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 <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81 #include <linux/proc_fs.h>
  82
  83 /* Uncomment to enable debugging */
  84 /* #define TUN_DEBUG 1 */
  85
  86 #ifdef TUN_DEBUG
  87 static int debug;
  88
  89 #define tun_debug(level, tun, fmt, args...)                     \
  90 do {                                                            \
  91         if (tun->debug)                                         \
  92                 netdev_printk(level, tun->dev, fmt, ##args);    \
  93 } while (0)
  94 #define DBG1(level, fmt, args...)                               \
  95 do {                                                            \
  96         if (debug == 2)                                         \
  97                 printk(level fmt, ##args);                      \
  98 } while (0)
  99 #else
 100 #define tun_debug(level, tun, fmt, args...)                     \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 netdev_printk(level, tun->dev, fmt, ##args);    \
 104 } while (0)
 105 #define DBG1(level, fmt, args...)                               \
 106 do {                                                            \
 107         if (0)                                                  \
 108                 printk(level fmt, ##args);                      \
 109 } while (0)
 110 #endif
 111
 112 #define TUN_HEADROOM 256
 113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 127
 128 #define GOODCOPY_LEN 128
 129
 130 #define FLT_EXACT_COUNT 8
 131 struct tap_filter {
 132         unsigned int    count;    /* Number of addrs. Zero means disabled */
 133         u32             mask[2];  /* Mask of the hashed addrs */
 134         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 135 };
 136
 137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 138  * to max number of VCPUs in guest. */
 139 #define MAX_TAP_QUEUES 256
 140 #define MAX_TAP_FLOWS  4096
 141
 142 #define TUN_FLOW_EXPIRE (3 * HZ)
 143
 144 struct tun_pcpu_stats {
 145         u64 rx_packets;
 146         u64 rx_bytes;
 147         u64 tx_packets;
 148         u64 tx_bytes;
 149         struct u64_stats_sync syncp;
 150         u32 rx_dropped;
 151         u32 tx_dropped;
 152         u32 rx_frame_errors;
 153 };
 154
 155 /* A tun_file connects an open character device to a tuntap netdevice. It
 156  * also contains all socket related structures (except sock_fprog and tap_filter)
 157  * to serve as one transmit queue for tuntap device. The sock_fprog and
 158  * tap_filter were kept in tun_struct since they were used for filtering for the
 159  * netdevice not for a specific queue (at least I didn't see the requirement for
 160  * this).
 161  *
 162  * RCU usage:
 163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 164  * other can only be read while rcu_read_lock or rtnl_lock is held.
 165  */
 166 struct tun_file {
 167         struct sock sk;
 168         struct socket socket;
 169         struct socket_wq wq;
 170         struct tun_struct __rcu *tun;
 171         struct fasync_struct *fasync;
 172         /* only used for fasnyc */
 173         unsigned int flags;
 174         union {
 175                 u16 queue_index;
 176                 unsigned int ifindex;
 177         };
 178         struct napi_struct napi;
 179         bool napi_enabled;
 180         struct mutex napi_mutex;        /* Protects access to the above napi */
 181         struct list_head next;
 182         struct tun_struct *detached;
 183         struct ptr_ring tx_ring;
 184         struct xdp_rxq_info xdp_rxq;
 185 };
 186
 187 struct tun_flow_entry {
 188         struct hlist_node hash_link;
 189         struct rcu_head rcu;
 190         struct tun_struct *tun;
 191
 192         u32 rxhash;
 193         u32 rps_rxhash;
 194         int queue_index;
 195         unsigned long updated;
 196 };
 197
 198 #define TUN_NUM_FLOW_ENTRIES 1024
 199
 200 struct tun_prog {
 201         struct rcu_head rcu;
 202         struct bpf_prog *prog;
 203 };
 204
 205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 206  * device, socket filter, sndbuf and vnet header size were restore when the
 207  * file were attached to a persist device.
 208  */
 209 struct tun_struct {
 210         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 211         unsigned int            numqueues;
 212         unsigned int            flags;
 213         kuid_t                  owner;
 214         kgid_t                  group;
 215
 216         struct net_device       *dev;
 217         netdev_features_t       set_features;
 218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 219                           NETIF_F_TSO6)
 220
 221         int                     align;
 222         int                     vnet_hdr_sz;
 223         int                     sndbuf;
 224         struct tap_filter       txflt;
 225         struct sock_fprog       fprog;
 226         /* protected by rtnl lock */
 227         bool                    filter_attached;
 228 #ifdef TUN_DEBUG
 229         int debug;
 230 #endif
 231         spinlock_t lock;
 232         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 233         struct timer_list flow_gc_timer;
 234         unsigned long ageing_time;
 235         unsigned int numdisabled;
 236         struct list_head disabled;
 237         void *security;
 238         u32 flow_count;
 239         u32 rx_batched;
 240         struct tun_pcpu_stats __percpu *pcpu_stats;
 241         struct bpf_prog __rcu *xdp_prog;
 242         struct tun_prog __rcu *steering_prog;
 243         struct tun_prog __rcu *filter_prog;
 244 };
 245
 246 struct veth {
 247         __be16 h_vlan_proto;
 248         __be16 h_vlan_TCI;
 249 };
 250
 251 bool tun_is_xdp_frame(void *ptr)
 252 {
 253         return (unsigned long)ptr & TUN_XDP_FLAG;
 254 }
 255 EXPORT_SYMBOL(tun_is_xdp_frame);
 256
 257 void *tun_xdp_to_ptr(void *ptr)
 258 {
 259         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 260 }
 261 EXPORT_SYMBOL(tun_xdp_to_ptr);
 262
 263 void *tun_ptr_to_xdp(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_ptr_to_xdp);
 268
 269 static int tun_napi_receive(struct napi_struct *napi, int budget)
 270 {
 271         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 272         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 273         struct sk_buff_head process_queue;
 274         struct sk_buff *skb;
 275         int received = 0;
 276
 277         __skb_queue_head_init(&process_queue);
 278
 279         spin_lock(&queue->lock);
 280         skb_queue_splice_tail_init(queue, &process_queue);
 281         spin_unlock(&queue->lock);
 282
 283         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 284                 napi_gro_receive(napi, skb);
 285                 ++received;
 286         }
 287
 288         if (!skb_queue_empty(&process_queue)) {
 289                 spin_lock(&queue->lock);
 290                 skb_queue_splice(&process_queue, queue);
 291                 spin_unlock(&queue->lock);
 292         }
 293
 294         return received;
 295 }
 296
 297 static int tun_napi_poll(struct napi_struct *napi, int budget)
 298 {
 299         unsigned int received;
 300
 301         received = tun_napi_receive(napi, budget);
 302
 303         if (received < budget)
 304                 napi_complete_done(napi, received);
 305
 306         return received;
 307 }
 308
 309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 310                           bool napi_en)
 311 {
 312         tfile->napi_enabled = napi_en;
 313         if (napi_en) {
 314                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 315                                NAPI_POLL_WEIGHT);
 316                 napi_enable(&tfile->napi);
 317                 mutex_init(&tfile->napi_mutex);
 318         }
 319 }
 320
 321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 322 {
 323         if (tfile->napi_enabled)
 324                 napi_disable(&tfile->napi);
 325 }
 326
 327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 netif_napi_del(&tfile->napi);
 331 }
 332
 333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 334 {
 335         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 336 }
 337
 338 #ifdef CONFIG_TUN_VNET_CROSS_LE
 339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 340 {
 341         return tun->flags & TUN_VNET_BE ? false :
 342                 virtio_legacy_is_little_endian();
 343 }
 344
 345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 346 {
 347         int be = !!(tun->flags & TUN_VNET_BE);
 348
 349         if (put_user(be, argp))
 350                 return -EFAULT;
 351
 352         return 0;
 353 }
 354
 355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 356 {
 357         int be;
 358
 359         if (get_user(be, argp))
 360                 return -EFAULT;
 361
 362         if (be)
 363                 tun->flags |= TUN_VNET_BE;
 364         else
 365                 tun->flags &= ~TUN_VNET_BE;
 366
 367         return 0;
 368 }
 369 #else
 370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 371 {
 372         return virtio_legacy_is_little_endian();
 373 }
 374
 375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 376 {
 377         return -EINVAL;
 378 }
 379
 380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 385
 386 static inline bool tun_is_little_endian(struct tun_struct *tun)
 387 {
 388         return tun->flags & TUN_VNET_LE ||
 389                 tun_legacy_is_little_endian(tun);
 390 }
 391
 392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 393 {
 394         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 395 }
 396
 397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 398 {
 399         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline u32 tun_hashfn(u32 rxhash)
 403 {
 404         return rxhash & 0x3ff;
 405 }
 406
 407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 408 {
 409         struct tun_flow_entry *e;
 410
 411         hlist_for_each_entry_rcu(e, head, hash_link) {
 412                 if (e->rxhash == rxhash)
 413                         return e;
 414         }
 415         return NULL;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 419                                               struct hlist_head *head,
 420                                               u32 rxhash, u16 queue_index)
 421 {
 422         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 423
 424         if (e) {
 425                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 426                           rxhash, queue_index);
 427                 e->updated = jiffies;
 428                 e->rxhash = rxhash;
 429                 e->rps_rxhash = 0;
 430                 e->queue_index = queue_index;
 431                 e->tun = tun;
 432                 hlist_add_head_rcu(&e->hash_link, head);
 433                 ++tun->flow_count;
 434         }
 435         return e;
 436 }
 437
 438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 439 {
 440         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 441                   e->rxhash, e->queue_index);
 442         hlist_del_rcu(&e->hash_link);
 443         kfree_rcu(e, rcu);
 444         --tun->flow_count;
 445 }
 446
 447 static void tun_flow_flush(struct tun_struct *tun)
 448 {
 449         int i;
 450
 451         spin_lock_bh(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 457                         tun_flow_delete(tun, e);
 458         }
 459         spin_unlock_bh(&tun->lock);
 460 }
 461
 462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 463 {
 464         int i;
 465
 466         spin_lock_bh(&tun->lock);
 467         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 468                 struct tun_flow_entry *e;
 469                 struct hlist_node *n;
 470
 471                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 472                         if (e->queue_index == queue_index)
 473                                 tun_flow_delete(tun, e);
 474                 }
 475         }
 476         spin_unlock_bh(&tun->lock);
 477 }
 478
 479 static void tun_flow_cleanup(struct timer_list *t)
 480 {
 481         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 482         unsigned long delay = tun->ageing_time;
 483         unsigned long next_timer = jiffies + delay;
 484         unsigned long count = 0;
 485         int i;
 486
 487         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 488
 489         spin_lock(&tun->lock);
 490         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 491                 struct tun_flow_entry *e;
 492                 struct hlist_node *n;
 493
 494                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 495                         unsigned long this_timer;
 496
 497                         this_timer = e->updated + delay;
 498                         if (time_before_eq(this_timer, jiffies)) {
 499                                 tun_flow_delete(tun, e);
 500                                 continue;
 501                         }
 502                         count++;
 503                         if (time_before(this_timer, next_timer))
 504                                 next_timer = this_timer;
 505                 }
 506         }
 507
 508         if (count)
 509                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 510         spin_unlock(&tun->lock);
 511 }
 512
 513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 514                             struct tun_file *tfile)
 515 {
 516         struct hlist_head *head;
 517         struct tun_flow_entry *e;
 518         unsigned long delay = tun->ageing_time;
 519         u16 queue_index = tfile->queue_index;
 520
 521         if (!rxhash)
 522                 return;
 523         else
 524                 head = &tun->flows[tun_hashfn(rxhash)];
 525
 526         rcu_read_lock();
 527
 528         /* We may get a very small possibility of OOO during switching, not
 529          * worth to optimize.*/
 530         if (tun->numqueues == 1 || tfile->detached)
 531                 goto unlock;
 532
 533         e = tun_flow_find(head, rxhash);
 534         if (likely(e)) {
 535                 /* TODO: keep queueing to old queue until it's empty? */
 536                 e->queue_index = queue_index;
 537                 e->updated = jiffies;
 538                 sock_rps_record_flow_hash(e->rps_rxhash);
 539         } else {
 540                 spin_lock_bh(&tun->lock);
 541                 if (!tun_flow_find(head, rxhash) &&
 542                     tun->flow_count < MAX_TAP_FLOWS)
 543                         tun_flow_create(tun, head, rxhash, queue_index);
 544
 545                 if (!timer_pending(&tun->flow_gc_timer))
 546                         mod_timer(&tun->flow_gc_timer,
 547                                   round_jiffies_up(jiffies + delay));
 548                 spin_unlock_bh(&tun->lock);
 549         }
 550
 551 unlock:
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash first. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here. If we could not get rxhash, then we would
 570  * hope the rxq no. may help here.
