drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   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 <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  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 <underley@underley.eu.org>
  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 <maxk@qualcomm.com>"
  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 /* Uncomment to enable debugging */
  85 /* #define TUN_DEBUG 1 */
  86
  87 #ifdef TUN_DEBUG
  88 static int debug;
  89
  90 #define tun_debug(level, tun, fmt, args...)                     \
  91 do {                                                            \
  92         if (tun->debug)                                         \
  93                 netdev_printk(level, tun->dev, fmt, ##args);    \
  94 } while (0)
  95 #define DBG1(level, fmt, args...)                               \
  96 do {                                                            \
  97         if (debug == 2)                                         \
  98                 printk(level fmt, ##args);                      \
  99 } while (0)
 100 #else
 101 #define tun_debug(level, tun, fmt, args...)                     \
 102 do {                                                            \
 103         if (0)                                                  \
 104                 netdev_printk(level, tun->dev, fmt, ##args);    \
 105 } while (0)
 106 #define DBG1(level, fmt, args...)                               \
 107 do {                                                            \
 108         if (0)                                                  \
 109                 printk(level fmt, ##args);                      \
 110 } while (0)
 111 #endif
 112
 113 #define TUN_HEADROOM 256
 114 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 115
 116 /* TUN device flags */
 117
 118 /* IFF_ATTACH_QUEUE is never stored in device flags,
 119  * overload it to mean fasync when stored there.
 120  */
 121 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 122 /* High bits in flags field are unused. */
 123 #define TUN_VNET_LE     0x80000000
 124 #define TUN_VNET_BE     0x40000000
 125
 126 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 127                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 128
 129 #define GOODCOPY_LEN 128
 130
 131 #define FLT_EXACT_COUNT 8
 132 struct tap_filter {
 133         unsigned int    count;    /* Number of addrs. Zero means disabled */
 134         u32             mask[2];  /* Mask of the hashed addrs */
 135         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 136 };
 137
 138 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 139  * to max number of VCPUs in guest. */
 140 #define MAX_TAP_QUEUES 256
 141 #define MAX_TAP_FLOWS  4096
 142
 143 #define TUN_FLOW_EXPIRE (3 * HZ)
 144
 145 struct tun_pcpu_stats {
 146         u64 rx_packets;
 147         u64 rx_bytes;
 148         u64 tx_packets;
 149         u64 tx_bytes;
 150         struct u64_stats_sync syncp;
 151         u32 rx_dropped;
 152         u32 tx_dropped;
 153         u32 rx_frame_errors;
 154 };
 155
 156 /* A tun_file connects an open character device to a tuntap netdevice. It
 157  * also contains all socket related structures (except sock_fprog and tap_filter)
 158  * to serve as one transmit queue for tuntap device. The sock_fprog and
 159  * tap_filter were kept in tun_struct since they were used for filtering for the
 160  * netdevice not for a specific queue (at least I didn't see the requirement for
 161  * this).
 162  *
 163  * RCU usage:
 164  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 165  * other can only be read while rcu_read_lock or rtnl_lock is held.
 166  */
 167 struct tun_file {
 168         struct sock sk;
 169         struct socket socket;
 170         struct socket_wq wq;
 171         struct tun_struct __rcu *tun;
 172         struct fasync_struct *fasync;
 173         /* only used for fasnyc */
 174         unsigned int flags;
 175         union {
 176                 u16 queue_index;
 177                 unsigned int ifindex;
 178         };
 179         struct napi_struct napi;
 180         bool napi_enabled;
 181         struct mutex napi_mutex;        /* Protects access to the above napi */
 182         struct list_head next;
 183         struct tun_struct *detached;
 184         struct ptr_ring tx_ring;
 185         struct xdp_rxq_info xdp_rxq;
 186 };
 187
 188 struct tun_flow_entry {
 189         struct hlist_node hash_link;
 190         struct rcu_head rcu;
 191         struct tun_struct *tun;
 192
 193         u32 rxhash;
 194         u32 rps_rxhash;
 195         int queue_index;
 196         unsigned long updated;
 197 };
 198
 199 #define TUN_NUM_FLOW_ENTRIES 1024
 200
 201 struct tun_prog {
 202         struct rcu_head rcu;
 203         struct bpf_prog *prog;
 204 };
 205
 206 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 207  * device, socket filter, sndbuf and vnet header size were restore when the
 208  * file were attached to a persist device.
 209  */
 210 struct tun_struct {
 211         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 212         unsigned int            numqueues;
 213         unsigned int            flags;
 214         kuid_t                  owner;
 215         kgid_t                  group;
 216
 217         struct net_device       *dev;
 218         netdev_features_t       set_features;
 219 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 220                           NETIF_F_TSO6)
 221
 222         int                     align;
 223         int                     vnet_hdr_sz;
 224         int                     sndbuf;
 225         struct tap_filter       txflt;
 226         struct sock_fprog       fprog;
 227         /* protected by rtnl lock */
 228         bool                    filter_attached;
 229 #ifdef TUN_DEBUG
 230         int debug;
 231 #endif
 232         spinlock_t lock;
 233         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 234         struct timer_list flow_gc_timer;
 235         unsigned long ageing_time;
 236         unsigned int numdisabled;
 237         struct list_head disabled;
 238         void *security;
 239         u32 flow_count;
 240         u32 rx_batched;
 241         struct tun_pcpu_stats __percpu *pcpu_stats;
 242         struct bpf_prog __rcu *xdp_prog;
 243         struct tun_prog __rcu *steering_prog;
 244         struct tun_prog __rcu *filter_prog;
 245 };
 246
 247 struct veth {
 248         __be16 h_vlan_proto;
 249         __be16 h_vlan_TCI;
 250 };
 251
 252 bool tun_is_xdp_frame(void *ptr)
 253 {
 254         return (unsigned long)ptr & TUN_XDP_FLAG;
 255 }
 256 EXPORT_SYMBOL(tun_is_xdp_frame);
 257
 258 void *tun_xdp_to_ptr(void *ptr)
 259 {
 260         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 261 }
 262 EXPORT_SYMBOL(tun_xdp_to_ptr);
 263
 264 void *tun_ptr_to_xdp(void *ptr)
 265 {
 266         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 267 }
 268 EXPORT_SYMBOL(tun_ptr_to_xdp);
 269
 270 static int tun_napi_receive(struct napi_struct *napi, int budget)
 271 {
 272         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 273         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 274         struct sk_buff_head process_queue;
 275         struct sk_buff *skb;
 276         int received = 0;
 277
 278         __skb_queue_head_init(&process_queue);
 279
 280         spin_lock(&queue->lock);
 281         skb_queue_splice_tail_init(queue, &process_queue);
 282         spin_unlock(&queue->lock);
 283
 284         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 285                 napi_gro_receive(napi, skb);
 286                 ++received;
 287         }
 288
 289         if (!skb_queue_empty(&process_queue)) {
 290                 spin_lock(&queue->lock);
 291                 skb_queue_splice(&process_queue, queue);
 292                 spin_unlock(&queue->lock);
 293         }
 294
 295         return received;
 296 }
 297
 298 static int tun_napi_poll(struct napi_struct *napi, int budget)
 299 {
 300         unsigned int received;
 301
 302         received = tun_napi_receive(napi, budget);
 303
 304         if (received < budget)
 305                 napi_complete_done(napi, received);
 306
 307         return received;
 308 }
 309
 310 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 311                           bool napi_en)
 312 {
 313         tfile->napi_enabled = napi_en;
 314         if (napi_en) {
 315                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 316                                NAPI_POLL_WEIGHT);
 317                 napi_enable(&tfile->napi);
 318                 mutex_init(&tfile->napi_mutex);
 319         }
 320 }
 321
 322 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 323 {
 324         if (tfile->napi_enabled)
 325                 napi_disable(&tfile->napi);
 326 }
 327
 328 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 329 {
 330         if (tfile->napi_enabled)
 331                 netif_napi_del(&tfile->napi);
 332 }
 333
 334 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 335 {
 336         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 337 }
 338
 339 #ifdef CONFIG_TUN_VNET_CROSS_LE
 340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 341 {
 342         return tun->flags & TUN_VNET_BE ? false :
 343                 virtio_legacy_is_little_endian();
 344 }
 345
 346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 347 {
 348         int be = !!(tun->flags & TUN_VNET_BE);
 349
 350         if (put_user(be, argp))
 351                 return -EFAULT;
 352
 353         return 0;
 354 }
 355
 356 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 357 {
 358         int be;
 359
 360         if (get_user(be, argp))
 361                 return -EFAULT;
 362
 363         if (be)
 364                 tun->flags |= TUN_VNET_BE;
 365         else
 366                 tun->flags &= ~TUN_VNET_BE;
 367
 368         return 0;
 369 }
 370 #else
 371 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 372 {
 373         return virtio_legacy_is_little_endian();
 374 }
 375
 376 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 377 {
 378         return -EINVAL;
 379 }
 380
 381 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 382 {
 383         return -EINVAL;
 384 }
 385 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 386
 387 static inline bool tun_is_little_endian(struct tun_struct *tun)
 388 {
 389         return tun->flags & TUN_VNET_LE ||
 390                 tun_legacy_is_little_endian(tun);
 391 }
 392
 393 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 394 {
 395         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 396 }
 397
 398 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 399 {
 400         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 401 }
 402
 403 static inline u32 tun_hashfn(u32 rxhash)
 404 {
 405         return rxhash & 0x3ff;
 406 }
 407
 408 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 409 {
 410         struct tun_flow_entry *e;
 411
 412         hlist_for_each_entry_rcu(e, head, hash_link) {
 413                 if (e->rxhash == rxhash)
 414                         return e;
 415         }
 416         return NULL;
 417 }
 418
 419 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 420                                               struct hlist_head *head,
 421                                               u32 rxhash, u16 queue_index)
 422 {
 423         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 424
 425         if (e) {
 426                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 427                           rxhash, queue_index);
 428                 e->updated = jiffies;
 429                 e->rxhash = rxhash;
 430                 e->rps_rxhash = 0;
 431                 e->queue_index = queue_index;
 432                 e->tun = tun;
 433                 hlist_add_head_rcu(&e->hash_link, head);
 434                 ++tun->flow_count;
 435         }
 436         return e;
 437 }
 438
 439 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 440 {
 441         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 442                   e->rxhash, e->queue_index);
 443         hlist_del_rcu(&e->hash_link);
 444         kfree_rcu(e, rcu);
 445         --tun->flow_count;
 446 }
 447
 448 static void tun_flow_flush(struct tun_struct *tun)
 449 {
 450         int i;
 451
 452         spin_lock_bh(&tun->lock);
 453         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 454                 struct tun_flow_entry *e;
 455                 struct hlist_node *n;
 456
 457                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 458                         tun_flow_delete(tun, e);
 459         }
 460         spin_unlock_bh(&tun->lock);
 461 }
 462
 463 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 464 {
 465         int i;
 466
 467         spin_lock_bh(&tun->lock);
 468         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 469                 struct tun_flow_entry *e;
 470                 struct hlist_node *n;
 471
 472                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 473                         if (e->queue_index == queue_index)
 474                                 tun_flow_delete(tun, e);
 475                 }
 476         }
 477         spin_unlock_bh(&tun->lock);
 478 }
 479
 480 static void tun_flow_cleanup(struct timer_list *t)
 481 {
 482         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 483         unsigned long delay = tun->ageing_time;
 484         unsigned long next_timer = jiffies + delay;
 485         unsigned long count = 0;
 486         int i;
 487
 488         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 489
 490         spin_lock(&tun->lock);
 491         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 492                 struct tun_flow_entry *e;
 493                 struct hlist_node *n;
 494
 495                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 496                         unsigned long this_timer;
 497
 498                         this_timer = e->updated + delay;
 499                         if (time_before_eq(this_timer, jiffies)) {
 500                                 tun_flow_delete(tun, e);
 501                                 continue;
 502                         }
 503                         count++;
 504                         if (time_before(this_timer, next_timer))
 505                                 next_timer = this_timer;
 506                 }
 507         }
 508
 509         if (count)
 510                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 511         spin_unlock(&tun->lock);
 512 }
 513
 514 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 515                             struct tun_file *tfile)
 516 {
 517         struct hlist_head *head;
 518         struct tun_flow_entry *e;
 519         unsigned long delay = tun->ageing_time;
 520         u16 queue_index = tfile->queue_index;
 521
 522         if (!rxhash)
 523                 return;
 524         else
 525                 head = &tun->flows[tun_hashfn(rxhash)];
 526
 527         rcu_read_lock();
 528
 529         e = tun_flow_find(head, rxhash);
 530         if (likely(e)) {
 531                 /* TODO: keep queueing to old queue until it's empty? */
 532                 e->queue_index = queue_index;
 533                 e->updated = jiffies;
 534                 sock_rps_record_flow_hash(e->rps_rxhash);
 535         } else {
 536                 spin_lock_bh(&tun->lock);
 537                 if (!tun_flow_find(head, rxhash) &&
 538                     tun->flow_count < MAX_TAP_FLOWS)
 539                         tun_flow_create(tun, head, rxhash, queue_index);
 540
 541                 if (!timer_pending(&tun->flow_gc_timer))
 542                         mod_timer(&tun->flow_gc_timer,
 543                                   round_jiffies_up(jiffies + delay));
 544                 spin_unlock_bh(&tun->lock);
 545         }
 546
 547         rcu_read_unlock();
 548 }
 549
 550 /**
 551  * Save the hash received in the stack receive path and update the
 552  * flow_hash table accordingly.
