drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <[email protected]>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <[email protected]> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <[email protected]>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <[email protected]>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <[email protected]>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <[email protected]>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 117
 118 /* TUN device flags */
 119
 120 /* IFF_ATTACH_QUEUE is never stored in device flags,
 121  * overload it to mean fasync when stored there.
 122  */
 123 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 124 /* High bits in flags field are unused. */
 125 #define TUN_VNET_LE     0x80000000
 126 #define TUN_VNET_BE     0x40000000
 127
 128 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 129                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 130
 131 #define GOODCOPY_LEN 128
 132
 133 #define FLT_EXACT_COUNT 8
 134 struct tap_filter {
 135         unsigned int    count;    /* Number of addrs. Zero means disabled */
 136         u32             mask[2];  /* Mask of the hashed addrs */
 137         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 138 };
 139
 140 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 141  * to max number of VCPUs in guest. */
 142 #define MAX_TAP_QUEUES 256
 143 #define MAX_TAP_FLOWS  4096
 144
 145 #define TUN_FLOW_EXPIRE (3 * HZ)
 146
 147 struct tun_pcpu_stats {
 148         u64 rx_packets;
 149         u64 rx_bytes;
 150         u64 tx_packets;
 151         u64 tx_bytes;
 152         struct u64_stats_sync syncp;
 153         u32 rx_dropped;
 154         u32 tx_dropped;
 155         u32 rx_frame_errors;
 156 };
 157
 158 /* A tun_file connects an open character device to a tuntap netdevice. It
 159  * also contains all socket related structures (except sock_fprog and tap_filter)
 160  * to serve as one transmit queue for tuntap device. The sock_fprog and
 161  * tap_filter were kept in tun_struct since they were used for filtering for the
 162  * netdevice not for a specific queue (at least I didn't see the requirement for
 163  * this).
 164  *
 165  * RCU usage:
 166  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 167  * other can only be read while rcu_read_lock or rtnl_lock is held.
 168  */
 169 struct tun_file {
 170         struct sock sk;
 171         struct socket socket;
 172         struct socket_wq wq;
 173         struct tun_struct __rcu *tun;
 174         struct fasync_struct *fasync;
 175         /* only used for fasnyc */
 176         unsigned int flags;
 177         union {
 178                 u16 queue_index;
 179                 unsigned int ifindex;
 180         };
 181         struct napi_struct napi;
 182         bool napi_enabled;
 183         bool napi_frags_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_flow_entry {
 192         struct hlist_node hash_link;
 193         struct rcu_head rcu;
 194         struct tun_struct *tun;
 195
 196         u32 rxhash;
 197         u32 rps_rxhash;
 198         int queue_index;
 199         unsigned long updated;
 200 };
 201
 202 #define TUN_NUM_FLOW_ENTRIES 1024
 203 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 204
 205 struct tun_prog {
 206         struct rcu_head rcu;
 207         struct bpf_prog *prog;
 208 };
 209
 210 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 211  * device, socket filter, sndbuf and vnet header size were restore when the
 212  * file were attached to a persist device.
 213  */
 214 struct tun_struct {
 215         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 216         unsigned int            numqueues;
 217         unsigned int            flags;
 218         kuid_t                  owner;
 219         kgid_t                  group;
 220
 221         struct net_device       *dev;
 222         netdev_features_t       set_features;
 223 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 224                           NETIF_F_TSO6)
 225
 226         int                     align;
 227         int                     vnet_hdr_sz;
 228         int                     sndbuf;
 229         struct tap_filter       txflt;
 230         struct sock_fprog       fprog;
 231         /* protected by rtnl lock */
 232         bool                    filter_attached;
 233 #ifdef TUN_DEBUG
 234         int debug;
 235 #endif
 236         spinlock_t lock;
 237         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 238         struct timer_list flow_gc_timer;
 239         unsigned long ageing_time;
 240         unsigned int numdisabled;
 241         struct list_head disabled;
 242         void *security;
 243         u32 flow_count;
 244         u32 rx_batched;
 245         struct tun_pcpu_stats __percpu *pcpu_stats;
 246         struct bpf_prog __rcu *xdp_prog;
 247         struct tun_prog __rcu *steering_prog;
 248         struct tun_prog __rcu *filter_prog;
 249         struct ethtool_link_ksettings link_ksettings;
 250 };
 251
 252 struct veth {
 253         __be16 h_vlan_proto;
 254         __be16 h_vlan_TCI;
 255 };
 256
 257 bool tun_is_xdp_frame(void *ptr)
 258 {
 259         return (unsigned long)ptr & TUN_XDP_FLAG;
 260 }
 261 EXPORT_SYMBOL(tun_is_xdp_frame);
 262
 263 void *tun_xdp_to_ptr(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_xdp_to_ptr);
 268
 269 void *tun_ptr_to_xdp(void *ptr)
 270 {
 271         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 272 }
 273 EXPORT_SYMBOL(tun_ptr_to_xdp);
 274
 275 static int tun_napi_receive(struct napi_struct *napi, int budget)
 276 {
 277         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 278         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 279         struct sk_buff_head process_queue;
 280         struct sk_buff *skb;
 281         int received = 0;
 282
 283         __skb_queue_head_init(&process_queue);
 284
 285         spin_lock(&queue->lock);
 286         skb_queue_splice_tail_init(queue, &process_queue);
 287         spin_unlock(&queue->lock);
 288
 289         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 290                 napi_gro_receive(napi, skb);
 291                 ++received;
 292         }
 293
 294         if (!skb_queue_empty(&process_queue)) {
 295                 spin_lock(&queue->lock);
 296                 skb_queue_splice(&process_queue, queue);
 297                 spin_unlock(&queue->lock);
 298         }
 299
 300         return received;
 301 }
 302
 303 static int tun_napi_poll(struct napi_struct *napi, int budget)
 304 {
 305         unsigned int received;
 306
 307         received = tun_napi_receive(napi, budget);
 308
 309         if (received < budget)
 310                 napi_complete_done(napi, received);
 311
 312         return received;
 313 }
 314
 315 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 316                           bool napi_en, bool napi_frags)
 317 {
 318         tfile->napi_enabled = napi_en;
 319         tfile->napi_frags_enabled = napi_en && napi_frags;
 320         if (napi_en) {
 321                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 322                                NAPI_POLL_WEIGHT);
 323                 napi_enable(&tfile->napi);
 324         }
 325 }
 326
 327 static void tun_napi_disable(struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 napi_disable(&tfile->napi);
 331 }
 332
 333 static void tun_napi_del(struct tun_file *tfile)
 334 {
 335         if (tfile->napi_enabled)
 336                 netif_napi_del(&tfile->napi);
 337 }
 338
 339 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 340 {
 341         return tfile->napi_frags_enabled;
 342 }
 343
 344 #ifdef CONFIG_TUN_VNET_CROSS_LE
 345 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 346 {
 347         return tun->flags & TUN_VNET_BE ? false :
 348                 virtio_legacy_is_little_endian();
 349 }
 350
 351 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 352 {
 353         int be = !!(tun->flags & TUN_VNET_BE);
 354
 355         if (put_user(be, argp))
 356                 return -EFAULT;
 357
 358         return 0;
 359 }
 360
 361 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 362 {
 363         int be;
 364
 365         if (get_user(be, argp))
 366                 return -EFAULT;
 367
 368         if (be)
 369                 tun->flags |= TUN_VNET_BE;
 370         else
 371                 tun->flags &= ~TUN_VNET_BE;
 372
 373         return 0;
 374 }
 375 #else
 376 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 377 {
 378         return virtio_legacy_is_little_endian();
 379 }
 380
 381 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 382 {
 383         return -EINVAL;
 384 }
 385
 386 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 387 {
 388         return -EINVAL;
 389 }
 390 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 391
 392 static inline bool tun_is_little_endian(struct tun_struct *tun)
 393 {
 394         return tun->flags & TUN_VNET_LE ||
 395                 tun_legacy_is_little_endian(tun);
 396 }
 397
 398 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 399 {
 400         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 401 }
 402
 403 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 404 {
 405         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 406 }
 407
 408 static inline u32 tun_hashfn(u32 rxhash)
 409 {
 410         return rxhash & TUN_MASK_FLOW_ENTRIES;
 411 }
 412
 413 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 414 {
 415         struct tun_flow_entry *e;
 416
 417         hlist_for_each_entry_rcu(e, head, hash_link) {
 418                 if (e->rxhash == rxhash)
 419                         return e;
 420         }
 421         return NULL;
 422 }
 423
 424 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 425                                               struct hlist_head *head,
 426                                               u32 rxhash, u16 queue_index)
 427 {
 428         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 429
 430         if (e) {
 431                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 432                           rxhash, queue_index);
 433                 e->updated = jiffies;
 434                 e->rxhash = rxhash;
 435                 e->rps_rxhash = 0;
 436                 e->queue_index = queue_index;
 437                 e->tun = tun;
 438                 hlist_add_head_rcu(&e->hash_link, head);
 439                 ++tun->flow_count;
 440         }
 441         return e;
 442 }
 443
 444 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 445 {
 446         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 447                   e->rxhash, e->queue_index);
 448         hlist_del_rcu(&e->hash_link);
 449         kfree_rcu(e, rcu);
 450         --tun->flow_count;
 451 }
 452
 453 static void tun_flow_flush(struct tun_struct *tun)
 454 {
 455         int i;
 456
 457         spin_lock_bh(&tun->lock);
 458         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 459                 struct tun_flow_entry *e;
 460                 struct hlist_node *n;
 461
 462                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 463                         tun_flow_delete(tun, e);
 464         }
 465         spin_unlock_bh(&tun->lock);
 466 }
 467
 468 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 469 {
 470         int i;
 471
 472         spin_lock_bh(&tun->lock);
 473         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 474                 struct tun_flow_entry *e;
 475                 struct hlist_node *n;
 476
 477                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 478                         if (e->queue_index == queue_index)
 479                                 tun_flow_delete(tun, e);
 480                 }
 481         }
 482         spin_unlock_bh(&tun->lock);
 483 }
 484
 485 static void tun_flow_cleanup(struct timer_list *t)
 486 {
 487         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 488         unsigned long delay = tun->ageing_time;
 489         unsigned long next_timer = jiffies + delay;
 490         unsigned long count = 0;
 491         int i;
 492
 493         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 494
 495         spin_lock(&tun->lock);
 496         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 497                 struct tun_flow_entry *e;
 498                 struct hlist_node *n;
 499
 500                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 501                         unsigned long this_timer;
 502
 503                         this_timer = e->updated + delay;
 504                         if (time_before_eq(this_timer, jiffies)) {
 505                                 tun_flow_delete(tun, e);
 506                                 continue;
 507                         }
 508                         count++;
 509                         if (time_before(this_timer, next_timer))
 510                                 next_timer = this_timer;
 511                 }
 512         }
 513
 514         if (count)
 515                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 516         spin_unlock(&tun->lock);
 517 }
 518
 519 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 520                             struct tun_file *tfile)
 521 {
 522         struct hlist_head *head;
 523         struct tun_flow_entry *e;
 524         unsigned long delay = tun->ageing_time;
 525         u16 queue_index = tfile->queue_index;
 526
 527         if (!rxhash)
 528                 return;
 529         else
 530                 head = &tun->flows[tun_hashfn(rxhash)];
 531
 532         rcu_read_lock();
 533
 534         e = tun_flow_find(head, rxhash);
 535         if (likely(e)) {
 536                 /* TODO: keep queueing to old queue until it's empty? */
 537                 e->queue_index = queue_index;
 538                 e->updated = jiffies;
 539                 sock_rps_record_flow_hash(e->rps_rxhash);
 540         } else {
 541                 spin_lock_bh(&tun->lock);
 542                 if (!tun_flow_find(head, rxhash) &&
 543                     tun->flow_count < MAX_TAP_FLOWS)
 544                         tun_flow_create(tun, head, rxhash, queue_index);
 545
 546                 if (!timer_pending(&tun->flow_gc_timer))
 547                         mod_timer(&tun->flow_gc_timer,
 548                                   round_jiffies_up(jiffies + delay));
 549                 spin_unlock_bh(&tun->lock);
 550         }
 551
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here.
