368         return NULL;
 369 }
 370
 371 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 372                                               struct hlist_head *head,
 373                                               u32 rxhash, u16 queue_index)
 374 {
 375         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 376
 377         if (e) {
 378                 netif_info(tun, tx_queued, tun->dev,
 379                            "create flow: hash %u index %u\n",
 380                            rxhash, queue_index);
 381                 e->updated = jiffies;
 382                 e->rxhash = rxhash;
 383                 e->rps_rxhash = 0;
 384                 e->queue_index = queue_index;
 385                 e->tun = tun;
 386                 hlist_add_head_rcu(&e->hash_link, head);
 387                 ++tun->flow_count;
 388         }
 389         return e;
 390 }
 391
 392 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 393 {
 394         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 395                    e->rxhash, e->queue_index);
 396         hlist_del_rcu(&e->hash_link);
 397         kfree_rcu(e, rcu);
 398         --tun->flow_count;
 399 }
 400
 401 static void tun_flow_flush(struct tun_struct *tun)
 402 {
 403         int i;
 404
 405         spin_lock_bh(&tun->lock);
 406         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 407                 struct tun_flow_entry *e;
 408                 struct hlist_node *n;
 409
 410                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 411                         tun_flow_delete(tun, e);
 412         }
 413         spin_unlock_bh(&tun->lock);
 414 }
 415
 416 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 417 {
 418         int i;
 419
 420         spin_lock_bh(&tun->lock);
 421         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 422                 struct tun_flow_entry *e;
 423                 struct hlist_node *n;
 424
 425                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 426                         if (e->queue_index == queue_index)
 427                                 tun_flow_delete(tun, e);
 428                 }
 429         }
 430         spin_unlock_bh(&tun->lock);
 431 }
 432
 433 static void tun_flow_cleanup(struct timer_list *t)
 434 {
 435         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 436         unsigned long delay = tun->ageing_time;
 437         unsigned long next_timer = jiffies + delay;
 438         unsigned long count = 0;
 439         int i;
 440
 441         spin_lock(&tun->lock);
 442         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 443                 struct tun_flow_entry *e;
 444                 struct hlist_node *n;
 445
 446                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 447                         unsigned long this_timer;
 448
 449                         this_timer = e->updated + delay;
 450                         if (time_before_eq(this_timer, jiffies)) {
 451                                 tun_flow_delete(tun, e);
 452                                 continue;
 453                         }
 454                         count++;
 455                         if (time_before(this_timer, next_timer))
 456                                 next_timer = this_timer;
 457                 }
 458         }
 459
 460         if (count)
 461                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 462         spin_unlock(&tun->lock);
 463 }
 464
 465 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 466                             struct tun_file *tfile)
 467 {
 468         struct hlist_head *head;
 469         struct tun_flow_entry *e;
 470         unsigned long delay = tun->ageing_time;
 471         u16 queue_index = tfile->queue_index;
 472
 473         head = &tun->flows[tun_hashfn(rxhash)];
 474
 475         rcu_read_lock();
 476
 477         e = tun_flow_find(head, rxhash);
 478         if (likely(e)) {
 479                 /* TODO: keep queueing to old queue until it's empty? */
 480                 if (READ_ONCE(e->queue_index) != queue_index)
 481                         WRITE_ONCE(e->queue_index, queue_index);
 482                 if (e->updated != jiffies)
 483                         e->updated = jiffies;
 484                 sock_rps_record_flow_hash(e->rps_rxhash);
 485         } else {
 486                 spin_lock_bh(&tun->lock);
 487                 if (!tun_flow_find(head, rxhash) &&
 488                     tun->flow_count < MAX_TAP_FLOWS)
 489                         tun_flow_create(tun, head, rxhash, queue_index);
 490
 491                 if (!timer_pending(&tun->flow_gc_timer))
 492                         mod_timer(&tun->flow_gc_timer,
 493                                   round_jiffies_up(jiffies + delay));
 494                 spin_unlock_bh(&tun->lock);
 495         }
 496
 497         rcu_read_unlock();
 498 }
 499
 500 /* Save the hash received in the stack receive path and update the
 501  * flow_hash table accordingly.
 502  */
 503 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 504 {
 505         if (unlikely(e->rps_rxhash != hash))
 506                 e->rps_rxhash = hash;
 507 }
 508
 509 /* We try to identify a flow through its rxhash. The reason that
 510  * we do not check rxq no. is because some cards(e.g 82599), chooses
 511  * the rxq based on the txq where the last packet of the flow comes. As
 512  * the userspace application move between processors, we may get a
 513  * different rxq no. here.
 514  */
 515 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 516 {
 517         struct tun_flow_entry *e;
 518         u32 txq = 0;
 519         u32 numqueues = 0;
 520
 521         numqueues = READ_ONCE(tun->numqueues);
 522
 523         txq = __skb_get_hash_symmetric(skb);
 524         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 525         if (e) {
 526                 tun_flow_save_rps_rxhash(e, txq);
 527                 txq = e->queue_index;
 528         } else {
 529                 /* use multiply and shift instead of expensive divide */
 530                 txq = ((u64)txq * numqueues) >> 32;
 531         }
 532
 533         return txq;
 534 }
 535
 536 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 537 {
 538         struct tun_prog *prog;
 539         u32 numqueues;
 540         u16 ret = 0;
 541
 542         numqueues = READ_ONCE(tun->numqueues);
 543         if (!numqueues)
 544                 return 0;
 545
 546         prog = rcu_dereference(tun->steering_prog);
 547         if (prog)
 548                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 549
 550         return ret % numqueues;
 551 }
 552
 553 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 554                             struct net_device *sb_dev)
 555 {
 556         struct tun_struct *tun = netdev_priv(dev);
 557         u16 ret;
 558
 559         rcu_read_lock();
 560         if (rcu_dereference(tun->steering_prog))
 561                 ret = tun_ebpf_select_queue(tun, skb);
 562         else
 563                 ret = tun_automq_select_queue(tun, skb);
 564         rcu_read_unlock();
 565
 566         return ret;
 567 }
 568
 569 static inline bool tun_not_capable(struct tun_struct *tun)
 570 {
 571         const struct cred *cred = current_cred();
 572         struct net *net = dev_net(tun->dev);
 573
 574         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 575                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 576                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 577 }
 578
 579 static void tun_set_real_num_queues(struct tun_struct *tun)
 580 {
 581         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 582         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 583 }
 584
 585 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 586 {
 587         tfile->detached = tun;
 588         list_add_tail(&tfile->next, &tun->disabled);
 589         ++tun->numdisabled;
 590 }
 591
 592 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 593 {
 594         struct tun_struct *tun = tfile->detached;
 595
 596         tfile->detached = NULL;
 597         list_del_init(&tfile->next);
 598         --tun->numdisabled;
 599         return tun;
 600 }
 601
 602 void tun_ptr_free(void *ptr)
 603 {
 604         if (!ptr)
 605                 return;
 606         if (tun_is_xdp_frame(ptr)) {
 607                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 608
 609                 xdp_return_frame(xdpf);
 610         } else {
 611                 __skb_array_destroy_skb(ptr);
 612         }
 613 }
 614 EXPORT_SYMBOL_GPL(tun_ptr_free);
 615
 616 static void tun_queue_purge(struct tun_file *tfile)
 617 {
 618         void *ptr;
 619
 620         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 621                 tun_ptr_free(ptr);
 622
 623         skb_queue_purge(&tfile->sk.sk_write_queue);
 624         skb_queue_purge(&tfile->sk.sk_error_queue);
 625 }
 626
 627 static void __tun_detach(struct tun_file *tfile, bool clean)
 628 {
 629         struct tun_file *ntfile;
 630         struct tun_struct *tun;
 631
 632         tun = rtnl_dereference(tfile->tun);
 633
 634         if (tun && clean) {
 635                 tun_napi_disable(tfile);
 636                 tun_napi_del(tfile);
 637         }
 638
 639         if (tun && !tfile->detached) {
 640                 u16 index = tfile->queue_index;
 641                 BUG_ON(index >= tun->numqueues);
 642
 643                 rcu_assign_pointer(tun->tfiles[index],
 644                                    tun->tfiles[tun->numqueues - 1]);
 645                 ntfile = rtnl_dereference(tun->tfiles[index]);
 646                 ntfile->queue_index = index;
 647                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 648                                    NULL);
 649
 650                 --tun->numqueues;
 651                 if (clean) {
 652                         RCU_INIT_POINTER(tfile->tun, NULL);
 653                         sock_put(&tfile->sk);
 654                 } else
 655                         tun_disable_queue(tun, tfile);
 656
 657                 synchronize_net();
 658                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 659                 /* Drop read queue */
 660                 tun_queue_purge(tfile);
 661                 tun_set_real_num_queues(tun);
 662         } else if (tfile->detached && clean) {
 663                 tun = tun_enable_queue(tfile);
 664                 sock_put(&tfile->sk);
 665         }
 666
 667         if (clean) {
 668                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 669                         netif_carrier_off(tun->dev);
 670
 671                         if (!(tun->flags & IFF_PERSIST) &&
 672                             tun->dev->reg_state == NETREG_REGISTERED)
 673                                 unregister_netdevice(tun->dev);
 674                 }
 675                 if (tun)
 676                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 677                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 678                 sock_put(&tfile->sk);
 679         }
 680 }
 681
 682 static void tun_detach(struct tun_file *tfile, bool clean)
 683 {
 684         struct tun_struct *tun;
 685         struct net_device *dev;
 686
 687         rtnl_lock();
 688         tun = rtnl_dereference(tfile->tun);
 689         dev = tun ? tun->dev : NULL;
 690         __tun_detach(tfile, clean);
 691         if (dev)
 692                 netdev_state_change(dev);
 693         rtnl_unlock();
 694 }
 695
 696 static void tun_detach_all(struct net_device *dev)
 697 {
 698         struct tun_struct *tun = netdev_priv(dev);
 699         struct tun_file *tfile, *tmp;
 700         int i, n = tun->numqueues;
 701
 702         for (i = 0; i < n; i++) {
 703                 tfile = rtnl_dereference(tun->tfiles[i]);
 704                 BUG_ON(!tfile);
 705                 tun_napi_disable(tfile);
 706                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 707                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 708                 RCU_INIT_POINTER(tfile->tun, NULL);
 709                 --tun->numqueues;
 710         }
 711         list_for_each_entry(tfile, &tun->disabled, next) {
 712                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 713                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 714                 RCU_INIT_POINTER(tfile->tun, NULL);
 715         }
 716         BUG_ON(tun->numqueues != 0);
 717
 718         synchronize_net();
 719         for (i = 0; i < n; i++) {
 720                 tfile = rtnl_dereference(tun->tfiles[i]);
 721                 tun_napi_del(tfile);
 722                 /* Drop read queue */
 723                 tun_queue_purge(tfile);
 724                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 725                 sock_put(&tfile->sk);
 726         }
 727         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 728                 tun_enable_queue(tfile);
 729                 tun_queue_purge(tfile);
 730                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 731                 sock_put(&tfile->sk);
 732         }
 733         BUG_ON(tun->numdisabled != 0);
 734
 735         if (tun->flags & IFF_PERSIST)
 736                 module_put(THIS_MODULE);
 737 }
 738
 739 static int tun_attach(struct tun_struct *tun, struct file *file,
 740                       bool skip_filter, bool napi, bool napi_frags,
 741                       bool publish_tun)
 742 {
 743         struct tun_file *tfile = file->private_data;
 744         struct net_device *dev = tun->dev;
 745         int err;
 746
 747         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 748         if (err < 0)
 749                 goto out;
 750
 751         err = -EINVAL;
 752         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 753                 goto out;
 754
 755         err = -EBUSY;
 756         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 757                 goto out;
 758
 759         err = -E2BIG;
 760         if (!tfile->detached &&
 761             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 762                 goto out;
 763
 764         err = 0;
 765
 766         /* Re-attach the filter to persist device */
 767         if (!skip_filter && (tun->filter_attached == true)) {
 768                 lock_sock(tfile->socket.sk);
 769                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 770                 release_sock(tfile->socket.sk);
 771                 if (!err)
 772                         goto out;
 773         }
 774
 775         if (!tfile->detached &&
 776             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 777                             GFP_KERNEL, tun_ptr_free)) {
 778                 err = -ENOMEM;
 779                 goto out;
 780         }
 781
 782         tfile->queue_index = tun->numqueues;
 783         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 784
 785         if (tfile->detached) {
 786                 /* Re-attach detached tfile, updating XDP queue_index */
 787                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 788
 789                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 790                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 791         } else {
 792                 /* Setup XDP RX-queue info, for new tfile getting attached */
 793                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 794                                        tun->dev, tfile->queue_index);
 795                 if (err < 0)
 796                         goto out;
 797                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 798                                                  MEM_TYPE_PAGE_SHARED, NULL);
 799                 if (err < 0) {
 800                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 801                         goto out;
 802                 }
 803                 err = 0;
 804         }
 805
 806         if (tfile->detached) {
 807                 tun_enable_queue(tfile);
 808         } else {
 809                 sock_hold(&tfile->sk);
 810                 tun_napi_init(tun, tfile, napi, napi_frags);
 811         }
 812
 813         if (rtnl_dereference(tun->xdp_prog))
 814                 sock_set_flag(&tfile->sk, SOCK_XDP);
 815
 816         /* device is allowed to go away first, so no need to hold extra
 817          * refcnt.