 553  */
 554 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 555 {
 556         if (unlikely(e->rps_rxhash != hash))
 557                 e->rps_rxhash = hash;
 558 }
 559
 560 /* We try to identify a flow through its rxhash first. The reason that
 561  * we do not check rxq no. is because some cards(e.g 82599), chooses
 562  * the rxq based on the txq where the last packet of the flow comes. As
 563  * the userspace application move between processors, we may get a
 564  * different rxq no. here. If we could not get rxhash, then we would
 565  * hope the rxq no. may help here.
 566  */
 567 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 568 {
 569         struct tun_flow_entry *e;
 570         u32 txq = 0;
 571         u32 numqueues = 0;
 572
 573         numqueues = READ_ONCE(tun->numqueues);
 574
 575         txq = __skb_get_hash_symmetric(skb);
 576         if (txq) {
 577                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 578                 if (e) {
 579                         tun_flow_save_rps_rxhash(e, txq);
 580                         txq = e->queue_index;
 581                 } else
 582                         /* use multiply and shift instead of expensive divide */
 583                         txq = ((u64)txq * numqueues) >> 32;
 584         } else if (likely(skb_rx_queue_recorded(skb))) {
 585                 txq = skb_get_rx_queue(skb);
 586                 while (unlikely(txq >= numqueues))
 587                         txq -= numqueues;
 588         }
 589
 590         return txq;
 591 }
 592
 593 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 594 {
 595         struct tun_prog *prog;
 596         u16 ret = 0;
 597
 598         prog = rcu_dereference(tun->steering_prog);
 599         if (prog)
 600                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 601
 602         return ret % tun->numqueues;
 603 }
 604
 605 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 606                             void *accel_priv, 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(tun, tfile);
 688                 tun_napi_del(tun, 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(tun, 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(tun, 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)
 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);
 864         }
 865
 866         tun_set_real_num_queues(tun);
 867
 868         /* device is allowed to go away first, so no need to hold extra
 869          * refcnt.
 870          */
 871
 872 out:
 873         return err;
 874 }
 875
 876 static struct tun_struct *tun_get(struct tun_file *tfile)
 877 {
 878         struct tun_struct *tun;
 879
 880         rcu_read_lock();
 881         tun = rcu_dereference(tfile->tun);
 882         if (tun)
 883                 dev_hold(tun->dev);
 884         rcu_read_unlock();
 885
 886         return tun;
 887 }
 888
 889 static void tun_put(struct tun_struct *tun)
 890 {
 891         dev_put(tun->dev);
 892 }
 893
 894 /* TAP filtering */
 895 static void addr_hash_set(u32 *mask, const u8 *addr)
 896 {
 897         int n = ether_crc(ETH_ALEN, addr) >> 26;
 898         mask[n >> 5] |= (1 << (n & 31));
 899 }
 900
 901 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 902 {
 903         int n = ether_crc(ETH_ALEN, addr) >> 26;
 904         return mask[n >> 5] & (1 << (n & 31));
 905 }
 906
 907 static int update_filter(struct tap_filter *filter, void __user *arg)
 908 {
 909         struct { u8 u[ETH_ALEN]; } *addr;
 910         struct tun_filter uf;
 911         int err, alen, n, nexact;
 912
 913         if (copy_from_user(&uf, arg, sizeof(uf)))
 914                 return -EFAULT;
 915
 916         if (!uf.count) {
 917                 /* Disabled */
 918                 filter->count = 0;
 919                 return 0;
 920         }
 921
 922         alen = ETH_ALEN * uf.count;
 923         addr = memdup_user(arg + sizeof(uf), alen);
 924         if (IS_ERR(addr))
 925                 return PTR_ERR(addr);
 926
 927         /* The filter is updated without holding any locks. Which is
 928          * perfectly safe. We disable it first and in the worst
 929          * case we'll accept a few undesired packets. */
 930         filter->count = 0;
 931         wmb();
 932
 933         /* Use first set of addresses as an exact filter */
 934         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 935                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 936
 937         nexact = n;
 938
 939         /* Remaining multicast addresses are hashed,
 940          * unicast will leave the filter disabled. */
 941         memset(filter->mask, 0, sizeof(filter->mask));
 942         for (; n < uf.count; n++) {
 943                 if (!is_multicast_ether_addr(addr[n].u)) {
 944                         err = 0; /* no filter */
 945                         goto free_addr;
 946                 }
 947                 addr_hash_set(filter->mask, addr[n].u);
 948         }
 949
 950         /* For ALLMULTI just set the mask to all ones.
 951          * This overrides the mask populated above. */
 952         if ((uf.flags & TUN_FLT_ALLMULTI))
 953                 memset(filter->mask, ~0, sizeof(filter->mask));
 954
 955         /* Now enable the filter */
 956         wmb();
 957         filter->count = nexact;
 958
 959         /* Return the number of exact filters */
 960         err = nexact;
 961 free_addr:
 962         kfree(addr);
 963         return err;
 964 }
 965
 966 /* Returns: 0 - drop, !=0 - accept */
 967 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 968 {
 969         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 970          * at this point. */
 971         struct ethhdr *eh = (struct ethhdr *) skb->data;
 972         int i;
 973
 974         /* Exact match */
 975         for (i = 0; i < filter->count; i++)
 976                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 977                         return 1;
 978
 979         /* Inexact match (multicast only) */
 980         if (is_multicast_ether_addr(eh->h_dest))
 981                 return addr_hash_test(filter->mask, eh->h_dest);
 982
 983         return 0;
 984 }
 985
 986 /*
 987  * Checks whether the packet is accepted or not.
 988  * Returns: 0 - drop, !=0 - accept
 989  */
 990 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 991 {
 992         if (!filter->count)
 993                 return 1;
 994
 995         return run_filter(filter, skb);
 996 }
 997
 998 /* Network device part of the driver */
 999
1000 static const struct ethtool_ops tun_ethtool_ops;
1001
1002 /* Net device detach from fd. */
1003 static void tun_net_uninit(struct net_device *dev)
1004 {
1005         tun_detach_all(dev);
1006 }
1007
1008 /* Net device open. */
1009 static int tun_net_open(struct net_device *dev)
1010 {
1011         struct tun_struct *tun = netdev_priv(dev);
1012         int i;
1013
1014         netif_tx_start_all_queues(dev);
1015
1016         for (i = 0; i < tun->numqueues; i++) {
1017                 struct tun_file *tfile;
1018
1019                 tfile = rtnl_dereference(tun->tfiles[i]);
1020                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1021         }
1022
1023         return 0;
1024 }
1025
1026 /* Net device close. */
1027 static int tun_net_close(struct net_device *dev)
1028 {
1029         netif_tx_stop_all_queues(dev);
1030         return 0;
1031 }
1032
1033 /* Net device start xmit */
1034 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1035 {
1036 #ifdef CONFIG_RPS
1037         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1038                 /* Select queue was not called for the skbuff, so we extract the
1039                  * RPS hash and save it into the flow_table here.
1040                  */
1041                 __u32 rxhash;
1042
1043                 rxhash = __skb_get_hash_symmetric(skb);
1044                 if (rxhash) {
1045                         struct tun_flow_entry *e;
1046                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1047                                         rxhash);
1048                         if (e)
1049                                 tun_flow_save_rps_rxhash(e, rxhash);
1050                 }
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 #ifdef CONFIG_NET_POLL_CONTROLLER
1151 static void tun_poll_controller(struct net_device *dev)
1152 {
1153         /*
1154          * Tun only receives frames when:
1155          * 1) the char device endpoint gets data from user space
1156          * 2) the tun socket gets a sendmsg call from user space
1157          * If NAPI is not enabled, since both of those are synchronous
1158          * operations, we are guaranteed never to have pending data when we poll
1159          * for it so there is nothing to do here but return.
1160          * We need this though so netpoll recognizes us as an interface that
1161          * supports polling, which enables bridge devices in virt setups to
1162          * still use netconsole
1163          * If NAPI is enabled, however, we need to schedule polling for all
1164          * queues unless we are using napi_gro_frags(), which we call in
1165          * process context and not in NAPI context.