 570  */
 571 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 572 {
 573         struct tun_flow_entry *e;
 574         u32 txq = 0;
 575         u32 numqueues = 0;
 576
 577         numqueues = READ_ONCE(tun->numqueues);
 578
 579         txq = __skb_get_hash_symmetric(skb);
 580         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 581         if (e) {
 582                 tun_flow_save_rps_rxhash(e, txq);
 583                 txq = e->queue_index;
 584         } else {
 585                 /* use multiply and shift instead of expensive divide */
 586                 txq = ((u64)txq * numqueues) >> 32;
 587         }
 588
 589         return txq;
 590 }
 591
 592 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 593 {
 594         struct tun_prog *prog;
 595         u16 ret = 0;
 596
 597         prog = rcu_dereference(tun->steering_prog);
 598         if (prog)
 599                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 600
 601         return ret % tun->numqueues;
 602 }
 603
 604 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 605                             struct net_device *sb_dev,
 606                             select_queue_fallback_t fallback)
 607 {
 608         struct tun_struct *tun = netdev_priv(dev);
 609         u16 ret;
 610
 611         rcu_read_lock();
 612         if (rcu_dereference(tun->steering_prog))
 613                 ret = tun_ebpf_select_queue(tun, skb);
 614         else
 615                 ret = tun_automq_select_queue(tun, skb);
 616         rcu_read_unlock();
 617
 618         return ret;
 619 }
 620
 621 static inline bool tun_not_capable(struct tun_struct *tun)
 622 {
 623         const struct cred *cred = current_cred();
 624         struct net *net = dev_net(tun->dev);
 625
 626         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 627                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 628                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 629 }
 630
 631 static void tun_set_real_num_queues(struct tun_struct *tun)
 632 {
 633         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 634         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 635 }
 636
 637 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 638 {
 639         tfile->detached = tun;
 640         list_add_tail(&tfile->next, &tun->disabled);
 641         ++tun->numdisabled;
 642 }
 643
 644 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 645 {
 646         struct tun_struct *tun = tfile->detached;
 647
 648         tfile->detached = NULL;
 649         list_del_init(&tfile->next);
 650         --tun->numdisabled;
 651         return tun;
 652 }
 653
 654 void tun_ptr_free(void *ptr)
 655 {
 656         if (!ptr)
 657                 return;
 658         if (tun_is_xdp_frame(ptr)) {
 659                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 660
 661                 xdp_return_frame(xdpf);
 662         } else {
 663                 __skb_array_destroy_skb(ptr);
 664         }
 665 }
 666 EXPORT_SYMBOL_GPL(tun_ptr_free);
 667
 668 static void tun_queue_purge(struct tun_file *tfile)
 669 {
 670         void *ptr;
 671
 672         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 673                 tun_ptr_free(ptr);
 674
 675         skb_queue_purge(&tfile->sk.sk_write_queue);
 676         skb_queue_purge(&tfile->sk.sk_error_queue);
 677 }
 678
 679 static void __tun_detach(struct tun_file *tfile, bool clean)
 680 {
 681         struct tun_file *ntfile;
 682         struct tun_struct *tun;
 683
 684         tun = rtnl_dereference(tfile->tun);
 685
 686         if (tun && clean) {
 687                 tun_napi_disable(tfile);
 688                 tun_napi_del(tfile);
 689         }
 690
 691         if (tun && !tfile->detached) {
 692                 u16 index = tfile->queue_index;
 693                 BUG_ON(index >= tun->numqueues);
 694
 695                 rcu_assign_pointer(tun->tfiles[index],
 696                                    tun->tfiles[tun->numqueues - 1]);
 697                 ntfile = rtnl_dereference(tun->tfiles[index]);
 698                 ntfile->queue_index = index;
 699
 700                 --tun->numqueues;
 701                 if (clean) {
 702                         RCU_INIT_POINTER(tfile->tun, NULL);
 703                         sock_put(&tfile->sk);
 704                 } else
 705                         tun_disable_queue(tun, tfile);
 706
 707                 synchronize_net();
 708                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 709                 /* Drop read queue */
 710                 tun_queue_purge(tfile);
 711                 tun_set_real_num_queues(tun);
 712         } else if (tfile->detached && clean) {
 713                 tun = tun_enable_queue(tfile);
 714                 sock_put(&tfile->sk);
 715         }
 716
 717         if (clean) {
 718                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 719                         netif_carrier_off(tun->dev);
 720
 721                         if (!(tun->flags & IFF_PERSIST) &&
 722                             tun->dev->reg_state == NETREG_REGISTERED)
 723                                 unregister_netdevice(tun->dev);
 724                 }
 725                 if (tun)
 726                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 727                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 728                 sock_put(&tfile->sk);
 729         }
 730 }
 731
 732 static void tun_detach(struct tun_file *tfile, bool clean)
 733 {
 734         struct tun_struct *tun;
 735         struct net_device *dev;
 736
 737         rtnl_lock();
 738         tun = rtnl_dereference(tfile->tun);
 739         dev = tun ? tun->dev : NULL;
 740         __tun_detach(tfile, clean);
 741         if (dev)
 742                 netdev_state_change(dev);
 743         rtnl_unlock();
 744 }
 745
 746 static void tun_detach_all(struct net_device *dev)
 747 {
 748         struct tun_struct *tun = netdev_priv(dev);
 749         struct tun_file *tfile, *tmp;
 750         int i, n = tun->numqueues;
 751
 752         for (i = 0; i < n; i++) {
 753                 tfile = rtnl_dereference(tun->tfiles[i]);
 754                 BUG_ON(!tfile);
 755                 tun_napi_disable(tfile);
 756                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 757                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 758                 RCU_INIT_POINTER(tfile->tun, NULL);
 759                 --tun->numqueues;
 760         }
 761         list_for_each_entry(tfile, &tun->disabled, next) {
 762                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 763                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 764                 RCU_INIT_POINTER(tfile->tun, NULL);
 765         }
 766         BUG_ON(tun->numqueues != 0);
 767
 768         synchronize_net();
 769         for (i = 0; i < n; i++) {
 770                 tfile = rtnl_dereference(tun->tfiles[i]);
 771                 tun_napi_del(tfile);
 772                 /* Drop read queue */
 773                 tun_queue_purge(tfile);
 774                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 775                 sock_put(&tfile->sk);
 776         }
 777         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 778                 tun_enable_queue(tfile);
 779                 tun_queue_purge(tfile);
 780                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 781                 sock_put(&tfile->sk);
 782         }
 783         BUG_ON(tun->numdisabled != 0);
 784
 785         if (tun->flags & IFF_PERSIST)
 786                 module_put(THIS_MODULE);
 787 }
 788
 789 static int tun_attach(struct tun_struct *tun, struct file *file,
 790                       bool skip_filter, bool napi, bool napi_frags)
 791 {
 792         struct tun_file *tfile = file->private_data;
 793         struct net_device *dev = tun->dev;
 794         int err;
 795
 796         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 797         if (err < 0)
 798                 goto out;
 799
 800         err = -EINVAL;
 801         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 802                 goto out;
 803
 804         err = -EBUSY;
 805         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 806                 goto out;
 807
 808         err = -E2BIG;
 809         if (!tfile->detached &&
 810             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 811                 goto out;
 812
 813         err = 0;
 814
 815         /* Re-attach the filter to persist device */
 816         if (!skip_filter && (tun->filter_attached == true)) {
 817                 lock_sock(tfile->socket.sk);
 818                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 819                 release_sock(tfile->socket.sk);
 820                 if (!err)
 821                         goto out;
 822         }
 823
 824         if (!tfile->detached &&
 825             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 826                             GFP_KERNEL, tun_ptr_free)) {
 827                 err = -ENOMEM;
 828                 goto out;
 829         }
 830
 831         tfile->queue_index = tun->numqueues;
 832         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 833
 834         if (tfile->detached) {
 835                 /* Re-attach detached tfile, updating XDP queue_index */
 836                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 837
 838                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 839                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 840         } else {
 841                 /* Setup XDP RX-queue info, for new tfile getting attached */
 842                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 843                                        tun->dev, tfile->queue_index);
 844                 if (err < 0)
 845                         goto out;
 846                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 847                                                  MEM_TYPE_PAGE_SHARED, NULL);
 848                 if (err < 0) {
 849                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 850                         goto out;
 851                 }
 852                 err = 0;
 853         }
 854
 855         rcu_assign_pointer(tfile->tun, tun);
 856         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 857         tun->numqueues++;
 858
 859         if (tfile->detached) {
 860                 tun_enable_queue(tfile);
 861         } else {
 862                 sock_hold(&tfile->sk);
 863                 tun_napi_init(tun, tfile, napi, napi_frags);
 864         }
 865
 866         if (rtnl_dereference(tun->xdp_prog))
 867                 sock_set_flag(&tfile->sk, SOCK_XDP);
 868
 869         tun_set_real_num_queues(tun);
 870
 871         /* device is allowed to go away first, so no need to hold extra
 872          * refcnt.
 873          */
 874
 875 out:
 876         return err;
 877 }
 878
 879 static struct tun_struct *tun_get(struct tun_file *tfile)
 880 {
 881         struct tun_struct *tun;
 882
 883         rcu_read_lock();
 884         tun = rcu_dereference(tfile->tun);
 885         if (tun)
 886                 dev_hold(tun->dev);
 887         rcu_read_unlock();
 888
 889         return tun;
 890 }
 891
 892 static void tun_put(struct tun_struct *tun)
 893 {
 894         dev_put(tun->dev);
 895 }
 896
 897 /* TAP filtering */
 898 static void addr_hash_set(u32 *mask, const u8 *addr)
 899 {
 900         int n = ether_crc(ETH_ALEN, addr) >> 26;
 901         mask[n >> 5] |= (1 << (n & 31));
 902 }
 903
 904 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 905 {
 906         int n = ether_crc(ETH_ALEN, addr) >> 26;
 907         return mask[n >> 5] & (1 << (n & 31));
 908 }
 909
 910 static int update_filter(struct tap_filter *filter, void __user *arg)
 911 {
 912         struct { u8 u[ETH_ALEN]; } *addr;
 913         struct tun_filter uf;
 914         int err, alen, n, nexact;
 915
 916         if (copy_from_user(&uf, arg, sizeof(uf)))
 917                 return -EFAULT;
 918
 919         if (!uf.count) {
 920                 /* Disabled */
 921                 filter->count = 0;
 922                 return 0;
 923         }
 924
 925         alen = ETH_ALEN * uf.count;
 926         addr = memdup_user(arg + sizeof(uf), alen);
 927         if (IS_ERR(addr))
 928                 return PTR_ERR(addr);
 929
 930         /* The filter is updated without holding any locks. Which is
 931          * perfectly safe. We disable it first and in the worst
 932          * case we'll accept a few undesired packets. */
 933         filter->count = 0;
 934         wmb();
 935
 936         /* Use first set of addresses as an exact filter */
 937         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 938                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 939
 940         nexact = n;
 941
 942         /* Remaining multicast addresses are hashed,
 943          * unicast will leave the filter disabled. */
 944         memset(filter->mask, 0, sizeof(filter->mask));
 945         for (; n < uf.count; n++) {
 946                 if (!is_multicast_ether_addr(addr[n].u)) {
 947                         err = 0; /* no filter */
 948                         goto free_addr;
 949                 }
 950                 addr_hash_set(filter->mask, addr[n].u);
 951         }
 952
 953         /* For ALLMULTI just set the mask to all ones.