 571  */
 572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 573 {
 574         struct tun_flow_entry *e;
 575         u32 txq = 0;
 576         u32 numqueues = 0;
 577
 578         numqueues = READ_ONCE(tun->numqueues);
 579
 580         txq = __skb_get_hash_symmetric(skb);
 581         if (txq) {
 582                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 583                 if (e) {
 584                         tun_flow_save_rps_rxhash(e, txq);
 585                         txq = e->queue_index;
 586                 } else
 587                         /* use multiply and shift instead of expensive divide */
 588                         txq = ((u64)txq * numqueues) >> 32;
 589         } else if (likely(skb_rx_queue_recorded(skb))) {
 590                 txq = skb_get_rx_queue(skb);
 591                 while (unlikely(txq >= numqueues))
 592                         txq -= numqueues;
 593         }
 594
 595         return txq;
 596 }
 597
 598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 599 {
 600         struct tun_prog *prog;
 601         u16 ret = 0;
 602
 603         prog = rcu_dereference(tun->steering_prog);
 604         if (prog)
 605                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 606
 607         return ret % tun->numqueues;
 608 }
 609
 610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 611                             void *accel_priv, select_queue_fallback_t fallback)
 612 {
 613         struct tun_struct *tun = netdev_priv(dev);
 614         u16 ret;
 615
 616         rcu_read_lock();
 617         if (rcu_dereference(tun->steering_prog))
 618                 ret = tun_ebpf_select_queue(tun, skb);
 619         else
 620                 ret = tun_automq_select_queue(tun, skb);
 621         rcu_read_unlock();
 622
 623         return ret;
 624 }
 625
 626 static inline bool tun_not_capable(struct tun_struct *tun)
 627 {
 628         const struct cred *cred = current_cred();
 629         struct net *net = dev_net(tun->dev);
 630
 631         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 632                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 633                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 634 }
 635
 636 static void tun_set_real_num_queues(struct tun_struct *tun)
 637 {
 638         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 639         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 640 }
 641
 642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 643 {
 644         tfile->detached = tun;
 645         list_add_tail(&tfile->next, &tun->disabled);
 646         ++tun->numdisabled;
 647 }
 648
 649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 650 {
 651         struct tun_struct *tun = tfile->detached;
 652
 653         tfile->detached = NULL;
 654         list_del_init(&tfile->next);
 655         --tun->numdisabled;
 656         return tun;
 657 }
 658
 659 void tun_ptr_free(void *ptr)
 660 {
 661         if (!ptr)
 662                 return;
 663         if (tun_is_xdp_frame(ptr)) {
 664                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 665
 666                 xdp_return_frame(xdpf);
 667         } else {
 668                 __skb_array_destroy_skb(ptr);
 669         }
 670 }
 671 EXPORT_SYMBOL_GPL(tun_ptr_free);
 672
 673 static void tun_queue_purge(struct tun_file *tfile)
 674 {
 675         void *ptr;
 676
 677         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 678                 tun_ptr_free(ptr);
 679
 680         skb_queue_purge(&tfile->sk.sk_write_queue);
 681         skb_queue_purge(&tfile->sk.sk_error_queue);
 682 }
 683
 684 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 685 {
 686         if (tfile->tx_ring.queue) {
 687                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 688                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 690         }
 691 }
 692
 693 static void __tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_file *ntfile;
 696         struct tun_struct *tun;
 697
 698         tun = rtnl_dereference(tfile->tun);
 699
 700         if (tun && clean) {
 701                 tun_napi_disable(tun, tfile);
 702                 tun_napi_del(tun, tfile);
 703         }
 704
 705         if (tun && !tfile->detached) {
 706                 u16 index = tfile->queue_index;
 707                 BUG_ON(index >= tun->numqueues);
 708
 709                 rcu_assign_pointer(tun->tfiles[index],
 710                                    tun->tfiles[tun->numqueues - 1]);
 711                 ntfile = rtnl_dereference(tun->tfiles[index]);
 712                 ntfile->queue_index = index;
 713
 714                 --tun->numqueues;
 715                 if (clean) {
 716                         RCU_INIT_POINTER(tfile->tun, NULL);
 717                         sock_put(&tfile->sk);
 718                 } else
 719                         tun_disable_queue(tun, tfile);
 720
 721                 synchronize_net();
 722                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 723                 /* Drop read queue */
 724                 tun_queue_purge(tfile);
 725                 tun_set_real_num_queues(tun);
 726         } else if (tfile->detached && clean) {
 727                 tun = tun_enable_queue(tfile);
 728                 sock_put(&tfile->sk);
 729         }
 730
 731         if (clean) {
 732                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 733                         netif_carrier_off(tun->dev);
 734
 735                         if (!(tun->flags & IFF_PERSIST) &&
 736                             tun->dev->reg_state == NETREG_REGISTERED)
 737                                 unregister_netdevice(tun->dev);
 738                 }
 739                 tun_cleanup_tx_ring(tfile);
 740                 sock_put(&tfile->sk);
 741         }
 742 }
 743
 744 static void tun_detach(struct tun_file *tfile, bool clean)
 745 {
 746         struct tun_struct *tun;
 747         struct net_device *dev;
 748
 749         rtnl_lock();
 750         tun = rtnl_dereference(tfile->tun);
 751         dev = tun ? tun->dev : NULL;
 752         __tun_detach(tfile, clean);
 753         if (dev)
 754                 netdev_state_change(dev);
 755         rtnl_unlock();
 756 }
 757
 758 static void tun_detach_all(struct net_device *dev)
 759 {
 760         struct tun_struct *tun = netdev_priv(dev);
 761         struct tun_file *tfile, *tmp;
 762         int i, n = tun->numqueues;
 763
 764         for (i = 0; i < n; i++) {
 765                 tfile = rtnl_dereference(tun->tfiles[i]);
 766                 BUG_ON(!tfile);
 767                 tun_napi_disable(tun, tfile);
 768                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 769                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 770                 RCU_INIT_POINTER(tfile->tun, NULL);
 771                 --tun->numqueues;
 772         }
 773         list_for_each_entry(tfile, &tun->disabled, next) {
 774                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 775                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 776                 RCU_INIT_POINTER(tfile->tun, NULL);
 777         }
 778         BUG_ON(tun->numqueues != 0);
 779
 780         synchronize_net();
 781         for (i = 0; i < n; i++) {
 782                 tfile = rtnl_dereference(tun->tfiles[i]);
 783                 tun_napi_del(tun, tfile);
 784                 /* Drop read queue */
 785                 tun_queue_purge(tfile);
 786                 sock_put(&tfile->sk);
 787                 tun_cleanup_tx_ring(tfile);
 788         }
 789         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 790                 tun_enable_queue(tfile);
 791                 tun_queue_purge(tfile);
 792                 sock_put(&tfile->sk);
 793                 tun_cleanup_tx_ring(tfile);
 794         }
 795         BUG_ON(tun->numdisabled != 0);
 796
 797         if (tun->flags & IFF_PERSIST)
 798                 module_put(THIS_MODULE);
 799 }
 800
 801 static int tun_attach(struct tun_struct *tun, struct file *file,
 802                       bool skip_filter, bool napi)
 803 {
 804         struct tun_file *tfile = file->private_data;
 805         struct net_device *dev = tun->dev;
 806         int err;
 807
 808         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 809         if (err < 0)
 810                 goto out;
 811
 812         err = -EINVAL;
 813         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 814                 goto out;
 815
 816         err = -EBUSY;
 817         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 818                 goto out;
 819
 820         err = -E2BIG;
 821         if (!tfile->detached &&
 822             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 823                 goto out;
 824
 825         err = 0;
 826
 827         /* Re-attach the filter to persist device */
 828         if (!skip_filter && (tun->filter_attached == true)) {
 829                 lock_sock(tfile->socket.sk);
 830                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 831                 release_sock(tfile->socket.sk);
 832                 if (!err)
 833                         goto out;
 834         }
 835
 836         if (!tfile->detached &&
 837             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 838                 err = -ENOMEM;
 839                 goto out;
 840         }
 841
 842         tfile->queue_index = tun->numqueues;
 843         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 844
 845         if (tfile->detached) {
 846                 /* Re-attach detached tfile, updating XDP queue_index */
 847                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 848
 849                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 850                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 851         } else {
 852                 /* Setup XDP RX-queue info, for new tfile getting attached */
 853                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 854                                        tun->dev, tfile->queue_index);
 855                 if (err < 0)
 856                         goto out;
 857                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 858                                                  MEM_TYPE_PAGE_SHARED, NULL);
 859                 if (err < 0) {
 860                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 861                         goto out;
 862                 }
 863                 err = 0;
 864         }
 865
 866         rcu_assign_pointer(tfile->tun, tun);
 867         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 868         tun->numqueues++;
 869
 870         if (tfile->detached) {
 871                 tun_enable_queue(tfile);
 872         } else {
 873                 sock_hold(&tfile->sk);
 874                 tun_napi_init(tun, tfile, napi);
 875         }
 876
 877         tun_set_real_num_queues(tun);
 878
 879         /* device is allowed to go away first, so no need to hold extra
 880          * refcnt.
 881          */
 882
 883 out:
 884         return err;
 885 }
 886
 887 static struct tun_struct *tun_get(struct tun_file *tfile)
 888 {
 889         struct tun_struct *tun;
 890
 891         rcu_read_lock();
 892         tun = rcu_dereference(tfile->tun);
 893         if (tun)
 894                 dev_hold(tun->dev);
 895         rcu_read_unlock();
 896
 897         return tun;
 898 }
 899
 900 static void tun_put(struct tun_struct *tun)
 901 {
 902         dev_put(tun->dev);
 903 }
 904
 905 /* TAP filtering */
 906 static void addr_hash_set(u32 *mask, const u8 *addr)
 907 {
 908         int n = ether_crc(ETH_ALEN, addr) >> 26;
 909         mask[n >> 5] |= (1 << (n & 31));
 910 }
 911
 912 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 913 {
 914         int n = ether_crc(ETH_ALEN, addr) >> 26;
 915         return mask[n >> 5] & (1 << (n & 31));
 916 }
 917
 918 static int update_filter(struct tap_filter *filter, void __user *arg)
 919 {
 920         struct { u8 u[ETH_ALEN]; } *addr;
 921         struct tun_filter uf;
 922         int err, alen, n, nexact;
 923
 924         if (copy_from_user(&uf, arg, sizeof(uf)))
 925                 return -EFAULT;
 926
 927         if (!uf.count) {
 928                 /* Disabled */
 929                 filter->count = 0;
 930                 return 0;
 931         }
 932
 933         alen = ETH_ALEN * uf.count;
 934         addr = memdup_user(arg + sizeof(uf), alen);
 935         if (IS_ERR(addr))
 936                 return PTR_ERR(addr);
 937
 938         /* The filter is updated without holding any locks. Which is
 939          * perfectly safe. We disable it first and in the worst
 940          * case we'll accept a few undesired packets. */
 941         filter->count = 0;
 942         wmb();
 943
 944         /* Use first set of addresses as an exact filter */
 945         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 946                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 947
 948         nexact = n;
 949
 950         /* Remaining multicast addresses are hashed,
 951          * unicast will leave the filter disabled. */
 952         memset(filter->mask, 0, sizeof(filter->mask));
 953         for (; n < uf.count; n++) {
 954                 if (!is_multicast_ether_addr(addr[n].u)) {
 955                         err = 0; /* no filter */
 956                         goto free_addr;
 957                 }
 958                 addr_hash_set(filter->mask, addr[n].u);
 959         }
 960
 961         /* For ALLMULTI just set the mask to all ones.