 818          */
 819
 820         /* Publish tfile->tun and tun->tfiles only after we've fully
 821          * initialized tfile; otherwise we risk using half-initialized
 822          * object.
 823          */
 824         if (publish_tun)
 825                 rcu_assign_pointer(tfile->tun, tun);
 826         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 827         tun->numqueues++;
 828         tun_set_real_num_queues(tun);
 829 out:
 830         return err;
 831 }
 832
 833 static struct tun_struct *tun_get(struct tun_file *tfile)
 834 {
 835         struct tun_struct *tun;
 836
 837         rcu_read_lock();
 838         tun = rcu_dereference(tfile->tun);
 839         if (tun)
 840                 dev_hold(tun->dev);
 841         rcu_read_unlock();
 842
 843         return tun;
 844 }
 845
 846 static void tun_put(struct tun_struct *tun)
 847 {
 848         dev_put(tun->dev);
 849 }
 850
 851 /* TAP filtering */
 852 static void addr_hash_set(u32 *mask, const u8 *addr)
 853 {
 854         int n = ether_crc(ETH_ALEN, addr) >> 26;
 855         mask[n >> 5] |= (1 << (n & 31));
 856 }
 857
 858 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 859 {
 860         int n = ether_crc(ETH_ALEN, addr) >> 26;
 861         return mask[n >> 5] & (1 << (n & 31));
 862 }
 863
 864 static int update_filter(struct tap_filter *filter, void __user *arg)
 865 {
 866         struct { u8 u[ETH_ALEN]; } *addr;
 867         struct tun_filter uf;
 868         int err, alen, n, nexact;
 869
 870         if (copy_from_user(&uf, arg, sizeof(uf)))
 871                 return -EFAULT;
 872
 873         if (!uf.count) {
 874                 /* Disabled */
 875                 filter->count = 0;
 876                 return 0;
 877         }
 878
 879         alen = ETH_ALEN * uf.count;
 880         addr = memdup_user(arg + sizeof(uf), alen);
 881         if (IS_ERR(addr))
 882                 return PTR_ERR(addr);
 883
 884         /* The filter is updated without holding any locks. Which is
 885          * perfectly safe. We disable it first and in the worst
 886          * case we'll accept a few undesired packets. */
 887         filter->count = 0;
 888         wmb();
 889
 890         /* Use first set of addresses as an exact filter */
 891         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 892                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 893
 894         nexact = n;
 895
 896         /* Remaining multicast addresses are hashed,
 897          * unicast will leave the filter disabled. */
 898         memset(filter->mask, 0, sizeof(filter->mask));
 899         for (; n < uf.count; n++) {
 900                 if (!is_multicast_ether_addr(addr[n].u)) {
 901                         err = 0; /* no filter */
 902                         goto free_addr;
 903                 }
 904                 addr_hash_set(filter->mask, addr[n].u);
 905         }
 906
 907         /* For ALLMULTI just set the mask to all ones.
 908          * This overrides the mask populated above. */
 909         if ((uf.flags & TUN_FLT_ALLMULTI))
 910                 memset(filter->mask, ~0, sizeof(filter->mask));
 911
 912         /* Now enable the filter */
 913         wmb();
 914         filter->count = nexact;
 915
 916         /* Return the number of exact filters */
 917         err = nexact;
 918 free_addr:
 919         kfree(addr);
 920         return err;
 921 }
 922
 923 /* Returns: 0 - drop, !=0 - accept */
 924 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 925 {
 926         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 927          * at this point. */
 928         struct ethhdr *eh = (struct ethhdr *) skb->data;
 929         int i;
 930
 931         /* Exact match */
 932         for (i = 0; i < filter->count; i++)
 933                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 934                         return 1;
 935
 936         /* Inexact match (multicast only) */
 937         if (is_multicast_ether_addr(eh->h_dest))
 938                 return addr_hash_test(filter->mask, eh->h_dest);
 939
 940         return 0;
 941 }
 942
 943 /*
 944  * Checks whether the packet is accepted or not.
 945  * Returns: 0 - drop, !=0 - accept
 946  */
 947 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 948 {
 949         if (!filter->count)
 950                 return 1;
 951
 952         return run_filter(filter, skb);
 953 }
 954
 955 /* Network device part of the driver */
 956
 957 static const struct ethtool_ops tun_ethtool_ops;
 958
 959 /* Net device detach from fd. */
 960 static void tun_net_uninit(struct net_device *dev)
 961 {
 962         tun_detach_all(dev);
 963 }
 964
 965 /* Net device open. */
 966 static int tun_net_open(struct net_device *dev)
 967 {
 968         netif_tx_start_all_queues(dev);
 969
 970         return 0;
 971 }
 972
 973 /* Net device close. */
 974 static int tun_net_close(struct net_device *dev)
 975 {
 976         netif_tx_stop_all_queues(dev);
 977         return 0;
 978 }
 979
 980 /* Net device start xmit */
 981 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
 982 {
 983 #ifdef CONFIG_RPS
 984         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
 985                 /* Select queue was not called for the skbuff, so we extract the
 986                  * RPS hash and save it into the flow_table here.
 987                  */
 988                 struct tun_flow_entry *e;
 989                 __u32 rxhash;
 990
 991                 rxhash = __skb_get_hash_symmetric(skb);
 992                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
 993                 if (e)
 994                         tun_flow_save_rps_rxhash(e, rxhash);
 995         }
 996 #endif
 997 }
 998
 999 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1000                                     struct sk_buff *skb,
1001                                     int len)
1002 {
1003         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1004
1005         if (prog)
1006                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1007
1008         return len;
1009 }
1010
1011 /* Net device start xmit */
1012 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1013 {
1014         struct tun_struct *tun = netdev_priv(dev);
1015         int txq = skb->queue_mapping;
1016         struct tun_file *tfile;
1017         int len = skb->len;
1018
1019         rcu_read_lock();
1020         tfile = rcu_dereference(tun->tfiles[txq]);
1021
1022         /* Drop packet if interface is not attached */
1023         if (!tfile)
1024                 goto drop;
1025
1026         if (!rcu_dereference(tun->steering_prog))
1027                 tun_automq_xmit(tun, skb);
1028
1029         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1030
1031         /* Drop if the filter does not like it.
1032          * This is a noop if the filter is disabled.
1033          * Filter can be enabled only for the TAP devices. */
1034         if (!check_filter(&tun->txflt, skb))
1035                 goto drop;
1036
1037         if (tfile->socket.sk->sk_filter &&
1038             sk_filter(tfile->socket.sk, skb))
1039                 goto drop;
1040
1041         len = run_ebpf_filter(tun, skb, len);
1042         if (len == 0 || pskb_trim(skb, len))
1043                 goto drop;
1044
1045         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1046                 goto drop;
1047
1048         skb_tx_timestamp(skb);
1049
1050         /* Orphan the skb - required as we might hang on to it
1051          * for indefinite time.
1052          */
1053         skb_orphan(skb);
1054
1055         nf_reset_ct(skb);
1056
1057         if (ptr_ring_produce(&tfile->tx_ring, skb))
1058                 goto drop;
1059
1060         /* Notify and wake up reader process */
1061         if (tfile->flags & TUN_FASYNC)
1062                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1063         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1064
1065         rcu_read_unlock();
1066         return NETDEV_TX_OK;
1067
1068 drop:
1069         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1070         skb_tx_error(skb);
1071         kfree_skb(skb);
1072         rcu_read_unlock();
1073         return NET_XMIT_DROP;
1074 }
1075
1076 static void tun_net_mclist(struct net_device *dev)
1077 {
1078         /*
1079          * This callback is supposed to deal with mc filter in
1080          * _rx_ path and has nothing to do with the _tx_ path.
1081          * In rx path we always accept everything userspace gives us.
1082          */
1083 }
1084
1085 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1086         netdev_features_t features)
1087 {
1088         struct tun_struct *tun = netdev_priv(dev);
1089
1090         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1091 }
1092
1093 static void tun_set_headroom(struct net_device *dev, int new_hr)
1094 {
1095         struct tun_struct *tun = netdev_priv(dev);
1096
1097         if (new_hr < NET_SKB_PAD)
1098                 new_hr = NET_SKB_PAD;
1099
1100         tun->align = new_hr;
1101 }
1102
1103 static void
1104 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1105 {
1106         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1107         struct tun_struct *tun = netdev_priv(dev);
1108         struct tun_pcpu_stats *p;
1109         int i;
1110
1111         for_each_possible_cpu(i) {
1112                 u64 rxpackets, rxbytes, txpackets, txbytes;
1113                 unsigned int start;
1114
1115                 p = per_cpu_ptr(tun->pcpu_stats, i);
1116                 do {
1117                         start = u64_stats_fetch_begin(&p->syncp);
1118                         rxpackets       = u64_stats_read(&p->rx_packets);
1119                         rxbytes         = u64_stats_read(&p->rx_bytes);
1120                         txpackets       = u64_stats_read(&p->tx_packets);
1121                         txbytes         = u64_stats_read(&p->tx_bytes);
1122                 } while (u64_stats_fetch_retry(&p->syncp, start));
1123
1124                 stats->rx_packets       += rxpackets;
1125                 stats->rx_bytes         += rxbytes;
1126                 stats->tx_packets       += txpackets;
1127                 stats->tx_bytes         += txbytes;
1128
1129                 /* u32 counters */
1130                 rx_dropped      += p->rx_dropped;
1131                 rx_frame_errors += p->rx_frame_errors;
1132                 tx_dropped      += p->tx_dropped;
1133         }
1134         stats->rx_dropped  = rx_dropped;
1135         stats->rx_frame_errors = rx_frame_errors;
1136         stats->tx_dropped = tx_dropped;
1137 }
1138
1139 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1140                        struct netlink_ext_ack *extack)
1141 {
1142         struct tun_struct *tun = netdev_priv(dev);
1143         struct tun_file *tfile;
1144         struct bpf_prog *old_prog;
1145         int i;
1146
1147         old_prog = rtnl_dereference(tun->xdp_prog);
1148         rcu_assign_pointer(tun->xdp_prog, prog);
1149         if (old_prog)
1150                 bpf_prog_put(old_prog);
1151
1152         for (i = 0; i < tun->numqueues; i++) {
1153                 tfile = rtnl_dereference(tun->tfiles[i]);
1154                 if (prog)
1155                         sock_set_flag(&tfile->sk, SOCK_XDP);
1156                 else
1157                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1158         }
1159         list_for_each_entry(tfile, &tun->disabled, next) {
1160                 if (prog)
1161                         sock_set_flag(&tfile->sk, SOCK_XDP);
1162                 else
1163                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1164         }
1165
1166         return 0;
1167 }
1168
1169 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1170 {
1171         switch (xdp->command) {
1172         case XDP_SETUP_PROG:
1173                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1174         default:
1175                 return -EINVAL;
1176         }
1177 }
1178
1179 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1180 {
1181         if (new_carrier) {
1182                 struct tun_struct *tun = netdev_priv(dev);
1183
1184                 if (!tun->numqueues)
1185                         return -EPERM;
1186
1187                 netif_carrier_on(dev);
1188         } else {
1189                 netif_carrier_off(dev);
1190         }
1191         return 0;
1192 }
1193
1194 static const struct net_device_ops tun_netdev_ops = {
1195         .ndo_uninit             = tun_net_uninit,
1196         .ndo_open               = tun_net_open,
1197         .ndo_stop               = tun_net_close,
1198         .ndo_start_xmit         = tun_net_xmit,
1199         .ndo_fix_features       = tun_net_fix_features,
1200         .ndo_select_queue       = tun_select_queue,
1201         .ndo_set_rx_headroom    = tun_set_headroom,
1202         .ndo_get_stats64        = tun_net_get_stats64,
1203         .ndo_change_carrier     = tun_net_change_carrier,
1204 };
1205
1206 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1207 {
1208         /* Notify and wake up reader process */
1209         if (tfile->flags & TUN_FASYNC)
1210                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1211         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1212 }
1213
1214 static int tun_xdp_xmit(struct net_device *dev, int n,
1215                         struct xdp_frame **frames, u32 flags)
1216 {
1217         struct tun_struct *tun = netdev_priv(dev);
1218         struct tun_file *tfile;
1219         u32 numqueues;
1220         int drops = 0;
1221         int cnt = n;
1222         int i;
1223
1224         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1225                 return -EINVAL;
1226
1227         rcu_read_lock();
1228
1229 resample:
1230         numqueues = READ_ONCE(tun->numqueues);
1231         if (!numqueues) {
1232                 rcu_read_unlock();
1233                 return -ENXIO; /* Caller will free/return all frames */
1234         }
1235
1236         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1237                                             numqueues]);
1238         if (unlikely(!tfile))
1239                 goto resample;
1240
1241         spin_lock(&tfile->tx_ring.producer_lock);
1242         for (i = 0; i < n; i++) {
1243                 struct xdp_frame *xdp = frames[i];
1244                 /* Encode the XDP flag into lowest bit for consumer to differ
1245                  * XDP buffer from sk_buff.