 954          * This overrides the mask populated above. */
 955         if ((uf.flags & TUN_FLT_ALLMULTI))
 956                 memset(filter->mask, ~0, sizeof(filter->mask));
 957
 958         /* Now enable the filter */
 959         wmb();
 960         filter->count = nexact;
 961
 962         /* Return the number of exact filters */
 963         err = nexact;
 964 free_addr:
 965         kfree(addr);
 966         return err;
 967 }
 968
 969 /* Returns: 0 - drop, !=0 - accept */
 970 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 971 {
 972         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 973          * at this point. */
 974         struct ethhdr *eh = (struct ethhdr *) skb->data;
 975         int i;
 976
 977         /* Exact match */
 978         for (i = 0; i < filter->count; i++)
 979                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 980                         return 1;
 981
 982         /* Inexact match (multicast only) */
 983         if (is_multicast_ether_addr(eh->h_dest))
 984                 return addr_hash_test(filter->mask, eh->h_dest);
 985
 986         return 0;
 987 }
 988
 989 /*
 990  * Checks whether the packet is accepted or not.
 991  * Returns: 0 - drop, !=0 - accept
 992  */
 993 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 994 {
 995         if (!filter->count)
 996                 return 1;
 997
 998         return run_filter(filter, skb);
 999 }
1000
1001 /* Network device part of the driver */
1002
1003 static const struct ethtool_ops tun_ethtool_ops;
1004
1005 /* Net device detach from fd. */
1006 static void tun_net_uninit(struct net_device *dev)
1007 {
1008         tun_detach_all(dev);
1009 }
1010
1011 /* Net device open. */
1012 static int tun_net_open(struct net_device *dev)
1013 {
1014         struct tun_struct *tun = netdev_priv(dev);
1015         int i;
1016
1017         netif_tx_start_all_queues(dev);
1018
1019         for (i = 0; i < tun->numqueues; i++) {
1020                 struct tun_file *tfile;
1021
1022                 tfile = rtnl_dereference(tun->tfiles[i]);
1023                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1024         }
1025
1026         return 0;
1027 }
1028
1029 /* Net device close. */
1030 static int tun_net_close(struct net_device *dev)
1031 {
1032         netif_tx_stop_all_queues(dev);
1033         return 0;
1034 }
1035
1036 /* Net device start xmit */
1037 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1038 {
1039 #ifdef CONFIG_RPS
1040         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1041                 /* Select queue was not called for the skbuff, so we extract the
1042                  * RPS hash and save it into the flow_table here.
1043                  */
1044                 struct tun_flow_entry *e;
1045                 __u32 rxhash;
1046
1047                 rxhash = __skb_get_hash_symmetric(skb);
1048                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1049                 if (e)
1050                         tun_flow_save_rps_rxhash(e, rxhash);
1051         }
1052 #endif
1053 }
1054
1055 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1056                                     struct sk_buff *skb,
1057                                     int len)
1058 {
1059         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1060
1061         if (prog)
1062                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1063
1064         return len;
1065 }
1066
1067 /* Net device start xmit */
1068 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1069 {
1070         struct tun_struct *tun = netdev_priv(dev);
1071         int txq = skb->queue_mapping;
1072         struct tun_file *tfile;
1073         int len = skb->len;
1074
1075         rcu_read_lock();
1076         tfile = rcu_dereference(tun->tfiles[txq]);
1077
1078         /* Drop packet if interface is not attached */
1079         if (txq >= tun->numqueues)
1080                 goto drop;
1081
1082         if (!rcu_dereference(tun->steering_prog))
1083                 tun_automq_xmit(tun, skb);
1084
1085         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1086
1087         BUG_ON(!tfile);
1088
1089         /* Drop if the filter does not like it.
1090          * This is a noop if the filter is disabled.
1091          * Filter can be enabled only for the TAP devices. */
1092         if (!check_filter(&tun->txflt, skb))
1093                 goto drop;
1094
1095         if (tfile->socket.sk->sk_filter &&
1096             sk_filter(tfile->socket.sk, skb))
1097                 goto drop;
1098
1099         len = run_ebpf_filter(tun, skb, len);
1100         if (len == 0 || pskb_trim(skb, len))
1101                 goto drop;
1102
1103         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1104                 goto drop;
1105
1106         skb_tx_timestamp(skb);
1107
1108         /* Orphan the skb - required as we might hang on to it
1109          * for indefinite time.
1110          */
1111         skb_orphan(skb);
1112
1113         nf_reset(skb);
1114
1115         if (ptr_ring_produce(&tfile->tx_ring, skb))
1116                 goto drop;
1117
1118         /* Notify and wake up reader process */
1119         if (tfile->flags & TUN_FASYNC)
1120                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1121         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1122
1123         rcu_read_unlock();
1124         return NETDEV_TX_OK;
1125
1126 drop:
1127         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1128         skb_tx_error(skb);
1129         kfree_skb(skb);
1130         rcu_read_unlock();
1131         return NET_XMIT_DROP;
1132 }
1133
1134 static void tun_net_mclist(struct net_device *dev)
1135 {
1136         /*
1137          * This callback is supposed to deal with mc filter in
1138          * _rx_ path and has nothing to do with the _tx_ path.
1139          * In rx path we always accept everything userspace gives us.
1140          */
1141 }
1142
1143 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1144         netdev_features_t features)
1145 {
1146         struct tun_struct *tun = netdev_priv(dev);
1147
1148         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1149 }
1150
1151 static void tun_set_headroom(struct net_device *dev, int new_hr)
1152 {
1153         struct tun_struct *tun = netdev_priv(dev);
1154
1155         if (new_hr < NET_SKB_PAD)
1156                 new_hr = NET_SKB_PAD;
1157
1158         tun->align = new_hr;
1159 }
1160
1161 static void
1162 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1163 {
1164         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1165         struct tun_struct *tun = netdev_priv(dev);
1166         struct tun_pcpu_stats *p;
1167         int i;
1168
1169         for_each_possible_cpu(i) {
1170                 u64 rxpackets, rxbytes, txpackets, txbytes;
1171                 unsigned int start;
1172
1173                 p = per_cpu_ptr(tun->pcpu_stats, i);
1174                 do {
1175                         start = u64_stats_fetch_begin(&p->syncp);
1176                         rxpackets       = p->rx_packets;
1177                         rxbytes         = p->rx_bytes;
1178                         txpackets       = p->tx_packets;
1179                         txbytes         = p->tx_bytes;
1180                 } while (u64_stats_fetch_retry(&p->syncp, start));
1181
1182                 stats->rx_packets       += rxpackets;
1183                 stats->rx_bytes         += rxbytes;
1184                 stats->tx_packets       += txpackets;
1185                 stats->tx_bytes         += txbytes;
1186
1187                 /* u32 counters */
1188                 rx_dropped      += p->rx_dropped;
1189                 rx_frame_errors += p->rx_frame_errors;
1190                 tx_dropped      += p->tx_dropped;
1191         }
1192         stats->rx_dropped  = rx_dropped;
1193         stats->rx_frame_errors = rx_frame_errors;
1194         stats->tx_dropped = tx_dropped;
1195 }
1196
1197 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1198                        struct netlink_ext_ack *extack)
1199 {
1200         struct tun_struct *tun = netdev_priv(dev);
1201         struct tun_file *tfile;
1202         struct bpf_prog *old_prog;
1203         int i;
1204
1205         old_prog = rtnl_dereference(tun->xdp_prog);
1206         rcu_assign_pointer(tun->xdp_prog, prog);
1207         if (old_prog)
1208                 bpf_prog_put(old_prog);
1209
1210         for (i = 0; i < tun->numqueues; i++) {
1211                 tfile = rtnl_dereference(tun->tfiles[i]);
1212                 if (prog)
1213                         sock_set_flag(&tfile->sk, SOCK_XDP);
1214                 else
1215                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1216         }
1217         list_for_each_entry(tfile, &tun->disabled, next) {
1218                 if (prog)
1219                         sock_set_flag(&tfile->sk, SOCK_XDP);
1220                 else
1221                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1222         }
1223
1224         return 0;
1225 }
1226
1227 static u32 tun_xdp_query(struct net_device *dev)
1228 {
1229         struct tun_struct *tun = netdev_priv(dev);
1230         const struct bpf_prog *xdp_prog;
1231
1232         xdp_prog = rtnl_dereference(tun->xdp_prog);
1233         if (xdp_prog)
1234                 return xdp_prog->aux->id;
1235
1236         return 0;
1237 }
1238
1239 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1240 {
1241         switch (xdp->command) {
1242         case XDP_SETUP_PROG:
1243                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1244         case XDP_QUERY_PROG:
1245                 xdp->prog_id = tun_xdp_query(dev);
1246                 return 0;
1247         default:
1248                 return -EINVAL;
1249         }
1250 }
1251
1252 static const struct net_device_ops tun_netdev_ops = {
1253         .ndo_uninit             = tun_net_uninit,
1254         .ndo_open               = tun_net_open,
1255         .ndo_stop               = tun_net_close,
1256         .ndo_start_xmit         = tun_net_xmit,
1257         .ndo_fix_features       = tun_net_fix_features,
1258         .ndo_select_queue       = tun_select_queue,
1259         .ndo_set_rx_headroom    = tun_set_headroom,
1260         .ndo_get_stats64        = tun_net_get_stats64,
1261 };
1262
1263 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1264 {
1265         /* Notify and wake up reader process */
1266         if (tfile->flags & TUN_FASYNC)
1267                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1268         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1269 }
1270
1271 static int tun_xdp_xmit(struct net_device *dev, int n,
1272                         struct xdp_frame **frames, u32 flags)
1273 {
1274         struct tun_struct *tun = netdev_priv(dev);
1275         struct tun_file *tfile;
1276         u32 numqueues;
1277         int drops = 0;
1278         int cnt = n;
1279         int i;
1280
1281         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1282                 return -EINVAL;
1283
1284         rcu_read_lock();
1285
1286         numqueues = READ_ONCE(tun->numqueues);
1287         if (!numqueues) {
1288                 rcu_read_unlock();
1289                 return -ENXIO; /* Caller will free/return all frames */
1290         }
1291
1292         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1293                                             numqueues]);
1294
1295         spin_lock(&tfile->tx_ring.producer_lock);
1296         for (i = 0; i < n; i++) {
1297                 struct xdp_frame *xdp = frames[i];
1298                 /* Encode the XDP flag into lowest bit for consumer to differ
1299                  * XDP buffer from sk_buff.