 962          * This overrides the mask populated above. */
 963         if ((uf.flags & TUN_FLT_ALLMULTI))
 964                 memset(filter->mask, ~0, sizeof(filter->mask));
 965
 966         /* Now enable the filter */
 967         wmb();
 968         filter->count = nexact;
 969
 970         /* Return the number of exact filters */
 971         err = nexact;
 972 free_addr:
 973         kfree(addr);
 974         return err;
 975 }
 976
 977 /* Returns: 0 - drop, !=0 - accept */
 978 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 979 {
 980         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 981          * at this point. */
 982         struct ethhdr *eh = (struct ethhdr *) skb->data;
 983         int i;
 984
 985         /* Exact match */
 986         for (i = 0; i < filter->count; i++)
 987                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 988                         return 1;
 989
 990         /* Inexact match (multicast only) */
 991         if (is_multicast_ether_addr(eh->h_dest))
 992                 return addr_hash_test(filter->mask, eh->h_dest);
 993
 994         return 0;
 995 }
 996
 997 /*
 998  * Checks whether the packet is accepted or not.
 999  * Returns: 0 - drop, !=0 - accept
1000  */
1001 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1002 {
1003         if (!filter->count)
1004                 return 1;
1005
1006         return run_filter(filter, skb);
1007 }
1008
1009 /* Network device part of the driver */
1010
1011 static const struct ethtool_ops tun_ethtool_ops;
1012
1013 /* Net device detach from fd. */
1014 static void tun_net_uninit(struct net_device *dev)
1015 {
1016         tun_detach_all(dev);
1017 }
1018
1019 /* Net device open. */
1020 static int tun_net_open(struct net_device *dev)
1021 {
1022         struct tun_struct *tun = netdev_priv(dev);
1023         int i;
1024
1025         netif_tx_start_all_queues(dev);
1026
1027         for (i = 0; i < tun->numqueues; i++) {
1028                 struct tun_file *tfile;
1029
1030                 tfile = rtnl_dereference(tun->tfiles[i]);
1031                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1032         }
1033
1034         return 0;
1035 }
1036
1037 /* Net device close. */
1038 static int tun_net_close(struct net_device *dev)
1039 {
1040         netif_tx_stop_all_queues(dev);
1041         return 0;
1042 }
1043
1044 /* Net device start xmit */
1045 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1046 {
1047 #ifdef CONFIG_RPS
1048         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1049                 /* Select queue was not called for the skbuff, so we extract the
1050                  * RPS hash and save it into the flow_table here.
1051                  */
1052                 __u32 rxhash;
1053
1054                 rxhash = __skb_get_hash_symmetric(skb);
1055                 if (rxhash) {
1056                         struct tun_flow_entry *e;
1057                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1058                                         rxhash);
1059                         if (e)
1060                                 tun_flow_save_rps_rxhash(e, rxhash);
1061                 }
1062         }
1063 #endif
1064 }
1065
1066 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1067                                     struct sk_buff *skb,
1068                                     int len)
1069 {
1070         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1071
1072         if (prog)
1073                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1074
1075         return len;
1076 }
1077
1078 /* Net device start xmit */
1079 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1080 {
1081         struct tun_struct *tun = netdev_priv(dev);
1082         int txq = skb->queue_mapping;
1083         struct tun_file *tfile;
1084         int len = skb->len;
1085
1086         rcu_read_lock();
1087         tfile = rcu_dereference(tun->tfiles[txq]);
1088
1089         /* Drop packet if interface is not attached */
1090         if (txq >= tun->numqueues)
1091                 goto drop;
1092
1093         if (!rcu_dereference(tun->steering_prog))
1094                 tun_automq_xmit(tun, skb);
1095
1096         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1097
1098         BUG_ON(!tfile);
1099
1100         /* Drop if the filter does not like it.
1101          * This is a noop if the filter is disabled.
1102          * Filter can be enabled only for the TAP devices. */
1103         if (!check_filter(&tun->txflt, skb))
1104                 goto drop;
1105
1106         if (tfile->socket.sk->sk_filter &&
1107             sk_filter(tfile->socket.sk, skb))
1108                 goto drop;
1109
1110         len = run_ebpf_filter(tun, skb, len);
1111
1112         /* Trim extra bytes since we may insert vlan proto & TCI
1113          * in tun_put_user().
1114          */
1115         len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1116         if (len <= 0 || pskb_trim(skb, len))
1117                 goto drop;
1118
1119         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1120                 goto drop;
1121
1122         skb_tx_timestamp(skb);
1123
1124         /* Orphan the skb - required as we might hang on to it
1125          * for indefinite time.
1126          */
1127         skb_orphan(skb);
1128
1129         nf_reset(skb);
1130
1131         if (ptr_ring_produce(&tfile->tx_ring, skb))
1132                 goto drop;
1133
1134         /* Notify and wake up reader process */
1135         if (tfile->flags & TUN_FASYNC)
1136                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1137         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1138
1139         rcu_read_unlock();
1140         return NETDEV_TX_OK;
1141
1142 drop:
1143         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1144         skb_tx_error(skb);
1145         kfree_skb(skb);
1146         rcu_read_unlock();
1147         return NET_XMIT_DROP;
1148 }
1149
1150 static void tun_net_mclist(struct net_device *dev)
1151 {
1152         /*
1153          * This callback is supposed to deal with mc filter in
1154          * _rx_ path and has nothing to do with the _tx_ path.
1155          * In rx path we always accept everything userspace gives us.
1156          */
1157 }
1158
1159 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1160         netdev_features_t features)
1161 {
1162         struct tun_struct *tun = netdev_priv(dev);
1163
1164         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1165 }
1166 #ifdef CONFIG_NET_POLL_CONTROLLER
1167 static void tun_poll_controller(struct net_device *dev)
1168 {
1169         /*
1170          * Tun only receives frames when:
1171          * 1) the char device endpoint gets data from user space
1172          * 2) the tun socket gets a sendmsg call from user space
1173          * If NAPI is not enabled, since both of those are synchronous
1174          * operations, we are guaranteed never to have pending data when we poll
1175          * for it so there is nothing to do here but return.
1176          * We need this though so netpoll recognizes us as an interface that
1177          * supports polling, which enables bridge devices in virt setups to
1178          * still use netconsole
1179          * If NAPI is enabled, however, we need to schedule polling for all
1180          * queues unless we are using napi_gro_frags(), which we call in
1181          * process context and not in NAPI context.
1182          */
1183         struct tun_struct *tun = netdev_priv(dev);
1184
1185         if (tun->flags & IFF_NAPI) {
1186                 struct tun_file *tfile;
1187                 int i;
1188
1189                 if (tun_napi_frags_enabled(tun))
1190                         return;
1191
1192                 rcu_read_lock();
1193                 for (i = 0; i < tun->numqueues; i++) {
1194                         tfile = rcu_dereference(tun->tfiles[i]);
1195                         if (tfile->napi_enabled)
1196                                 napi_schedule(&tfile->napi);
1197                 }
1198                 rcu_read_unlock();
1199         }
1200         return;
1201 }
1202 #endif
1203
1204 static void tun_set_headroom(struct net_device *dev, int new_hr)
1205 {
1206         struct tun_struct *tun = netdev_priv(dev);
1207
1208         if (new_hr < NET_SKB_PAD)
1209                 new_hr = NET_SKB_PAD;
1210
1211         tun->align = new_hr;
1212 }
1213
1214 static void
1215 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1216 {
1217         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1218         struct tun_struct *tun = netdev_priv(dev);
1219         struct tun_pcpu_stats *p;
1220         int i;
1221
1222         for_each_possible_cpu(i) {
1223                 u64 rxpackets, rxbytes, txpackets, txbytes;
1224                 unsigned int start;
1225
1226                 p = per_cpu_ptr(tun->pcpu_stats, i);
1227                 do {
1228                         start = u64_stats_fetch_begin(&p->syncp);
1229                         rxpackets       = p->rx_packets;
1230                         rxbytes         = p->rx_bytes;
1231                         txpackets       = p->tx_packets;
1232                         txbytes         = p->tx_bytes;
1233                 } while (u64_stats_fetch_retry(&p->syncp, start));
1234
1235                 stats->rx_packets       += rxpackets;
1236                 stats->rx_bytes         += rxbytes;
1237                 stats->tx_packets       += txpackets;
1238                 stats->tx_bytes         += txbytes;
1239
1240                 /* u32 counters */
1241                 rx_dropped      += p->rx_dropped;
1242                 rx_frame_errors += p->rx_frame_errors;
1243                 tx_dropped      += p->tx_dropped;
1244         }
1245         stats->rx_dropped  = rx_dropped;
1246         stats->rx_frame_errors = rx_frame_errors;
1247         stats->tx_dropped = tx_dropped;
1248 }
1249
1250 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1251                        struct netlink_ext_ack *extack)
1252 {
1253         struct tun_struct *tun = netdev_priv(dev);
1254         struct bpf_prog *old_prog;
1255
1256         old_prog = rtnl_dereference(tun->xdp_prog);
1257         rcu_assign_pointer(tun->xdp_prog, prog);
1258         if (old_prog)
1259                 bpf_prog_put(old_prog);
1260
1261         return 0;
1262 }
1263
1264 static u32 tun_xdp_query(struct net_device *dev)
1265 {
1266         struct tun_struct *tun = netdev_priv(dev);
1267         const struct bpf_prog *xdp_prog;
1268
1269         xdp_prog = rtnl_dereference(tun->xdp_prog);
1270         if (xdp_prog)
1271                 return xdp_prog->aux->id;
1272
1273         return 0;
1274 }
1275
1276 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1277 {
1278         switch (xdp->command) {
1279         case XDP_SETUP_PROG:
1280                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1281         case XDP_QUERY_PROG:
1282                 xdp->prog_id = tun_xdp_query(dev);
1283                 xdp->prog_attached = !!xdp->prog_id;
1284                 return 0;
1285         default:
1286                 return -EINVAL;
1287         }
1288 }
1289
1290 static const struct net_device_ops tun_netdev_ops = {
1291         .ndo_uninit             = tun_net_uninit,
1292         .ndo_open               = tun_net_open,
1293         .ndo_stop               = tun_net_close,
1294         .ndo_start_xmit         = tun_net_xmit,
1295         .ndo_fix_features       = tun_net_fix_features,
1296         .ndo_select_queue       = tun_select_queue,
1297 #ifdef CONFIG_NET_POLL_CONTROLLER
1298         .ndo_poll_controller    = tun_poll_controller,
1299 #endif
1300         .ndo_set_rx_headroom    = tun_set_headroom,
1301         .ndo_get_stats64        = tun_net_get_stats64,
1302 };
1303
1304 static int tun_xdp_xmit(struct net_device *dev, struct xdp_frame *frame)
1305 {
1306         struct tun_struct *tun = netdev_priv(dev);
1307         struct tun_file *tfile;
1308         u32 numqueues;
1309         int ret = 0;
1310
1311         rcu_read_lock();
1312
1313         numqueues = READ_ONCE(tun->numqueues);
1314         if (!numqueues) {
1315                 ret = -ENOSPC;
1316                 goto out;
1317         }
1318
1319         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1320                                             numqueues]);
1321         /* Encode the XDP flag into lowest bit for consumer to differ
1322          * XDP buffer from sk_buff.