1246                  */
1247                 void *frame = tun_xdp_to_ptr(xdp);
1248
1249                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1250                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1251                         xdp_return_frame_rx_napi(xdp);
1252                         drops++;
1253                 }
1254         }
1255         spin_unlock(&tfile->tx_ring.producer_lock);
1256
1257         if (flags & XDP_XMIT_FLUSH)
1258                 __tun_xdp_flush_tfile(tfile);
1259
1260         rcu_read_unlock();
1261         return cnt - drops;
1262 }
1263
1264 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1265 {
1266         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1267
1268         if (unlikely(!frame))
1269                 return -EOVERFLOW;
1270
1271         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1272 }
1273
1274 static const struct net_device_ops tap_netdev_ops = {
1275         .ndo_uninit             = tun_net_uninit,
1276         .ndo_open               = tun_net_open,
1277         .ndo_stop               = tun_net_close,
1278         .ndo_start_xmit         = tun_net_xmit,
1279         .ndo_fix_features       = tun_net_fix_features,
1280         .ndo_set_rx_mode        = tun_net_mclist,
1281         .ndo_set_mac_address    = eth_mac_addr,
1282         .ndo_validate_addr      = eth_validate_addr,
1283         .ndo_select_queue       = tun_select_queue,
1284         .ndo_features_check     = passthru_features_check,
1285         .ndo_set_rx_headroom    = tun_set_headroom,
1286         .ndo_get_stats64        = tun_net_get_stats64,
1287         .ndo_bpf                = tun_xdp,
1288         .ndo_xdp_xmit           = tun_xdp_xmit,
1289         .ndo_change_carrier     = tun_net_change_carrier,
1290 };
1291
1292 static void tun_flow_init(struct tun_struct *tun)
1293 {
1294         int i;
1295
1296         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1297                 INIT_HLIST_HEAD(&tun->flows[i]);
1298
1299         tun->ageing_time = TUN_FLOW_EXPIRE;
1300         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1301         mod_timer(&tun->flow_gc_timer,
1302                   round_jiffies_up(jiffies + tun->ageing_time));
1303 }
1304
1305 static void tun_flow_uninit(struct tun_struct *tun)
1306 {
1307         del_timer_sync(&tun->flow_gc_timer);
1308         tun_flow_flush(tun);
1309 }
1310
1311 #define MIN_MTU 68
1312 #define MAX_MTU 65535
1313
1314 /* Initialize net device. */
1315 static void tun_net_init(struct net_device *dev)
1316 {
1317         struct tun_struct *tun = netdev_priv(dev);
1318
1319         switch (tun->flags & TUN_TYPE_MASK) {
1320         case IFF_TUN:
1321                 dev->netdev_ops = &tun_netdev_ops;
1322                 dev->header_ops = &ip_tunnel_header_ops;
1323
1324                 /* Point-to-Point TUN Device */
1325                 dev->hard_header_len = 0;
1326                 dev->addr_len = 0;
1327                 dev->mtu = 1500;
1328
1329                 /* Zero header length */
1330                 dev->type = ARPHRD_NONE;
1331                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1332                 break;
1333
1334         case IFF_TAP:
1335                 dev->netdev_ops = &tap_netdev_ops;
1336                 /* Ethernet TAP Device */
1337                 ether_setup(dev);
1338                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1339                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1340
1341                 eth_hw_addr_random(dev);
1342
1343                 break;
1344         }
1345
1346         dev->min_mtu = MIN_MTU;
1347         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1348 }
1349
1350 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1351 {
1352         struct sock *sk = tfile->socket.sk;
1353
1354         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1355 }
1356
1357 /* Character device part */
1358
1359 /* Poll */
1360 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1361 {
1362         struct tun_file *tfile = file->private_data;
1363         struct tun_struct *tun = tun_get(tfile);
1364         struct sock *sk;
1365         __poll_t mask = 0;
1366
1367         if (!tun)
1368                 return EPOLLERR;
1369
1370         sk = tfile->socket.sk;
1371
1372         poll_wait(file, sk_sleep(sk), wait);
1373
1374         if (!ptr_ring_empty(&tfile->tx_ring))
1375                 mask |= EPOLLIN | EPOLLRDNORM;
1376
1377         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1378          * guarantee EPOLLOUT to be raised by either here or
1379          * tun_sock_write_space(). Then process could get notification
1380          * after it writes to a down device and meets -EIO.
1381          */
1382         if (tun_sock_writeable(tun, tfile) ||
1383             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1384              tun_sock_writeable(tun, tfile)))
1385                 mask |= EPOLLOUT | EPOLLWRNORM;
1386
1387         if (tun->dev->reg_state != NETREG_REGISTERED)
1388                 mask = EPOLLERR;
1389
1390         tun_put(tun);
1391         return mask;
1392 }
1393
1394 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1395                                             size_t len,
1396                                             const struct iov_iter *it)
1397 {
1398         struct sk_buff *skb;
1399         size_t linear;
1400         int err;
1401         int i;
1402
1403         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1404                 return ERR_PTR(-ENOMEM);
1405
1406         local_bh_disable();
1407         skb = napi_get_frags(&tfile->napi);
1408         local_bh_enable();
1409         if (!skb)
1410                 return ERR_PTR(-ENOMEM);
1411
1412         linear = iov_iter_single_seg_count(it);
1413         err = __skb_grow(skb, linear);
1414         if (err)
1415                 goto free;
1416
1417         skb->len = len;
1418         skb->data_len = len - linear;
1419         skb->truesize += skb->data_len;
1420
1421         for (i = 1; i < it->nr_segs; i++) {
1422                 size_t fragsz = it->iov[i].iov_len;
1423                 struct page *page;
1424                 void *frag;
1425
1426                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1427                         err = -EINVAL;
1428                         goto free;
1429                 }
1430                 frag = netdev_alloc_frag(fragsz);
1431                 if (!frag) {
1432                         err = -ENOMEM;
1433                         goto free;
1434                 }
1435                 page = virt_to_head_page(frag);
1436                 skb_fill_page_desc(skb, i - 1, page,
1437                                    frag - page_address(page), fragsz);
1438         }
1439
1440         return skb;
1441 free:
1442         /* frees skb and all frags allocated with napi_alloc_frag() */
1443         napi_free_frags(&tfile->napi);
1444         return ERR_PTR(err);
1445 }
1446
1447 /* prepad is the amount to reserve at front.  len is length after that.
1448  * linear is a hint as to how much to copy (usually headers). */
1449 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1450                                      size_t prepad, size_t len,
1451                                      size_t linear, int noblock)
1452 {
1453         struct sock *sk = tfile->socket.sk;
1454         struct sk_buff *skb;
1455         int err;
1456
1457         /* Under a page?  Don't bother with paged skb. */
1458         if (prepad + len < PAGE_SIZE || !linear)
1459                 linear = len;
1460
1461         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1462                                    &err, 0);
1463         if (!skb)
1464                 return ERR_PTR(err);
1465
1466         skb_reserve(skb, prepad);
1467         skb_put(skb, linear);
1468         skb->data_len = len - linear;
1469         skb->len += len - linear;
1470
1471         return skb;
1472 }
1473
1474 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1475                            struct sk_buff *skb, int more)
1476 {
1477         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1478         struct sk_buff_head process_queue;
1479         u32 rx_batched = tun->rx_batched;
1480         bool rcv = false;
1481
1482         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1483                 local_bh_disable();
1484                 skb_record_rx_queue(skb, tfile->queue_index);
1485                 netif_receive_skb(skb);
1486                 local_bh_enable();
1487                 return;
1488         }
1489
1490         spin_lock(&queue->lock);
1491         if (!more || skb_queue_len(queue) == rx_batched) {
1492                 __skb_queue_head_init(&process_queue);
1493                 skb_queue_splice_tail_init(queue, &process_queue);
1494                 rcv = true;
1495         } else {
1496                 __skb_queue_tail(queue, skb);
1497         }
1498         spin_unlock(&queue->lock);
1499
1500         if (rcv) {
1501                 struct sk_buff *nskb;
1502
1503                 local_bh_disable();
1504                 while ((nskb = __skb_dequeue(&process_queue))) {
1505                         skb_record_rx_queue(nskb, tfile->queue_index);
1506                         netif_receive_skb(nskb);
1507                 }
1508                 skb_record_rx_queue(skb, tfile->queue_index);
1509                 netif_receive_skb(skb);
1510                 local_bh_enable();
1511         }
1512 }
1513
1514 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1515                               int len, int noblock, bool zerocopy)
1516 {
1517         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1518                 return false;
1519
1520         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1521                 return false;
1522
1523         if (!noblock)
1524                 return false;
1525
1526         if (zerocopy)
1527                 return false;
1528
1529         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1530             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1531                 return false;
1532
1533         return true;
1534 }
1535
1536 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1537                                        struct page_frag *alloc_frag, char *buf,
1538                                        int buflen, int len, int pad)
1539 {
1540         struct sk_buff *skb = build_skb(buf, buflen);
1541
1542         if (!skb)
1543                 return ERR_PTR(-ENOMEM);
1544
1545         skb_reserve(skb, pad);
1546         skb_put(skb, len);
1547         skb_set_owner_w(skb, tfile->socket.sk);
1548
1549         get_page(alloc_frag->page);
1550         alloc_frag->offset += buflen;
1551
1552         return skb;
1553 }
1554
1555 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1556                        struct xdp_buff *xdp, u32 act)
1557 {
1558         int err;
1559
1560         switch (act) {
1561         case XDP_REDIRECT:
1562                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1563                 if (err)
1564                         return err;
1565                 break;
1566         case XDP_TX:
1567                 err = tun_xdp_tx(tun->dev, xdp);
1568                 if (err < 0)
1569                         return err;
1570                 break;
1571         case XDP_PASS:
1572                 break;
1573         default:
1574                 bpf_warn_invalid_xdp_action(act);
1575                 fallthrough;
1576         case XDP_ABORTED:
1577                 trace_xdp_exception(tun->dev, xdp_prog, act);
1578                 fallthrough;
1579         case XDP_DROP:
1580                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1581                 break;
1582         }
1583
1584         return act;
1585 }
1586
1587 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1588                                      struct tun_file *tfile,
1589                                      struct iov_iter *from,
1590                                      struct virtio_net_hdr *hdr,
1591                                      int len, int *skb_xdp)
1592 {
1593         struct page_frag *alloc_frag = &current->task_frag;
1594         struct bpf_prog *xdp_prog;
1595         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1596         char *buf;
1597         size_t copied;
1598         int pad = TUN_RX_PAD;
1599         int err = 0;
1600
1601         rcu_read_lock();
1602         xdp_prog = rcu_dereference(tun->xdp_prog);
1603         if (xdp_prog)
1604                 pad += XDP_PACKET_HEADROOM;
1605         buflen += SKB_DATA_ALIGN(len + pad);
1606         rcu_read_unlock();
1607
1608         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1609         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1610                 return ERR_PTR(-ENOMEM);
1611
1612         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1613         copied = copy_page_from_iter(alloc_frag->page,
1614                                      alloc_frag->offset + pad,
1615                                      len, from);
1616         if (copied != len)
1617                 return ERR_PTR(-EFAULT);
1618
1619         /* There's a small window that XDP may be set after the check
1620          * of xdp_prog above, this should be rare and for simplicity
1621          * we do XDP on skb in case the headroom is not enough.