1166          */
1167         struct tun_struct *tun = netdev_priv(dev);
1168
1169         if (tun->flags & IFF_NAPI) {
1170                 struct tun_file *tfile;
1171                 int i;
1172
1173                 if (tun_napi_frags_enabled(tun))
1174                         return;
1175
1176                 rcu_read_lock();
1177                 for (i = 0; i < tun->numqueues; i++) {
1178                         tfile = rcu_dereference(tun->tfiles[i]);
1179                         if (tfile->napi_enabled)
1180                                 napi_schedule(&tfile->napi);
1181                 }
1182                 rcu_read_unlock();
1183         }
1184         return;
1185 }
1186 #endif
1187
1188 static void tun_set_headroom(struct net_device *dev, int new_hr)
1189 {
1190         struct tun_struct *tun = netdev_priv(dev);
1191
1192         if (new_hr < NET_SKB_PAD)
1193                 new_hr = NET_SKB_PAD;
1194
1195         tun->align = new_hr;
1196 }
1197
1198 static void
1199 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1200 {
1201         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1202         struct tun_struct *tun = netdev_priv(dev);
1203         struct tun_pcpu_stats *p;
1204         int i;
1205
1206         for_each_possible_cpu(i) {
1207                 u64 rxpackets, rxbytes, txpackets, txbytes;
1208                 unsigned int start;
1209
1210                 p = per_cpu_ptr(tun->pcpu_stats, i);
1211                 do {
1212                         start = u64_stats_fetch_begin(&p->syncp);
1213                         rxpackets       = p->rx_packets;
1214                         rxbytes         = p->rx_bytes;
1215                         txpackets       = p->tx_packets;
1216                         txbytes         = p->tx_bytes;
1217                 } while (u64_stats_fetch_retry(&p->syncp, start));
1218
1219                 stats->rx_packets       += rxpackets;
1220                 stats->rx_bytes         += rxbytes;
1221                 stats->tx_packets       += txpackets;
1222                 stats->tx_bytes         += txbytes;
1223
1224                 /* u32 counters */
1225                 rx_dropped      += p->rx_dropped;
1226                 rx_frame_errors += p->rx_frame_errors;
1227                 tx_dropped      += p->tx_dropped;
1228         }
1229         stats->rx_dropped  = rx_dropped;
1230         stats->rx_frame_errors = rx_frame_errors;
1231         stats->tx_dropped = tx_dropped;
1232 }
1233
1234 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1235                        struct netlink_ext_ack *extack)
1236 {
1237         struct tun_struct *tun = netdev_priv(dev);
1238         struct bpf_prog *old_prog;
1239
1240         old_prog = rtnl_dereference(tun->xdp_prog);
1241         rcu_assign_pointer(tun->xdp_prog, prog);
1242         if (old_prog)
1243                 bpf_prog_put(old_prog);
1244
1245         return 0;
1246 }
1247
1248 static u32 tun_xdp_query(struct net_device *dev)
1249 {
1250         struct tun_struct *tun = netdev_priv(dev);
1251         const struct bpf_prog *xdp_prog;
1252
1253         xdp_prog = rtnl_dereference(tun->xdp_prog);
1254         if (xdp_prog)
1255                 return xdp_prog->aux->id;
1256
1257         return 0;
1258 }
1259
1260 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1261 {
1262         switch (xdp->command) {
1263         case XDP_SETUP_PROG:
1264                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1265         case XDP_QUERY_PROG:
1266                 xdp->prog_id = tun_xdp_query(dev);
1267                 xdp->prog_attached = !!xdp->prog_id;
1268                 return 0;
1269         default:
1270                 return -EINVAL;
1271         }
1272 }
1273
1274 static const struct net_device_ops tun_netdev_ops = {
1275         .ndo_uninit             = tun_net_uninit,
1276         .ndo_open               = tun_net_open,
1277         .ndo_stop               = tun_net_close,
1278         .ndo_start_xmit         = tun_net_xmit,
1279         .ndo_fix_features       = tun_net_fix_features,
1280         .ndo_select_queue       = tun_select_queue,
1281 #ifdef CONFIG_NET_POLL_CONTROLLER
1282         .ndo_poll_controller    = tun_poll_controller,
1283 #endif
1284         .ndo_set_rx_headroom    = tun_set_headroom,
1285         .ndo_get_stats64        = tun_net_get_stats64,
1286 };
1287
1288 static int tun_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames)
1289 {
1290         struct tun_struct *tun = netdev_priv(dev);
1291         struct tun_file *tfile;
1292         u32 numqueues;
1293         int drops = 0;
1294         int cnt = n;
1295         int i;
1296
1297         rcu_read_lock();
1298
1299         numqueues = READ_ONCE(tun->numqueues);
1300         if (!numqueues) {
1301                 rcu_read_unlock();
1302                 return -ENXIO; /* Caller will free/return all frames */
1303         }
1304
1305         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1306                                             numqueues]);
1307
1308         spin_lock(&tfile->tx_ring.producer_lock);
1309         for (i = 0; i < n; i++) {
1310                 struct xdp_frame *xdp = frames[i];
1311                 /* Encode the XDP flag into lowest bit for consumer to differ
1312                  * XDP buffer from sk_buff.
1313                  */
1314                 void *frame = tun_xdp_to_ptr(xdp);
1315
1316                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1317                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1318                         xdp_return_frame_rx_napi(xdp);
1319                         drops++;
1320                 }
1321         }
1322         spin_unlock(&tfile->tx_ring.producer_lock);
1323
1324         rcu_read_unlock();
1325         return cnt - drops;
1326 }
1327
1328 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1329 {
1330         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1331
1332         if (unlikely(!frame))
1333                 return -EOVERFLOW;
1334
1335         return tun_xdp_xmit(dev, 1, &frame);
1336 }
1337
1338 static void tun_xdp_flush(struct net_device *dev)
1339 {
1340         struct tun_struct *tun = netdev_priv(dev);
1341         struct tun_file *tfile;
1342         u32 numqueues;
1343
1344         rcu_read_lock();
1345
1346         numqueues = READ_ONCE(tun->numqueues);
1347         if (!numqueues)
1348                 goto out;
1349
1350         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1351                                             numqueues]);
1352         /* Notify and wake up reader process */
1353         if (tfile->flags & TUN_FASYNC)
1354                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1355         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1356
1357 out:
1358         rcu_read_unlock();
1359 }
1360
1361 static const struct net_device_ops tap_netdev_ops = {
1362         .ndo_uninit             = tun_net_uninit,
1363         .ndo_open               = tun_net_open,
1364         .ndo_stop               = tun_net_close,
1365         .ndo_start_xmit         = tun_net_xmit,
1366         .ndo_fix_features       = tun_net_fix_features,
1367         .ndo_set_rx_mode        = tun_net_mclist,
1368         .ndo_set_mac_address    = eth_mac_addr,
1369         .ndo_validate_addr      = eth_validate_addr,
1370         .ndo_select_queue       = tun_select_queue,
1371 #ifdef CONFIG_NET_POLL_CONTROLLER
1372         .ndo_poll_controller    = tun_poll_controller,
1373 #endif
1374         .ndo_features_check     = passthru_features_check,
1375         .ndo_set_rx_headroom    = tun_set_headroom,
1376         .ndo_get_stats64        = tun_net_get_stats64,
1377         .ndo_bpf                = tun_xdp,
1378         .ndo_xdp_xmit           = tun_xdp_xmit,
1379         .ndo_xdp_flush          = tun_xdp_flush,
1380 };
1381
1382 static void tun_flow_init(struct tun_struct *tun)
1383 {
1384         int i;
1385
1386         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1387                 INIT_HLIST_HEAD(&tun->flows[i]);
1388
1389         tun->ageing_time = TUN_FLOW_EXPIRE;
1390         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1391         mod_timer(&tun->flow_gc_timer,
1392                   round_jiffies_up(jiffies + tun->ageing_time));
1393 }
1394
1395 static void tun_flow_uninit(struct tun_struct *tun)
1396 {
1397         del_timer_sync(&tun->flow_gc_timer);
1398         tun_flow_flush(tun);
1399 }
1400
1401 #define MIN_MTU 68
1402 #define MAX_MTU 65535
1403
1404 /* Initialize net device. */
1405 static void tun_net_init(struct net_device *dev)
1406 {
1407         struct tun_struct *tun = netdev_priv(dev);
1408
1409         switch (tun->flags & TUN_TYPE_MASK) {
1410         case IFF_TUN:
1411                 dev->netdev_ops = &tun_netdev_ops;
1412
1413                 /* Point-to-Point TUN Device */
1414                 dev->hard_header_len = 0;
1415                 dev->addr_len = 0;
1416                 dev->mtu = 1500;
1417
1418                 /* Zero header length */
1419                 dev->type = ARPHRD_NONE;
1420                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1421                 break;
1422
1423         case IFF_TAP:
1424                 dev->netdev_ops = &tap_netdev_ops;
1425                 /* Ethernet TAP Device */
1426                 ether_setup(dev);
1427                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1428                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1429
1430                 eth_hw_addr_random(dev);
1431
1432                 break;
1433         }
1434
1435         dev->min_mtu = MIN_MTU;
1436         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1437 }
1438
1439 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1440 {
1441         struct sock *sk = tfile->socket.sk;
1442
1443         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1444 }
1445
1446 /* Character device part */
1447
1448 /* Poll */
1449 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1450 {
1451         struct tun_file *tfile = file->private_data;
1452         struct tun_struct *tun = tun_get(tfile);
1453         struct sock *sk;
1454         __poll_t mask = 0;
1455
1456         if (!tun)
1457                 return EPOLLERR;
1458
1459         sk = tfile->socket.sk;
1460
1461         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1462
1463         poll_wait(file, sk_sleep(sk), wait);
1464
1465         if (!ptr_ring_empty(&tfile->tx_ring))
1466                 mask |= EPOLLIN | EPOLLRDNORM;
1467
1468         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1469          * guarantee EPOLLOUT to be raised by either here or
1470          * tun_sock_write_space(). Then process could get notification
1471          * after it writes to a down device and meets -EIO.
1472          */
1473         if (tun_sock_writeable(tun, tfile) ||
1474             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1475              tun_sock_writeable(tun, tfile)))
1476                 mask |= EPOLLOUT | EPOLLWRNORM;
1477
1478         if (tun->dev->reg_state != NETREG_REGISTERED)
1479                 mask = EPOLLERR;
1480
1481         tun_put(tun);
1482         return mask;
1483 }
1484
1485 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1486                                             size_t len,
1487                                             const struct iov_iter *it)
1488 {
1489         struct sk_buff *skb;
1490         size_t linear;
1491         int err;
1492         int i;
1493
1494         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1495                 return ERR_PTR(-ENOMEM);
1496
1497         local_bh_disable();
1498         skb = napi_get_frags(&tfile->napi);
1499         local_bh_enable();
1500         if (!skb)
1501                 return ERR_PTR(-ENOMEM);
1502
1503         linear = iov_iter_single_seg_count(it);
1504         err = __skb_grow(skb, linear);
1505         if (err)
1506                 goto free;
1507
1508         skb->len = len;
1509         skb->data_len = len - linear;
1510         skb->truesize += skb->data_len;
1511
1512         for (i = 1; i < it->nr_segs; i++) {
1513                 struct page_frag *pfrag = &current->task_frag;
1514                 size_t fragsz = it->iov[i].iov_len;
1515
1516                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1517                         err = -EINVAL;
1518                         goto free;
1519                 }
1520
1521                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1522                         err = -ENOMEM;
1523                         goto free;
1524                 }
1525
1526                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1527                                    pfrag->offset, fragsz);
1528                 page_ref_inc(pfrag->page);
1529                 pfrag->offset += fragsz;
1530         }
1531
1532         return skb;
1533 free:
1534         /* frees skb and all frags allocated with napi_alloc_frag() */
1535         napi_free_frags(&tfile->napi);
1536         return ERR_PTR(err);
1537 }
1538
1539 /* prepad is the amount to reserve at front.  len is length after that.