1300                  */
1301                 void *frame = tun_xdp_to_ptr(xdp);
1302
1303                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1304                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1305                         xdp_return_frame_rx_napi(xdp);
1306                         drops++;
1307                 }
1308         }
1309         spin_unlock(&tfile->tx_ring.producer_lock);
1310
1311         if (flags & XDP_XMIT_FLUSH)
1312                 __tun_xdp_flush_tfile(tfile);
1313
1314         rcu_read_unlock();
1315         return cnt - drops;
1316 }
1317
1318 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1319 {
1320         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1321
1322         if (unlikely(!frame))
1323                 return -EOVERFLOW;
1324
1325         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1326 }
1327
1328 static const struct net_device_ops tap_netdev_ops = {
1329         .ndo_uninit             = tun_net_uninit,
1330         .ndo_open               = tun_net_open,
1331         .ndo_stop               = tun_net_close,
1332         .ndo_start_xmit         = tun_net_xmit,
1333         .ndo_fix_features       = tun_net_fix_features,
1334         .ndo_set_rx_mode        = tun_net_mclist,
1335         .ndo_set_mac_address    = eth_mac_addr,
1336         .ndo_validate_addr      = eth_validate_addr,
1337         .ndo_select_queue       = tun_select_queue,
1338         .ndo_features_check     = passthru_features_check,
1339         .ndo_set_rx_headroom    = tun_set_headroom,
1340         .ndo_get_stats64        = tun_net_get_stats64,
1341         .ndo_bpf                = tun_xdp,
1342         .ndo_xdp_xmit           = tun_xdp_xmit,
1343 };
1344
1345 static void tun_flow_init(struct tun_struct *tun)
1346 {
1347         int i;
1348
1349         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1350                 INIT_HLIST_HEAD(&tun->flows[i]);
1351
1352         tun->ageing_time = TUN_FLOW_EXPIRE;
1353         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1354         mod_timer(&tun->flow_gc_timer,
1355                   round_jiffies_up(jiffies + tun->ageing_time));
1356 }
1357
1358 static void tun_flow_uninit(struct tun_struct *tun)
1359 {
1360         del_timer_sync(&tun->flow_gc_timer);
1361         tun_flow_flush(tun);
1362 }
1363
1364 #define MIN_MTU 68
1365 #define MAX_MTU 65535
1366
1367 /* Initialize net device. */
1368 static void tun_net_init(struct net_device *dev)
1369 {
1370         struct tun_struct *tun = netdev_priv(dev);
1371
1372         switch (tun->flags & TUN_TYPE_MASK) {
1373         case IFF_TUN:
1374                 dev->netdev_ops = &tun_netdev_ops;
1375
1376                 /* Point-to-Point TUN Device */
1377                 dev->hard_header_len = 0;
1378                 dev->addr_len = 0;
1379                 dev->mtu = 1500;
1380
1381                 /* Zero header length */
1382                 dev->type = ARPHRD_NONE;
1383                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1384                 break;
1385
1386         case IFF_TAP:
1387                 dev->netdev_ops = &tap_netdev_ops;
1388                 /* Ethernet TAP Device */
1389                 ether_setup(dev);
1390                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1391                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1392
1393                 eth_hw_addr_random(dev);
1394
1395                 break;
1396         }
1397
1398         dev->min_mtu = MIN_MTU;
1399         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1400 }
1401
1402 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1403 {
1404         struct sock *sk = tfile->socket.sk;
1405
1406         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1407 }
1408
1409 /* Character device part */
1410
1411 /* Poll */
1412 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1413 {
1414         struct tun_file *tfile = file->private_data;
1415         struct tun_struct *tun = tun_get(tfile);
1416         struct sock *sk;
1417         __poll_t mask = 0;
1418
1419         if (!tun)
1420                 return EPOLLERR;
1421
1422         sk = tfile->socket.sk;
1423
1424         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1425
1426         poll_wait(file, sk_sleep(sk), wait);
1427
1428         if (!ptr_ring_empty(&tfile->tx_ring))
1429                 mask |= EPOLLIN | EPOLLRDNORM;
1430
1431         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1432          * guarantee EPOLLOUT to be raised by either here or
1433          * tun_sock_write_space(). Then process could get notification
1434          * after it writes to a down device and meets -EIO.
1435          */
1436         if (tun_sock_writeable(tun, tfile) ||
1437             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1438              tun_sock_writeable(tun, tfile)))
1439                 mask |= EPOLLOUT | EPOLLWRNORM;
1440
1441         if (tun->dev->reg_state != NETREG_REGISTERED)
1442                 mask = EPOLLERR;
1443
1444         tun_put(tun);
1445         return mask;
1446 }
1447
1448 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1449                                             size_t len,
1450                                             const struct iov_iter *it)
1451 {
1452         struct sk_buff *skb;
1453         size_t linear;
1454         int err;
1455         int i;
1456
1457         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1458                 return ERR_PTR(-ENOMEM);
1459
1460         local_bh_disable();
1461         skb = napi_get_frags(&tfile->napi);
1462         local_bh_enable();
1463         if (!skb)
1464                 return ERR_PTR(-ENOMEM);
1465
1466         linear = iov_iter_single_seg_count(it);
1467         err = __skb_grow(skb, linear);
1468         if (err)
1469                 goto free;
1470
1471         skb->len = len;
1472         skb->data_len = len - linear;
1473         skb->truesize += skb->data_len;
1474
1475         for (i = 1; i < it->nr_segs; i++) {
1476                 struct page_frag *pfrag = &current->task_frag;
1477                 size_t fragsz = it->iov[i].iov_len;
1478
1479                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1480                         err = -EINVAL;
1481                         goto free;
1482                 }
1483
1484                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1485                         err = -ENOMEM;
1486                         goto free;
1487                 }
1488
1489                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1490                                    pfrag->offset, fragsz);
1491                 page_ref_inc(pfrag->page);
1492                 pfrag->offset += fragsz;
1493         }
1494
1495         return skb;
1496 free:
1497         /* frees skb and all frags allocated with napi_alloc_frag() */
1498         napi_free_frags(&tfile->napi);
1499         return ERR_PTR(err);
1500 }
1501
1502 /* prepad is the amount to reserve at front.  len is length after that.
1503  * linear is a hint as to how much to copy (usually headers). */
1504 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1505                                      size_t prepad, size_t len,
1506                                      size_t linear, int noblock)
1507 {
1508         struct sock *sk = tfile->socket.sk;
1509         struct sk_buff *skb;
1510         int err;
1511
1512         /* Under a page?  Don't bother with paged skb. */
1513         if (prepad + len < PAGE_SIZE || !linear)
1514                 linear = len;
1515
1516         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1517                                    &err, 0);
1518         if (!skb)
1519                 return ERR_PTR(err);
1520
1521         skb_reserve(skb, prepad);
1522         skb_put(skb, linear);
1523         skb->data_len = len - linear;
1524         skb->len += len - linear;
1525
1526         return skb;
1527 }
1528
1529 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1530                            struct sk_buff *skb, int more)
1531 {
1532         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1533         struct sk_buff_head process_queue;
1534         u32 rx_batched = tun->rx_batched;
1535         bool rcv = false;
1536
1537         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1538                 local_bh_disable();
1539                 skb_record_rx_queue(skb, tfile->queue_index);
1540                 netif_receive_skb(skb);
1541                 local_bh_enable();
1542                 return;
1543         }
1544
1545         spin_lock(&queue->lock);
1546         if (!more || skb_queue_len(queue) == rx_batched) {
1547                 __skb_queue_head_init(&process_queue);
1548                 skb_queue_splice_tail_init(queue, &process_queue);
1549                 rcv = true;
1550         } else {
1551                 __skb_queue_tail(queue, skb);
1552         }
1553         spin_unlock(&queue->lock);
1554
1555         if (rcv) {
1556                 struct sk_buff *nskb;
1557
1558                 local_bh_disable();
1559                 while ((nskb = __skb_dequeue(&process_queue))) {
1560                         skb_record_rx_queue(nskb, tfile->queue_index);
1561                         netif_receive_skb(nskb);
1562                 }
1563                 skb_record_rx_queue(skb, tfile->queue_index);
1564                 netif_receive_skb(skb);
1565                 local_bh_enable();
1566         }
1567 }
1568
1569 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1570                               int len, int noblock, bool zerocopy)
1571 {
1572         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1573                 return false;
1574
1575         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1576                 return false;
1577
1578         if (!noblock)
1579                 return false;
1580
1581         if (zerocopy)
1582                 return false;
1583
1584         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1585             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1586                 return false;
1587
1588         return true;
1589 }
1590
1591 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1592                                        int buflen, int len, int pad)
1593 {
1594         struct sk_buff *skb = build_skb(buf, buflen);
1595
1596         if (!skb)
1597                 return ERR_PTR(-ENOMEM);
1598
1599         skb_reserve(skb, pad);
1600         skb_put(skb, len);
1601
1602         get_page(alloc_frag->page);
1603         alloc_frag->offset += buflen;
1604
1605         return skb;
1606 }
1607
1608 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1609                        struct xdp_buff *xdp, u32 act)
1610 {
1611         int err;
1612
1613         switch (act) {
1614         case XDP_REDIRECT:
1615                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1616                 if (err)
1617                         return err;
1618                 break;
1619         case XDP_TX:
1620                 err = tun_xdp_tx(tun->dev, xdp);
1621                 if (err < 0)
1622                         return err;
1623                 break;
1624         case XDP_PASS:
1625                 break;
1626         default:
1627                 bpf_warn_invalid_xdp_action(act);
1628                 /* fall through */
1629         case XDP_ABORTED:
1630                 trace_xdp_exception(tun->dev, xdp_prog, act);
1631                 /* fall through */
1632         case XDP_DROP:
1633                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1634                 break;
1635         }
1636
1637         return act;
1638 }
1639
1640 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1641                                      struct tun_file *tfile,
1642                                      struct iov_iter *from,
1643                                      struct virtio_net_hdr *hdr,
1644                                      int len, int *skb_xdp)
1645 {
1646         struct page_frag *alloc_frag = &current->task_frag;
1647         struct bpf_prog *xdp_prog;
1648         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1649         char *buf;
1650         size_t copied;
1651         int pad = TUN_RX_PAD;
1652         int err = 0;
1653
1654         rcu_read_lock();
1655         xdp_prog = rcu_dereference(tun->xdp_prog);
1656         if (xdp_prog)
1657                 pad += XDP_PACKET_HEADROOM;
1658         buflen += SKB_DATA_ALIGN(len + pad);
1659         rcu_read_unlock();
1660
1661         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1662         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1663                 return ERR_PTR(-ENOMEM);
1664
1665         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1666         copied = copy_page_from_iter(alloc_frag->page,
1667                                      alloc_frag->offset + pad,
1668                                      len, from);
1669         if (copied != len)
1670                 return ERR_PTR(-EFAULT);
1671
1672         /* There's a small window that XDP may be set after the check
1673          * of xdp_prog above, this should be rare and for simplicity
1674          * we do XDP on skb in case the headroom is not enough.