1622          */
1623         if (hdr->gso_type || !xdp_prog) {
1624                 *skb_xdp = 1;
1625                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1626                                        pad);
1627         }
1628
1629         *skb_xdp = 0;
1630
1631         local_bh_disable();
1632         rcu_read_lock();
1633         xdp_prog = rcu_dereference(tun->xdp_prog);
1634         if (xdp_prog) {
1635                 struct xdp_buff xdp;
1636                 u32 act;
1637
1638                 xdp.data_hard_start = buf;
1639                 xdp.data = buf + pad;
1640                 xdp_set_data_meta_invalid(&xdp);
1641                 xdp.data_end = xdp.data + len;
1642                 xdp.rxq = &tfile->xdp_rxq;
1643                 xdp.frame_sz = buflen;
1644
1645                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1646                 if (act == XDP_REDIRECT || act == XDP_TX) {
1647                         get_page(alloc_frag->page);
1648                         alloc_frag->offset += buflen;
1649                 }
1650                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1651                 if (err < 0) {
1652                         if (act == XDP_REDIRECT || act == XDP_TX)
1653                                 put_page(alloc_frag->page);
1654                         goto out;
1655                 }
1656
1657                 if (err == XDP_REDIRECT)
1658                         xdp_do_flush();
1659                 if (err != XDP_PASS)
1660                         goto out;
1661
1662                 pad = xdp.data - xdp.data_hard_start;
1663                 len = xdp.data_end - xdp.data;
1664         }
1665         rcu_read_unlock();
1666         local_bh_enable();
1667
1668         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1669
1670 out:
1671         rcu_read_unlock();
1672         local_bh_enable();
1673         return NULL;
1674 }
1675
1676 /* Get packet from user space buffer */
1677 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1678                             void *msg_control, struct iov_iter *from,
1679                             int noblock, bool more)
1680 {
1681         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1682         struct sk_buff *skb;
1683         size_t total_len = iov_iter_count(from);
1684         size_t len = total_len, align = tun->align, linear;
1685         struct virtio_net_hdr gso = { 0 };
1686         struct tun_pcpu_stats *stats;
1687         int good_linear;
1688         int copylen;
1689         bool zerocopy = false;
1690         int err;
1691         u32 rxhash = 0;
1692         int skb_xdp = 1;
1693         bool frags = tun_napi_frags_enabled(tfile);
1694
1695         if (!(tun->flags & IFF_NO_PI)) {
1696                 if (len < sizeof(pi))
1697                         return -EINVAL;
1698                 len -= sizeof(pi);
1699
1700                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1701                         return -EFAULT;
1702         }
1703
1704         if (tun->flags & IFF_VNET_HDR) {
1705                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1706
1707                 if (len < vnet_hdr_sz)
1708                         return -EINVAL;
1709                 len -= vnet_hdr_sz;
1710
1711                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1712                         return -EFAULT;
1713
1714                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1715                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1716                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1717
1718                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1719                         return -EINVAL;
1720                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1721         }
1722
1723         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1724                 align += NET_IP_ALIGN;
1725                 if (unlikely(len < ETH_HLEN ||
1726                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1727                         return -EINVAL;
1728         }
1729
1730         good_linear = SKB_MAX_HEAD(align);
1731
1732         if (msg_control) {
1733                 struct iov_iter i = *from;
1734
1735                 /* There are 256 bytes to be copied in skb, so there is
1736                  * enough room for skb expand head in case it is used.
1737                  * The rest of the buffer is mapped from userspace.
1738                  */
1739                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1740                 if (copylen > good_linear)
1741                         copylen = good_linear;
1742                 linear = copylen;
1743                 iov_iter_advance(&i, copylen);
1744                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1745                         zerocopy = true;
1746         }
1747
1748         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1749                 /* For the packet that is not easy to be processed
1750                  * (e.g gso or jumbo packet), we will do it at after
1751                  * skb was created with generic XDP routine.
1752                  */
1753                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1754                 if (IS_ERR(skb)) {
1755                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1756                         return PTR_ERR(skb);
1757                 }
1758                 if (!skb)
1759                         return total_len;
1760         } else {
1761                 if (!zerocopy) {
1762                         copylen = len;
1763                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1764                                 linear = good_linear;
1765                         else
1766                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1767                 }
1768
1769                 if (frags) {
1770                         mutex_lock(&tfile->napi_mutex);
1771                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1772                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1773                          * If zerocopy is enabled, then this layout will be
1774                          * overwritten by zerocopy_sg_from_iter().
1775                          */
1776                         zerocopy = false;
1777                 } else {
1778                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1779                                             noblock);
1780                 }
1781
1782                 if (IS_ERR(skb)) {
1783                         if (PTR_ERR(skb) != -EAGAIN)
1784                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1785                         if (frags)
1786                                 mutex_unlock(&tfile->napi_mutex);
1787                         return PTR_ERR(skb);
1788                 }
1789
1790                 if (zerocopy)
1791                         err = zerocopy_sg_from_iter(skb, from);
1792                 else
1793                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1794
1795                 if (err) {
1796                         err = -EFAULT;
1797 drop:
1798                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1799                         kfree_skb(skb);
1800                         if (frags) {
1801                                 tfile->napi.skb = NULL;
1802                                 mutex_unlock(&tfile->napi_mutex);
1803                         }
1804
1805                         return err;
1806                 }
1807         }
1808
1809         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1810                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1811                 kfree_skb(skb);
1812                 if (frags) {
1813                         tfile->napi.skb = NULL;
1814                         mutex_unlock(&tfile->napi_mutex);
1815                 }
1816
1817                 return -EINVAL;
1818         }
1819
1820         switch (tun->flags & TUN_TYPE_MASK) {
1821         case IFF_TUN:
1822                 if (tun->flags & IFF_NO_PI) {
1823                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1824
1825                         switch (ip_version) {
1826                         case 4:
1827                                 pi.proto = htons(ETH_P_IP);
1828                                 break;
1829                         case 6:
1830                                 pi.proto = htons(ETH_P_IPV6);
1831                                 break;
1832                         default:
1833                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1834                                 kfree_skb(skb);
1835                                 return -EINVAL;
1836                         }
1837                 }
1838
1839                 skb_reset_mac_header(skb);
1840                 skb->protocol = pi.proto;
1841                 skb->dev = tun->dev;
1842                 break;
1843         case IFF_TAP:
1844                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1845                         err = -ENOMEM;
1846                         goto drop;
1847                 }
1848                 skb->protocol = eth_type_trans(skb, tun->dev);
1849                 break;
1850         }
1851
1852         /* copy skb_ubuf_info for callback when skb has no error */
1853         if (zerocopy) {
1854                 skb_shinfo(skb)->destructor_arg = msg_control;
1855                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1856                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1857         } else if (msg_control) {
1858                 struct ubuf_info *uarg = msg_control;
1859                 uarg->callback(uarg, false);
1860         }
1861
1862         skb_reset_network_header(skb);
1863         skb_probe_transport_header(skb);
1864         skb_record_rx_queue(skb, tfile->queue_index);
1865
1866         if (skb_xdp) {
1867                 struct bpf_prog *xdp_prog;
1868                 int ret;
1869
1870                 local_bh_disable();
1871                 rcu_read_lock();
1872                 xdp_prog = rcu_dereference(tun->xdp_prog);
1873                 if (xdp_prog) {
1874                         ret = do_xdp_generic(xdp_prog, skb);
1875                         if (ret != XDP_PASS) {
1876                                 rcu_read_unlock();
1877                                 local_bh_enable();
1878                                 if (frags) {
1879                                         tfile->napi.skb = NULL;
1880                                         mutex_unlock(&tfile->napi_mutex);
1881                                 }
1882                                 return total_len;
1883                         }
1884                 }
1885                 rcu_read_unlock();
1886                 local_bh_enable();
1887         }
1888
1889         /* Compute the costly rx hash only if needed for flow updates.
1890          * We may get a very small possibility of OOO during switching, not
1891          * worth to optimize.
1892          */
1893         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1894             !tfile->detached)
1895                 rxhash = __skb_get_hash_symmetric(skb);
1896
1897         rcu_read_lock();
1898         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1899                 err = -EIO;
1900                 rcu_read_unlock();
1901                 goto drop;
1902         }
1903
1904         if (frags) {
1905                 u32 headlen;
1906
1907                 /* Exercise flow dissector code path. */
1908                 skb_push(skb, ETH_HLEN);
1909                 headlen = eth_get_headlen(tun->dev, skb->data,
1910                                           skb_headlen(skb));
1911
1912                 if (unlikely(headlen > skb_headlen(skb))) {
1913                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1914                         napi_free_frags(&tfile->napi);
1915                         rcu_read_unlock();
1916                         mutex_unlock(&tfile->napi_mutex);
1917                         WARN_ON(1);
1918                         return -ENOMEM;
1919                 }
1920
1921                 local_bh_disable();
1922                 napi_gro_frags(&tfile->napi);
1923                 local_bh_enable();
1924                 mutex_unlock(&tfile->napi_mutex);
1925         } else if (tfile->napi_enabled) {
1926                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1927                 int queue_len;
1928
1929                 spin_lock_bh(&queue->lock);
1930                 __skb_queue_tail(queue, skb);
1931                 queue_len = skb_queue_len(queue);
1932                 spin_unlock(&queue->lock);
1933
1934                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1935                         napi_schedule(&tfile->napi);
1936
1937                 local_bh_enable();
1938         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1939                 tun_rx_batched(tun, tfile, skb, more);
1940         } else {
1941                 netif_rx_ni(skb);
1942         }
1943         rcu_read_unlock();
1944
1945         stats = get_cpu_ptr(tun->pcpu_stats);
1946         u64_stats_update_begin(&stats->syncp);
1947         u64_stats_inc(&stats->rx_packets);
1948         u64_stats_add(&stats->rx_bytes, len);
1949         u64_stats_update_end(&stats->syncp);
1950         put_cpu_ptr(stats);
1951
1952         if (rxhash)
1953                 tun_flow_update(tun, rxhash, tfile);
1954
1955         return total_len;
1956 }
1957
1958 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1959 {
1960         struct file *file = iocb->ki_filp;
1961         struct tun_file *tfile = file->private_data;
1962         struct tun_struct *tun = tun_get(tfile);
1963         ssize_t result;
1964
1965         if (!tun)
1966                 return -EBADFD;
1967
1968         result = tun_get_user(tun, tfile, NULL, from,
1969                               file->f_flags & O_NONBLOCK, false);
1970
1971         tun_put(tun);
1972         return result;
1973 }
1974
1975 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1976                                 struct tun_file *tfile,
1977                                 struct xdp_frame *xdp_frame,
1978                                 struct iov_iter *iter)
1979 {
1980         int vnet_hdr_sz = 0;
1981         size_t size = xdp_frame->len;
1982         struct tun_pcpu_stats *stats;
1983         size_t ret;
1984
1985         if (tun->flags & IFF_VNET_HDR) {
1986                 struct virtio_net_hdr gso = { 0 };
1987
1988                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1989                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1990                         return -EINVAL;
1991                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1992                              sizeof(gso)))
1993                         return -EFAULT;
1994                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1995         }