1540  * linear is a hint as to how much to copy (usually headers). */
1541 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1542                                      size_t prepad, size_t len,
1543                                      size_t linear, int noblock)
1544 {
1545         struct sock *sk = tfile->socket.sk;
1546         struct sk_buff *skb;
1547         int err;
1548
1549         /* Under a page?  Don't bother with paged skb. */
1550         if (prepad + len < PAGE_SIZE || !linear)
1551                 linear = len;
1552
1553         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1554                                    &err, 0);
1555         if (!skb)
1556                 return ERR_PTR(err);
1557
1558         skb_reserve(skb, prepad);
1559         skb_put(skb, linear);
1560         skb->data_len = len - linear;
1561         skb->len += len - linear;
1562
1563         return skb;
1564 }
1565
1566 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1567                            struct sk_buff *skb, int more)
1568 {
1569         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1570         struct sk_buff_head process_queue;
1571         u32 rx_batched = tun->rx_batched;
1572         bool rcv = false;
1573
1574         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1575                 local_bh_disable();
1576                 netif_receive_skb(skb);
1577                 local_bh_enable();
1578                 return;
1579         }
1580
1581         spin_lock(&queue->lock);
1582         if (!more || skb_queue_len(queue) == rx_batched) {
1583                 __skb_queue_head_init(&process_queue);
1584                 skb_queue_splice_tail_init(queue, &process_queue);
1585                 rcv = true;
1586         } else {
1587                 __skb_queue_tail(queue, skb);
1588         }
1589         spin_unlock(&queue->lock);
1590
1591         if (rcv) {
1592                 struct sk_buff *nskb;
1593
1594                 local_bh_disable();
1595                 while ((nskb = __skb_dequeue(&process_queue)))
1596                         netif_receive_skb(nskb);
1597                 netif_receive_skb(skb);
1598                 local_bh_enable();
1599         }
1600 }
1601
1602 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1603                               int len, int noblock, bool zerocopy)
1604 {
1605         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1606                 return false;
1607
1608         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1609                 return false;
1610
1611         if (!noblock)
1612                 return false;
1613
1614         if (zerocopy)
1615                 return false;
1616
1617         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1618             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1619                 return false;
1620
1621         return true;
1622 }
1623
1624 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1625                                      struct tun_file *tfile,
1626                                      struct iov_iter *from,
1627                                      struct virtio_net_hdr *hdr,
1628                                      int len, int *skb_xdp)
1629 {
1630         struct page_frag *alloc_frag = &current->task_frag;
1631         struct sk_buff *skb;
1632         struct bpf_prog *xdp_prog;
1633         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1634         unsigned int delta = 0;
1635         char *buf;
1636         size_t copied;
1637         int err, pad = TUN_RX_PAD;
1638
1639         rcu_read_lock();
1640         xdp_prog = rcu_dereference(tun->xdp_prog);
1641         if (xdp_prog)
1642                 pad += TUN_HEADROOM;
1643         buflen += SKB_DATA_ALIGN(len + pad);
1644         rcu_read_unlock();
1645
1646         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1647         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1648                 return ERR_PTR(-ENOMEM);
1649
1650         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1651         copied = copy_page_from_iter(alloc_frag->page,
1652                                      alloc_frag->offset + pad,
1653                                      len, from);
1654         if (copied != len)
1655                 return ERR_PTR(-EFAULT);
1656
1657         /* There's a small window that XDP may be set after the check
1658          * of xdp_prog above, this should be rare and for simplicity
1659          * we do XDP on skb in case the headroom is not enough.
1660          */
1661         if (hdr->gso_type || !xdp_prog)
1662                 *skb_xdp = 1;
1663         else
1664                 *skb_xdp = 0;
1665
1666         preempt_disable();
1667         rcu_read_lock();
1668         xdp_prog = rcu_dereference(tun->xdp_prog);
1669         if (xdp_prog && !*skb_xdp) {
1670                 struct xdp_buff xdp;
1671                 void *orig_data;
1672                 u32 act;
1673
1674                 xdp.data_hard_start = buf;
1675                 xdp.data = buf + pad;
1676                 xdp_set_data_meta_invalid(&xdp);
1677                 xdp.data_end = xdp.data + len;
1678                 xdp.rxq = &tfile->xdp_rxq;
1679                 orig_data = xdp.data;
1680                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1681
1682                 switch (act) {
1683                 case XDP_REDIRECT:
1684                         get_page(alloc_frag->page);
1685                         alloc_frag->offset += buflen;
1686                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1687                         xdp_do_flush_map();
1688                         if (err)
1689                                 goto err_redirect;
1690                         rcu_read_unlock();
1691                         preempt_enable();
1692                         return NULL;
1693                 case XDP_TX:
1694                         get_page(alloc_frag->page);
1695                         alloc_frag->offset += buflen;
1696                         if (tun_xdp_tx(tun->dev, &xdp))
1697                                 goto err_redirect;
1698                         tun_xdp_flush(tun->dev);
1699                         rcu_read_unlock();
1700                         preempt_enable();
1701                         return NULL;
1702                 case XDP_PASS:
1703                         delta = orig_data - xdp.data;
1704                         len = xdp.data_end - xdp.data;
1705                         break;
1706                 default:
1707                         bpf_warn_invalid_xdp_action(act);
1708                         /* fall through */
1709                 case XDP_ABORTED:
1710                         trace_xdp_exception(tun->dev, xdp_prog, act);
1711                         /* fall through */
1712                 case XDP_DROP:
1713                         goto err_xdp;
1714                 }
1715         }
1716
1717         skb = build_skb(buf, buflen);
1718         if (!skb) {
1719                 rcu_read_unlock();
1720                 preempt_enable();
1721                 return ERR_PTR(-ENOMEM);
1722         }
1723
1724         skb_reserve(skb, pad - delta);
1725         skb_put(skb, len);
1726         get_page(alloc_frag->page);
1727         alloc_frag->offset += buflen;
1728
1729         rcu_read_unlock();
1730         preempt_enable();
1731
1732         return skb;
1733
1734 err_redirect:
1735         put_page(alloc_frag->page);
1736 err_xdp:
1737         rcu_read_unlock();
1738         preempt_enable();
1739         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1740         return NULL;
1741 }
1742
1743 /* Get packet from user space buffer */
1744 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1745                             void *msg_control, struct iov_iter *from,
1746                             int noblock, bool more)
1747 {
1748         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1749         struct sk_buff *skb;
1750         size_t total_len = iov_iter_count(from);
1751         size_t len = total_len, align = tun->align, linear;
1752         struct virtio_net_hdr gso = { 0 };
1753         struct tun_pcpu_stats *stats;
1754         int good_linear;
1755         int copylen;
1756         bool zerocopy = false;
1757         int err;
1758         u32 rxhash = 0;
1759         int skb_xdp = 1;
1760         bool frags = tun_napi_frags_enabled(tun);
1761
1762         if (!(tun->dev->flags & IFF_UP))
1763                 return -EIO;
1764
1765         if (!(tun->flags & IFF_NO_PI)) {
1766                 if (len < sizeof(pi))
1767                         return -EINVAL;
1768                 len -= sizeof(pi);
1769
1770                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1771                         return -EFAULT;
1772         }
1773
1774         if (tun->flags & IFF_VNET_HDR) {
1775                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1776
1777                 if (len < vnet_hdr_sz)
1778                         return -EINVAL;
1779                 len -= vnet_hdr_sz;
1780
1781                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1782                         return -EFAULT;
1783
1784                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1785                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1786                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1787
1788                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1789                         return -EINVAL;
1790                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1791         }
1792
1793         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1794                 align += NET_IP_ALIGN;
1795                 if (unlikely(len < ETH_HLEN ||
1796                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1797                         return -EINVAL;
1798         }
1799
1800         good_linear = SKB_MAX_HEAD(align);
1801
1802         if (msg_control) {
1803                 struct iov_iter i = *from;
1804
1805                 /* There are 256 bytes to be copied in skb, so there is
1806                  * enough room for skb expand head in case it is used.
1807                  * The rest of the buffer is mapped from userspace.
1808                  */
1809                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1810                 if (copylen > good_linear)
1811                         copylen = good_linear;
1812                 linear = copylen;
1813                 iov_iter_advance(&i, copylen);
1814                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1815                         zerocopy = true;
1816         }
1817
1818         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1819                 /* For the packet that is not easy to be processed
1820                  * (e.g gso or jumbo packet), we will do it at after
1821                  * skb was created with generic XDP routine.
1822                  */
1823                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1824                 if (IS_ERR(skb)) {
1825                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1826                         return PTR_ERR(skb);
1827                 }
1828                 if (!skb)
1829                         return total_len;
1830         } else {
1831                 if (!zerocopy) {
1832                         copylen = len;
1833                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1834                                 linear = good_linear;
1835                         else
1836                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1837                 }
1838
1839                 if (frags) {
1840                         mutex_lock(&tfile->napi_mutex);
1841                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1842                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1843                          * If zerocopy is enabled, then this layout will be
1844                          * overwritten by zerocopy_sg_from_iter().
1845                          */
1846                         zerocopy = false;
1847                 } else {
1848                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1849                                             noblock);
1850                 }
1851
1852                 if (IS_ERR(skb)) {
1853                         if (PTR_ERR(skb) != -EAGAIN)
1854                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1855                         if (frags)
1856                                 mutex_unlock(&tfile->napi_mutex);
1857                         return PTR_ERR(skb);
1858                 }
1859
1860                 if (zerocopy)
1861                         err = zerocopy_sg_from_iter(skb, from);
1862                 else
1863                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1864
1865                 if (err) {
1866                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1867                         kfree_skb(skb);
1868                         if (frags) {
1869                                 tfile->napi.skb = NULL;
1870                                 mutex_unlock(&tfile->napi_mutex);
1871                         }
1872
1873                         return -EFAULT;
1874                 }
1875         }
1876
1877         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1878                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1879                 kfree_skb(skb);
1880                 if (frags) {
1881                         tfile->napi.skb = NULL;
1882                         mutex_unlock(&tfile->napi_mutex);
1883                 }
1884
1885                 return -EINVAL;
1886         }
1887
1888         switch (tun->flags & TUN_TYPE_MASK) {
1889         case IFF_TUN:
1890                 if (tun->flags & IFF_NO_PI) {
1891                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1892
1893                         switch (ip_version) {
1894                         case 4:
1895                                 pi.proto = htons(ETH_P_IP);
1896                                 break;
1897                         case 6:
1898                                 pi.proto = htons(ETH_P_IPV6);
1899                                 break;
1900                         default:
1901                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1902                                 kfree_skb(skb);
1903                                 return -EINVAL;
1904                         }
1905                 }
1906
1907                 skb_reset_mac_header(skb);
1908                 skb->protocol = pi.proto;
1909                 skb->dev = tun->dev;
1910                 break;
1911         case IFF_TAP:
1912                 if (!frags)
1913                         skb->protocol = eth_type_trans(skb, tun->dev);
1914                 break;
1915         }
1916
1917         /* copy skb_ubuf_info for callback when skb has no error */
1918         if (zerocopy) {
1919                 skb_shinfo(skb)->destructor_arg = msg_control;
1920                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1921                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1922         } else if (msg_control) {
1923                 struct ubuf_info *uarg = msg_control;
1924                 uarg->callback(uarg, false);
1925         }
1926
1927         skb_reset_network_header(skb);
1928         skb_probe_transport_header(skb, 0);
1929
1930         if (skb_xdp) {
1931                 struct bpf_prog *xdp_prog;
1932                 int ret;
1933
1934                 rcu_read_lock();
1935                 xdp_prog = rcu_dereference(tun->xdp_prog);
1936                 if (xdp_prog) {
1937                         ret = do_xdp_generic(xdp_prog, skb);
1938                         if (ret != XDP_PASS) {
1939                                 rcu_read_unlock();
1940                                 return total_len;
1941                         }
1942                 }
1943                 rcu_read_unlock();
1944         }
1945
1946         /* Compute the costly rx hash only if needed for flow updates.
1947          * We may get a very small possibility of OOO during switching, not
1948          * worth to optimize.