1323          */
1324         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(frame))) {
1325                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1326                 ret = -ENOSPC;
1327         }
1328
1329 out:
1330         rcu_read_unlock();
1331         return ret;
1332 }
1333
1334 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1335 {
1336         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1337
1338         if (unlikely(!frame))
1339                 return -EOVERFLOW;
1340
1341         return tun_xdp_xmit(dev, frame);
1342 }
1343
1344 static void tun_xdp_flush(struct net_device *dev)
1345 {
1346         struct tun_struct *tun = netdev_priv(dev);
1347         struct tun_file *tfile;
1348         u32 numqueues;
1349
1350         rcu_read_lock();
1351
1352         numqueues = READ_ONCE(tun->numqueues);
1353         if (!numqueues)
1354                 goto out;
1355
1356         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1357                                             numqueues]);
1358         /* Notify and wake up reader process */
1359         if (tfile->flags & TUN_FASYNC)
1360                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1361         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1362
1363 out:
1364         rcu_read_unlock();
1365 }
1366
1367 static const struct net_device_ops tap_netdev_ops = {
1368         .ndo_uninit             = tun_net_uninit,
1369         .ndo_open               = tun_net_open,
1370         .ndo_stop               = tun_net_close,
1371         .ndo_start_xmit         = tun_net_xmit,
1372         .ndo_fix_features       = tun_net_fix_features,
1373         .ndo_set_rx_mode        = tun_net_mclist,
1374         .ndo_set_mac_address    = eth_mac_addr,
1375         .ndo_validate_addr      = eth_validate_addr,
1376         .ndo_select_queue       = tun_select_queue,
1377 #ifdef CONFIG_NET_POLL_CONTROLLER
1378         .ndo_poll_controller    = tun_poll_controller,
1379 #endif
1380         .ndo_features_check     = passthru_features_check,
1381         .ndo_set_rx_headroom    = tun_set_headroom,
1382         .ndo_get_stats64        = tun_net_get_stats64,
1383         .ndo_bpf                = tun_xdp,
1384         .ndo_xdp_xmit           = tun_xdp_xmit,
1385         .ndo_xdp_flush          = tun_xdp_flush,
1386 };
1387
1388 static void tun_flow_init(struct tun_struct *tun)
1389 {
1390         int i;
1391
1392         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1393                 INIT_HLIST_HEAD(&tun->flows[i]);
1394
1395         tun->ageing_time = TUN_FLOW_EXPIRE;
1396         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1397         mod_timer(&tun->flow_gc_timer,
1398                   round_jiffies_up(jiffies + tun->ageing_time));
1399 }
1400
1401 static void tun_flow_uninit(struct tun_struct *tun)
1402 {
1403         del_timer_sync(&tun->flow_gc_timer);
1404         tun_flow_flush(tun);
1405 }
1406
1407 #define MIN_MTU 68
1408 #define MAX_MTU 65535
1409
1410 /* Initialize net device. */
1411 static void tun_net_init(struct net_device *dev)
1412 {
1413         struct tun_struct *tun = netdev_priv(dev);
1414
1415         switch (tun->flags & TUN_TYPE_MASK) {
1416         case IFF_TUN:
1417                 dev->netdev_ops = &tun_netdev_ops;
1418
1419                 /* Point-to-Point TUN Device */
1420                 dev->hard_header_len = 0;
1421                 dev->addr_len = 0;
1422                 dev->mtu = 1500;
1423
1424                 /* Zero header length */
1425                 dev->type = ARPHRD_NONE;
1426                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1427                 break;
1428
1429         case IFF_TAP:
1430                 dev->netdev_ops = &tap_netdev_ops;
1431                 /* Ethernet TAP Device */
1432                 ether_setup(dev);
1433                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1434                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1435
1436                 eth_hw_addr_random(dev);
1437
1438                 break;
1439         }
1440
1441         dev->min_mtu = MIN_MTU;
1442         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1443 }
1444
1445 /* Character device part */
1446
1447 /* Poll */
1448 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1449 {
1450         struct tun_file *tfile = file->private_data;
1451         struct tun_struct *tun = tun_get(tfile);
1452         struct sock *sk;
1453         __poll_t mask = 0;
1454
1455         if (!tun)
1456                 return EPOLLERR;
1457
1458         sk = tfile->socket.sk;
1459
1460         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1461
1462         poll_wait(file, sk_sleep(sk), wait);
1463
1464         if (!ptr_ring_empty(&tfile->tx_ring))
1465                 mask |= EPOLLIN | EPOLLRDNORM;
1466
1467         if (tun->dev->flags & IFF_UP &&
1468             (sock_writeable(sk) ||
1469              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1470               sock_writeable(sk))))
1471                 mask |= EPOLLOUT | EPOLLWRNORM;
1472
1473         if (tun->dev->reg_state != NETREG_REGISTERED)
1474                 mask = EPOLLERR;
1475
1476         tun_put(tun);
1477         return mask;
1478 }
1479
1480 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1481                                             size_t len,
1482                                             const struct iov_iter *it)
1483 {
1484         struct sk_buff *skb;
1485         size_t linear;
1486         int err;
1487         int i;
1488
1489         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1490                 return ERR_PTR(-ENOMEM);
1491
1492         local_bh_disable();
1493         skb = napi_get_frags(&tfile->napi);
1494         local_bh_enable();
1495         if (!skb)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         linear = iov_iter_single_seg_count(it);
1499         err = __skb_grow(skb, linear);
1500         if (err)
1501                 goto free;
1502
1503         skb->len = len;
1504         skb->data_len = len - linear;
1505         skb->truesize += skb->data_len;
1506
1507         for (i = 1; i < it->nr_segs; i++) {
1508                 struct page_frag *pfrag = &current->task_frag;
1509                 size_t fragsz = it->iov[i].iov_len;
1510
1511                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1512                         err = -EINVAL;
1513                         goto free;
1514                 }
1515
1516                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1517                         err = -ENOMEM;
1518                         goto free;
1519                 }
1520
1521                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1522                                    pfrag->offset, fragsz);
1523                 page_ref_inc(pfrag->page);
1524                 pfrag->offset += fragsz;
1525         }
1526
1527         return skb;
1528 free:
1529         /* frees skb and all frags allocated with napi_alloc_frag() */
1530         napi_free_frags(&tfile->napi);
1531         return ERR_PTR(err);
1532 }
1533
1534 /* prepad is the amount to reserve at front.  len is length after that.
1535  * linear is a hint as to how much to copy (usually headers). */
1536 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1537                                      size_t prepad, size_t len,
1538                                      size_t linear, int noblock)
1539 {
1540         struct sock *sk = tfile->socket.sk;
1541         struct sk_buff *skb;
1542         int err;
1543
1544         /* Under a page?  Don't bother with paged skb. */
1545         if (prepad + len < PAGE_SIZE || !linear)
1546                 linear = len;
1547
1548         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1549                                    &err, 0);
1550         if (!skb)
1551                 return ERR_PTR(err);
1552
1553         skb_reserve(skb, prepad);
1554         skb_put(skb, linear);
1555         skb->data_len = len - linear;
1556         skb->len += len - linear;
1557
1558         return skb;
1559 }
1560
1561 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1562                            struct sk_buff *skb, int more)
1563 {
1564         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1565         struct sk_buff_head process_queue;
1566         u32 rx_batched = tun->rx_batched;
1567         bool rcv = false;
1568
1569         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1570                 local_bh_disable();
1571                 netif_receive_skb(skb);
1572                 local_bh_enable();
1573                 return;
1574         }
1575
1576         spin_lock(&queue->lock);
1577         if (!more || skb_queue_len(queue) == rx_batched) {
1578                 __skb_queue_head_init(&process_queue);
1579                 skb_queue_splice_tail_init(queue, &process_queue);
1580                 rcv = true;
1581         } else {
1582                 __skb_queue_tail(queue, skb);
1583         }
1584         spin_unlock(&queue->lock);
1585
1586         if (rcv) {
1587                 struct sk_buff *nskb;
1588
1589                 local_bh_disable();
1590                 while ((nskb = __skb_dequeue(&process_queue)))
1591                         netif_receive_skb(nskb);
1592                 netif_receive_skb(skb);
1593                 local_bh_enable();
1594         }
1595 }
1596
1597 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1598                               int len, int noblock, bool zerocopy)
1599 {
1600         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1601                 return false;
1602
1603         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1604                 return false;
1605
1606         if (!noblock)
1607                 return false;
1608
1609         if (zerocopy)
1610                 return false;
1611
1612         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1613             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1614                 return false;
1615
1616         return true;
1617 }
1618
1619 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1620                                      struct tun_file *tfile,
1621                                      struct iov_iter *from,
1622                                      struct virtio_net_hdr *hdr,
1623                                      int len, int *skb_xdp)
1624 {
1625         struct page_frag *alloc_frag = &current->task_frag;
1626         struct sk_buff *skb;
1627         struct bpf_prog *xdp_prog;
1628         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1629         unsigned int delta = 0;
1630         char *buf;
1631         size_t copied;
1632         int err, pad = TUN_RX_PAD;
1633
1634         rcu_read_lock();
1635         xdp_prog = rcu_dereference(tun->xdp_prog);
1636         if (xdp_prog)
1637                 pad += TUN_HEADROOM;
1638         buflen += SKB_DATA_ALIGN(len + pad);
1639         rcu_read_unlock();
1640
1641         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1642         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1643                 return ERR_PTR(-ENOMEM);
1644
1645         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1646         copied = copy_page_from_iter(alloc_frag->page,
1647                                      alloc_frag->offset + pad,
1648                                      len, from);
1649         if (copied != len)
1650                 return ERR_PTR(-EFAULT);
1651
1652         /* There's a small window that XDP may be set after the check
1653          * of xdp_prog above, this should be rare and for simplicity
1654          * we do XDP on skb in case the headroom is not enough.
1655          */
1656         if (hdr->gso_type || !xdp_prog)
1657                 *skb_xdp = 1;
1658         else
1659                 *skb_xdp = 0;
1660
1661         preempt_disable();
1662         rcu_read_lock();
1663         xdp_prog = rcu_dereference(tun->xdp_prog);
1664         if (xdp_prog && !*skb_xdp) {
1665                 struct xdp_buff xdp;
1666                 void *orig_data;
1667                 u32 act;
1668
1669                 xdp.data_hard_start = buf;
1670                 xdp.data = buf + pad;
1671                 xdp_set_data_meta_invalid(&xdp);
1672                 xdp.data_end = xdp.data + len;
1673                 xdp.rxq = &tfile->xdp_rxq;
1674                 orig_data = xdp.data;
1675                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1676
1677                 switch (act) {
1678                 case XDP_REDIRECT:
1679                         get_page(alloc_frag->page);
1680                         alloc_frag->offset += buflen;
1681                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1682                         xdp_do_flush_map();
1683                         if (err)
1684                                 goto err_redirect;
1685                         rcu_read_unlock();
1686                         preempt_enable();
1687                         return NULL;
1688                 case XDP_TX:
1689                         get_page(alloc_frag->page);
1690                         alloc_frag->offset += buflen;
1691                         if (tun_xdp_tx(tun->dev, &xdp))
1692                                 goto err_redirect;
1693                         tun_xdp_flush(tun->dev);
1694                         rcu_read_unlock();
1695                         preempt_enable();
1696                         return NULL;
1697                 case XDP_PASS:
1698                         delta = orig_data - xdp.data;
1699                         break;
1700                 default:
1701                         bpf_warn_invalid_xdp_action(act);
1702                         /* fall through */
1703                 case XDP_ABORTED:
1704                         trace_xdp_exception(tun->dev, xdp_prog, act);
1705                         /* fall through */
1706                 case XDP_DROP:
1707                         goto err_xdp;
1708                 }
1709         }
1710
1711         skb = build_skb(buf, buflen);
1712         if (!skb) {
1713                 rcu_read_unlock();
1714                 preempt_enable();
1715                 return ERR_PTR(-ENOMEM);
1716         }
1717
1718         skb_reserve(skb, pad - delta);
1719         skb_put(skb, len + delta);
1720         get_page(alloc_frag->page);
1721         alloc_frag->offset += buflen;
1722
1723         rcu_read_unlock();
1724         preempt_enable();
1725
1726         return skb;
1727
1728 err_redirect:
1729         put_page(alloc_frag->page);
1730 err_xdp:
1731         rcu_read_unlock();
1732         preempt_enable();
1733         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1734         return NULL;
1735 }
1736
1737 /* Get packet from user space buffer */
1738 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1739                             void *msg_control, struct iov_iter *from,
1740                             int noblock, bool more)
1741 {
1742         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1743         struct sk_buff *skb;
1744         size_t total_len = iov_iter_count(from);
1745         size_t len = total_len, align = tun->align, linear;
1746         struct virtio_net_hdr gso = { 0 };
1747         struct tun_pcpu_stats *stats;
1748         int good_linear;
1749         int copylen;
1750         bool zerocopy = false;
1751         int err;
1752         u32 rxhash = 0;
1753         int skb_xdp = 1;
1754         bool frags = tun_napi_frags_enabled(tun);
1755
1756         if (!(tun->dev->flags & IFF_UP))
1757                 return -EIO;
1758
1759         if (!(tun->flags & IFF_NO_PI)) {
1760                 if (len < sizeof(pi))
1761                         return -EINVAL;
1762                 len -= sizeof(pi);
1763
1764                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1765                         return -EFAULT;
1766         }
1767
1768         if (tun->flags & IFF_VNET_HDR) {
1769                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1770
1771                 if (len < vnet_hdr_sz)
1772                         return -EINVAL;
1773                 len -= vnet_hdr_sz;
1774
1775                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1776                         return -EFAULT;
1777
1778                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1779                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1780                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1781
1782                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1783                         return -EINVAL;
1784                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1785         }
1786
1787         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1788                 align += NET_IP_ALIGN;
1789                 if (unlikely(len < ETH_HLEN ||
1790                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1791                         return -EINVAL;
1792         }
1793
1794         good_linear = SKB_MAX_HEAD(align);
1795
1796         if (msg_control) {
1797                 struct iov_iter i = *from;
1798
1799                 /* There are 256 bytes to be copied in skb, so there is
1800                  * enough room for skb expand head in case it is used.