1996
1997         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1998
1999         stats = get_cpu_ptr(tun->pcpu_stats);
2000         u64_stats_update_begin(&stats->syncp);
2001         u64_stats_inc(&stats->tx_packets);
2002         u64_stats_add(&stats->tx_bytes, ret);
2003         u64_stats_update_end(&stats->syncp);
2004         put_cpu_ptr(tun->pcpu_stats);
2005
2006         return ret;
2007 }
2008
2009 /* Put packet to the user space buffer */
2010 static ssize_t tun_put_user(struct tun_struct *tun,
2011                             struct tun_file *tfile,
2012                             struct sk_buff *skb,
2013                             struct iov_iter *iter)
2014 {
2015         struct tun_pi pi = { 0, skb->protocol };
2016         struct tun_pcpu_stats *stats;
2017         ssize_t total;
2018         int vlan_offset = 0;
2019         int vlan_hlen = 0;
2020         int vnet_hdr_sz = 0;
2021
2022         if (skb_vlan_tag_present(skb))
2023                 vlan_hlen = VLAN_HLEN;
2024
2025         if (tun->flags & IFF_VNET_HDR)
2026                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2027
2028         total = skb->len + vlan_hlen + vnet_hdr_sz;
2029
2030         if (!(tun->flags & IFF_NO_PI)) {
2031                 if (iov_iter_count(iter) < sizeof(pi))
2032                         return -EINVAL;
2033
2034                 total += sizeof(pi);
2035                 if (iov_iter_count(iter) < total) {
2036                         /* Packet will be striped */
2037                         pi.flags |= TUN_PKT_STRIP;
2038                 }
2039
2040                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2041                         return -EFAULT;
2042         }
2043
2044         if (vnet_hdr_sz) {
2045                 struct virtio_net_hdr gso;
2046
2047                 if (iov_iter_count(iter) < vnet_hdr_sz)
2048                         return -EINVAL;
2049
2050                 if (virtio_net_hdr_from_skb(skb, &gso,
2051                                             tun_is_little_endian(tun), true,
2052                                             vlan_hlen)) {
2053                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2054                         pr_err("unexpected GSO type: "
2055                                "0x%x, gso_size %d, hdr_len %d\n",
2056                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2057                                tun16_to_cpu(tun, gso.hdr_len));
2058                         print_hex_dump(KERN_ERR, "tun: ",
2059                                        DUMP_PREFIX_NONE,
2060                                        16, 1, skb->head,
2061                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2062                         WARN_ON_ONCE(1);
2063                         return -EINVAL;
2064                 }
2065
2066                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2067                         return -EFAULT;
2068
2069                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2070         }
2071
2072         if (vlan_hlen) {
2073                 int ret;
2074                 struct veth veth;
2075
2076                 veth.h_vlan_proto = skb->vlan_proto;
2077                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2078
2079                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2080
2081                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2082                 if (ret || !iov_iter_count(iter))
2083                         goto done;
2084
2085                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2086                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2087                         goto done;
2088         }
2089
2090         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2091
2092 done:
2093         /* caller is in process context, */
2094         stats = get_cpu_ptr(tun->pcpu_stats);
2095         u64_stats_update_begin(&stats->syncp);
2096         u64_stats_inc(&stats->tx_packets);
2097         u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2098         u64_stats_update_end(&stats->syncp);
2099         put_cpu_ptr(tun->pcpu_stats);
2100
2101         return total;
2102 }
2103
2104 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2105 {
2106         DECLARE_WAITQUEUE(wait, current);
2107         void *ptr = NULL;
2108         int error = 0;
2109
2110         ptr = ptr_ring_consume(&tfile->tx_ring);
2111         if (ptr)
2112                 goto out;
2113         if (noblock) {
2114                 error = -EAGAIN;
2115                 goto out;
2116         }
2117
2118         add_wait_queue(&tfile->socket.wq.wait, &wait);
2119
2120         while (1) {
2121                 set_current_state(TASK_INTERRUPTIBLE);
2122                 ptr = ptr_ring_consume(&tfile->tx_ring);
2123                 if (ptr)
2124                         break;
2125                 if (signal_pending(current)) {
2126                         error = -ERESTARTSYS;
2127                         break;
2128                 }
2129                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2130                         error = -EFAULT;
2131                         break;
2132                 }
2133
2134                 schedule();
2135         }
2136
2137         __set_current_state(TASK_RUNNING);
2138         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2139
2140 out:
2141         *err = error;
2142         return ptr;
2143 }
2144
2145 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2146                            struct iov_iter *to,
2147                            int noblock, void *ptr)
2148 {
2149         ssize_t ret;
2150         int err;
2151
2152         if (!iov_iter_count(to)) {
2153                 tun_ptr_free(ptr);
2154                 return 0;
2155         }
2156
2157         if (!ptr) {
2158                 /* Read frames from ring */
2159                 ptr = tun_ring_recv(tfile, noblock, &err);
2160                 if (!ptr)
2161                         return err;
2162         }
2163
2164         if (tun_is_xdp_frame(ptr)) {
2165                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2166
2167                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2168                 xdp_return_frame(xdpf);
2169         } else {
2170                 struct sk_buff *skb = ptr;
2171
2172                 ret = tun_put_user(tun, tfile, skb, to);
2173                 if (unlikely(ret < 0))
2174                         kfree_skb(skb);
2175                 else
2176                         consume_skb(skb);
2177         }
2178
2179         return ret;
2180 }
2181
2182 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2183 {
2184         struct file *file = iocb->ki_filp;
2185         struct tun_file *tfile = file->private_data;
2186         struct tun_struct *tun = tun_get(tfile);
2187         ssize_t len = iov_iter_count(to), ret;
2188
2189         if (!tun)
2190                 return -EBADFD;
2191         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2192         ret = min_t(ssize_t, ret, len);
2193         if (ret > 0)
2194                 iocb->ki_pos = ret;
2195         tun_put(tun);
2196         return ret;
2197 }
2198
2199 static void tun_prog_free(struct rcu_head *rcu)
2200 {
2201         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2202
2203         bpf_prog_destroy(prog->prog);
2204         kfree(prog);
2205 }
2206
2207 static int __tun_set_ebpf(struct tun_struct *tun,
2208                           struct tun_prog __rcu **prog_p,
2209                           struct bpf_prog *prog)
2210 {
2211         struct tun_prog *old, *new = NULL;
2212
2213         if (prog) {
2214                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2215                 if (!new)
2216                         return -ENOMEM;
2217                 new->prog = prog;
2218         }
2219
2220         spin_lock_bh(&tun->lock);
2221         old = rcu_dereference_protected(*prog_p,
2222                                         lockdep_is_held(&tun->lock));
2223         rcu_assign_pointer(*prog_p, new);
2224         spin_unlock_bh(&tun->lock);
2225
2226         if (old)
2227                 call_rcu(&old->rcu, tun_prog_free);
2228
2229         return 0;
2230 }
2231
2232 static void tun_free_netdev(struct net_device *dev)
2233 {
2234         struct tun_struct *tun = netdev_priv(dev);
2235
2236         BUG_ON(!(list_empty(&tun->disabled)));
2237
2238         free_percpu(tun->pcpu_stats);
2239         /* We clear pcpu_stats so that tun_set_iff() can tell if
2240          * tun_free_netdev() has been called from register_netdevice().
2241          */
2242         tun->pcpu_stats = NULL;
2243
2244         tun_flow_uninit(tun);
2245         security_tun_dev_free_security(tun->security);
2246         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2247         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2248 }
2249
2250 static void tun_setup(struct net_device *dev)
2251 {
2252         struct tun_struct *tun = netdev_priv(dev);
2253
2254         tun->owner = INVALID_UID;
2255         tun->group = INVALID_GID;
2256         tun_default_link_ksettings(dev, &tun->link_ksettings);
2257
2258         dev->ethtool_ops = &tun_ethtool_ops;
2259         dev->needs_free_netdev = true;
2260         dev->priv_destructor = tun_free_netdev;
2261         /* We prefer our own queue length */
2262         dev->tx_queue_len = TUN_READQ_SIZE;
2263 }
2264
2265 /* Trivial set of netlink ops to allow deleting tun or tap
2266  * device with netlink.
2267  */
2268 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2269                         struct netlink_ext_ack *extack)
2270 {
2271         NL_SET_ERR_MSG(extack,
2272                        "tun/tap creation via rtnetlink is not supported.");
2273         return -EOPNOTSUPP;
2274 }
2275
2276 static size_t tun_get_size(const struct net_device *dev)
2277 {
2278         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2279         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2280
2281         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2282                nla_total_size(sizeof(gid_t)) + /* GROUP */
2283                nla_total_size(sizeof(u8)) + /* TYPE */
2284                nla_total_size(sizeof(u8)) + /* PI */
2285                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2286                nla_total_size(sizeof(u8)) + /* PERSIST */
2287                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2288                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2289                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2290                0;
2291 }
2292
2293 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2294 {
2295         struct tun_struct *tun = netdev_priv(dev);
2296
2297         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2298                 goto nla_put_failure;
2299         if (uid_valid(tun->owner) &&
2300             nla_put_u32(skb, IFLA_TUN_OWNER,
2301                         from_kuid_munged(current_user_ns(), tun->owner)))
2302                 goto nla_put_failure;
2303         if (gid_valid(tun->group) &&
2304             nla_put_u32(skb, IFLA_TUN_GROUP,
2305                         from_kgid_munged(current_user_ns(), tun->group)))
2306                 goto nla_put_failure;
2307         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2308                 goto nla_put_failure;
2309         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2310                 goto nla_put_failure;
2311         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2312                 goto nla_put_failure;
2313         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2314                        !!(tun->flags & IFF_MULTI_QUEUE)))
2315                 goto nla_put_failure;
2316         if (tun->flags & IFF_MULTI_QUEUE) {
2317                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2318                         goto nla_put_failure;
2319                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2320                                 tun->numdisabled))
2321                         goto nla_put_failure;
2322         }
2323
2324         return 0;
2325
2326 nla_put_failure:
2327         return -EMSGSIZE;
2328 }
2329
2330 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2331         .kind           = DRV_NAME,
2332         .priv_size      = sizeof(struct tun_struct),
2333         .setup          = tun_setup,
2334         .validate       = tun_validate,
2335         .get_size       = tun_get_size,
2336         .fill_info      = tun_fill_info,
2337 };
2338
2339 static void tun_sock_write_space(struct sock *sk)
2340 {
2341         struct tun_file *tfile;
2342         wait_queue_head_t *wqueue;
2343
2344         if (!sock_writeable(sk))
2345                 return;
2346
2347         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2348                 return;
2349
2350         wqueue = sk_sleep(sk);
2351         if (wqueue && waitqueue_active(wqueue))
2352                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2353                                                 EPOLLWRNORM | EPOLLWRBAND);
2354
2355         tfile = container_of(sk, struct tun_file, sk);
2356         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2357 }
2358
2359 static void tun_put_page(struct tun_page *tpage)
2360 {
2361         if (tpage->page)
2362                 __page_frag_cache_drain(tpage->page, tpage->count);
2363 }
2364
2365 static int tun_xdp_one(struct tun_struct *tun,
2366                        struct tun_file *tfile,
2367                        struct xdp_buff *xdp, int *flush,
2368                        struct tun_page *tpage)
2369 {
2370         unsigned int datasize = xdp->data_end - xdp->data;
2371         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2372         struct virtio_net_hdr *gso = &hdr->gso;
2373         struct tun_pcpu_stats *stats;
2374         struct bpf_prog *xdp_prog;
2375         struct sk_buff *skb = NULL;
2376         u32 rxhash = 0, act;
2377         int buflen = hdr->buflen;
2378         int err = 0;
2379         bool skb_xdp = false;
2380         struct page *page;
2381
2382         xdp_prog = rcu_dereference(tun->xdp_prog);
2383         if (xdp_prog) {
2384                 if (gso->gso_type) {
2385                         skb_xdp = true;
2386                         goto build;
2387                 }
2388                 xdp_set_data_meta_invalid(xdp);
2389                 xdp->rxq = &tfile->xdp_rxq;
2390                 xdp->frame_sz = buflen;
2391
2392                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2393                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2394                 if (err < 0) {
2395                         put_page(virt_to_head_page(xdp->data));
2396                         return err;
2397                 }
2398
2399                 switch (err) {
2400                 case XDP_REDIRECT:
2401                         *flush = true;