1949          */
1950         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1951             !tfile->detached)
1952                 rxhash = __skb_get_hash_symmetric(skb);
1953
1954         if (frags) {
1955                 /* Exercise flow dissector code path. */
1956                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1957
1958                 if (unlikely(headlen > skb_headlen(skb))) {
1959                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1960                         napi_free_frags(&tfile->napi);
1961                         mutex_unlock(&tfile->napi_mutex);
1962                         WARN_ON(1);
1963                         return -ENOMEM;
1964                 }
1965
1966                 local_bh_disable();
1967                 napi_gro_frags(&tfile->napi);
1968                 local_bh_enable();
1969                 mutex_unlock(&tfile->napi_mutex);
1970         } else if (tfile->napi_enabled) {
1971                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1972                 int queue_len;
1973
1974                 spin_lock_bh(&queue->lock);
1975                 __skb_queue_tail(queue, skb);
1976                 queue_len = skb_queue_len(queue);
1977                 spin_unlock(&queue->lock);
1978
1979                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1980                         napi_schedule(&tfile->napi);
1981
1982                 local_bh_enable();
1983         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1984                 tun_rx_batched(tun, tfile, skb, more);
1985         } else {
1986                 netif_rx_ni(skb);
1987         }
1988
1989         stats = get_cpu_ptr(tun->pcpu_stats);
1990         u64_stats_update_begin(&stats->syncp);
1991         stats->rx_packets++;
1992         stats->rx_bytes += len;
1993         u64_stats_update_end(&stats->syncp);
1994         put_cpu_ptr(stats);
1995
1996         if (rxhash)
1997                 tun_flow_update(tun, rxhash, tfile);
1998
1999         return total_len;
2000 }
2001
2002 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2003 {
2004         struct file *file = iocb->ki_filp;
2005         struct tun_file *tfile = file->private_data;
2006         struct tun_struct *tun = tun_get(tfile);
2007         ssize_t result;
2008
2009         if (!tun)
2010                 return -EBADFD;
2011
2012         result = tun_get_user(tun, tfile, NULL, from,
2013                               file->f_flags & O_NONBLOCK, false);
2014
2015         tun_put(tun);
2016         return result;
2017 }
2018
2019 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2020                                 struct tun_file *tfile,
2021                                 struct xdp_frame *xdp_frame,
2022                                 struct iov_iter *iter)
2023 {
2024         int vnet_hdr_sz = 0;
2025         size_t size = xdp_frame->len;
2026         struct tun_pcpu_stats *stats;
2027         size_t ret;
2028
2029         if (tun->flags & IFF_VNET_HDR) {
2030                 struct virtio_net_hdr gso = { 0 };
2031
2032                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2033                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2034                         return -EINVAL;
2035                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2036                              sizeof(gso)))
2037                         return -EFAULT;
2038                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2039         }
2040
2041         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2042
2043         stats = get_cpu_ptr(tun->pcpu_stats);
2044         u64_stats_update_begin(&stats->syncp);
2045         stats->tx_packets++;
2046         stats->tx_bytes += ret;
2047         u64_stats_update_end(&stats->syncp);
2048         put_cpu_ptr(tun->pcpu_stats);
2049
2050         return ret;
2051 }
2052
2053 /* Put packet to the user space buffer */
2054 static ssize_t tun_put_user(struct tun_struct *tun,
2055                             struct tun_file *tfile,
2056                             struct sk_buff *skb,
2057                             struct iov_iter *iter)
2058 {
2059         struct tun_pi pi = { 0, skb->protocol };
2060         struct tun_pcpu_stats *stats;
2061         ssize_t total;
2062         int vlan_offset = 0;
2063         int vlan_hlen = 0;
2064         int vnet_hdr_sz = 0;
2065
2066         if (skb_vlan_tag_present(skb))
2067                 vlan_hlen = VLAN_HLEN;
2068
2069         if (tun->flags & IFF_VNET_HDR)
2070                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2071
2072         total = skb->len + vlan_hlen + vnet_hdr_sz;
2073
2074         if (!(tun->flags & IFF_NO_PI)) {
2075                 if (iov_iter_count(iter) < sizeof(pi))
2076                         return -EINVAL;
2077
2078                 total += sizeof(pi);
2079                 if (iov_iter_count(iter) < total) {
2080                         /* Packet will be striped */
2081                         pi.flags |= TUN_PKT_STRIP;
2082                 }
2083
2084                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2085                         return -EFAULT;
2086         }
2087
2088         if (vnet_hdr_sz) {
2089                 struct virtio_net_hdr gso;
2090
2091                 if (iov_iter_count(iter) < vnet_hdr_sz)
2092                         return -EINVAL;
2093
2094                 if (virtio_net_hdr_from_skb(skb, &gso,
2095                                             tun_is_little_endian(tun), true)) {
2096                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2097                         pr_err("unexpected GSO type: "
2098                                "0x%x, gso_size %d, hdr_len %d\n",
2099                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2100                                tun16_to_cpu(tun, gso.hdr_len));
2101                         print_hex_dump(KERN_ERR, "tun: ",
2102                                        DUMP_PREFIX_NONE,
2103                                        16, 1, skb->head,
2104                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2105                         WARN_ON_ONCE(1);
2106                         return -EINVAL;
2107                 }
2108
2109                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2110                         return -EFAULT;
2111
2112                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2113         }
2114
2115         if (vlan_hlen) {
2116                 int ret;
2117                 struct veth veth;
2118
2119                 veth.h_vlan_proto = skb->vlan_proto;
2120                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2121
2122                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2123
2124                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2125                 if (ret || !iov_iter_count(iter))
2126                         goto done;
2127
2128                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2129                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2130                         goto done;
2131         }
2132
2133         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2134
2135 done:
2136         /* caller is in process context, */
2137         stats = get_cpu_ptr(tun->pcpu_stats);
2138         u64_stats_update_begin(&stats->syncp);
2139         stats->tx_packets++;
2140         stats->tx_bytes += skb->len + vlan_hlen;
2141         u64_stats_update_end(&stats->syncp);
2142         put_cpu_ptr(tun->pcpu_stats);
2143
2144         return total;
2145 }
2146
2147 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2148 {
2149         DECLARE_WAITQUEUE(wait, current);
2150         void *ptr = NULL;
2151         int error = 0;
2152
2153         ptr = ptr_ring_consume(&tfile->tx_ring);
2154         if (ptr)
2155                 goto out;
2156         if (noblock) {
2157                 error = -EAGAIN;
2158                 goto out;
2159         }
2160
2161         add_wait_queue(&tfile->wq.wait, &wait);
2162         current->state = TASK_INTERRUPTIBLE;
2163
2164         while (1) {
2165                 ptr = ptr_ring_consume(&tfile->tx_ring);
2166                 if (ptr)
2167                         break;
2168                 if (signal_pending(current)) {
2169                         error = -ERESTARTSYS;
2170                         break;
2171                 }
2172                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2173                         error = -EFAULT;
2174                         break;
2175                 }
2176
2177                 schedule();
2178         }
2179
2180         current->state = TASK_RUNNING;
2181         remove_wait_queue(&tfile->wq.wait, &wait);
2182
2183 out:
2184         *err = error;
2185         return ptr;
2186 }
2187
2188 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2189                            struct iov_iter *to,
2190                            int noblock, void *ptr)
2191 {
2192         ssize_t ret;
2193         int err;
2194
2195         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2196
2197         if (!iov_iter_count(to)) {
2198                 tun_ptr_free(ptr);
2199                 return 0;
2200         }
2201
2202         if (!ptr) {
2203                 /* Read frames from ring */
2204                 ptr = tun_ring_recv(tfile, noblock, &err);
2205                 if (!ptr)
2206                         return err;
2207         }
2208
2209         if (tun_is_xdp_frame(ptr)) {
2210                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2211
2212                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2213                 xdp_return_frame(xdpf);
2214         } else {
2215                 struct sk_buff *skb = ptr;
2216
2217                 ret = tun_put_user(tun, tfile, skb, to);
2218                 if (unlikely(ret < 0))
2219                         kfree_skb(skb);
2220                 else
2221                         consume_skb(skb);
2222         }
2223
2224         return ret;
2225 }
2226
2227 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2228 {
2229         struct file *file = iocb->ki_filp;
2230         struct tun_file *tfile = file->private_data;
2231         struct tun_struct *tun = tun_get(tfile);
2232         ssize_t len = iov_iter_count(to), ret;
2233
2234         if (!tun)
2235                 return -EBADFD;
2236         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2237         ret = min_t(ssize_t, ret, len);
2238         if (ret > 0)
2239                 iocb->ki_pos = ret;
2240         tun_put(tun);
2241         return ret;
2242 }
2243
2244 static void tun_prog_free(struct rcu_head *rcu)
2245 {
2246         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2247
2248         bpf_prog_destroy(prog->prog);
2249         kfree(prog);
2250 }
2251
2252 static int __tun_set_ebpf(struct tun_struct *tun,
2253                           struct tun_prog __rcu **prog_p,
2254                           struct bpf_prog *prog)
2255 {
2256         struct tun_prog *old, *new = NULL;
2257
2258         if (prog) {
2259                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2260                 if (!new)
2261                         return -ENOMEM;
2262                 new->prog = prog;
2263         }
2264
2265         spin_lock_bh(&tun->lock);
2266         old = rcu_dereference_protected(*prog_p,
2267                                         lockdep_is_held(&tun->lock));
2268         rcu_assign_pointer(*prog_p, new);
2269         spin_unlock_bh(&tun->lock);
2270
2271         if (old)
2272                 call_rcu(&old->rcu, tun_prog_free);
2273
2274         return 0;
2275 }
2276
2277 static void tun_free_netdev(struct net_device *dev)
2278 {
2279         struct tun_struct *tun = netdev_priv(dev);
2280
2281         BUG_ON(!(list_empty(&tun->disabled)));
2282         free_percpu(tun->pcpu_stats);
2283         tun_flow_uninit(tun);
2284         security_tun_dev_free_security(tun->security);
2285         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2286         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2287 }
2288
2289 static void tun_setup(struct net_device *dev)
2290 {
2291         struct tun_struct *tun = netdev_priv(dev);
2292
2293         tun->owner = INVALID_UID;
2294         tun->group = INVALID_GID;
2295
2296         dev->ethtool_ops = &tun_ethtool_ops;
2297         dev->needs_free_netdev = true;
2298         dev->priv_destructor = tun_free_netdev;
2299         /* We prefer our own queue length */
2300         dev->tx_queue_len = TUN_READQ_SIZE;
2301 }
2302
2303 /* Trivial set of netlink ops to allow deleting tun or tap
2304  * device with netlink.