1801                  * The rest of the buffer is mapped from userspace.
1802                  */
1803                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1804                 if (copylen > good_linear)
1805                         copylen = good_linear;
1806                 linear = copylen;
1807                 iov_iter_advance(&i, copylen);
1808                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1809                         zerocopy = true;
1810         }
1811
1812         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1813                 /* For the packet that is not easy to be processed
1814                  * (e.g gso or jumbo packet), we will do it at after
1815                  * skb was created with generic XDP routine.
1816                  */
1817                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1818                 if (IS_ERR(skb)) {
1819                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1820                         return PTR_ERR(skb);
1821                 }
1822                 if (!skb)
1823                         return total_len;
1824         } else {
1825                 if (!zerocopy) {
1826                         copylen = len;
1827                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1828                                 linear = good_linear;
1829                         else
1830                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1831                 }
1832
1833                 if (frags) {
1834                         mutex_lock(&tfile->napi_mutex);
1835                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1836                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1837                          * If zerocopy is enabled, then this layout will be
1838                          * overwritten by zerocopy_sg_from_iter().
1839                          */
1840                         zerocopy = false;
1841                 } else {
1842                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1843                                             noblock);
1844                 }
1845
1846                 if (IS_ERR(skb)) {
1847                         if (PTR_ERR(skb) != -EAGAIN)
1848                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1849                         if (frags)
1850                                 mutex_unlock(&tfile->napi_mutex);
1851                         return PTR_ERR(skb);
1852                 }
1853
1854                 if (zerocopy)
1855                         err = zerocopy_sg_from_iter(skb, from);
1856                 else
1857                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1858
1859                 if (err) {
1860                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1861                         kfree_skb(skb);
1862                         if (frags) {
1863                                 tfile->napi.skb = NULL;
1864                                 mutex_unlock(&tfile->napi_mutex);
1865                         }
1866
1867                         return -EFAULT;
1868                 }
1869         }
1870
1871         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1872                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1873                 kfree_skb(skb);
1874                 if (frags) {
1875                         tfile->napi.skb = NULL;
1876                         mutex_unlock(&tfile->napi_mutex);
1877                 }
1878
1879                 return -EINVAL;
1880         }
1881
1882         switch (tun->flags & TUN_TYPE_MASK) {
1883         case IFF_TUN:
1884                 if (tun->flags & IFF_NO_PI) {
1885                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1886
1887                         switch (ip_version) {
1888                         case 4:
1889                                 pi.proto = htons(ETH_P_IP);
1890                                 break;
1891                         case 6:
1892                                 pi.proto = htons(ETH_P_IPV6);
1893                                 break;
1894                         default:
1895                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1896                                 kfree_skb(skb);
1897                                 return -EINVAL;
1898                         }
1899                 }
1900
1901                 skb_reset_mac_header(skb);
1902                 skb->protocol = pi.proto;
1903                 skb->dev = tun->dev;
1904                 break;
1905         case IFF_TAP:
1906                 if (!frags)
1907                         skb->protocol = eth_type_trans(skb, tun->dev);
1908                 break;
1909         }
1910
1911         /* copy skb_ubuf_info for callback when skb has no error */
1912         if (zerocopy) {
1913                 skb_shinfo(skb)->destructor_arg = msg_control;
1914                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1915                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1916         } else if (msg_control) {
1917                 struct ubuf_info *uarg = msg_control;
1918                 uarg->callback(uarg, false);
1919         }
1920
1921         skb_reset_network_header(skb);
1922         skb_probe_transport_header(skb, 0);
1923
1924         if (skb_xdp) {
1925                 struct bpf_prog *xdp_prog;
1926                 int ret;
1927
1928                 rcu_read_lock();
1929                 xdp_prog = rcu_dereference(tun->xdp_prog);
1930                 if (xdp_prog) {
1931                         ret = do_xdp_generic(xdp_prog, skb);
1932                         if (ret != XDP_PASS) {
1933                                 rcu_read_unlock();
1934                                 return total_len;
1935                         }
1936                 }
1937                 rcu_read_unlock();
1938         }
1939
1940         rcu_read_lock();
1941         if (!rcu_dereference(tun->steering_prog))
1942                 rxhash = __skb_get_hash_symmetric(skb);
1943         rcu_read_unlock();
1944
1945         if (frags) {
1946                 /* Exercise flow dissector code path. */
1947                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1948
1949                 if (unlikely(headlen > skb_headlen(skb))) {
1950                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1951                         napi_free_frags(&tfile->napi);
1952                         mutex_unlock(&tfile->napi_mutex);
1953                         WARN_ON(1);
1954                         return -ENOMEM;
1955                 }
1956
1957                 local_bh_disable();
1958                 napi_gro_frags(&tfile->napi);
1959                 local_bh_enable();
1960                 mutex_unlock(&tfile->napi_mutex);
1961         } else if (tfile->napi_enabled) {
1962                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1963                 int queue_len;
1964
1965                 spin_lock_bh(&queue->lock);
1966                 __skb_queue_tail(queue, skb);
1967                 queue_len = skb_queue_len(queue);
1968                 spin_unlock(&queue->lock);
1969
1970                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1971                         napi_schedule(&tfile->napi);
1972
1973                 local_bh_enable();
1974         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1975                 tun_rx_batched(tun, tfile, skb, more);
1976         } else {
1977                 netif_rx_ni(skb);
1978         }
1979
1980         stats = get_cpu_ptr(tun->pcpu_stats);
1981         u64_stats_update_begin(&stats->syncp);
1982         stats->rx_packets++;
1983         stats->rx_bytes += len;
1984         u64_stats_update_end(&stats->syncp);
1985         put_cpu_ptr(stats);
1986
1987         if (rxhash)
1988                 tun_flow_update(tun, rxhash, tfile);
1989
1990         return total_len;
1991 }
1992
1993 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1994 {
1995         struct file *file = iocb->ki_filp;
1996         struct tun_file *tfile = file->private_data;
1997         struct tun_struct *tun = tun_get(tfile);
1998         ssize_t result;
1999
2000         if (!tun)
2001                 return -EBADFD;
2002
2003         result = tun_get_user(tun, tfile, NULL, from,
2004                               file->f_flags & O_NONBLOCK, false);
2005
2006         tun_put(tun);
2007         return result;
2008 }
2009
2010 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2011                                 struct tun_file *tfile,
2012                                 struct xdp_frame *xdp_frame,
2013                                 struct iov_iter *iter)
2014 {
2015         int vnet_hdr_sz = 0;
2016         size_t size = xdp_frame->len;
2017         struct tun_pcpu_stats *stats;
2018         size_t ret;
2019
2020         if (tun->flags & IFF_VNET_HDR) {
2021                 struct virtio_net_hdr gso = { 0 };
2022
2023                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2024                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2025                         return -EINVAL;
2026                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2027                              sizeof(gso)))
2028                         return -EFAULT;
2029                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2030         }
2031
2032         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2033
2034         stats = get_cpu_ptr(tun->pcpu_stats);
2035         u64_stats_update_begin(&stats->syncp);
2036         stats->tx_packets++;
2037         stats->tx_bytes += ret;
2038         u64_stats_update_end(&stats->syncp);
2039         put_cpu_ptr(tun->pcpu_stats);
2040
2041         return ret;
2042 }
2043
2044 /* Put packet to the user space buffer */
2045 static ssize_t tun_put_user(struct tun_struct *tun,
2046                             struct tun_file *tfile,
2047                             struct sk_buff *skb,
2048                             struct iov_iter *iter)
2049 {
2050         struct tun_pi pi = { 0, skb->protocol };
2051         struct tun_pcpu_stats *stats;
2052         ssize_t total;
2053         int vlan_offset = 0;
2054         int vlan_hlen = 0;
2055         int vnet_hdr_sz = 0;
2056
2057         if (skb_vlan_tag_present(skb))
2058                 vlan_hlen = VLAN_HLEN;
2059
2060         if (tun->flags & IFF_VNET_HDR)
2061                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2062
2063         total = skb->len + vlan_hlen + vnet_hdr_sz;
2064
2065         if (!(tun->flags & IFF_NO_PI)) {
2066                 if (iov_iter_count(iter) < sizeof(pi))
2067                         return -EINVAL;
2068
2069                 total += sizeof(pi);
2070                 if (iov_iter_count(iter) < total) {
2071                         /* Packet will be striped */
2072                         pi.flags |= TUN_PKT_STRIP;
2073                 }
2074
2075                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2076                         return -EFAULT;
2077         }
2078
2079         if (vnet_hdr_sz) {
2080                 struct virtio_net_hdr gso;
2081
2082                 if (iov_iter_count(iter) < vnet_hdr_sz)
2083                         return -EINVAL;
2084
2085                 if (virtio_net_hdr_from_skb(skb, &gso,
2086                                             tun_is_little_endian(tun), true)) {
2087                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2088                         pr_err("unexpected GSO type: "
2089                                "0x%x, gso_size %d, hdr_len %d\n",
2090                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2091                                tun16_to_cpu(tun, gso.hdr_len));
2092                         print_hex_dump(KERN_ERR, "tun: ",
2093                                        DUMP_PREFIX_NONE,
2094                                        16, 1, skb->head,
2095                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2096                         WARN_ON_ONCE(1);
2097                         return -EINVAL;
2098                 }
2099
2100                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2101                         return -EFAULT;
2102
2103                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2104         }
2105
2106         if (vlan_hlen) {
2107                 int ret;
2108                 struct veth veth;
2109
2110                 veth.h_vlan_proto = skb->vlan_proto;
2111                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2112
2113                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2114
2115                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2116                 if (ret || !iov_iter_count(iter))
2117                         goto done;
2118
2119                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2120                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2121                         goto done;
2122         }
2123
2124         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2125
2126 done:
2127         /* caller is in process context, */
2128         stats = get_cpu_ptr(tun->pcpu_stats);
2129         u64_stats_update_begin(&stats->syncp);
2130         stats->tx_packets++;
2131         stats->tx_bytes += skb->len + vlan_hlen;
2132         u64_stats_update_end(&stats->syncp);
2133         put_cpu_ptr(tun->pcpu_stats);
2134
2135         return total;
2136 }
2137
2138 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2139 {
2140         DECLARE_WAITQUEUE(wait, current);
2141         void *ptr = NULL;
2142         int error = 0;
2143
2144         ptr = ptr_ring_consume(&tfile->tx_ring);
2145         if (ptr)
2146                 goto out;
2147         if (noblock) {
2148                 error = -EAGAIN;
2149                 goto out;
2150         }
2151
2152         add_wait_queue(&tfile->wq.wait, &wait);
2153         current->state = TASK_INTERRUPTIBLE;
2154
2155         while (1) {
2156                 ptr = ptr_ring_consume(&tfile->tx_ring);
2157                 if (ptr)
2158                         break;
2159                 if (signal_pending(current)) {
2160                         error = -ERESTARTSYS;
2161                         break;
2162                 }
2163                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2164                         error = -EFAULT;
2165                         break;
2166                 }