1675          */
1676         if (hdr->gso_type || !xdp_prog) {
1677                 *skb_xdp = 1;
1678                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1679         }
1680
1681         *skb_xdp = 0;
1682
1683         local_bh_disable();
1684         rcu_read_lock();
1685         xdp_prog = rcu_dereference(tun->xdp_prog);
1686         if (xdp_prog) {
1687                 struct xdp_buff xdp;
1688                 u32 act;
1689
1690                 xdp.data_hard_start = buf;
1691                 xdp.data = buf + pad;
1692                 xdp_set_data_meta_invalid(&xdp);
1693                 xdp.data_end = xdp.data + len;
1694                 xdp.rxq = &tfile->xdp_rxq;
1695
1696                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1697                 if (act == XDP_REDIRECT || act == XDP_TX) {
1698                         get_page(alloc_frag->page);
1699                         alloc_frag->offset += buflen;
1700                 }
1701                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1702                 if (err < 0)
1703                         goto err_xdp;
1704                 if (err == XDP_REDIRECT)
1705                         xdp_do_flush_map();
1706                 if (err != XDP_PASS)
1707                         goto out;
1708
1709                 pad = xdp.data - xdp.data_hard_start;
1710                 len = xdp.data_end - xdp.data;
1711         }
1712         rcu_read_unlock();
1713         local_bh_enable();
1714
1715         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1716
1717 err_xdp:
1718         put_page(alloc_frag->page);
1719 out:
1720         rcu_read_unlock();
1721         local_bh_enable();
1722         return NULL;
1723 }
1724
1725 /* Get packet from user space buffer */
1726 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1727                             void *msg_control, struct iov_iter *from,
1728                             int noblock, bool more)
1729 {
1730         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1731         struct sk_buff *skb;
1732         size_t total_len = iov_iter_count(from);
1733         size_t len = total_len, align = tun->align, linear;
1734         struct virtio_net_hdr gso = { 0 };
1735         struct tun_pcpu_stats *stats;
1736         int good_linear;
1737         int copylen;
1738         bool zerocopy = false;
1739         int err;
1740         u32 rxhash = 0;
1741         int skb_xdp = 1;
1742         bool frags = tun_napi_frags_enabled(tfile);
1743
1744         if (!(tun->dev->flags & IFF_UP))
1745                 return -EIO;
1746
1747         if (!(tun->flags & IFF_NO_PI)) {
1748                 if (len < sizeof(pi))
1749                         return -EINVAL;
1750                 len -= sizeof(pi);
1751
1752                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1753                         return -EFAULT;
1754         }
1755
1756         if (tun->flags & IFF_VNET_HDR) {
1757                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1758
1759                 if (len < vnet_hdr_sz)
1760                         return -EINVAL;
1761                 len -= vnet_hdr_sz;
1762
1763                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1764                         return -EFAULT;
1765
1766                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1767                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1768                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1769
1770                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1771                         return -EINVAL;
1772                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1773         }
1774
1775         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1776                 align += NET_IP_ALIGN;
1777                 if (unlikely(len < ETH_HLEN ||
1778                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1779                         return -EINVAL;
1780         }
1781
1782         good_linear = SKB_MAX_HEAD(align);
1783
1784         if (msg_control) {
1785                 struct iov_iter i = *from;
1786
1787                 /* There are 256 bytes to be copied in skb, so there is
1788                  * enough room for skb expand head in case it is used.
1789                  * The rest of the buffer is mapped from userspace.
1790                  */
1791                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1792                 if (copylen > good_linear)
1793                         copylen = good_linear;
1794                 linear = copylen;
1795                 iov_iter_advance(&i, copylen);
1796                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1797                         zerocopy = true;
1798         }
1799
1800         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1801                 /* For the packet that is not easy to be processed
1802                  * (e.g gso or jumbo packet), we will do it at after
1803                  * skb was created with generic XDP routine.
1804                  */
1805                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1806                 if (IS_ERR(skb)) {
1807                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1808                         return PTR_ERR(skb);
1809                 }
1810                 if (!skb)
1811                         return total_len;
1812         } else {
1813                 if (!zerocopy) {
1814                         copylen = len;
1815                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1816                                 linear = good_linear;
1817                         else
1818                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1819                 }
1820
1821                 if (frags) {
1822                         mutex_lock(&tfile->napi_mutex);
1823                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1824                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1825                          * If zerocopy is enabled, then this layout will be
1826                          * overwritten by zerocopy_sg_from_iter().
1827                          */
1828                         zerocopy = false;
1829                 } else {
1830                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1831                                             noblock);
1832                 }
1833
1834                 if (IS_ERR(skb)) {
1835                         if (PTR_ERR(skb) != -EAGAIN)
1836                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1837                         if (frags)
1838                                 mutex_unlock(&tfile->napi_mutex);
1839                         return PTR_ERR(skb);
1840                 }
1841
1842                 if (zerocopy)
1843                         err = zerocopy_sg_from_iter(skb, from);
1844                 else
1845                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1846
1847                 if (err) {
1848                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1849                         kfree_skb(skb);
1850                         if (frags) {
1851                                 tfile->napi.skb = NULL;
1852                                 mutex_unlock(&tfile->napi_mutex);
1853                         }
1854
1855                         return -EFAULT;
1856                 }
1857         }
1858
1859         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1860                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1861                 kfree_skb(skb);
1862                 if (frags) {
1863                         tfile->napi.skb = NULL;
1864                         mutex_unlock(&tfile->napi_mutex);
1865                 }
1866
1867                 return -EINVAL;
1868         }
1869
1870         switch (tun->flags & TUN_TYPE_MASK) {
1871         case IFF_TUN:
1872                 if (tun->flags & IFF_NO_PI) {
1873                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1874
1875                         switch (ip_version) {
1876                         case 4:
1877                                 pi.proto = htons(ETH_P_IP);
1878                                 break;
1879                         case 6:
1880                                 pi.proto = htons(ETH_P_IPV6);
1881                                 break;
1882                         default:
1883                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1884                                 kfree_skb(skb);
1885                                 return -EINVAL;
1886                         }
1887                 }
1888
1889                 skb_reset_mac_header(skb);
1890                 skb->protocol = pi.proto;
1891                 skb->dev = tun->dev;
1892                 break;
1893         case IFF_TAP:
1894                 if (!frags)
1895                         skb->protocol = eth_type_trans(skb, tun->dev);
1896                 break;
1897         }
1898
1899         /* copy skb_ubuf_info for callback when skb has no error */
1900         if (zerocopy) {
1901                 skb_shinfo(skb)->destructor_arg = msg_control;
1902                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1903                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1904         } else if (msg_control) {
1905                 struct ubuf_info *uarg = msg_control;
1906                 uarg->callback(uarg, false);
1907         }
1908
1909         skb_reset_network_header(skb);
1910         skb_probe_transport_header(skb, 0);
1911
1912         if (skb_xdp) {
1913                 struct bpf_prog *xdp_prog;
1914                 int ret;
1915
1916                 local_bh_disable();
1917                 rcu_read_lock();
1918                 xdp_prog = rcu_dereference(tun->xdp_prog);
1919                 if (xdp_prog) {
1920                         ret = do_xdp_generic(xdp_prog, skb);
1921                         if (ret != XDP_PASS) {
1922                                 rcu_read_unlock();
1923                                 local_bh_enable();
1924                                 return total_len;
1925                         }
1926                 }
1927                 rcu_read_unlock();
1928                 local_bh_enable();
1929         }
1930
1931         /* Compute the costly rx hash only if needed for flow updates.
1932          * We may get a very small possibility of OOO during switching, not
1933          * worth to optimize.
1934          */
1935         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1936             !tfile->detached)
1937                 rxhash = __skb_get_hash_symmetric(skb);
1938
1939         if (frags) {
1940                 /* Exercise flow dissector code path. */
1941                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1942
1943                 if (unlikely(headlen > skb_headlen(skb))) {
1944                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1945                         napi_free_frags(&tfile->napi);
1946                         mutex_unlock(&tfile->napi_mutex);
1947                         WARN_ON(1);
1948                         return -ENOMEM;
1949                 }
1950
1951                 local_bh_disable();
1952                 napi_gro_frags(&tfile->napi);
1953                 local_bh_enable();
1954                 mutex_unlock(&tfile->napi_mutex);
1955         } else if (tfile->napi_enabled) {
1956                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1957                 int queue_len;
1958
1959                 spin_lock_bh(&queue->lock);
1960                 __skb_queue_tail(queue, skb);
1961                 queue_len = skb_queue_len(queue);
1962                 spin_unlock(&queue->lock);
1963
1964                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1965                         napi_schedule(&tfile->napi);
1966
1967                 local_bh_enable();
1968         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1969                 tun_rx_batched(tun, tfile, skb, more);
1970         } else {
1971                 netif_rx_ni(skb);
1972         }
1973
1974         stats = get_cpu_ptr(tun->pcpu_stats);
1975         u64_stats_update_begin(&stats->syncp);
1976         stats->rx_packets++;
1977         stats->rx_bytes += len;
1978         u64_stats_update_end(&stats->syncp);
1979         put_cpu_ptr(stats);
1980
1981         if (rxhash)
1982                 tun_flow_update(tun, rxhash, tfile);
1983
1984         return total_len;
1985 }
1986
1987 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1988 {
1989         struct file *file = iocb->ki_filp;
1990         struct tun_file *tfile = file->private_data;
1991         struct tun_struct *tun = tun_get(tfile);
1992         ssize_t result;
1993
1994         if (!tun)
1995                 return -EBADFD;
1996
1997         result = tun_get_user(tun, tfile, NULL, from,
1998                               file->f_flags & O_NONBLOCK, false);
1999
2000         tun_put(tun);
2001         return result;
2002 }
2003
2004 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2005                                 struct tun_file *tfile,
2006                                 struct xdp_frame *xdp_frame,
2007                                 struct iov_iter *iter)
2008 {
2009         int vnet_hdr_sz = 0;
2010         size_t size = xdp_frame->len;
2011         struct tun_pcpu_stats *stats;
2012         size_t ret;
2013
2014         if (tun->flags & IFF_VNET_HDR) {
2015                 struct virtio_net_hdr gso = { 0 };
2016
2017                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2018                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2019                         return -EINVAL;
2020                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2021                              sizeof(gso)))
2022                         return -EFAULT;
2023                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2024         }
2025
2026         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2027
2028         stats = get_cpu_ptr(tun->pcpu_stats);
2029         u64_stats_update_begin(&stats->syncp);
2030         stats->tx_packets++;
2031         stats->tx_bytes += ret;
2032         u64_stats_update_end(&stats->syncp);
2033         put_cpu_ptr(tun->pcpu_stats);
2034
2035         return ret;
2036 }
2037
2038 /* Put packet to the user space buffer */
2039 static ssize_t tun_put_user(struct tun_struct *tun,
2040                             struct tun_file *tfile,
2041                             struct sk_buff *skb,
2042                             struct iov_iter *iter)
2043 {
2044         struct tun_pi pi = { 0, skb->protocol };
2045         struct tun_pcpu_stats *stats;