2305  */
2306 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2307                         struct netlink_ext_ack *extack)
2308 {
2309         return -EINVAL;
2310 }
2311
2312 static size_t tun_get_size(const struct net_device *dev)
2313 {
2314         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2315         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2316
2317         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2318                nla_total_size(sizeof(gid_t)) + /* GROUP */
2319                nla_total_size(sizeof(u8)) + /* TYPE */
2320                nla_total_size(sizeof(u8)) + /* PI */
2321                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2322                nla_total_size(sizeof(u8)) + /* PERSIST */
2323                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2324                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2325                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2326                0;
2327 }
2328
2329 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2330 {
2331         struct tun_struct *tun = netdev_priv(dev);
2332
2333         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2334                 goto nla_put_failure;
2335         if (uid_valid(tun->owner) &&
2336             nla_put_u32(skb, IFLA_TUN_OWNER,
2337                         from_kuid_munged(current_user_ns(), tun->owner)))
2338                 goto nla_put_failure;
2339         if (gid_valid(tun->group) &&
2340             nla_put_u32(skb, IFLA_TUN_GROUP,
2341                         from_kgid_munged(current_user_ns(), tun->group)))
2342                 goto nla_put_failure;
2343         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2344                 goto nla_put_failure;
2345         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2346                 goto nla_put_failure;
2347         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2348                 goto nla_put_failure;
2349         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2350                        !!(tun->flags & IFF_MULTI_QUEUE)))
2351                 goto nla_put_failure;
2352         if (tun->flags & IFF_MULTI_QUEUE) {
2353                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2354                         goto nla_put_failure;
2355                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2356                                 tun->numdisabled))
2357                         goto nla_put_failure;
2358         }
2359
2360         return 0;
2361
2362 nla_put_failure:
2363         return -EMSGSIZE;
2364 }
2365
2366 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2367         .kind           = DRV_NAME,
2368         .priv_size      = sizeof(struct tun_struct),
2369         .setup          = tun_setup,
2370         .validate       = tun_validate,
2371         .get_size       = tun_get_size,
2372         .fill_info      = tun_fill_info,
2373 };
2374
2375 static void tun_sock_write_space(struct sock *sk)
2376 {
2377         struct tun_file *tfile;
2378         wait_queue_head_t *wqueue;
2379
2380         if (!sock_writeable(sk))
2381                 return;
2382
2383         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2384                 return;
2385
2386         wqueue = sk_sleep(sk);
2387         if (wqueue && waitqueue_active(wqueue))
2388                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2389                                                 EPOLLWRNORM | EPOLLWRBAND);
2390
2391         tfile = container_of(sk, struct tun_file, sk);
2392         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2393 }
2394
2395 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2396 {
2397         int ret;
2398         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2399         struct tun_struct *tun = tun_get(tfile);
2400
2401         if (!tun)
2402                 return -EBADFD;
2403
2404         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2405                            m->msg_flags & MSG_DONTWAIT,
2406                            m->msg_flags & MSG_MORE);
2407         tun_put(tun);
2408         return ret;
2409 }
2410
2411 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2412                        int flags)
2413 {
2414         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2415         struct tun_struct *tun = tun_get(tfile);
2416         void *ptr = m->msg_control;
2417         int ret;
2418
2419         if (!tun) {
2420                 ret = -EBADFD;
2421                 goto out_free;
2422         }
2423
2424         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2425                 ret = -EINVAL;
2426                 goto out_put_tun;
2427         }
2428         if (flags & MSG_ERRQUEUE) {
2429                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2430                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2431                 goto out;
2432         }
2433         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2434         if (ret > (ssize_t)total_len) {
2435                 m->msg_flags |= MSG_TRUNC;
2436                 ret = flags & MSG_TRUNC ? ret : total_len;
2437         }
2438 out:
2439         tun_put(tun);
2440         return ret;
2441
2442 out_put_tun:
2443         tun_put(tun);
2444 out_free:
2445         tun_ptr_free(ptr);
2446         return ret;
2447 }
2448
2449 static int tun_ptr_peek_len(void *ptr)
2450 {
2451         if (likely(ptr)) {
2452                 if (tun_is_xdp_frame(ptr)) {
2453                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2454
2455                         return xdpf->len;
2456                 }
2457                 return __skb_array_len_with_tag(ptr);
2458         } else {
2459                 return 0;
2460         }
2461 }
2462
2463 static int tun_peek_len(struct socket *sock)
2464 {
2465         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2466         struct tun_struct *tun;
2467         int ret = 0;
2468
2469         tun = tun_get(tfile);
2470         if (!tun)
2471                 return 0;
2472
2473         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2474         tun_put(tun);
2475
2476         return ret;
2477 }
2478
2479 /* Ops structure to mimic raw sockets with tun */
2480 static const struct proto_ops tun_socket_ops = {
2481         .peek_len = tun_peek_len,
2482         .sendmsg = tun_sendmsg,
2483         .recvmsg = tun_recvmsg,
2484 };
2485
2486 static struct proto tun_proto = {
2487         .name           = "tun",
2488         .owner          = THIS_MODULE,
2489         .obj_size       = sizeof(struct tun_file),
2490 };
2491
2492 static int tun_flags(struct tun_struct *tun)
2493 {
2494         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2495 }
2496
2497 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2498                               char *buf)
2499 {
2500         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2501         return sprintf(buf, "0x%x\n", tun_flags(tun));
2502 }
2503
2504 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2505                               char *buf)
2506 {
2507         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2508         return uid_valid(tun->owner)?
2509                 sprintf(buf, "%u\n",
2510                         from_kuid_munged(current_user_ns(), tun->owner)):
2511                 sprintf(buf, "-1\n");
2512 }
2513
2514 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2515                               char *buf)
2516 {
2517         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2518         return gid_valid(tun->group) ?
2519                 sprintf(buf, "%u\n",
2520                         from_kgid_munged(current_user_ns(), tun->group)):
2521                 sprintf(buf, "-1\n");
2522 }
2523
2524 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2525 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2526 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2527
2528 static struct attribute *tun_dev_attrs[] = {
2529         &dev_attr_tun_flags.attr,
2530         &dev_attr_owner.attr,
2531         &dev_attr_group.attr,
2532         NULL
2533 };
2534
2535 static const struct attribute_group tun_attr_group = {
2536         .attrs = tun_dev_attrs
2537 };
2538
2539 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2540 {
2541         struct tun_struct *tun;
2542         struct tun_file *tfile = file->private_data;
2543         struct net_device *dev;
2544         int err;
2545
2546         if (tfile->detached)
2547                 return -EINVAL;
2548
2549         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2550                 if (!capable(CAP_NET_ADMIN))
2551                         return -EPERM;
2552
2553                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2554                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2555                         return -EINVAL;
2556         }
2557
2558         dev = __dev_get_by_name(net, ifr->ifr_name);
2559         if (dev) {
2560                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2561                         return -EBUSY;
2562                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2563                         tun = netdev_priv(dev);
2564                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2565                         tun = netdev_priv(dev);
2566                 else
2567                         return -EINVAL;
2568
2569                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2570                     !!(tun->flags & IFF_MULTI_QUEUE))
2571                         return -EINVAL;
2572
2573                 if (tun_not_capable(tun))
2574                         return -EPERM;
2575                 err = security_tun_dev_open(tun->security);
2576                 if (err < 0)
2577                         return err;
2578
2579                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2580                                  ifr->ifr_flags & IFF_NAPI);
2581                 if (err < 0)
2582                         return err;
2583
2584                 if (tun->flags & IFF_MULTI_QUEUE &&
2585                     (tun->numqueues + tun->numdisabled > 1)) {
2586                         /* One or more queue has already been attached, no need
2587                          * to initialize the device again.
2588                          */
2589                         netdev_state_change(dev);
2590                         return 0;
2591                 }
2592
2593                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2594                               (ifr->ifr_flags & TUN_FEATURES);
2595
2596                 netdev_state_change(dev);
2597         } else {
2598                 char *name;
2599                 unsigned long flags = 0;
2600                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2601                              MAX_TAP_QUEUES : 1;
2602
2603                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2604                         return -EPERM;
2605                 err = security_tun_dev_create();
2606                 if (err < 0)
2607                         return err;
2608
2609                 /* Set dev type */
2610                 if (ifr->ifr_flags & IFF_TUN) {
2611                         /* TUN device */
2612                         flags |= IFF_TUN;
2613                         name = "tun%d";
2614                 } else if (ifr->ifr_flags & IFF_TAP) {
2615                         /* TAP device */
2616                         flags |= IFF_TAP;
2617                         name = "tap%d";
2618                 } else
2619                         return -EINVAL;
2620
2621                 if (*ifr->ifr_name)
2622                         name = ifr->ifr_name;
2623
2624                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2625                                        NET_NAME_UNKNOWN, tun_setup, queues,
2626                                        queues);
2627
2628                 if (!dev)
2629                         return -ENOMEM;
2630                 err = dev_get_valid_name(net, dev, name);
2631                 if (err < 0)
2632                         goto err_free_dev;
2633
2634                 dev_net_set(dev, net);
2635                 dev->rtnl_link_ops = &tun_link_ops;
2636                 dev->ifindex = tfile->ifindex;
2637                 dev->sysfs_groups[0] = &tun_attr_group;
2638
2639                 tun = netdev_priv(dev);
2640                 tun->dev = dev;
2641                 tun->flags = flags;
2642                 tun->txflt.count = 0;
2643                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2644
2645                 tun->align = NET_SKB_PAD;
2646                 tun->filter_attached = false;
2647                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2648                 tun->rx_batched = 0;
2649                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2650
2651                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2652                 if (!tun->pcpu_stats) {
2653                         err = -ENOMEM;
2654                         goto err_free_dev;
2655                 }
2656
2657                 spin_lock_init(&tun->lock);
2658
2659                 err = security_tun_dev_alloc_security(&tun->security);
2660                 if (err < 0)
2661                         goto err_free_stat;
2662
2663                 tun_net_init(dev);
2664                 tun_flow_init(tun);
2665
2666                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2667                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2668                                    NETIF_F_HW_VLAN_STAG_TX;
2669                 dev->features = dev->hw_features | NETIF_F_LLTX;
2670                 dev->vlan_features = dev->features &
2671                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2672                                        NETIF_F_HW_VLAN_STAG_TX);
2673
2674                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2675                               (ifr->ifr_flags & TUN_FEATURES);
2676
2677                 INIT_LIST_HEAD(&tun->disabled);
2678                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2679                 if (err < 0)
2680                         goto err_free_flow;
2681
2682                 err = register_netdevice(tun->dev);
2683                 if (err < 0)
2684                         goto err_detach;
2685         }
2686
2687         netif_carrier_on(tun->dev);
2688
2689         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2690
2691         /* Make sure persistent devices do not get stuck in
2692          * xoff state.