2402                         fallthrough;
2403                 case XDP_TX:
2404                         return 0;
2405                 case XDP_PASS:
2406                         break;
2407                 default:
2408                         page = virt_to_head_page(xdp->data);
2409                         if (tpage->page == page) {
2410                                 ++tpage->count;
2411                         } else {
2412                                 tun_put_page(tpage);
2413                                 tpage->page = page;
2414                                 tpage->count = 1;
2415                         }
2416                         return 0;
2417                 }
2418         }
2419
2420 build:
2421         skb = build_skb(xdp->data_hard_start, buflen);
2422         if (!skb) {
2423                 err = -ENOMEM;
2424                 goto out;
2425         }
2426
2427         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2428         skb_put(skb, xdp->data_end - xdp->data);
2429
2430         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2431                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2432                 kfree_skb(skb);
2433                 err = -EINVAL;
2434                 goto out;
2435         }
2436
2437         skb->protocol = eth_type_trans(skb, tun->dev);
2438         skb_reset_network_header(skb);
2439         skb_probe_transport_header(skb);
2440         skb_record_rx_queue(skb, tfile->queue_index);
2441
2442         if (skb_xdp) {
2443                 err = do_xdp_generic(xdp_prog, skb);
2444                 if (err != XDP_PASS)
2445                         goto out;
2446         }
2447
2448         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2449             !tfile->detached)
2450                 rxhash = __skb_get_hash_symmetric(skb);
2451
2452         netif_receive_skb(skb);
2453
2454         /* No need for get_cpu_ptr() here since this function is
2455          * always called with bh disabled
2456          */
2457         stats = this_cpu_ptr(tun->pcpu_stats);
2458         u64_stats_update_begin(&stats->syncp);
2459         u64_stats_inc(&stats->rx_packets);
2460         u64_stats_add(&stats->rx_bytes, datasize);
2461         u64_stats_update_end(&stats->syncp);
2462
2463         if (rxhash)
2464                 tun_flow_update(tun, rxhash, tfile);
2465
2466 out:
2467         return err;
2468 }
2469
2470 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2471 {
2472         int ret, i;
2473         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2474         struct tun_struct *tun = tun_get(tfile);
2475         struct tun_msg_ctl *ctl = m->msg_control;
2476         struct xdp_buff *xdp;
2477
2478         if (!tun)
2479                 return -EBADFD;
2480
2481         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2482                 struct tun_page tpage;
2483                 int n = ctl->num;
2484                 int flush = 0;
2485
2486                 memset(&tpage, 0, sizeof(tpage));
2487
2488                 local_bh_disable();
2489                 rcu_read_lock();
2490
2491                 for (i = 0; i < n; i++) {
2492                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2493                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2494                 }
2495
2496                 if (flush)
2497                         xdp_do_flush();
2498
2499                 rcu_read_unlock();
2500                 local_bh_enable();
2501
2502                 tun_put_page(&tpage);
2503
2504                 ret = total_len;
2505                 goto out;
2506         }
2507
2508         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2509                            m->msg_flags & MSG_DONTWAIT,
2510                            m->msg_flags & MSG_MORE);
2511 out:
2512         tun_put(tun);
2513         return ret;
2514 }
2515
2516 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2517                        int flags)
2518 {
2519         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2520         struct tun_struct *tun = tun_get(tfile);
2521         void *ptr = m->msg_control;
2522         int ret;
2523
2524         if (!tun) {
2525                 ret = -EBADFD;
2526                 goto out_free;
2527         }
2528
2529         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2530                 ret = -EINVAL;
2531                 goto out_put_tun;
2532         }
2533         if (flags & MSG_ERRQUEUE) {
2534                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2535                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2536                 goto out;
2537         }
2538         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2539         if (ret > (ssize_t)total_len) {
2540                 m->msg_flags |= MSG_TRUNC;
2541                 ret = flags & MSG_TRUNC ? ret : total_len;
2542         }
2543 out:
2544         tun_put(tun);
2545         return ret;
2546
2547 out_put_tun:
2548         tun_put(tun);
2549 out_free:
2550         tun_ptr_free(ptr);
2551         return ret;
2552 }
2553
2554 static int tun_ptr_peek_len(void *ptr)
2555 {
2556         if (likely(ptr)) {
2557                 if (tun_is_xdp_frame(ptr)) {
2558                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2559
2560                         return xdpf->len;
2561                 }
2562                 return __skb_array_len_with_tag(ptr);
2563         } else {
2564                 return 0;
2565         }
2566 }
2567
2568 static int tun_peek_len(struct socket *sock)
2569 {
2570         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2571         struct tun_struct *tun;
2572         int ret = 0;
2573
2574         tun = tun_get(tfile);
2575         if (!tun)
2576                 return 0;
2577
2578         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2579         tun_put(tun);
2580
2581         return ret;
2582 }
2583
2584 /* Ops structure to mimic raw sockets with tun */
2585 static const struct proto_ops tun_socket_ops = {
2586         .peek_len = tun_peek_len,
2587         .sendmsg = tun_sendmsg,
2588         .recvmsg = tun_recvmsg,
2589 };
2590
2591 static struct proto tun_proto = {
2592         .name           = "tun",
2593         .owner          = THIS_MODULE,
2594         .obj_size       = sizeof(struct tun_file),
2595 };
2596
2597 static int tun_flags(struct tun_struct *tun)
2598 {
2599         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2600 }
2601
2602 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2603                               char *buf)
2604 {
2605         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2606         return sprintf(buf, "0x%x\n", tun_flags(tun));
2607 }
2608
2609 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2610                               char *buf)
2611 {
2612         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2613         return uid_valid(tun->owner)?
2614                 sprintf(buf, "%u\n",
2615                         from_kuid_munged(current_user_ns(), tun->owner)):
2616                 sprintf(buf, "-1\n");
2617 }
2618
2619 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2620                               char *buf)
2621 {
2622         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2623         return gid_valid(tun->group) ?
2624                 sprintf(buf, "%u\n",
2625                         from_kgid_munged(current_user_ns(), tun->group)):
2626                 sprintf(buf, "-1\n");
2627 }
2628
2629 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2630 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2631 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2632
2633 static struct attribute *tun_dev_attrs[] = {
2634         &dev_attr_tun_flags.attr,
2635         &dev_attr_owner.attr,
2636         &dev_attr_group.attr,
2637         NULL
2638 };
2639
2640 static const struct attribute_group tun_attr_group = {
2641         .attrs = tun_dev_attrs
2642 };
2643
2644 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2645 {
2646         struct tun_struct *tun;
2647         struct tun_file *tfile = file->private_data;
2648         struct net_device *dev;
2649         int err;
2650
2651         if (tfile->detached)
2652                 return -EINVAL;
2653
2654         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2655                 if (!capable(CAP_NET_ADMIN))
2656                         return -EPERM;
2657
2658                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2659                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2660                         return -EINVAL;
2661         }
2662
2663         dev = __dev_get_by_name(net, ifr->ifr_name);
2664         if (dev) {
2665                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2666                         return -EBUSY;
2667                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2668                         tun = netdev_priv(dev);
2669                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2670                         tun = netdev_priv(dev);
2671                 else
2672                         return -EINVAL;
2673
2674                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2675                     !!(tun->flags & IFF_MULTI_QUEUE))
2676                         return -EINVAL;
2677
2678                 if (tun_not_capable(tun))
2679                         return -EPERM;
2680                 err = security_tun_dev_open(tun->security);
2681                 if (err < 0)
2682                         return err;
2683
2684                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2685                                  ifr->ifr_flags & IFF_NAPI,
2686                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2687                 if (err < 0)
2688                         return err;
2689
2690                 if (tun->flags & IFF_MULTI_QUEUE &&
2691                     (tun->numqueues + tun->numdisabled > 1)) {
2692                         /* One or more queue has already been attached, no need
2693                          * to initialize the device again.
2694                          */
2695                         netdev_state_change(dev);
2696                         return 0;
2697                 }
2698
2699                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2700                               (ifr->ifr_flags & TUN_FEATURES);
2701
2702                 netdev_state_change(dev);
2703         } else {
2704                 char *name;
2705                 unsigned long flags = 0;
2706                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2707                              MAX_TAP_QUEUES : 1;
2708
2709                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2710                         return -EPERM;
2711                 err = security_tun_dev_create();
2712                 if (err < 0)
2713                         return err;
2714
2715                 /* Set dev type */
2716                 if (ifr->ifr_flags & IFF_TUN) {
2717                         /* TUN device */
2718                         flags |= IFF_TUN;
2719                         name = "tun%d";
2720                 } else if (ifr->ifr_flags & IFF_TAP) {
2721                         /* TAP device */
2722                         flags |= IFF_TAP;
2723                         name = "tap%d";
2724                 } else
2725                         return -EINVAL;
2726
2727                 if (*ifr->ifr_name)
2728                         name = ifr->ifr_name;
2729
2730                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2731                                        NET_NAME_UNKNOWN, tun_setup, queues,
2732                                        queues);
2733
2734                 if (!dev)
2735                         return -ENOMEM;
2736
2737                 dev_net_set(dev, net);
2738                 dev->rtnl_link_ops = &tun_link_ops;
2739                 dev->ifindex = tfile->ifindex;
2740                 dev->sysfs_groups[0] = &tun_attr_group;
2741
2742                 tun = netdev_priv(dev);
2743                 tun->dev = dev;
2744                 tun->flags = flags;
2745                 tun->txflt.count = 0;
2746                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2747
2748                 tun->align = NET_SKB_PAD;
2749                 tun->filter_attached = false;
2750                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2751                 tun->rx_batched = 0;
2752                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2753
2754                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2755                 if (!tun->pcpu_stats) {
2756                         err = -ENOMEM;
2757                         goto err_free_dev;
2758                 }
2759
2760                 spin_lock_init(&tun->lock);
2761
2762                 err = security_tun_dev_alloc_security(&tun->security);
2763                 if (err < 0)
2764                         goto err_free_stat;
2765
2766                 tun_net_init(dev);
2767                 tun_flow_init(tun);
2768
2769                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2770                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2771                                    NETIF_F_HW_VLAN_STAG_TX;
2772                 dev->features = dev->hw_features | NETIF_F_LLTX;
2773                 dev->vlan_features = dev->features &
2774                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2775                                        NETIF_F_HW_VLAN_STAG_TX);
2776
2777                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2778                               (ifr->ifr_flags & TUN_FEATURES);
2779
2780                 INIT_LIST_HEAD(&tun->disabled);
2781                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2782                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2783                 if (err < 0)
2784                         goto err_free_flow;
2785
2786                 err = register_netdevice(tun->dev);
2787                 if (err < 0)
2788                         goto err_detach;
2789                 /* free_netdev() won't check refcnt, to aovid race
2790                  * with dev_put() we need publish tun after registration.
2791                  */
2792                 rcu_assign_pointer(tfile->tun, tun);
2793         }
2794
2795         netif_carrier_on(tun->dev);
2796
2797         /* Make sure persistent devices do not get stuck in
2798          * xoff state.
2799          */
2800         if (netif_running(tun->dev))
2801                 netif_tx_wake_all_queues(tun->dev);
2802
2803         strcpy(ifr->ifr_name, tun->dev->name);
2804         return 0;
2805
2806 err_detach:
2807         tun_detach_all(dev);
2808         /* We are here because register_netdevice() has failed.
2809          * If register_netdevice() already called tun_free_netdev()
2810          * while dealing with the error, tun->pcpu_stats has been cleared.