2046         ssize_t total;
2047         int vlan_offset = 0;
2048         int vlan_hlen = 0;
2049         int vnet_hdr_sz = 0;
2050
2051         if (skb_vlan_tag_present(skb))
2052                 vlan_hlen = VLAN_HLEN;
2053
2054         if (tun->flags & IFF_VNET_HDR)
2055                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2056
2057         total = skb->len + vlan_hlen + vnet_hdr_sz;
2058
2059         if (!(tun->flags & IFF_NO_PI)) {
2060                 if (iov_iter_count(iter) < sizeof(pi))
2061                         return -EINVAL;
2062
2063                 total += sizeof(pi);
2064                 if (iov_iter_count(iter) < total) {
2065                         /* Packet will be striped */
2066                         pi.flags |= TUN_PKT_STRIP;
2067                 }
2068
2069                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2070                         return -EFAULT;
2071         }
2072
2073         if (vnet_hdr_sz) {
2074                 struct virtio_net_hdr gso;
2075
2076                 if (iov_iter_count(iter) < vnet_hdr_sz)
2077                         return -EINVAL;
2078
2079                 if (virtio_net_hdr_from_skb(skb, &gso,
2080                                             tun_is_little_endian(tun), true,
2081                                             vlan_hlen)) {
2082                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2083                         pr_err("unexpected GSO type: "
2084                                "0x%x, gso_size %d, hdr_len %d\n",
2085                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2086                                tun16_to_cpu(tun, gso.hdr_len));
2087                         print_hex_dump(KERN_ERR, "tun: ",
2088                                        DUMP_PREFIX_NONE,
2089                                        16, 1, skb->head,
2090                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2091                         WARN_ON_ONCE(1);
2092                         return -EINVAL;
2093                 }
2094
2095                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2096                         return -EFAULT;
2097
2098                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2099         }
2100
2101         if (vlan_hlen) {
2102                 int ret;
2103                 struct veth veth;
2104
2105                 veth.h_vlan_proto = skb->vlan_proto;
2106                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2107
2108                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2109
2110                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2111                 if (ret || !iov_iter_count(iter))
2112                         goto done;
2113
2114                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2115                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2116                         goto done;
2117         }
2118
2119         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2120
2121 done:
2122         /* caller is in process context, */
2123         stats = get_cpu_ptr(tun->pcpu_stats);
2124         u64_stats_update_begin(&stats->syncp);
2125         stats->tx_packets++;
2126         stats->tx_bytes += skb->len + vlan_hlen;
2127         u64_stats_update_end(&stats->syncp);
2128         put_cpu_ptr(tun->pcpu_stats);
2129
2130         return total;
2131 }
2132
2133 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2134 {
2135         DECLARE_WAITQUEUE(wait, current);
2136         void *ptr = NULL;
2137         int error = 0;
2138
2139         ptr = ptr_ring_consume(&tfile->tx_ring);
2140         if (ptr)
2141                 goto out;
2142         if (noblock) {
2143                 error = -EAGAIN;
2144                 goto out;
2145         }
2146
2147         add_wait_queue(&tfile->wq.wait, &wait);
2148         current->state = TASK_INTERRUPTIBLE;
2149
2150         while (1) {
2151                 ptr = ptr_ring_consume(&tfile->tx_ring);
2152                 if (ptr)
2153                         break;
2154                 if (signal_pending(current)) {
2155                         error = -ERESTARTSYS;
2156                         break;
2157                 }
2158                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2159                         error = -EFAULT;
2160                         break;
2161                 }
2162
2163                 schedule();
2164         }
2165
2166         current->state = TASK_RUNNING;
2167         remove_wait_queue(&tfile->wq.wait, &wait);
2168
2169 out:
2170         *err = error;
2171         return ptr;
2172 }
2173
2174 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2175                            struct iov_iter *to,
2176                            int noblock, void *ptr)
2177 {
2178         ssize_t ret;
2179         int err;
2180
2181         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2182
2183         if (!iov_iter_count(to)) {
2184                 tun_ptr_free(ptr);
2185                 return 0;
2186         }
2187
2188         if (!ptr) {
2189                 /* Read frames from ring */
2190                 ptr = tun_ring_recv(tfile, noblock, &err);
2191                 if (!ptr)
2192                         return err;
2193         }
2194
2195         if (tun_is_xdp_frame(ptr)) {
2196                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2197
2198                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2199                 xdp_return_frame(xdpf);
2200         } else {
2201                 struct sk_buff *skb = ptr;
2202
2203                 ret = tun_put_user(tun, tfile, skb, to);
2204                 if (unlikely(ret < 0))
2205                         kfree_skb(skb);
2206                 else
2207                         consume_skb(skb);
2208         }
2209
2210         return ret;
2211 }
2212
2213 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2214 {
2215         struct file *file = iocb->ki_filp;
2216         struct tun_file *tfile = file->private_data;
2217         struct tun_struct *tun = tun_get(tfile);
2218         ssize_t len = iov_iter_count(to), ret;
2219
2220         if (!tun)
2221                 return -EBADFD;
2222         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2223         ret = min_t(ssize_t, ret, len);
2224         if (ret > 0)
2225                 iocb->ki_pos = ret;
2226         tun_put(tun);
2227         return ret;
2228 }
2229
2230 static void tun_prog_free(struct rcu_head *rcu)
2231 {
2232         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2233
2234         bpf_prog_destroy(prog->prog);
2235         kfree(prog);
2236 }
2237
2238 static int __tun_set_ebpf(struct tun_struct *tun,
2239                           struct tun_prog __rcu **prog_p,
2240                           struct bpf_prog *prog)
2241 {
2242         struct tun_prog *old, *new = NULL;
2243
2244         if (prog) {
2245                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2246                 if (!new)
2247                         return -ENOMEM;
2248                 new->prog = prog;
2249         }
2250
2251         spin_lock_bh(&tun->lock);
2252         old = rcu_dereference_protected(*prog_p,
2253                                         lockdep_is_held(&tun->lock));
2254         rcu_assign_pointer(*prog_p, new);
2255         spin_unlock_bh(&tun->lock);
2256
2257         if (old)
2258                 call_rcu(&old->rcu, tun_prog_free);
2259
2260         return 0;
2261 }
2262
2263 static void tun_free_netdev(struct net_device *dev)
2264 {
2265         struct tun_struct *tun = netdev_priv(dev);
2266
2267         BUG_ON(!(list_empty(&tun->disabled)));
2268         free_percpu(tun->pcpu_stats);
2269         tun_flow_uninit(tun);
2270         security_tun_dev_free_security(tun->security);
2271         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2272         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2273 }
2274
2275 static void tun_setup(struct net_device *dev)
2276 {
2277         struct tun_struct *tun = netdev_priv(dev);
2278
2279         tun->owner = INVALID_UID;
2280         tun->group = INVALID_GID;
2281         tun_default_link_ksettings(dev, &tun->link_ksettings);
2282
2283         dev->ethtool_ops = &tun_ethtool_ops;
2284         dev->needs_free_netdev = true;
2285         dev->priv_destructor = tun_free_netdev;
2286         /* We prefer our own queue length */
2287         dev->tx_queue_len = TUN_READQ_SIZE;
2288 }
2289
2290 /* Trivial set of netlink ops to allow deleting tun or tap
2291  * device with netlink.
2292  */
2293 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2294                         struct netlink_ext_ack *extack)
2295 {
2296         if (!data)
2297                 return 0;
2298         return -EINVAL;
2299 }
2300
2301 static size_t tun_get_size(const struct net_device *dev)
2302 {
2303         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2304         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2305
2306         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2307                nla_total_size(sizeof(gid_t)) + /* GROUP */
2308                nla_total_size(sizeof(u8)) + /* TYPE */
2309                nla_total_size(sizeof(u8)) + /* PI */
2310                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2311                nla_total_size(sizeof(u8)) + /* PERSIST */
2312                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2313                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2314                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2315                0;
2316 }
2317
2318 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2319 {
2320         struct tun_struct *tun = netdev_priv(dev);
2321
2322         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2323                 goto nla_put_failure;
2324         if (uid_valid(tun->owner) &&
2325             nla_put_u32(skb, IFLA_TUN_OWNER,
2326                         from_kuid_munged(current_user_ns(), tun->owner)))
2327                 goto nla_put_failure;
2328         if (gid_valid(tun->group) &&
2329             nla_put_u32(skb, IFLA_TUN_GROUP,
2330                         from_kgid_munged(current_user_ns(), tun->group)))
2331                 goto nla_put_failure;
2332         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2333                 goto nla_put_failure;
2334         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2335                 goto nla_put_failure;
2336         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2337                 goto nla_put_failure;
2338         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2339                        !!(tun->flags & IFF_MULTI_QUEUE)))
2340                 goto nla_put_failure;
2341         if (tun->flags & IFF_MULTI_QUEUE) {
2342                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2343                         goto nla_put_failure;
2344                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2345                                 tun->numdisabled))
2346                         goto nla_put_failure;
2347         }
2348
2349         return 0;
2350
2351 nla_put_failure:
2352         return -EMSGSIZE;
2353 }
2354
2355 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2356         .kind           = DRV_NAME,
2357         .priv_size      = sizeof(struct tun_struct),
2358         .setup          = tun_setup,
2359         .validate       = tun_validate,
2360         .get_size       = tun_get_size,
2361         .fill_info      = tun_fill_info,
2362 };
2363
2364 static void tun_sock_write_space(struct sock *sk)
2365 {
2366         struct tun_file *tfile;
2367         wait_queue_head_t *wqueue;
2368
2369         if (!sock_writeable(sk))
2370                 return;
2371
2372         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2373                 return;
2374
2375         wqueue = sk_sleep(sk);
2376         if (wqueue && waitqueue_active(wqueue))
2377                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2378                                                 EPOLLWRNORM | EPOLLWRBAND);
2379
2380         tfile = container_of(sk, struct tun_file, sk);
2381         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2382 }
2383
2384 static int tun_xdp_one(struct tun_struct *tun,
2385                        struct tun_file *tfile,
2386                        struct xdp_buff *xdp, int *flush)
2387 {
2388         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2389         struct virtio_net_hdr *gso = &hdr->gso;
2390         struct tun_pcpu_stats *stats;
2391         struct bpf_prog *xdp_prog;
2392         struct sk_buff *skb = NULL;
2393         u32 rxhash = 0, act;
2394         int buflen = hdr->buflen;
2395         int err = 0;
2396         bool skb_xdp = false;
2397
2398         xdp_prog = rcu_dereference(tun->xdp_prog);
2399         if (xdp_prog) {
2400                 if (gso->gso_type) {
2401                         skb_xdp = true;
2402                         goto build;
2403                 }
2404                 xdp_set_data_meta_invalid(xdp);
2405                 xdp->rxq = &tfile->xdp_rxq;
2406
2407                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2408                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2409                 if (err < 0) {
2410                         put_page(virt_to_head_page(xdp->data));
2411                         return err;
2412                 }
2413
2414                 switch (err) {
2415                 case XDP_REDIRECT:
2416                         *flush = true;
2417                         /* fall through */
2418                 case XDP_TX:
2419                         return 0;
2420                 case XDP_PASS:
2421                         break;
2422                 default:
2423                         put_page(virt_to_head_page(xdp->data));
2424                         return 0;
2425                 }
2426         }
2427
2428 build:
2429         skb = build_skb(xdp->data_hard_start, buflen);
2430         if (!skb) {
2431                 err = -ENOMEM;
2432                 goto out;
2433         }
2434
2435         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2436         skb_put(skb, xdp->data_end - xdp->data);
2437
2438         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2439                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2440                 kfree_skb(skb);
2441                 err = -EINVAL;
2442                 goto out;
2443         }
2444
2445         skb->protocol = eth_type_trans(skb, tun->dev);
2446         skb_reset_network_header(skb);
2447         skb_probe_transport_header(skb, 0);
2448
2449         if (skb_xdp) {
2450                 err = do_xdp_generic(xdp_prog, skb);