2693          */
2694         if (netif_running(tun->dev))
2695                 netif_tx_wake_all_queues(tun->dev);
2696
2697         strcpy(ifr->ifr_name, tun->dev->name);
2698         return 0;
2699
2700 err_detach:
2701         tun_detach_all(dev);
2702         /* register_netdevice() already called tun_free_netdev() */
2703         goto err_free_dev;
2704
2705 err_free_flow:
2706         tun_flow_uninit(tun);
2707         security_tun_dev_free_security(tun->security);
2708 err_free_stat:
2709         free_percpu(tun->pcpu_stats);
2710 err_free_dev:
2711         free_netdev(dev);
2712         return err;
2713 }
2714
2715 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2716                        struct ifreq *ifr)
2717 {
2718         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2719
2720         strcpy(ifr->ifr_name, tun->dev->name);
2721
2722         ifr->ifr_flags = tun_flags(tun);
2723
2724 }
2725
2726 /* This is like a cut-down ethtool ops, except done via tun fd so no
2727  * privs required. */
2728 static int set_offload(struct tun_struct *tun, unsigned long arg)
2729 {
2730         netdev_features_t features = 0;
2731
2732         if (arg & TUN_F_CSUM) {
2733                 features |= NETIF_F_HW_CSUM;
2734                 arg &= ~TUN_F_CSUM;
2735
2736                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2737                         if (arg & TUN_F_TSO_ECN) {
2738                                 features |= NETIF_F_TSO_ECN;
2739                                 arg &= ~TUN_F_TSO_ECN;
2740                         }
2741                         if (arg & TUN_F_TSO4)
2742                                 features |= NETIF_F_TSO;
2743                         if (arg & TUN_F_TSO6)
2744                                 features |= NETIF_F_TSO6;
2745                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2746                 }
2747
2748                 arg &= ~TUN_F_UFO;
2749         }
2750
2751         /* This gives the user a way to test for new features in future by
2752          * trying to set them. */
2753         if (arg)
2754                 return -EINVAL;
2755
2756         tun->set_features = features;
2757         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2758         tun->dev->wanted_features |= features;
2759         netdev_update_features(tun->dev);
2760
2761         return 0;
2762 }
2763
2764 static void tun_detach_filter(struct tun_struct *tun, int n)
2765 {
2766         int i;
2767         struct tun_file *tfile;
2768
2769         for (i = 0; i < n; i++) {
2770                 tfile = rtnl_dereference(tun->tfiles[i]);
2771                 lock_sock(tfile->socket.sk);
2772                 sk_detach_filter(tfile->socket.sk);
2773                 release_sock(tfile->socket.sk);
2774         }
2775
2776         tun->filter_attached = false;
2777 }
2778
2779 static int tun_attach_filter(struct tun_struct *tun)
2780 {
2781         int i, ret = 0;
2782         struct tun_file *tfile;
2783
2784         for (i = 0; i < tun->numqueues; i++) {
2785                 tfile = rtnl_dereference(tun->tfiles[i]);
2786                 lock_sock(tfile->socket.sk);
2787                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2788                 release_sock(tfile->socket.sk);
2789                 if (ret) {
2790                         tun_detach_filter(tun, i);
2791                         return ret;
2792                 }
2793         }
2794
2795         tun->filter_attached = true;
2796         return ret;
2797 }
2798
2799 static void tun_set_sndbuf(struct tun_struct *tun)
2800 {
2801         struct tun_file *tfile;
2802         int i;
2803
2804         for (i = 0; i < tun->numqueues; i++) {
2805                 tfile = rtnl_dereference(tun->tfiles[i]);
2806                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2807         }
2808 }
2809
2810 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2811 {
2812         struct tun_file *tfile = file->private_data;
2813         struct tun_struct *tun;
2814         int ret = 0;
2815
2816         rtnl_lock();
2817
2818         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2819                 tun = tfile->detached;
2820                 if (!tun) {
2821                         ret = -EINVAL;
2822                         goto unlock;
2823                 }
2824                 ret = security_tun_dev_attach_queue(tun->security);
2825                 if (ret < 0)
2826                         goto unlock;
2827                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2828         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2829                 tun = rtnl_dereference(tfile->tun);
2830                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2831                         ret = -EINVAL;
2832                 else
2833                         __tun_detach(tfile, false);
2834         } else
2835                 ret = -EINVAL;
2836
2837         if (ret >= 0)
2838                 netdev_state_change(tun->dev);
2839
2840 unlock:
2841         rtnl_unlock();
2842         return ret;
2843 }
2844
2845 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2846                         void __user *data)
2847 {
2848         struct bpf_prog *prog;
2849         int fd;
2850
2851         if (copy_from_user(&fd, data, sizeof(fd)))
2852                 return -EFAULT;
2853
2854         if (fd == -1) {
2855                 prog = NULL;
2856         } else {
2857                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2858                 if (IS_ERR(prog))
2859                         return PTR_ERR(prog);
2860         }
2861
2862         return __tun_set_ebpf(tun, prog_p, prog);
2863 }
2864
2865 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2866                             unsigned long arg, int ifreq_len)
2867 {
2868         struct tun_file *tfile = file->private_data;
2869         struct net *net = sock_net(&tfile->sk);
2870         struct tun_struct *tun;
2871         void __user* argp = (void __user*)arg;
2872         struct ifreq ifr;
2873         kuid_t owner;
2874         kgid_t group;
2875         int sndbuf;
2876         int vnet_hdr_sz;
2877         unsigned int ifindex;
2878         int le;
2879         int ret;
2880         bool do_notify = false;
2881
2882         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2883             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2884                 if (copy_from_user(&ifr, argp, ifreq_len))
2885                         return -EFAULT;
2886         } else {
2887                 memset(&ifr, 0, sizeof(ifr));
2888         }
2889         if (cmd == TUNGETFEATURES) {
2890                 /* Currently this just means: "what IFF flags are valid?".
2891                  * This is needed because we never checked for invalid flags on
2892                  * TUNSETIFF.
2893                  */
2894                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2895                                 (unsigned int __user*)argp);
2896         } else if (cmd == TUNSETQUEUE) {
2897                 return tun_set_queue(file, &ifr);
2898         } else if (cmd == SIOCGSKNS) {
2899                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2900                         return -EPERM;
2901                 return open_related_ns(&net->ns, get_net_ns);
2902         }
2903
2904         ret = 0;
2905         rtnl_lock();
2906
2907         tun = tun_get(tfile);
2908         if (cmd == TUNSETIFF) {
2909                 ret = -EEXIST;
2910                 if (tun)
2911                         goto unlock;
2912
2913                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2914
2915                 ret = tun_set_iff(net, file, &ifr);
2916
2917                 if (ret)
2918                         goto unlock;
2919
2920                 if (copy_to_user(argp, &ifr, ifreq_len))
2921                         ret = -EFAULT;
2922                 goto unlock;
2923         }
2924         if (cmd == TUNSETIFINDEX) {
2925                 ret = -EPERM;
2926                 if (tun)
2927                         goto unlock;
2928
2929                 ret = -EFAULT;
2930                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2931                         goto unlock;
2932
2933                 ret = 0;
2934                 tfile->ifindex = ifindex;
2935                 goto unlock;
2936         }
2937
2938         ret = -EBADFD;
2939         if (!tun)
2940                 goto unlock;
2941
2942         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2943
2944         ret = 0;
2945         switch (cmd) {
2946         case TUNGETIFF:
2947                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2948
2949                 if (tfile->detached)
2950                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2951                 if (!tfile->socket.sk->sk_filter)
2952                         ifr.ifr_flags |= IFF_NOFILTER;
2953
2954                 if (copy_to_user(argp, &ifr, ifreq_len))
2955                         ret = -EFAULT;
2956                 break;
2957
2958         case TUNSETNOCSUM:
2959                 /* Disable/Enable checksum */
2960
2961                 /* [unimplemented] */
2962                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2963                           arg ? "disabled" : "enabled");
2964                 break;
2965
2966         case TUNSETPERSIST:
2967                 /* Disable/Enable persist mode. Keep an extra reference to the
2968                  * module to prevent the module being unprobed.
2969                  */
2970                 if (arg && !(tun->flags & IFF_PERSIST)) {
2971                         tun->flags |= IFF_PERSIST;
2972                         __module_get(THIS_MODULE);
2973                         do_notify = true;
2974                 }
2975                 if (!arg && (tun->flags & IFF_PERSIST)) {
2976                         tun->flags &= ~IFF_PERSIST;
2977                         module_put(THIS_MODULE);
2978                         do_notify = true;
2979                 }
2980
2981                 tun_debug(KERN_INFO, tun, "persist %s\n",
2982                           arg ? "enabled" : "disabled");
2983                 break;
2984
2985         case TUNSETOWNER:
2986                 /* Set owner of the device */
2987                 owner = make_kuid(current_user_ns(), arg);
2988                 if (!uid_valid(owner)) {
2989                         ret = -EINVAL;
2990                         break;
2991                 }
2992                 tun->owner = owner;
2993                 do_notify = true;
2994                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2995                           from_kuid(&init_user_ns, tun->owner));
2996                 break;
2997
2998         case TUNSETGROUP:
2999                 /* Set group of the device */
3000                 group = make_kgid(current_user_ns(), arg);
3001                 if (!gid_valid(group)) {
3002                         ret = -EINVAL;
3003                         break;
3004                 }
3005                 tun->group = group;
3006                 do_notify = true;
3007                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3008                           from_kgid(&init_user_ns, tun->group));
3009                 break;
3010
3011         case TUNSETLINK:
3012                 /* Only allow setting the type when the interface is down */
3013                 if (tun->dev->flags & IFF_UP) {
3014                         tun_debug(KERN_INFO, tun,
3015                                   "Linktype set failed because interface is up\n");
3016                         ret = -EBUSY;
3017                 } else {
3018                         tun->dev->type = (int) arg;
3019                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3020                                   tun->dev->type);
3021                         ret = 0;
3022                 }
3023                 break;
3024
3025 #ifdef TUN_DEBUG
3026         case TUNSETDEBUG:
3027                 tun->debug = arg;
3028                 break;
3029 #endif
3030         case TUNSETOFFLOAD:
3031                 ret = set_offload(tun, arg);
3032                 break;
3033
3034         case TUNSETTXFILTER:
3035                 /* Can be set only for TAPs */
3036                 ret = -EINVAL;
3037                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3038                         break;
3039                 ret = update_filter(&tun->txflt, (void __user *)arg);
3040                 break;
3041
3042         case SIOCGIFHWADDR:
3043                 /* Get hw address */
3044                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3045                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3046                 if (copy_to_user(argp, &ifr, ifreq_len))
3047                         ret = -EFAULT;
3048                 break;
3049
3050         case SIOCSIFHWADDR:
3051                 /* Set hw address */
3052                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3053                           ifr.ifr_hwaddr.sa_data);
3054
3055                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3056                 break;
3057
3058         case TUNGETSNDBUF:
3059                 sndbuf = tfile->socket.sk->sk_sndbuf;
3060                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3061                         ret = -EFAULT;
3062                 break;
3063
3064         case TUNSETSNDBUF:
3065                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3066                         ret = -EFAULT;
3067                         break;
3068                 }
3069                 if (sndbuf <= 0) {
3070                         ret = -EINVAL;
3071                         break;
3072                 }
3073
3074                 tun->sndbuf = sndbuf;
3075                 tun_set_sndbuf(tun);
3076                 break;
3077
3078         case TUNGETVNETHDRSZ:
3079                 vnet_hdr_sz = tun->vnet_hdr_sz;
3080                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3081                         ret = -EFAULT;
3082                 break;
3083
3084         case TUNSETVNETHDRSZ:
3085                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3086                         ret = -EFAULT;
3087                         break;
3088                 }
3089                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3090                         ret = -EINVAL;
3091                         break;
3092                 }
3093
3094                 tun->vnet_hdr_sz = vnet_hdr_sz;
3095                 break;
3096
3097         case TUNGETVNETLE:
3098                 le = !!(tun->flags & TUN_VNET_LE);
3099                 if (put_user(le, (int __user *)argp))
3100                         ret = -EFAULT;
3101                 break;
3102
3103         case TUNSETVNETLE:
3104                 if (get_user(le, (int __user *)argp)) {
3105                         ret = -EFAULT;
3106                         break;
3107                 }
3108                 if (le)
3109                         tun->flags |= TUN_VNET_LE;
3110                 else
3111                         tun->flags &= ~TUN_VNET_LE;
3112                 break;
3113
3114         case TUNGETVNETBE:
3115                 ret = tun_get_vnet_be(tun, argp);
3116                 break;
3117
3118         case TUNSETVNETBE:
3119                 ret = tun_set_vnet_be(tun, argp);
3120                 break;
3121
3122         case TUNATTACHFILTER:
3123                 /* Can be set only for TAPs */
3124                 ret = -EINVAL;
3125                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3126                         break;
3127                 ret = -EFAULT;
3128                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3129                         break;
3130
3131                 ret = tun_attach_filter(tun);
3132                 break;
3133
3134         case TUNDETACHFILTER:
3135                 /* Can be set only for TAPs */
3136                 ret = -EINVAL;
3137                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3138                         break;
3139                 ret = 0;
3140                 tun_detach_filter(tun, tun->numqueues);
3141                 break;
3142
3143         case TUNGETFILTER:
3144                 ret = -EINVAL;
3145                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3146                         break;
3147                 ret = -EFAULT;
3148                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3149                         break;
3150                 ret = 0;
3151                 break;
3152
3153         case TUNSETSTEERINGEBPF:
3154                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3155                 break;
3156
3157         case TUNSETFILTEREBPF:
3158                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3159                 break;
3160
3161         default:
3162                 ret = -EINVAL;
3163                 break;
3164         }
3165
3166         if (do_notify)
3167                 netdev_state_change(tun->dev);
3168
3169 unlock:
3170         rtnl_unlock();
3171         if (tun)
3172                 tun_put(tun);
3173         return ret;
3174 }
3175
3176 static long tun_chr_ioctl(struct file *file,
3177                           unsigned int cmd, unsigned long arg)
3178 {
3179         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3180 }
3181
3182 #ifdef CONFIG_COMPAT
3183 static long tun_chr_compat_ioctl(struct file *file,
3184                          unsigned int cmd, unsigned long arg)
3185 {
3186         switch (cmd) {
3187         case TUNSETIFF:
3188         case TUNGETIFF:
3189         case TUNSETTXFILTER:
3190         case TUNGETSNDBUF:
3191         case TUNSETSNDBUF:
3192         case SIOCGIFHWADDR:
3193         case SIOCSIFHWADDR:
3194                 arg = (unsigned long)compat_ptr(arg);
3195                 break;
3196         default:
3197                 arg = (compat_ulong_t)arg;
3198                 break;
3199         }
3200
3201         /*
3202          * compat_ifreq is shorter than ifreq, so we must not access beyond
3203          * the end of that structure. All fields that are used in this
3204          * driver are compatible though, we don't need to convert the
3205          * contents.