2167
2168                 schedule();
2169         }
2170
2171         current->state = TASK_RUNNING;
2172         remove_wait_queue(&tfile->wq.wait, &wait);
2173
2174 out:
2175         *err = error;
2176         return ptr;
2177 }
2178
2179 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2180                            struct iov_iter *to,
2181                            int noblock, void *ptr)
2182 {
2183         ssize_t ret;
2184         int err;
2185
2186         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2187
2188         if (!iov_iter_count(to)) {
2189                 tun_ptr_free(ptr);
2190                 return 0;
2191         }
2192
2193         if (!ptr) {
2194                 /* Read frames from ring */
2195                 ptr = tun_ring_recv(tfile, noblock, &err);
2196                 if (!ptr)
2197                         return err;
2198         }
2199
2200         if (tun_is_xdp_frame(ptr)) {
2201                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2202
2203                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2204                 xdp_return_frame(xdpf);
2205         } else {
2206                 struct sk_buff *skb = ptr;
2207
2208                 ret = tun_put_user(tun, tfile, skb, to);
2209                 if (unlikely(ret < 0))
2210                         kfree_skb(skb);
2211                 else
2212                         consume_skb(skb);
2213         }
2214
2215         return ret;
2216 }
2217
2218 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2219 {
2220         struct file *file = iocb->ki_filp;
2221         struct tun_file *tfile = file->private_data;
2222         struct tun_struct *tun = tun_get(tfile);
2223         ssize_t len = iov_iter_count(to), ret;
2224
2225         if (!tun)
2226                 return -EBADFD;
2227         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2228         ret = min_t(ssize_t, ret, len);
2229         if (ret > 0)
2230                 iocb->ki_pos = ret;
2231         tun_put(tun);
2232         return ret;
2233 }
2234
2235 static void tun_prog_free(struct rcu_head *rcu)
2236 {
2237         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2238
2239         bpf_prog_destroy(prog->prog);
2240         kfree(prog);
2241 }
2242
2243 static int __tun_set_ebpf(struct tun_struct *tun,
2244                           struct tun_prog __rcu **prog_p,
2245                           struct bpf_prog *prog)
2246 {
2247         struct tun_prog *old, *new = NULL;
2248
2249         if (prog) {
2250                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2251                 if (!new)
2252                         return -ENOMEM;
2253                 new->prog = prog;
2254         }
2255
2256         spin_lock_bh(&tun->lock);
2257         old = rcu_dereference_protected(*prog_p,
2258                                         lockdep_is_held(&tun->lock));
2259         rcu_assign_pointer(*prog_p, new);
2260         spin_unlock_bh(&tun->lock);
2261
2262         if (old)
2263                 call_rcu(&old->rcu, tun_prog_free);
2264
2265         return 0;
2266 }
2267
2268 static void tun_free_netdev(struct net_device *dev)
2269 {
2270         struct tun_struct *tun = netdev_priv(dev);
2271
2272         BUG_ON(!(list_empty(&tun->disabled)));
2273         free_percpu(tun->pcpu_stats);
2274         tun_flow_uninit(tun);
2275         security_tun_dev_free_security(tun->security);
2276         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2277         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2278 }
2279
2280 static void tun_setup(struct net_device *dev)
2281 {
2282         struct tun_struct *tun = netdev_priv(dev);
2283
2284         tun->owner = INVALID_UID;
2285         tun->group = INVALID_GID;
2286
2287         dev->ethtool_ops = &tun_ethtool_ops;
2288         dev->needs_free_netdev = true;
2289         dev->priv_destructor = tun_free_netdev;
2290         /* We prefer our own queue length */
2291         dev->tx_queue_len = TUN_READQ_SIZE;
2292 }
2293
2294 /* Trivial set of netlink ops to allow deleting tun or tap
2295  * device with netlink.
2296  */
2297 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2298                         struct netlink_ext_ack *extack)
2299 {
2300         return -EINVAL;
2301 }
2302
2303 static size_t tun_get_size(const struct net_device *dev)
2304 {
2305         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2306         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2307
2308         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2309                nla_total_size(sizeof(gid_t)) + /* GROUP */
2310                nla_total_size(sizeof(u8)) + /* TYPE */
2311                nla_total_size(sizeof(u8)) + /* PI */
2312                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2313                nla_total_size(sizeof(u8)) + /* PERSIST */
2314                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2315                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2316                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2317                0;
2318 }
2319
2320 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2321 {
2322         struct tun_struct *tun = netdev_priv(dev);
2323
2324         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2325                 goto nla_put_failure;
2326         if (uid_valid(tun->owner) &&
2327             nla_put_u32(skb, IFLA_TUN_OWNER,
2328                         from_kuid_munged(current_user_ns(), tun->owner)))
2329                 goto nla_put_failure;
2330         if (gid_valid(tun->group) &&
2331             nla_put_u32(skb, IFLA_TUN_GROUP,
2332                         from_kgid_munged(current_user_ns(), tun->group)))
2333                 goto nla_put_failure;
2334         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2335                 goto nla_put_failure;
2336         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2337                 goto nla_put_failure;
2338         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2339                 goto nla_put_failure;
2340         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2341                        !!(tun->flags & IFF_MULTI_QUEUE)))
2342                 goto nla_put_failure;
2343         if (tun->flags & IFF_MULTI_QUEUE) {
2344                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2345                         goto nla_put_failure;
2346                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2347                                 tun->numdisabled))
2348                         goto nla_put_failure;
2349         }
2350
2351         return 0;
2352
2353 nla_put_failure:
2354         return -EMSGSIZE;
2355 }
2356
2357 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2358         .kind           = DRV_NAME,
2359         .priv_size      = sizeof(struct tun_struct),
2360         .setup          = tun_setup,
2361         .validate       = tun_validate,
2362         .get_size       = tun_get_size,
2363         .fill_info      = tun_fill_info,
2364 };
2365
2366 static void tun_sock_write_space(struct sock *sk)
2367 {
2368         struct tun_file *tfile;
2369         wait_queue_head_t *wqueue;
2370
2371         if (!sock_writeable(sk))
2372                 return;
2373
2374         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2375                 return;
2376
2377         wqueue = sk_sleep(sk);
2378         if (wqueue && waitqueue_active(wqueue))
2379                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2380                                                 EPOLLWRNORM | EPOLLWRBAND);
2381
2382         tfile = container_of(sk, struct tun_file, sk);
2383         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2384 }
2385
2386 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2387 {
2388         int ret;
2389         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2390         struct tun_struct *tun = tun_get(tfile);
2391
2392         if (!tun)
2393                 return -EBADFD;
2394
2395         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2396                            m->msg_flags & MSG_DONTWAIT,
2397                            m->msg_flags & MSG_MORE);
2398         tun_put(tun);
2399         return ret;
2400 }
2401
2402 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2403                        int flags)
2404 {
2405         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2406         struct tun_struct *tun = tun_get(tfile);
2407         void *ptr = m->msg_control;
2408         int ret;
2409
2410         if (!tun) {
2411                 ret = -EBADFD;
2412                 goto out_free;
2413         }
2414
2415         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2416                 ret = -EINVAL;
2417                 goto out_put_tun;
2418         }
2419         if (flags & MSG_ERRQUEUE) {
2420                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2421                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2422                 goto out;
2423         }
2424         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2425         if (ret > (ssize_t)total_len) {
2426                 m->msg_flags |= MSG_TRUNC;
2427                 ret = flags & MSG_TRUNC ? ret : total_len;
2428         }
2429 out:
2430         tun_put(tun);
2431         return ret;
2432
2433 out_put_tun:
2434         tun_put(tun);
2435 out_free:
2436         tun_ptr_free(ptr);
2437         return ret;
2438 }
2439
2440 static int tun_ptr_peek_len(void *ptr)
2441 {
2442         if (likely(ptr)) {
2443                 if (tun_is_xdp_frame(ptr)) {
2444                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2445
2446                         return xdpf->len;
2447                 }
2448                 return __skb_array_len_with_tag(ptr);
2449         } else {
2450                 return 0;
2451         }
2452 }
2453
2454 static int tun_peek_len(struct socket *sock)
2455 {
2456         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2457         struct tun_struct *tun;
2458         int ret = 0;
2459
2460         tun = tun_get(tfile);
2461         if (!tun)
2462                 return 0;
2463
2464         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2465         tun_put(tun);
2466
2467         return ret;
2468 }
2469
2470 /* Ops structure to mimic raw sockets with tun */
2471 static const struct proto_ops tun_socket_ops = {
2472         .peek_len = tun_peek_len,
2473         .sendmsg = tun_sendmsg,
2474         .recvmsg = tun_recvmsg,
2475 };
2476
2477 static struct proto tun_proto = {
2478         .name           = "tun",
2479         .owner          = THIS_MODULE,
2480         .obj_size       = sizeof(struct tun_file),
2481 };
2482
2483 static int tun_flags(struct tun_struct *tun)
2484 {
2485         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2486 }
2487
2488 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2489                               char *buf)
2490 {
2491         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2492         return sprintf(buf, "0x%x\n", tun_flags(tun));
2493 }
2494
2495 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2496                               char *buf)
2497 {
2498         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2499         return uid_valid(tun->owner)?
2500                 sprintf(buf, "%u\n",
2501                         from_kuid_munged(current_user_ns(), tun->owner)):
2502                 sprintf(buf, "-1\n");
2503 }
2504
2505 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2506                               char *buf)
2507 {
2508         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2509         return gid_valid(tun->group) ?
2510                 sprintf(buf, "%u\n",
2511                         from_kgid_munged(current_user_ns(), tun->group)):
2512                 sprintf(buf, "-1\n");
2513 }
2514
2515 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2516 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2517 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2518
2519 static struct attribute *tun_dev_attrs[] = {
2520         &dev_attr_tun_flags.attr,
2521         &dev_attr_owner.attr,
2522         &dev_attr_group.attr,
2523         NULL
2524 };
2525
2526 static const struct attribute_group tun_attr_group = {
2527         .attrs = tun_dev_attrs
2528 };
2529
2530 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2531 {
2532         struct tun_struct *tun;
2533         struct tun_file *tfile = file->private_data;
2534         struct net_device *dev;
2535         int err;
2536
2537         if (tfile->detached)
2538                 return -EINVAL;
2539
2540         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2541                 if (!capable(CAP_NET_ADMIN))
2542                         return -EPERM;
2543
2544                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2545                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2546                         return -EINVAL;
2547         }
2548
2549         dev = __dev_get_by_name(net, ifr->ifr_name);
2550         if (dev) {
2551                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2552                         return -EBUSY;
2553                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2554                         tun = netdev_priv(dev);
2555                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2556                         tun = netdev_priv(dev);
2557                 else
2558                         return -EINVAL;
2559
2560                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2561                     !!(tun->flags & IFF_MULTI_QUEUE))
2562                         return -EINVAL;
2563
2564                 if (tun_not_capable(tun))
2565                         return -EPERM;
2566                 err = security_tun_dev_open(tun->security);
2567                 if (err < 0)
2568                         return err;
2569
2570                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2571                                  ifr->ifr_flags & IFF_NAPI);
2572                 if (err < 0)
2573                         return err;
2574
2575                 if (tun->flags & IFF_MULTI_QUEUE &&
2576                     (tun->numqueues + tun->numdisabled > 1)) {
2577                         /* One or more queue has already been attached, no need
2578                          * to initialize the device again.