2451                 if (err != XDP_PASS)
2452                         goto out;
2453         }
2454
2455         if (!rcu_dereference(tun->steering_prog))
2456                 rxhash = __skb_get_hash_symmetric(skb);
2457
2458         skb_record_rx_queue(skb, tfile->queue_index);
2459         netif_receive_skb(skb);
2460
2461         stats = get_cpu_ptr(tun->pcpu_stats);
2462         u64_stats_update_begin(&stats->syncp);
2463         stats->rx_packets++;
2464         stats->rx_bytes += skb->len;
2465         u64_stats_update_end(&stats->syncp);
2466         put_cpu_ptr(stats);
2467
2468         if (rxhash)
2469                 tun_flow_update(tun, rxhash, tfile);
2470
2471 out:
2472         return err;
2473 }
2474
2475 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2476 {
2477         int ret, i;
2478         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2479         struct tun_struct *tun = tun_get(tfile);
2480         struct tun_msg_ctl *ctl = m->msg_control;
2481         struct xdp_buff *xdp;
2482
2483         if (!tun)
2484                 return -EBADFD;
2485
2486         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2487                 int n = ctl->num;
2488                 int flush = 0;
2489
2490                 local_bh_disable();
2491                 rcu_read_lock();
2492
2493                 for (i = 0; i < n; i++) {
2494                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2495                         tun_xdp_one(tun, tfile, xdp, &flush);
2496                 }
2497
2498                 if (flush)
2499                         xdp_do_flush_map();
2500
2501                 rcu_read_unlock();
2502                 local_bh_enable();
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);
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                 err = dev_get_valid_name(net, dev, name);
2737                 if (err < 0)
2738                         goto err_free_dev;
2739
2740                 dev_net_set(dev, net);
2741                 dev->rtnl_link_ops = &tun_link_ops;
2742                 dev->ifindex = tfile->ifindex;
2743                 dev->sysfs_groups[0] = &tun_attr_group;
2744
2745                 tun = netdev_priv(dev);
2746                 tun->dev = dev;
2747                 tun->flags = flags;
2748                 tun->txflt.count = 0;
2749                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2750
2751                 tun->align = NET_SKB_PAD;
2752                 tun->filter_attached = false;
2753                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2754                 tun->rx_batched = 0;
2755                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2756
2757                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2758                 if (!tun->pcpu_stats) {
2759                         err = -ENOMEM;
2760                         goto err_free_dev;
2761                 }
2762
2763                 spin_lock_init(&tun->lock);
2764
2765                 err = security_tun_dev_alloc_security(&tun->security);
2766                 if (err < 0)
2767                         goto err_free_stat;
2768
2769                 tun_net_init(dev);
2770                 tun_flow_init(tun);
2771
2772                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2773                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2774                                    NETIF_F_HW_VLAN_STAG_TX;
2775                 dev->features = dev->hw_features | NETIF_F_LLTX;
2776                 dev->vlan_features = dev->features &
2777                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2778                                        NETIF_F_HW_VLAN_STAG_TX);
2779
2780                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2781                               (ifr->ifr_flags & TUN_FEATURES);
2782
2783                 INIT_LIST_HEAD(&tun->disabled);
2784                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2785                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2786                 if (err < 0)
2787                         goto err_free_flow;
2788
2789                 err = register_netdevice(tun->dev);
2790                 if (err < 0)
2791                         goto err_detach;
2792         }
2793
2794         netif_carrier_on(tun->dev);
2795
2796         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2797
2798         /* Make sure persistent devices do not get stuck in
2799          * xoff state.
2800          */
2801         if (netif_running(tun->dev))
2802                 netif_tx_wake_all_queues(tun->dev);
2803
2804         strcpy(ifr->ifr_name, tun->dev->name);
2805         return 0;
2806
2807 err_detach:
2808         tun_detach_all(dev);
2809         /* register_netdevice() already called tun_free_netdev() */
2810         goto err_free_dev;
2811
2812 err_free_flow:
2813         tun_flow_uninit(tun);
2814         security_tun_dev_free_security(tun->security);
2815 err_free_stat:
2816         free_percpu(tun->pcpu_stats);
2817 err_free_dev:
2818         free_netdev(dev);
2819         return err;
2820 }
2821
2822 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2823                        struct ifreq *ifr)
2824 {
2825         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
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);
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 **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         struct ifreq ifr;
2981         kuid_t owner;
2982         kgid_t group;
2983         int sndbuf;
2984         int vnet_hdr_sz;
2985         unsigned int ifindex;
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         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3051
3052         ret = 0;
3053         switch (cmd) {
3054         case TUNGETIFF:
3055                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3056
3057                 if (tfile->detached)
3058                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3059                 if (!tfile->socket.sk->sk_filter)
3060                         ifr.ifr_flags |= IFF_NOFILTER;
3061
3062                 if (copy_to_user(argp, &ifr, ifreq_len))
3063                         ret = -EFAULT;
3064                 break;
3065
3066         case TUNSETNOCSUM:
3067                 /* Disable/Enable checksum */
3068
3069                 /* [unimplemented] */
3070                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3071                           arg ? "disabled" : "enabled");
3072                 break;
3073
3074         case TUNSETPERSIST:
3075                 /* Disable/Enable persist mode. Keep an extra reference to the
3076                  * module to prevent the module being unprobed.
3077                  */
3078                 if (arg && !(tun->flags & IFF_PERSIST)) {
3079                         tun->flags |= IFF_PERSIST;
3080                         __module_get(THIS_MODULE);
3081                         do_notify = true;
3082                 }
3083                 if (!arg && (tun->flags & IFF_PERSIST)) {
3084                         tun->flags &= ~IFF_PERSIST;
3085                         module_put(THIS_MODULE);
3086                         do_notify = true;
3087                 }
3088
3089                 tun_debug(KERN_INFO, tun, "persist %s\n",
3090                           arg ? "enabled" : "disabled");
3091                 break;
3092
3093         case TUNSETOWNER:
3094                 /* Set owner of the device */
3095                 owner = make_kuid(current_user_ns(), arg);
3096                 if (!uid_valid(owner)) {
3097                         ret = -EINVAL;
3098                         break;
3099                 }
3100                 tun->owner = owner;
3101                 do_notify = true;
3102                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3103                           from_kuid(&init_user_ns, tun->owner));
3104                 break;
3105
3106         case TUNSETGROUP:
3107                 /* Set group of the device */
3108                 group = make_kgid(current_user_ns(), arg);
3109                 if (!gid_valid(group)) {
3110                         ret = -EINVAL;
3111                         break;
3112                 }
3113                 tun->group = group;
3114                 do_notify = true;
3115                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3116                           from_kgid(&init_user_ns, tun->group));
3117                 break;
3118
3119         case TUNSETLINK:
3120                 /* Only allow setting the type when the interface is down */
3121                 if (tun->dev->flags & IFF_UP) {
3122                         tun_debug(KERN_INFO, tun,
3123                                   "Linktype set failed because interface is up\n");
3124                         ret = -EBUSY;
3125                 } else {
3126                         tun->dev->type = (int) arg;
3127                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3128                                   tun->dev->type);
3129                         ret = 0;
3130                 }
3131                 break;
3132
3133 #ifdef TUN_DEBUG
3134         case TUNSETDEBUG:
3135                 tun->debug = arg;
3136                 break;
3137 #endif
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                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3161                           ifr.ifr_hwaddr.sa_data);
3162
3163                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3164                 break;
3165
3166         case TUNGETSNDBUF:
3167                 sndbuf = tfile->socket.sk->sk_sndbuf;
3168                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3169                         ret = -EFAULT;
3170                 break;
3171
3172         case TUNSETSNDBUF:
3173                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3174                         ret = -EFAULT;
3175                         break;
3176                 }
3177                 if (sndbuf <= 0) {
3178                         ret = -EINVAL;
3179                         break;
3180                 }
3181
3182                 tun->sndbuf = sndbuf;
3183                 tun_set_sndbuf(tun);
3184                 break;
3185
3186         case TUNGETVNETHDRSZ:
3187                 vnet_hdr_sz = tun->vnet_hdr_sz;
3188                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3189                         ret = -EFAULT;
3190                 break;
3191
3192         case TUNSETVNETHDRSZ:
3193                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3194                         ret = -EFAULT;
3195                         break;
3196                 }
3197                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3198                         ret = -EINVAL;
3199                         break;
3200                 }
3201
3202                 tun->vnet_hdr_sz = vnet_hdr_sz;
3203                 break;
3204
3205         case TUNGETVNETLE:
3206                 le = !!(tun->flags & TUN_VNET_LE);
3207                 if (put_user(le, (int __user *)argp))
3208                         ret = -EFAULT;
3209                 break;
3210
3211         case TUNSETVNETLE:
3212                 if (get_user(le, (int __user *)argp)) {
3213                         ret = -EFAULT;
3214                         break;
3215                 }
3216                 if (le)
3217                         tun->flags |= TUN_VNET_LE;
3218                 else
3219                         tun->flags &= ~TUN_VNET_LE;
3220                 break;
3221
3222         case TUNGETVNETBE:
3223                 ret = tun_get_vnet_be(tun, argp);
3224                 break;
3225
3226         case TUNSETVNETBE:
3227                 ret = tun_set_vnet_be(tun, argp);
3228                 break;
3229
3230         case TUNATTACHFILTER:
3231                 /* Can be set only for TAPs */
3232                 ret = -EINVAL;
3233                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3234                         break;
3235                 ret = -EFAULT;
3236                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3237                         break;
3238
3239                 ret = tun_attach_filter(tun);
3240                 break;
3241
3242         case TUNDETACHFILTER:
3243                 /* Can be set only for TAPs */
3244                 ret = -EINVAL;
3245                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3246                         break;
3247                 ret = 0;
3248                 tun_detach_filter(tun, tun->numqueues);
3249                 break;
3250
3251         case TUNGETFILTER:
3252                 ret = -EINVAL;
3253                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3254                         break;
3255                 ret = -EFAULT;
3256                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3257                         break;
3258                 ret = 0;
3259                 break;
3260
3261         case TUNSETSTEERINGEBPF:
3262                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3263                 break;
3264
3265         case TUNSETFILTEREBPF:
3266                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3267                 break;
3268
3269         default:
3270                 ret = -EINVAL;
3271                 break;
3272         }
3273
3274         if (do_notify)
3275                 netdev_state_change(tun->dev);
3276
3277 unlock:
3278         rtnl_unlock();
3279         if (tun)
3280                 tun_put(tun);
3281         return ret;
3282 }
3283
3284 static long tun_chr_ioctl(struct file *file,
3285                           unsigned int cmd, unsigned long arg)
3286 {
3287         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3288 }
3289
3290 #ifdef CONFIG_COMPAT
3291 static long tun_chr_compat_ioctl(struct file *file,
3292                          unsigned int cmd, unsigned long arg)
3293 {
3294         switch (cmd) {
3295         case TUNSETIFF:
3296         case TUNGETIFF:
3297         case TUNSETTXFILTER:
3298         case TUNGETSNDBUF:
3299         case TUNSETSNDBUF:
3300         case SIOCGIFHWADDR:
3301         case SIOCSIFHWADDR:
3302                 arg = (unsigned long)compat_ptr(arg);
3303                 break;
3304         default:
3305                 arg = (compat_ulong_t)arg;
3306                 break;
3307         }
3308
3309         /*
3310          * compat_ifreq is shorter than ifreq, so we must not access beyond
3311          * the end of that structure. All fields that are used in this
3312          * driver are compatible though, we don't need to convert the
3313          * contents.