3206          */
3207         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3208 }
3209 #endif /* CONFIG_COMPAT */
3210
3211 static int tun_chr_fasync(int fd, struct file *file, int on)
3212 {
3213         struct tun_file *tfile = file->private_data;
3214         int ret;
3215
3216         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3217                 goto out;
3218
3219         if (on) {
3220                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3221                 tfile->flags |= TUN_FASYNC;
3222         } else
3223                 tfile->flags &= ~TUN_FASYNC;
3224         ret = 0;
3225 out:
3226         return ret;
3227 }
3228
3229 static int tun_chr_open(struct inode *inode, struct file * file)
3230 {
3231         struct net *net = current->nsproxy->net_ns;
3232         struct tun_file *tfile;
3233
3234         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3235
3236         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3237                                             &tun_proto, 0);
3238         if (!tfile)
3239                 return -ENOMEM;
3240         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3241                 sk_free(&tfile->sk);
3242                 return -ENOMEM;
3243         }
3244
3245         RCU_INIT_POINTER(tfile->tun, NULL);
3246         tfile->flags = 0;
3247         tfile->ifindex = 0;
3248
3249         init_waitqueue_head(&tfile->wq.wait);
3250         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3251
3252         tfile->socket.file = file;
3253         tfile->socket.ops = &tun_socket_ops;
3254
3255         sock_init_data(&tfile->socket, &tfile->sk);
3256
3257         tfile->sk.sk_write_space = tun_sock_write_space;
3258         tfile->sk.sk_sndbuf = INT_MAX;
3259
3260         file->private_data = tfile;
3261         INIT_LIST_HEAD(&tfile->next);
3262
3263         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3264
3265         return 0;
3266 }
3267
3268 static int tun_chr_close(struct inode *inode, struct file *file)
3269 {
3270         struct tun_file *tfile = file->private_data;
3271
3272         tun_detach(tfile, true);
3273
3274         return 0;
3275 }
3276
3277 #ifdef CONFIG_PROC_FS
3278 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3279 {
3280         struct tun_file *tfile = file->private_data;
3281         struct tun_struct *tun;
3282         struct ifreq ifr;
3283
3284         memset(&ifr, 0, sizeof(ifr));
3285
3286         rtnl_lock();
3287         tun = tun_get(tfile);
3288         if (tun)
3289                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3290         rtnl_unlock();
3291
3292         if (tun)
3293                 tun_put(tun);
3294
3295         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3296 }
3297 #endif
3298
3299 static const struct file_operations tun_fops = {
3300         .owner  = THIS_MODULE,
3301         .llseek = no_llseek,
3302         .read_iter  = tun_chr_read_iter,
3303         .write_iter = tun_chr_write_iter,
3304         .poll   = tun_chr_poll,
3305         .unlocked_ioctl = tun_chr_ioctl,
3306 #ifdef CONFIG_COMPAT
3307         .compat_ioctl = tun_chr_compat_ioctl,
3308 #endif
3309         .open   = tun_chr_open,
3310         .release = tun_chr_close,
3311         .fasync = tun_chr_fasync,
3312 #ifdef CONFIG_PROC_FS
3313         .show_fdinfo = tun_chr_show_fdinfo,
3314 #endif
3315 };
3316
3317 static struct miscdevice tun_miscdev = {
3318         .minor = TUN_MINOR,
3319         .name = "tun",
3320         .nodename = "net/tun",
3321         .fops = &tun_fops,
3322 };
3323
3324 /* ethtool interface */
3325
3326 static int tun_get_link_ksettings(struct net_device *dev,
3327                                   struct ethtool_link_ksettings *cmd)
3328 {
3329         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3330         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3331         cmd->base.speed         = SPEED_10;
3332         cmd->base.duplex        = DUPLEX_FULL;
3333         cmd->base.port          = PORT_TP;
3334         cmd->base.phy_address   = 0;
3335         cmd->base.autoneg       = AUTONEG_DISABLE;
3336         return 0;
3337 }
3338
3339 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3340 {
3341         struct tun_struct *tun = netdev_priv(dev);
3342
3343         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3344         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3345
3346         switch (tun->flags & TUN_TYPE_MASK) {
3347         case IFF_TUN:
3348                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3349                 break;
3350         case IFF_TAP:
3351                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3352                 break;
3353         }
3354 }
3355
3356 static u32 tun_get_msglevel(struct net_device *dev)
3357 {
3358 #ifdef TUN_DEBUG
3359         struct tun_struct *tun = netdev_priv(dev);
3360         return tun->debug;
3361 #else
3362         return -EOPNOTSUPP;
3363 #endif
3364 }
3365
3366 static void tun_set_msglevel(struct net_device *dev, u32 value)
3367 {
3368 #ifdef TUN_DEBUG
3369         struct tun_struct *tun = netdev_priv(dev);
3370         tun->debug = value;
3371 #endif
3372 }
3373
3374 static int tun_get_coalesce(struct net_device *dev,
3375                             struct ethtool_coalesce *ec)
3376 {
3377         struct tun_struct *tun = netdev_priv(dev);
3378
3379         ec->rx_max_coalesced_frames = tun->rx_batched;
3380
3381         return 0;
3382 }
3383
3384 static int tun_set_coalesce(struct net_device *dev,
3385                             struct ethtool_coalesce *ec)
3386 {
3387         struct tun_struct *tun = netdev_priv(dev);
3388
3389         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3390                 tun->rx_batched = NAPI_POLL_WEIGHT;
3391         else
3392                 tun->rx_batched = ec->rx_max_coalesced_frames;
3393
3394         return 0;
3395 }
3396
3397 static const struct ethtool_ops tun_ethtool_ops = {
3398         .get_drvinfo    = tun_get_drvinfo,
3399         .get_msglevel   = tun_get_msglevel,
3400         .set_msglevel   = tun_set_msglevel,
3401         .get_link       = ethtool_op_get_link,
3402         .get_ts_info    = ethtool_op_get_ts_info,
3403         .get_coalesce   = tun_get_coalesce,
3404         .set_coalesce   = tun_set_coalesce,
3405         .get_link_ksettings = tun_get_link_ksettings,
3406 };
3407
3408 static int tun_queue_resize(struct tun_struct *tun)
3409 {
3410         struct net_device *dev = tun->dev;
3411         struct tun_file *tfile;
3412         struct ptr_ring **rings;
3413         int n = tun->numqueues + tun->numdisabled;
3414         int ret, i;
3415
3416         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3417         if (!rings)
3418                 return -ENOMEM;
3419
3420         for (i = 0; i < tun->numqueues; i++) {
3421                 tfile = rtnl_dereference(tun->tfiles[i]);
3422                 rings[i] = &tfile->tx_ring;
3423         }
3424         list_for_each_entry(tfile, &tun->disabled, next)
3425                 rings[i++] = &tfile->tx_ring;
3426
3427         ret = ptr_ring_resize_multiple(rings, n,
3428                                        dev->tx_queue_len, GFP_KERNEL,
3429                                        tun_ptr_free);
3430
3431         kfree(rings);
3432         return ret;
3433 }
3434
3435 static int tun_device_event(struct notifier_block *unused,
3436                             unsigned long event, void *ptr)
3437 {
3438         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3439         struct tun_struct *tun = netdev_priv(dev);
3440
3441         if (dev->rtnl_link_ops != &tun_link_ops)
3442                 return NOTIFY_DONE;
3443
3444         switch (event) {
3445         case NETDEV_CHANGE_TX_QUEUE_LEN:
3446                 if (tun_queue_resize(tun))
3447                         return NOTIFY_BAD;
3448                 break;
3449         default:
3450                 break;
3451         }
3452
3453         return NOTIFY_DONE;
3454 }
3455
3456 static struct notifier_block tun_notifier_block __read_mostly = {
3457         .notifier_call  = tun_device_event,
3458 };
3459
3460 static int __init tun_init(void)
3461 {
3462         int ret = 0;
3463
3464         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3465
3466         ret = rtnl_link_register(&tun_link_ops);
3467         if (ret) {
3468                 pr_err("Can't register link_ops\n");
3469                 goto err_linkops;
3470         }
3471
3472         ret = misc_register(&tun_miscdev);
3473         if (ret) {
3474                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3475                 goto err_misc;
3476         }
3477
3478         ret = register_netdevice_notifier(&tun_notifier_block);
3479         if (ret) {
3480                 pr_err("Can't register netdevice notifier\n");
3481                 goto err_notifier;
3482         }
3483
3484         return  0;
3485
3486 err_notifier:
3487         misc_deregister(&tun_miscdev);
3488 err_misc:
3489         rtnl_link_unregister(&tun_link_ops);
3490 err_linkops:
3491         return ret;
3492 }
3493
3494 static void tun_cleanup(void)
3495 {
3496         misc_deregister(&tun_miscdev);
3497         rtnl_link_unregister(&tun_link_ops);
3498         unregister_netdevice_notifier(&tun_notifier_block);
3499 }
3500
3501 /* Get an underlying socket object from tun file.  Returns error unless file is
3502  * attached to a device.  The returned object works like a packet socket, it
3503  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3504  * holding a reference to the file for as long as the socket is in use. */
3505 struct socket *tun_get_socket(struct file *file)
3506 {
3507         struct tun_file *tfile;
3508         if (file->f_op != &tun_fops)
3509                 return ERR_PTR(-EINVAL);
3510         tfile = file->private_data;
3511         if (!tfile)
3512                 return ERR_PTR(-EBADFD);
3513         return &tfile->socket;
3514 }
3515 EXPORT_SYMBOL_GPL(tun_get_socket);
3516
3517 struct ptr_ring *tun_get_tx_ring(struct file *file)
3518 {
3519         struct tun_file *tfile;
3520
3521         if (file->f_op != &tun_fops)
3522                 return ERR_PTR(-EINVAL);
3523         tfile = file->private_data;
3524         if (!tfile)
3525                 return ERR_PTR(-EBADFD);
3526         return &tfile->tx_ring;
3527 }
3528 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3529
3530 module_init(tun_init);
3531 module_exit(tun_cleanup);
3532 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3533 MODULE_AUTHOR(DRV_COPYRIGHT);
3534 MODULE_LICENSE("GPL");
3535 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3536 MODULE_ALIAS("devname:net/tun");