2579                          */
2580                         netdev_state_change(dev);
2581                         return 0;
2582                 }
2583
2584                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2585                               (ifr->ifr_flags & TUN_FEATURES);
2586
2587                 netdev_state_change(dev);
2588         } else {
2589                 char *name;
2590                 unsigned long flags = 0;
2591                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2592                              MAX_TAP_QUEUES : 1;
2593
2594                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2595                         return -EPERM;
2596                 err = security_tun_dev_create();
2597                 if (err < 0)
2598                         return err;
2599
2600                 /* Set dev type */
2601                 if (ifr->ifr_flags & IFF_TUN) {
2602                         /* TUN device */
2603                         flags |= IFF_TUN;
2604                         name = "tun%d";
2605                 } else if (ifr->ifr_flags & IFF_TAP) {
2606                         /* TAP device */
2607                         flags |= IFF_TAP;
2608                         name = "tap%d";
2609                 } else
2610                         return -EINVAL;
2611
2612                 if (*ifr->ifr_name)
2613                         name = ifr->ifr_name;
2614
2615                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2616                                        NET_NAME_UNKNOWN, tun_setup, queues,
2617                                        queues);
2618
2619                 if (!dev)
2620                         return -ENOMEM;
2621                 err = dev_get_valid_name(net, dev, name);
2622                 if (err < 0)
2623                         goto err_free_dev;
2624
2625                 dev_net_set(dev, net);
2626                 dev->rtnl_link_ops = &tun_link_ops;
2627                 dev->ifindex = tfile->ifindex;
2628                 dev->sysfs_groups[0] = &tun_attr_group;
2629
2630                 tun = netdev_priv(dev);
2631                 tun->dev = dev;
2632                 tun->flags = flags;
2633                 tun->txflt.count = 0;
2634                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2635
2636                 tun->align = NET_SKB_PAD;
2637                 tun->filter_attached = false;
2638                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2639                 tun->rx_batched = 0;
2640                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2641
2642                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2643                 if (!tun->pcpu_stats) {
2644                         err = -ENOMEM;
2645                         goto err_free_dev;
2646                 }
2647
2648                 spin_lock_init(&tun->lock);
2649
2650                 err = security_tun_dev_alloc_security(&tun->security);
2651                 if (err < 0)
2652                         goto err_free_stat;
2653
2654                 tun_net_init(dev);
2655                 tun_flow_init(tun);
2656
2657                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2658                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2659                                    NETIF_F_HW_VLAN_STAG_TX;
2660                 dev->features = dev->hw_features | NETIF_F_LLTX;
2661                 dev->vlan_features = dev->features &
2662                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2663                                        NETIF_F_HW_VLAN_STAG_TX);
2664
2665                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2666                               (ifr->ifr_flags & TUN_FEATURES);
2667
2668                 INIT_LIST_HEAD(&tun->disabled);
2669                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2670                 if (err < 0)
2671                         goto err_free_flow;
2672
2673                 err = register_netdevice(tun->dev);
2674                 if (err < 0)
2675                         goto err_detach;
2676         }
2677
2678         netif_carrier_on(tun->dev);
2679
2680         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2681
2682         /* Make sure persistent devices do not get stuck in
2683          * xoff state.
2684          */
2685         if (netif_running(tun->dev))
2686                 netif_tx_wake_all_queues(tun->dev);
2687
2688         strcpy(ifr->ifr_name, tun->dev->name);
2689         return 0;
2690
2691 err_detach:
2692         tun_detach_all(dev);
2693         /* register_netdevice() already called tun_free_netdev() */
2694         goto err_free_dev;
2695
2696 err_free_flow:
2697         tun_flow_uninit(tun);
2698         security_tun_dev_free_security(tun->security);
2699 err_free_stat:
2700         free_percpu(tun->pcpu_stats);
2701 err_free_dev:
2702         free_netdev(dev);
2703         return err;
2704 }
2705
2706 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2707                        struct ifreq *ifr)
2708 {
2709         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2710
2711         strcpy(ifr->ifr_name, tun->dev->name);
2712
2713         ifr->ifr_flags = tun_flags(tun);
2714
2715 }
2716
2717 /* This is like a cut-down ethtool ops, except done via tun fd so no
2718  * privs required. */
2719 static int set_offload(struct tun_struct *tun, unsigned long arg)
2720 {
2721         netdev_features_t features = 0;
2722
2723         if (arg & TUN_F_CSUM) {
2724                 features |= NETIF_F_HW_CSUM;
2725                 arg &= ~TUN_F_CSUM;
2726
2727                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2728                         if (arg & TUN_F_TSO_ECN) {
2729                                 features |= NETIF_F_TSO_ECN;
2730                                 arg &= ~TUN_F_TSO_ECN;
2731                         }
2732                         if (arg & TUN_F_TSO4)
2733                                 features |= NETIF_F_TSO;
2734                         if (arg & TUN_F_TSO6)
2735                                 features |= NETIF_F_TSO6;
2736                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2737                 }
2738
2739                 arg &= ~TUN_F_UFO;
2740         }
2741
2742         /* This gives the user a way to test for new features in future by
2743          * trying to set them. */
2744         if (arg)
2745                 return -EINVAL;
2746
2747         tun->set_features = features;
2748         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2749         tun->dev->wanted_features |= features;
2750         netdev_update_features(tun->dev);
2751
2752         return 0;
2753 }
2754
2755 static void tun_detach_filter(struct tun_struct *tun, int n)
2756 {
2757         int i;
2758         struct tun_file *tfile;
2759
2760         for (i = 0; i < n; i++) {
2761                 tfile = rtnl_dereference(tun->tfiles[i]);
2762                 lock_sock(tfile->socket.sk);
2763                 sk_detach_filter(tfile->socket.sk);
2764                 release_sock(tfile->socket.sk);
2765         }
2766
2767         tun->filter_attached = false;
2768 }
2769
2770 static int tun_attach_filter(struct tun_struct *tun)
2771 {
2772         int i, ret = 0;
2773         struct tun_file *tfile;
2774
2775         for (i = 0; i < tun->numqueues; i++) {
2776                 tfile = rtnl_dereference(tun->tfiles[i]);
2777                 lock_sock(tfile->socket.sk);
2778                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2779                 release_sock(tfile->socket.sk);
2780                 if (ret) {
2781                         tun_detach_filter(tun, i);
2782                         return ret;
2783                 }
2784         }
2785
2786         tun->filter_attached = true;
2787         return ret;
2788 }
2789
2790 static void tun_set_sndbuf(struct tun_struct *tun)
2791 {
2792         struct tun_file *tfile;
2793         int i;
2794
2795         for (i = 0; i < tun->numqueues; i++) {
2796                 tfile = rtnl_dereference(tun->tfiles[i]);
2797                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2798         }
2799 }
2800
2801 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2802 {
2803         struct tun_file *tfile = file->private_data;
2804         struct tun_struct *tun;
2805         int ret = 0;
2806
2807         rtnl_lock();
2808
2809         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2810                 tun = tfile->detached;
2811                 if (!tun) {
2812                         ret = -EINVAL;
2813                         goto unlock;
2814                 }
2815                 ret = security_tun_dev_attach_queue(tun->security);
2816                 if (ret < 0)
2817                         goto unlock;
2818                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2819         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2820                 tun = rtnl_dereference(tfile->tun);
2821                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2822                         ret = -EINVAL;
2823                 else
2824                         __tun_detach(tfile, false);
2825         } else
2826                 ret = -EINVAL;
2827
2828         if (ret >= 0)
2829                 netdev_state_change(tun->dev);
2830
2831 unlock:
2832         rtnl_unlock();
2833         return ret;
2834 }
2835
2836 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2837                         void __user *data)
2838 {
2839         struct bpf_prog *prog;
2840         int fd;
2841
2842         if (copy_from_user(&fd, data, sizeof(fd)))
2843                 return -EFAULT;
2844
2845         if (fd == -1) {
2846                 prog = NULL;
2847         } else {
2848                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2849                 if (IS_ERR(prog))
2850                         return PTR_ERR(prog);
2851         }
2852
2853         return __tun_set_ebpf(tun, prog_p, prog);
2854 }
2855
2856 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2857                             unsigned long arg, int ifreq_len)
2858 {
2859         struct tun_file *tfile = file->private_data;
2860         struct tun_struct *tun;
2861         void __user* argp = (void __user*)arg;
2862         struct ifreq ifr;
2863         struct net *net;
2864         kuid_t owner;
2865         kgid_t group;
2866         int sndbuf;
2867         int vnet_hdr_sz;
2868         unsigned int ifindex;
2869         int le;
2870         int ret;
2871         bool do_notify = false;
2872
2873         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2874             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2875                 if (copy_from_user(&ifr, argp, ifreq_len))
2876                         return -EFAULT;
2877         } else {
2878                 memset(&ifr, 0, sizeof(ifr));
2879         }
2880         if (cmd == TUNGETFEATURES) {
2881                 /* Currently this just means: "what IFF flags are valid?".
2882                  * This is needed because we never checked for invalid flags on
2883                  * TUNSETIFF.
2884                  */
2885                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2886                                 (unsigned int __user*)argp);
2887         } else if (cmd == TUNSETQUEUE)
2888                 return tun_set_queue(file, &ifr);
2889
2890         ret = 0;
2891         rtnl_lock();
2892
2893         tun = tun_get(tfile);
2894         net = sock_net(&tfile->sk);
2895         if (cmd == TUNSETIFF) {
2896                 ret = -EEXIST;
2897                 if (tun)
2898                         goto unlock;
2899
2900                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2901
2902                 ret = tun_set_iff(net, file, &ifr);
2903
2904                 if (ret)
2905                         goto unlock;
2906
2907                 if (copy_to_user(argp, &ifr, ifreq_len))
2908                         ret = -EFAULT;
2909                 goto unlock;
2910         }
2911         if (cmd == TUNSETIFINDEX) {
2912                 ret = -EPERM;
2913                 if (tun)
2914                         goto unlock;
2915
2916                 ret = -EFAULT;
2917                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2918                         goto unlock;
2919
2920                 ret = 0;
2921                 tfile->ifindex = ifindex;
2922                 goto unlock;
2923         }
2924         if (cmd == SIOCGSKNS) {
2925                 ret = -EPERM;
2926                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2927                         goto unlock;
2928
2929                 ret = open_related_ns(&net->ns, get_net_ns);
2930                 goto unlock;
2931         }
2932
2933         ret = -EBADFD;
2934         if (!tun)
2935                 goto unlock;
2936
2937         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2938
2939         ret = 0;
2940         switch (cmd) {
2941         case TUNGETIFF:
2942                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2943
2944                 if (tfile->detached)
2945                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2946                 if (!tfile->socket.sk->sk_filter)
2947                         ifr.ifr_flags |= IFF_NOFILTER;
2948
2949                 if (copy_to_user(argp, &ifr, ifreq_len))
2950                         ret = -EFAULT;
2951                 break;
2952
2953         case TUNSETNOCSUM:
2954                 /* Disable/Enable checksum */
2955
2956                 /* [unimplemented] */
2957                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2958                           arg ? "disabled" : "enabled");
2959                 break;
2960
2961         case TUNSETPERSIST:
2962                 /* Disable/Enable persist mode. Keep an extra reference to the
2963                  * module to prevent the module being unprobed.