2811          */
2812         if (!tun->pcpu_stats)
2813                 goto err_free_dev;
2814
2815 err_free_flow:
2816         tun_flow_uninit(tun);
2817         security_tun_dev_free_security(tun->security);
2818 err_free_stat:
2819         free_percpu(tun->pcpu_stats);
2820 err_free_dev:
2821         free_netdev(dev);
2822         return err;
2823 }
2824
2825 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2826 {
2827         strcpy(ifr->ifr_name, tun->dev->name);
2828
2829         ifr->ifr_flags = tun_flags(tun);
2830
2831 }
2832
2833 /* This is like a cut-down ethtool ops, except done via tun fd so no
2834  * privs required. */
2835 static int set_offload(struct tun_struct *tun, unsigned long arg)
2836 {
2837         netdev_features_t features = 0;
2838
2839         if (arg & TUN_F_CSUM) {
2840                 features |= NETIF_F_HW_CSUM;
2841                 arg &= ~TUN_F_CSUM;
2842
2843                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2844                         if (arg & TUN_F_TSO_ECN) {
2845                                 features |= NETIF_F_TSO_ECN;
2846                                 arg &= ~TUN_F_TSO_ECN;
2847                         }
2848                         if (arg & TUN_F_TSO4)
2849                                 features |= NETIF_F_TSO;
2850                         if (arg & TUN_F_TSO6)
2851                                 features |= NETIF_F_TSO6;
2852                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2853                 }
2854
2855                 arg &= ~TUN_F_UFO;
2856         }
2857
2858         /* This gives the user a way to test for new features in future by
2859          * trying to set them. */
2860         if (arg)
2861                 return -EINVAL;
2862
2863         tun->set_features = features;
2864         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2865         tun->dev->wanted_features |= features;
2866         netdev_update_features(tun->dev);
2867
2868         return 0;
2869 }
2870
2871 static void tun_detach_filter(struct tun_struct *tun, int n)
2872 {
2873         int i;
2874         struct tun_file *tfile;
2875
2876         for (i = 0; i < n; i++) {
2877                 tfile = rtnl_dereference(tun->tfiles[i]);
2878                 lock_sock(tfile->socket.sk);
2879                 sk_detach_filter(tfile->socket.sk);
2880                 release_sock(tfile->socket.sk);
2881         }
2882
2883         tun->filter_attached = false;
2884 }
2885
2886 static int tun_attach_filter(struct tun_struct *tun)
2887 {
2888         int i, ret = 0;
2889         struct tun_file *tfile;
2890
2891         for (i = 0; i < tun->numqueues; i++) {
2892                 tfile = rtnl_dereference(tun->tfiles[i]);
2893                 lock_sock(tfile->socket.sk);
2894                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2895                 release_sock(tfile->socket.sk);
2896                 if (ret) {
2897                         tun_detach_filter(tun, i);
2898                         return ret;
2899                 }
2900         }
2901
2902         tun->filter_attached = true;
2903         return ret;
2904 }
2905
2906 static void tun_set_sndbuf(struct tun_struct *tun)
2907 {
2908         struct tun_file *tfile;
2909         int i;
2910
2911         for (i = 0; i < tun->numqueues; i++) {
2912                 tfile = rtnl_dereference(tun->tfiles[i]);
2913                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2914         }
2915 }
2916
2917 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2918 {
2919         struct tun_file *tfile = file->private_data;
2920         struct tun_struct *tun;
2921         int ret = 0;
2922
2923         rtnl_lock();
2924
2925         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2926                 tun = tfile->detached;
2927                 if (!tun) {
2928                         ret = -EINVAL;
2929                         goto unlock;
2930                 }
2931                 ret = security_tun_dev_attach_queue(tun->security);
2932                 if (ret < 0)
2933                         goto unlock;
2934                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2935                                  tun->flags & IFF_NAPI_FRAGS, true);
2936         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2937                 tun = rtnl_dereference(tfile->tun);
2938                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2939                         ret = -EINVAL;
2940                 else
2941                         __tun_detach(tfile, false);
2942         } else
2943                 ret = -EINVAL;
2944
2945         if (ret >= 0)
2946                 netdev_state_change(tun->dev);
2947
2948 unlock:
2949         rtnl_unlock();
2950         return ret;
2951 }
2952
2953 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2954                         void __user *data)
2955 {
2956         struct bpf_prog *prog;
2957         int fd;
2958
2959         if (copy_from_user(&fd, data, sizeof(fd)))
2960                 return -EFAULT;
2961
2962         if (fd == -1) {
2963                 prog = NULL;
2964         } else {
2965                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2966                 if (IS_ERR(prog))
2967                         return PTR_ERR(prog);
2968         }
2969
2970         return __tun_set_ebpf(tun, prog_p, prog);
2971 }
2972
2973 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2974                             unsigned long arg, int ifreq_len)
2975 {
2976         struct tun_file *tfile = file->private_data;
2977         struct net *net = sock_net(&tfile->sk);
2978         struct tun_struct *tun;
2979         void __user* argp = (void __user*)arg;
2980         unsigned int ifindex, carrier;
2981         struct ifreq ifr;
2982         kuid_t owner;
2983         kgid_t group;
2984         int sndbuf;
2985         int vnet_hdr_sz;
2986         int le;
2987         int ret;
2988         bool do_notify = false;
2989
2990         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2991             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2992                 if (copy_from_user(&ifr, argp, ifreq_len))
2993                         return -EFAULT;
2994         } else {
2995                 memset(&ifr, 0, sizeof(ifr));
2996         }
2997         if (cmd == TUNGETFEATURES) {
2998                 /* Currently this just means: "what IFF flags are valid?".
2999                  * This is needed because we never checked for invalid flags on
3000                  * TUNSETIFF.
3001                  */
3002                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3003                                 (unsigned int __user*)argp);
3004         } else if (cmd == TUNSETQUEUE) {
3005                 return tun_set_queue(file, &ifr);
3006         } else if (cmd == SIOCGSKNS) {
3007                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3008                         return -EPERM;
3009                 return open_related_ns(&net->ns, get_net_ns);
3010         }
3011
3012         ret = 0;
3013         rtnl_lock();
3014
3015         tun = tun_get(tfile);
3016         if (cmd == TUNSETIFF) {
3017                 ret = -EEXIST;
3018                 if (tun)
3019                         goto unlock;
3020
3021                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3022
3023                 ret = tun_set_iff(net, file, &ifr);
3024
3025                 if (ret)
3026                         goto unlock;
3027
3028                 if (copy_to_user(argp, &ifr, ifreq_len))
3029                         ret = -EFAULT;
3030                 goto unlock;
3031         }
3032         if (cmd == TUNSETIFINDEX) {
3033                 ret = -EPERM;
3034                 if (tun)
3035                         goto unlock;
3036
3037                 ret = -EFAULT;
3038                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3039                         goto unlock;
3040
3041                 ret = 0;
3042                 tfile->ifindex = ifindex;
3043                 goto unlock;
3044         }
3045
3046         ret = -EBADFD;
3047         if (!tun)
3048                 goto unlock;
3049
3050         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3051
3052         net = dev_net(tun->dev);
3053         ret = 0;
3054         switch (cmd) {
3055         case TUNGETIFF:
3056                 tun_get_iff(tun, &ifr);
3057
3058                 if (tfile->detached)
3059                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3060                 if (!tfile->socket.sk->sk_filter)
3061                         ifr.ifr_flags |= IFF_NOFILTER;
3062
3063                 if (copy_to_user(argp, &ifr, ifreq_len))
3064                         ret = -EFAULT;
3065                 break;
3066
3067         case TUNSETNOCSUM:
3068                 /* Disable/Enable checksum */
3069
3070                 /* [unimplemented] */
3071                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3072                            arg ? "disabled" : "enabled");
3073                 break;
3074
3075         case TUNSETPERSIST:
3076                 /* Disable/Enable persist mode. Keep an extra reference to the
3077                  * module to prevent the module being unprobed.
3078                  */
3079                 if (arg && !(tun->flags & IFF_PERSIST)) {
3080                         tun->flags |= IFF_PERSIST;
3081                         __module_get(THIS_MODULE);
3082                         do_notify = true;
3083                 }
3084                 if (!arg && (tun->flags & IFF_PERSIST)) {
3085                         tun->flags &= ~IFF_PERSIST;
3086                         module_put(THIS_MODULE);
3087                         do_notify = true;
3088                 }
3089
3090                 netif_info(tun, drv, tun->dev, "persist %s\n",
3091                            arg ? "enabled" : "disabled");
3092                 break;
3093
3094         case TUNSETOWNER:
3095                 /* Set owner of the device */
3096                 owner = make_kuid(current_user_ns(), arg);
3097                 if (!uid_valid(owner)) {
3098                         ret = -EINVAL;
3099                         break;
3100                 }
3101                 tun->owner = owner;
3102                 do_notify = true;
3103                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3104                            from_kuid(&init_user_ns, tun->owner));
3105                 break;
3106
3107         case TUNSETGROUP:
3108                 /* Set group of the device */
3109                 group = make_kgid(current_user_ns(), arg);
3110                 if (!gid_valid(group)) {
3111                         ret = -EINVAL;
3112                         break;
3113                 }
3114                 tun->group = group;
3115                 do_notify = true;
3116                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3117                            from_kgid(&init_user_ns, tun->group));
3118                 break;
3119
3120         case TUNSETLINK:
3121                 /* Only allow setting the type when the interface is down */
3122                 if (tun->dev->flags & IFF_UP) {
3123                         netif_info(tun, drv, tun->dev,
3124                                    "Linktype set failed because interface is up\n");
3125                         ret = -EBUSY;
3126                 } else {
3127                         tun->dev->type = (int) arg;
3128                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3129                                    tun->dev->type);
3130                         ret = 0;
3131                 }
3132                 break;
3133
3134         case TUNSETDEBUG:
3135                 tun->msg_enable = (u32)arg;
3136                 break;
3137
3138         case TUNSETOFFLOAD:
3139                 ret = set_offload(tun, arg);
3140                 break;
3141
3142         case TUNSETTXFILTER:
3143                 /* Can be set only for TAPs */
3144                 ret = -EINVAL;
3145                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3146                         break;
3147                 ret = update_filter(&tun->txflt, (void __user *)arg);
3148                 break;
3149
3150         case SIOCGIFHWADDR:
3151                 /* Get hw address */
3152                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3153                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3154                 if (copy_to_user(argp, &ifr, ifreq_len))
3155                         ret = -EFAULT;
3156                 break;
3157
3158         case SIOCSIFHWADDR:
3159                 /* Set hw address */
3160                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3161                 break;
3162
3163         case TUNGETSNDBUF:
3164                 sndbuf = tfile->socket.sk->sk_sndbuf;
3165                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3166                         ret = -EFAULT;
3167                 break;
3168
3169         case TUNSETSNDBUF:
3170                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3171                         ret = -EFAULT;
3172                         break;
3173                 }
3174                 if (sndbuf <= 0) {
3175                         ret = -EINVAL;
3176                         break;
3177                 }
3178
3179                 tun->sndbuf = sndbuf;
3180                 tun_set_sndbuf(tun);
3181                 break;
3182
3183         case TUNGETVNETHDRSZ:
3184                 vnet_hdr_sz = tun->vnet_hdr_sz;
3185                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3186                         ret = -EFAULT;
3187                 break;
3188
3189         case TUNSETVNETHDRSZ:
3190                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3191                         ret = -EFAULT;
3192                         break;
3193                 }
3194                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3195                         ret = -EINVAL;
3196                         break;
3197                 }
3198
3199                 tun->vnet_hdr_sz = vnet_hdr_sz;
3200                 break;
3201
3202         case TUNGETVNETLE:
3203                 le = !!(tun->flags & TUN_VNET_LE);
3204                 if (put_user(le, (int __user *)argp))
3205                         ret = -EFAULT;
3206                 break;
3207
3208         case TUNSETVNETLE:
3209                 if (get_user(le, (int __user *)argp)) {
3210                         ret = -EFAULT;
3211                         break;
3212                 }
3213                 if (le)
3214                         tun->flags |= TUN_VNET_LE;
3215                 else
3216                         tun->flags &= ~TUN_VNET_LE;
3217                 break;
3218
3219         case TUNGETVNETBE:
3220                 ret = tun_get_vnet_be(tun, argp);
3221                 break;
3222
3223         case TUNSETVNETBE:
3224                 ret = tun_set_vnet_be(tun, argp);
3225                 break;
3226
3227         case TUNATTACHFILTER:
3228                 /* Can be set only for TAPs */
3229                 ret = -EINVAL;
3230                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3231                         break;
3232                 ret = -EFAULT;
3233                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3234                         break;
3235
3236                 ret = tun_attach_filter(tun);
3237                 break;
3238
3239         case TUNDETACHFILTER:
3240                 /* Can be set only for TAPs */
3241                 ret = -EINVAL;
3242                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3243                         break;
3244                 ret = 0;
3245                 tun_detach_filter(tun, tun->numqueues);
3246                 break;
3247
3248         case TUNGETFILTER:
3249                 ret = -EINVAL;
3250                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3251                         break;
3252                 ret = -EFAULT;
3253                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3254                         break;
3255                 ret = 0;
3256                 break;
3257
3258         case TUNSETSTEERINGEBPF:
3259                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3260                 break;
3261
3262         case TUNSETFILTEREBPF:
3263                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3264                 break;
3265
3266         case TUNSETCARRIER:
3267                 ret = -EFAULT;
3268                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3269                         goto unlock;
3270
3271                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3272                 break;
3273
3274         case TUNGETDEVNETNS:
3275                 ret = -EPERM;
3276                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3277                         goto unlock;
3278                 ret = open_related_ns(&net->ns, get_net_ns);
3279                 break;
3280
3281         default:
3282                 ret = -EINVAL;
3283                 break;
3284         }
3285
3286         if (do_notify)
3287                 netdev_state_change(tun->dev);
3288
3289 unlock:
3290         rtnl_unlock();
3291         if (tun)
3292                 tun_put(tun);
3293         return ret;
3294 }
3295
3296 static long tun_chr_ioctl(struct file *file,
3297                           unsigned int cmd, unsigned long arg)
3298 {
3299         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3300 }
3301
3302 #ifdef CONFIG_COMPAT
3303 static long tun_chr_compat_ioctl(struct file *file,
3304                          unsigned int cmd, unsigned long arg)
3305 {
3306         switch (cmd) {
3307         case TUNSETIFF:
3308         case TUNGETIFF:
3309         case TUNSETTXFILTER:
3310         case TUNGETSNDBUF:
3311         case TUNSETSNDBUF:
3312         case SIOCGIFHWADDR:
3313         case SIOCSIFHWADDR:
3314                 arg = (unsigned long)compat_ptr(arg);
3315                 break;
3316         default:
3317                 arg = (compat_ulong_t)arg;
3318                 break;
3319         }
3320
3321         /*
3322          * compat_ifreq is shorter than ifreq, so we must not access beyond
3323          * the end of that structure. All fields that are used in this
3324          * driver are compatible though, we don't need to convert the
3325          * contents.