3314          */
3315         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3316 }
3317 #endif /* CONFIG_COMPAT */
3318
3319 static int tun_chr_fasync(int fd, struct file *file, int on)
3320 {
3321         struct tun_file *tfile = file->private_data;
3322         int ret;
3323
3324         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3325                 goto out;
3326
3327         if (on) {
3328                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3329                 tfile->flags |= TUN_FASYNC;
3330         } else
3331                 tfile->flags &= ~TUN_FASYNC;
3332         ret = 0;
3333 out:
3334         return ret;
3335 }
3336
3337 static int tun_chr_open(struct inode *inode, struct file * file)
3338 {
3339         struct net *net = current->nsproxy->net_ns;
3340         struct tun_file *tfile;
3341
3342         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3343
3344         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3345                                             &tun_proto, 0);
3346         if (!tfile)
3347                 return -ENOMEM;
3348         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3349                 sk_free(&tfile->sk);
3350                 return -ENOMEM;
3351         }
3352
3353         mutex_init(&tfile->napi_mutex);
3354         RCU_INIT_POINTER(tfile->tun, NULL);
3355         tfile->flags = 0;
3356         tfile->ifindex = 0;
3357
3358         init_waitqueue_head(&tfile->wq.wait);
3359         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3360
3361         tfile->socket.file = file;
3362         tfile->socket.ops = &tun_socket_ops;
3363
3364         sock_init_data(&tfile->socket, &tfile->sk);
3365
3366         tfile->sk.sk_write_space = tun_sock_write_space;
3367         tfile->sk.sk_sndbuf = INT_MAX;
3368
3369         file->private_data = tfile;
3370         INIT_LIST_HEAD(&tfile->next);
3371
3372         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3373
3374         return 0;
3375 }
3376
3377 static int tun_chr_close(struct inode *inode, struct file *file)
3378 {
3379         struct tun_file *tfile = file->private_data;
3380
3381         tun_detach(tfile, true);
3382
3383         return 0;
3384 }
3385
3386 #ifdef CONFIG_PROC_FS
3387 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3388 {
3389         struct tun_file *tfile = file->private_data;
3390         struct tun_struct *tun;
3391         struct ifreq ifr;
3392
3393         memset(&ifr, 0, sizeof(ifr));
3394
3395         rtnl_lock();
3396         tun = tun_get(tfile);
3397         if (tun)
3398                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3399         rtnl_unlock();
3400
3401         if (tun)
3402                 tun_put(tun);
3403
3404         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3405 }
3406 #endif
3407
3408 static const struct file_operations tun_fops = {
3409         .owner  = THIS_MODULE,
3410         .llseek = no_llseek,
3411         .read_iter  = tun_chr_read_iter,
3412         .write_iter = tun_chr_write_iter,
3413         .poll   = tun_chr_poll,
3414         .unlocked_ioctl = tun_chr_ioctl,
3415 #ifdef CONFIG_COMPAT
3416         .compat_ioctl = tun_chr_compat_ioctl,
3417 #endif
3418         .open   = tun_chr_open,
3419         .release = tun_chr_close,
3420         .fasync = tun_chr_fasync,
3421 #ifdef CONFIG_PROC_FS
3422         .show_fdinfo = tun_chr_show_fdinfo,
3423 #endif
3424 };
3425
3426 static struct miscdevice tun_miscdev = {
3427         .minor = TUN_MINOR,
3428         .name = "tun",
3429         .nodename = "net/tun",
3430         .fops = &tun_fops,
3431 };
3432
3433 /* ethtool interface */
3434
3435 static void tun_default_link_ksettings(struct net_device *dev,
3436                                        struct ethtool_link_ksettings *cmd)
3437 {
3438         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3439         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3440         cmd->base.speed         = SPEED_10;
3441         cmd->base.duplex        = DUPLEX_FULL;
3442         cmd->base.port          = PORT_TP;
3443         cmd->base.phy_address   = 0;
3444         cmd->base.autoneg       = AUTONEG_DISABLE;
3445 }
3446
3447 static int tun_get_link_ksettings(struct net_device *dev,
3448                                   struct ethtool_link_ksettings *cmd)
3449 {
3450         struct tun_struct *tun = netdev_priv(dev);
3451
3452         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3453         return 0;
3454 }
3455
3456 static int tun_set_link_ksettings(struct net_device *dev,
3457                                   const struct ethtool_link_ksettings *cmd)
3458 {
3459         struct tun_struct *tun = netdev_priv(dev);
3460
3461         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3462         return 0;
3463 }
3464
3465 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3466 {
3467         struct tun_struct *tun = netdev_priv(dev);
3468
3469         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3470         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3471
3472         switch (tun->flags & TUN_TYPE_MASK) {
3473         case IFF_TUN:
3474                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3475                 break;
3476         case IFF_TAP:
3477                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3478                 break;
3479         }
3480 }
3481
3482 static u32 tun_get_msglevel(struct net_device *dev)
3483 {
3484 #ifdef TUN_DEBUG
3485         struct tun_struct *tun = netdev_priv(dev);
3486         return tun->debug;
3487 #else
3488         return -EOPNOTSUPP;
3489 #endif
3490 }
3491
3492 static void tun_set_msglevel(struct net_device *dev, u32 value)
3493 {
3494 #ifdef TUN_DEBUG
3495         struct tun_struct *tun = netdev_priv(dev);
3496         tun->debug = value;
3497 #endif
3498 }
3499
3500 static int tun_get_coalesce(struct net_device *dev,
3501                             struct ethtool_coalesce *ec)
3502 {
3503         struct tun_struct *tun = netdev_priv(dev);
3504
3505         ec->rx_max_coalesced_frames = tun->rx_batched;
3506
3507         return 0;
3508 }
3509
3510 static int tun_set_coalesce(struct net_device *dev,
3511                             struct ethtool_coalesce *ec)
3512 {
3513         struct tun_struct *tun = netdev_priv(dev);
3514
3515         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3516                 tun->rx_batched = NAPI_POLL_WEIGHT;
3517         else
3518                 tun->rx_batched = ec->rx_max_coalesced_frames;
3519
3520         return 0;
3521 }
3522
3523 static const struct ethtool_ops tun_ethtool_ops = {
3524         .get_drvinfo    = tun_get_drvinfo,
3525         .get_msglevel   = tun_get_msglevel,
3526         .set_msglevel   = tun_set_msglevel,
3527         .get_link       = ethtool_op_get_link,
3528         .get_ts_info    = ethtool_op_get_ts_info,
3529         .get_coalesce   = tun_get_coalesce,
3530         .set_coalesce   = tun_set_coalesce,
3531         .get_link_ksettings = tun_get_link_ksettings,
3532         .set_link_ksettings = tun_set_link_ksettings,
3533 };
3534
3535 static int tun_queue_resize(struct tun_struct *tun)
3536 {
3537         struct net_device *dev = tun->dev;
3538         struct tun_file *tfile;
3539         struct ptr_ring **rings;
3540         int n = tun->numqueues + tun->numdisabled;
3541         int ret, i;
3542
3543         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3544         if (!rings)
3545                 return -ENOMEM;
3546
3547         for (i = 0; i < tun->numqueues; i++) {
3548                 tfile = rtnl_dereference(tun->tfiles[i]);
3549                 rings[i] = &tfile->tx_ring;
3550         }
3551         list_for_each_entry(tfile, &tun->disabled, next)
3552                 rings[i++] = &tfile->tx_ring;
3553
3554         ret = ptr_ring_resize_multiple(rings, n,
3555                                        dev->tx_queue_len, GFP_KERNEL,
3556                                        tun_ptr_free);
3557
3558         kfree(rings);
3559         return ret;
3560 }
3561
3562 static int tun_device_event(struct notifier_block *unused,
3563                             unsigned long event, void *ptr)
3564 {
3565         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3566         struct tun_struct *tun = netdev_priv(dev);
3567
3568         if (dev->rtnl_link_ops != &tun_link_ops)
3569                 return NOTIFY_DONE;
3570
3571         switch (event) {
3572         case NETDEV_CHANGE_TX_QUEUE_LEN:
3573                 if (tun_queue_resize(tun))
3574                         return NOTIFY_BAD;
3575                 break;
3576         default:
3577                 break;
3578         }
3579
3580         return NOTIFY_DONE;
3581 }
3582
3583 static struct notifier_block tun_notifier_block __read_mostly = {
3584         .notifier_call  = tun_device_event,
3585 };
3586
3587 static int __init tun_init(void)
3588 {
3589         int ret = 0;
3590
3591         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3592
3593         ret = rtnl_link_register(&tun_link_ops);
3594         if (ret) {
3595                 pr_err("Can't register link_ops\n");
3596                 goto err_linkops;
3597         }
3598
3599         ret = misc_register(&tun_miscdev);
3600         if (ret) {
3601                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3602                 goto err_misc;
3603         }
3604
3605         ret = register_netdevice_notifier(&tun_notifier_block);
3606         if (ret) {
3607                 pr_err("Can't register netdevice notifier\n");
3608                 goto err_notifier;
3609         }
3610
3611         return  0;
3612
3613 err_notifier:
3614         misc_deregister(&tun_miscdev);
3615 err_misc:
3616         rtnl_link_unregister(&tun_link_ops);
3617 err_linkops:
3618         return ret;
3619 }
3620
3621 static void tun_cleanup(void)
3622 {
3623         misc_deregister(&tun_miscdev);
3624         rtnl_link_unregister(&tun_link_ops);
3625         unregister_netdevice_notifier(&tun_notifier_block);
3626 }
3627
3628 /* Get an underlying socket object from tun file.  Returns error unless file is
3629  * attached to a device.  The returned object works like a packet socket, it
3630  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3631  * holding a reference to the file for as long as the socket is in use. */
3632 struct socket *tun_get_socket(struct file *file)
3633 {
3634         struct tun_file *tfile;
3635         if (file->f_op != &tun_fops)
3636                 return ERR_PTR(-EINVAL);
3637         tfile = file->private_data;
3638         if (!tfile)
3639                 return ERR_PTR(-EBADFD);
3640         return &tfile->socket;
3641 }
3642 EXPORT_SYMBOL_GPL(tun_get_socket);
3643
3644 struct ptr_ring *tun_get_tx_ring(struct file *file)
3645 {
3646         struct tun_file *tfile;
3647
3648         if (file->f_op != &tun_fops)
3649                 return ERR_PTR(-EINVAL);
3650         tfile = file->private_data;
3651         if (!tfile)
3652                 return ERR_PTR(-EBADFD);
3653         return &tfile->tx_ring;
3654 }
3655 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3656
3657 module_init(tun_init);
3658 module_exit(tun_cleanup);
3659 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3660 MODULE_AUTHOR(DRV_COPYRIGHT);
3661 MODULE_LICENSE("GPL");
3662 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3663 MODULE_ALIAS("devname:net/tun");