3326          */
3327         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3328 }
3329 #endif /* CONFIG_COMPAT */
3330
3331 static int tun_chr_fasync(int fd, struct file *file, int on)
3332 {
3333         struct tun_file *tfile = file->private_data;
3334         int ret;
3335
3336         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3337                 goto out;
3338
3339         if (on) {
3340                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3341                 tfile->flags |= TUN_FASYNC;
3342         } else
3343                 tfile->flags &= ~TUN_FASYNC;
3344         ret = 0;
3345 out:
3346         return ret;
3347 }
3348
3349 static int tun_chr_open(struct inode *inode, struct file * file)
3350 {
3351         struct net *net = current->nsproxy->net_ns;
3352         struct tun_file *tfile;
3353
3354         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3355                                             &tun_proto, 0);
3356         if (!tfile)
3357                 return -ENOMEM;
3358         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3359                 sk_free(&tfile->sk);
3360                 return -ENOMEM;
3361         }
3362
3363         mutex_init(&tfile->napi_mutex);
3364         RCU_INIT_POINTER(tfile->tun, NULL);
3365         tfile->flags = 0;
3366         tfile->ifindex = 0;
3367
3368         init_waitqueue_head(&tfile->socket.wq.wait);
3369
3370         tfile->socket.file = file;
3371         tfile->socket.ops = &tun_socket_ops;
3372
3373         sock_init_data(&tfile->socket, &tfile->sk);
3374
3375         tfile->sk.sk_write_space = tun_sock_write_space;
3376         tfile->sk.sk_sndbuf = INT_MAX;
3377
3378         file->private_data = tfile;
3379         INIT_LIST_HEAD(&tfile->next);
3380
3381         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3382
3383         return 0;
3384 }
3385
3386 static int tun_chr_close(struct inode *inode, struct file *file)
3387 {
3388         struct tun_file *tfile = file->private_data;
3389
3390         tun_detach(tfile, true);
3391
3392         return 0;
3393 }
3394
3395 #ifdef CONFIG_PROC_FS
3396 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3397 {
3398         struct tun_file *tfile = file->private_data;
3399         struct tun_struct *tun;
3400         struct ifreq ifr;
3401
3402         memset(&ifr, 0, sizeof(ifr));
3403
3404         rtnl_lock();
3405         tun = tun_get(tfile);
3406         if (tun)
3407                 tun_get_iff(tun, &ifr);
3408         rtnl_unlock();
3409
3410         if (tun)
3411                 tun_put(tun);
3412
3413         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3414 }
3415 #endif
3416
3417 static const struct file_operations tun_fops = {
3418         .owner  = THIS_MODULE,
3419         .llseek = no_llseek,
3420         .read_iter  = tun_chr_read_iter,
3421         .write_iter = tun_chr_write_iter,
3422         .poll   = tun_chr_poll,
3423         .unlocked_ioctl = tun_chr_ioctl,
3424 #ifdef CONFIG_COMPAT
3425         .compat_ioctl = tun_chr_compat_ioctl,
3426 #endif
3427         .open   = tun_chr_open,
3428         .release = tun_chr_close,
3429         .fasync = tun_chr_fasync,
3430 #ifdef CONFIG_PROC_FS
3431         .show_fdinfo = tun_chr_show_fdinfo,
3432 #endif
3433 };
3434
3435 static struct miscdevice tun_miscdev = {
3436         .minor = TUN_MINOR,
3437         .name = "tun",
3438         .nodename = "net/tun",
3439         .fops = &tun_fops,
3440 };
3441
3442 /* ethtool interface */
3443
3444 static void tun_default_link_ksettings(struct net_device *dev,
3445                                        struct ethtool_link_ksettings *cmd)
3446 {
3447         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3448         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3449         cmd->base.speed         = SPEED_10;
3450         cmd->base.duplex        = DUPLEX_FULL;
3451         cmd->base.port          = PORT_TP;
3452         cmd->base.phy_address   = 0;
3453         cmd->base.autoneg       = AUTONEG_DISABLE;
3454 }
3455
3456 static int tun_get_link_ksettings(struct net_device *dev,
3457                                   struct ethtool_link_ksettings *cmd)
3458 {
3459         struct tun_struct *tun = netdev_priv(dev);
3460
3461         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3462         return 0;
3463 }
3464
3465 static int tun_set_link_ksettings(struct net_device *dev,
3466                                   const struct ethtool_link_ksettings *cmd)
3467 {
3468         struct tun_struct *tun = netdev_priv(dev);
3469
3470         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3471         return 0;
3472 }
3473
3474 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3475 {
3476         struct tun_struct *tun = netdev_priv(dev);
3477
3478         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3479         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3480
3481         switch (tun->flags & TUN_TYPE_MASK) {
3482         case IFF_TUN:
3483                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3484                 break;
3485         case IFF_TAP:
3486                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3487                 break;
3488         }
3489 }
3490
3491 static u32 tun_get_msglevel(struct net_device *dev)
3492 {
3493         struct tun_struct *tun = netdev_priv(dev);
3494
3495         return tun->msg_enable;
3496 }
3497
3498 static void tun_set_msglevel(struct net_device *dev, u32 value)
3499 {
3500         struct tun_struct *tun = netdev_priv(dev);
3501
3502         tun->msg_enable = value;
3503 }
3504
3505 static int tun_get_coalesce(struct net_device *dev,
3506                             struct ethtool_coalesce *ec)
3507 {
3508         struct tun_struct *tun = netdev_priv(dev);
3509
3510         ec->rx_max_coalesced_frames = tun->rx_batched;
3511
3512         return 0;
3513 }
3514
3515 static int tun_set_coalesce(struct net_device *dev,
3516                             struct ethtool_coalesce *ec)
3517 {
3518         struct tun_struct *tun = netdev_priv(dev);
3519
3520         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3521                 tun->rx_batched = NAPI_POLL_WEIGHT;
3522         else
3523                 tun->rx_batched = ec->rx_max_coalesced_frames;
3524
3525         return 0;
3526 }
3527
3528 static const struct ethtool_ops tun_ethtool_ops = {
3529         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3530         .get_drvinfo    = tun_get_drvinfo,
3531         .get_msglevel   = tun_get_msglevel,
3532         .set_msglevel   = tun_set_msglevel,
3533         .get_link       = ethtool_op_get_link,
3534         .get_ts_info    = ethtool_op_get_ts_info,
3535         .get_coalesce   = tun_get_coalesce,
3536         .set_coalesce   = tun_set_coalesce,
3537         .get_link_ksettings = tun_get_link_ksettings,
3538         .set_link_ksettings = tun_set_link_ksettings,
3539 };
3540
3541 static int tun_queue_resize(struct tun_struct *tun)
3542 {
3543         struct net_device *dev = tun->dev;
3544         struct tun_file *tfile;
3545         struct ptr_ring **rings;
3546         int n = tun->numqueues + tun->numdisabled;
3547         int ret, i;
3548
3549         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3550         if (!rings)
3551                 return -ENOMEM;
3552
3553         for (i = 0; i < tun->numqueues; i++) {
3554                 tfile = rtnl_dereference(tun->tfiles[i]);
3555                 rings[i] = &tfile->tx_ring;
3556         }
3557         list_for_each_entry(tfile, &tun->disabled, next)
3558                 rings[i++] = &tfile->tx_ring;
3559
3560         ret = ptr_ring_resize_multiple(rings, n,
3561                                        dev->tx_queue_len, GFP_KERNEL,
3562                                        tun_ptr_free);
3563
3564         kfree(rings);
3565         return ret;
3566 }
3567
3568 static int tun_device_event(struct notifier_block *unused,
3569                             unsigned long event, void *ptr)
3570 {
3571         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3572         struct tun_struct *tun = netdev_priv(dev);
3573         int i;
3574
3575         if (dev->rtnl_link_ops != &tun_link_ops)
3576                 return NOTIFY_DONE;
3577
3578         switch (event) {
3579         case NETDEV_CHANGE_TX_QUEUE_LEN:
3580                 if (tun_queue_resize(tun))
3581                         return NOTIFY_BAD;
3582                 break;
3583         case NETDEV_UP:
3584                 for (i = 0; i < tun->numqueues; i++) {
3585                         struct tun_file *tfile;
3586
3587                         tfile = rtnl_dereference(tun->tfiles[i]);
3588                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3589                 }
3590                 break;
3591         default:
3592                 break;
3593         }
3594
3595         return NOTIFY_DONE;
3596 }
3597
3598 static struct notifier_block tun_notifier_block __read_mostly = {
3599         .notifier_call  = tun_device_event,
3600 };
3601
3602 static int __init tun_init(void)
3603 {
3604         int ret = 0;
3605
3606         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3607
3608         ret = rtnl_link_register(&tun_link_ops);
3609         if (ret) {
3610                 pr_err("Can't register link_ops\n");
3611                 goto err_linkops;
3612         }
3613
3614         ret = misc_register(&tun_miscdev);
3615         if (ret) {
3616                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3617                 goto err_misc;
3618         }
3619
3620         ret = register_netdevice_notifier(&tun_notifier_block);
3621         if (ret) {
3622                 pr_err("Can't register netdevice notifier\n");
3623                 goto err_notifier;
3624         }
3625
3626         return  0;
3627
3628 err_notifier:
3629         misc_deregister(&tun_miscdev);
3630 err_misc:
3631         rtnl_link_unregister(&tun_link_ops);
3632 err_linkops:
3633         return ret;
3634 }
3635
3636 static void tun_cleanup(void)
3637 {
3638         misc_deregister(&tun_miscdev);
3639         rtnl_link_unregister(&tun_link_ops);
3640         unregister_netdevice_notifier(&tun_notifier_block);
3641 }
3642
3643 /* Get an underlying socket object from tun file.  Returns error unless file is
3644  * attached to a device.  The returned object works like a packet socket, it
3645  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3646  * holding a reference to the file for as long as the socket is in use. */
3647 struct socket *tun_get_socket(struct file *file)
3648 {
3649         struct tun_file *tfile;
3650         if (file->f_op != &tun_fops)
3651                 return ERR_PTR(-EINVAL);
3652         tfile = file->private_data;
3653         if (!tfile)
3654                 return ERR_PTR(-EBADFD);
3655         return &tfile->socket;
3656 }
3657 EXPORT_SYMBOL_GPL(tun_get_socket);
3658
3659 struct ptr_ring *tun_get_tx_ring(struct file *file)
3660 {
3661         struct tun_file *tfile;
3662
3663         if (file->f_op != &tun_fops)
3664                 return ERR_PTR(-EINVAL);
3665         tfile = file->private_data;
3666         if (!tfile)
3667                 return ERR_PTR(-EBADFD);
3668         return &tfile->tx_ring;
3669 }
3670 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3671
3672 module_init(tun_init);
3673 module_exit(tun_cleanup);
3674 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3675 MODULE_AUTHOR(DRV_COPYRIGHT);
3676 MODULE_LICENSE("GPL");
3677 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3678 MODULE_ALIAS("devname:net/tun");