2964                  */
2965                 if (arg && !(tun->flags & IFF_PERSIST)) {
2966                         tun->flags |= IFF_PERSIST;
2967                         __module_get(THIS_MODULE);
2968                         do_notify = true;
2969                 }
2970                 if (!arg && (tun->flags & IFF_PERSIST)) {
2971                         tun->flags &= ~IFF_PERSIST;
2972                         module_put(THIS_MODULE);
2973                         do_notify = true;
2974                 }
2975
2976                 tun_debug(KERN_INFO, tun, "persist %s\n",
2977                           arg ? "enabled" : "disabled");
2978                 break;
2979
2980         case TUNSETOWNER:
2981                 /* Set owner of the device */
2982                 owner = make_kuid(current_user_ns(), arg);
2983                 if (!uid_valid(owner)) {
2984                         ret = -EINVAL;
2985                         break;
2986                 }
2987                 tun->owner = owner;
2988                 do_notify = true;
2989                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2990                           from_kuid(&init_user_ns, tun->owner));
2991                 break;
2992
2993         case TUNSETGROUP:
2994                 /* Set group of the device */
2995                 group = make_kgid(current_user_ns(), arg);
2996                 if (!gid_valid(group)) {
2997                         ret = -EINVAL;
2998                         break;
2999                 }
3000                 tun->group = group;
3001                 do_notify = true;
3002                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3003                           from_kgid(&init_user_ns, tun->group));
3004                 break;
3005
3006         case TUNSETLINK:
3007                 /* Only allow setting the type when the interface is down */
3008                 if (tun->dev->flags & IFF_UP) {
3009                         tun_debug(KERN_INFO, tun,
3010                                   "Linktype set failed because interface is up\n");
3011                         ret = -EBUSY;
3012                 } else {
3013                         tun->dev->type = (int) arg;
3014                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3015                                   tun->dev->type);
3016                         ret = 0;
3017                 }
3018                 break;
3019
3020 #ifdef TUN_DEBUG
3021         case TUNSETDEBUG:
3022                 tun->debug = arg;
3023                 break;
3024 #endif
3025         case TUNSETOFFLOAD:
3026                 ret = set_offload(tun, arg);
3027                 break;
3028
3029         case TUNSETTXFILTER:
3030                 /* Can be set only for TAPs */
3031                 ret = -EINVAL;
3032                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3033                         break;
3034                 ret = update_filter(&tun->txflt, (void __user *)arg);
3035                 break;
3036
3037         case SIOCGIFHWADDR:
3038                 /* Get hw address */
3039                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3040                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3041                 if (copy_to_user(argp, &ifr, ifreq_len))
3042                         ret = -EFAULT;
3043                 break;
3044
3045         case SIOCSIFHWADDR:
3046                 /* Set hw address */
3047                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3048                           ifr.ifr_hwaddr.sa_data);
3049
3050                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3051                 break;
3052
3053         case TUNGETSNDBUF:
3054                 sndbuf = tfile->socket.sk->sk_sndbuf;
3055                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3056                         ret = -EFAULT;
3057                 break;
3058
3059         case TUNSETSNDBUF:
3060                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3061                         ret = -EFAULT;
3062                         break;
3063                 }
3064                 if (sndbuf <= 0) {
3065                         ret = -EINVAL;
3066                         break;
3067                 }
3068
3069                 tun->sndbuf = sndbuf;
3070                 tun_set_sndbuf(tun);
3071                 break;
3072
3073         case TUNGETVNETHDRSZ:
3074                 vnet_hdr_sz = tun->vnet_hdr_sz;
3075                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3076                         ret = -EFAULT;
3077                 break;
3078
3079         case TUNSETVNETHDRSZ:
3080                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3081                         ret = -EFAULT;
3082                         break;
3083                 }
3084                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3085                         ret = -EINVAL;
3086                         break;
3087                 }
3088
3089                 tun->vnet_hdr_sz = vnet_hdr_sz;
3090                 break;
3091
3092         case TUNGETVNETLE:
3093                 le = !!(tun->flags & TUN_VNET_LE);
3094                 if (put_user(le, (int __user *)argp))
3095                         ret = -EFAULT;
3096                 break;
3097
3098         case TUNSETVNETLE:
3099                 if (get_user(le, (int __user *)argp)) {
3100                         ret = -EFAULT;
3101                         break;
3102                 }
3103                 if (le)
3104                         tun->flags |= TUN_VNET_LE;
3105                 else
3106                         tun->flags &= ~TUN_VNET_LE;
3107                 break;
3108
3109         case TUNGETVNETBE:
3110                 ret = tun_get_vnet_be(tun, argp);
3111                 break;
3112
3113         case TUNSETVNETBE:
3114                 ret = tun_set_vnet_be(tun, argp);
3115                 break;
3116
3117         case TUNATTACHFILTER:
3118                 /* Can be set only for TAPs */
3119                 ret = -EINVAL;
3120                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3121                         break;
3122                 ret = -EFAULT;
3123                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3124                         break;
3125
3126                 ret = tun_attach_filter(tun);
3127                 break;
3128
3129         case TUNDETACHFILTER:
3130                 /* Can be set only for TAPs */
3131                 ret = -EINVAL;
3132                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3133                         break;
3134                 ret = 0;
3135                 tun_detach_filter(tun, tun->numqueues);
3136                 break;
3137
3138         case TUNGETFILTER:
3139                 ret = -EINVAL;
3140                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3141                         break;
3142                 ret = -EFAULT;
3143                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3144                         break;
3145                 ret = 0;
3146                 break;
3147
3148         case TUNSETSTEERINGEBPF:
3149                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3150                 break;
3151
3152         case TUNSETFILTEREBPF:
3153                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3154                 break;
3155
3156         default:
3157                 ret = -EINVAL;
3158                 break;
3159         }
3160
3161         if (do_notify)
3162                 netdev_state_change(tun->dev);
3163
3164 unlock:
3165         rtnl_unlock();
3166         if (tun)
3167                 tun_put(tun);
3168         return ret;
3169 }
3170
3171 static long tun_chr_ioctl(struct file *file,
3172                           unsigned int cmd, unsigned long arg)
3173 {
3174         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3175 }
3176
3177 #ifdef CONFIG_COMPAT
3178 static long tun_chr_compat_ioctl(struct file *file,
3179                          unsigned int cmd, unsigned long arg)
3180 {
3181         switch (cmd) {
3182         case TUNSETIFF:
3183         case TUNGETIFF:
3184         case TUNSETTXFILTER:
3185         case TUNGETSNDBUF:
3186         case TUNSETSNDBUF:
3187         case SIOCGIFHWADDR:
3188         case SIOCSIFHWADDR:
3189                 arg = (unsigned long)compat_ptr(arg);
3190                 break;
3191         default:
3192                 arg = (compat_ulong_t)arg;
3193                 break;
3194         }
3195
3196         /*
3197          * compat_ifreq is shorter than ifreq, so we must not access beyond
3198          * the end of that structure. All fields that are used in this
3199          * driver are compatible though, we don't need to convert the
3200          * contents.
3201          */
3202         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3203 }
3204 #endif /* CONFIG_COMPAT */
3205
3206 static int tun_chr_fasync(int fd, struct file *file, int on)
3207 {
3208         struct tun_file *tfile = file->private_data;
3209         int ret;
3210
3211         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3212                 goto out;
3213
3214         if (on) {
3215                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3216                 tfile->flags |= TUN_FASYNC;
3217         } else
3218                 tfile->flags &= ~TUN_FASYNC;
3219         ret = 0;
3220 out:
3221         return ret;
3222 }
3223
3224 static int tun_chr_open(struct inode *inode, struct file * file)
3225 {
3226         struct net *net = current->nsproxy->net_ns;
3227         struct tun_file *tfile;
3228
3229         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3230
3231         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3232                                             &tun_proto, 0);
3233         if (!tfile)
3234                 return -ENOMEM;
3235         RCU_INIT_POINTER(tfile->tun, NULL);
3236         tfile->flags = 0;
3237         tfile->ifindex = 0;
3238
3239         init_waitqueue_head(&tfile->wq.wait);
3240         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3241
3242         tfile->socket.file = file;
3243         tfile->socket.ops = &tun_socket_ops;
3244
3245         sock_init_data(&tfile->socket, &tfile->sk);
3246
3247         tfile->sk.sk_write_space = tun_sock_write_space;
3248         tfile->sk.sk_sndbuf = INT_MAX;
3249
3250         file->private_data = tfile;
3251         INIT_LIST_HEAD(&tfile->next);
3252
3253         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3254
3255         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3256
3257         return 0;
3258 }
3259
3260 static int tun_chr_close(struct inode *inode, struct file *file)
3261 {
3262         struct tun_file *tfile = file->private_data;
3263
3264         tun_detach(tfile, true);
3265
3266         return 0;
3267 }
3268
3269 #ifdef CONFIG_PROC_FS
3270 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3271 {
3272         struct tun_file *tfile = file->private_data;
3273         struct tun_struct *tun;
3274         struct ifreq ifr;
3275
3276         memset(&ifr, 0, sizeof(ifr));
3277
3278         rtnl_lock();
3279         tun = tun_get(tfile);
3280         if (tun)
3281                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3282         rtnl_unlock();
3283
3284         if (tun)
3285                 tun_put(tun);
3286
3287         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3288 }
3289 #endif
3290
3291 static const struct file_operations tun_fops = {
3292         .owner  = THIS_MODULE,
3293         .llseek = no_llseek,
3294         .read_iter  = tun_chr_read_iter,
3295         .write_iter = tun_chr_write_iter,
3296         .poll   = tun_chr_poll,
3297         .unlocked_ioctl = tun_chr_ioctl,
3298 #ifdef CONFIG_COMPAT
3299         .compat_ioctl = tun_chr_compat_ioctl,
3300 #endif
3301         .open   = tun_chr_open,
3302         .release = tun_chr_close,
3303         .fasync = tun_chr_fasync,
3304 #ifdef CONFIG_PROC_FS
3305         .show_fdinfo = tun_chr_show_fdinfo,
3306 #endif
3307 };
3308
3309 static struct miscdevice tun_miscdev = {
3310         .minor = TUN_MINOR,
3311         .name = "tun",
3312         .nodename = "net/tun",
3313         .fops = &tun_fops,
3314 };
3315
3316 /* ethtool interface */
3317
3318 static int tun_get_link_ksettings(struct net_device *dev,
3319                                   struct ethtool_link_ksettings *cmd)
3320 {
3321         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3322         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3323         cmd->base.speed         = SPEED_10;
3324         cmd->base.duplex        = DUPLEX_FULL;
3325         cmd->base.port          = PORT_TP;
3326         cmd->base.phy_address   = 0;
3327         cmd->base.autoneg       = AUTONEG_DISABLE;
3328         return 0;
3329 }
3330
3331 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3332 {
3333         struct tun_struct *tun = netdev_priv(dev);
3334
3335         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3336         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3337
3338         switch (tun->flags & TUN_TYPE_MASK) {
3339         case IFF_TUN:
3340                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3341                 break;
3342         case IFF_TAP:
3343                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3344                 break;
3345         }
3346 }
3347
3348 static u32 tun_get_msglevel(struct net_device *dev)
3349 {
3350 #ifdef TUN_DEBUG
3351         struct tun_struct *tun = netdev_priv(dev);
3352         return tun->debug;
3353 #else
3354         return -EOPNOTSUPP;
3355 #endif
3356 }
3357
3358 static void tun_set_msglevel(struct net_device *dev, u32 value)
3359 {
3360 #ifdef TUN_DEBUG
3361         struct tun_struct *tun = netdev_priv(dev);
3362         tun->debug = value;
3363 #endif
3364 }
3365
3366 static int tun_get_coalesce(struct net_device *dev,
3367                             struct ethtool_coalesce *ec)
3368 {
3369         struct tun_struct *tun = netdev_priv(dev);
3370
3371         ec->rx_max_coalesced_frames = tun->rx_batched;
3372
3373         return 0;
3374 }
3375
3376 static int tun_set_coalesce(struct net_device *dev,
3377                             struct ethtool_coalesce *ec)
3378 {
3379         struct tun_struct *tun = netdev_priv(dev);
3380
3381         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3382                 tun->rx_batched = NAPI_POLL_WEIGHT;
3383         else
3384                 tun->rx_batched = ec->rx_max_coalesced_frames;
3385
3386         return 0;
3387 }
3388
3389 static const struct ethtool_ops tun_ethtool_ops = {
3390         .get_drvinfo    = tun_get_drvinfo,
3391         .get_msglevel   = tun_get_msglevel,
3392         .set_msglevel   = tun_set_msglevel,
3393         .get_link       = ethtool_op_get_link,
3394         .get_ts_info    = ethtool_op_get_ts_info,
3395         .get_coalesce   = tun_get_coalesce,
3396         .set_coalesce   = tun_set_coalesce,
3397         .get_link_ksettings = tun_get_link_ksettings,
3398 };
3399
3400 static int tun_queue_resize(struct tun_struct *tun)
3401 {
3402         struct net_device *dev = tun->dev;
3403         struct tun_file *tfile;
3404         struct ptr_ring **rings;
3405         int n = tun->numqueues + tun->numdisabled;
3406         int ret, i;
3407
3408         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3409         if (!rings)
3410                 return -ENOMEM;
3411
3412         for (i = 0; i < tun->numqueues; i++) {
3413                 tfile = rtnl_dereference(tun->tfiles[i]);
3414                 rings[i] = &tfile->tx_ring;
3415         }
3416         list_for_each_entry(tfile, &tun->disabled, next)
3417                 rings[i++] = &tfile->tx_ring;
3418
3419         ret = ptr_ring_resize_multiple(rings, n,
3420                                        dev->tx_queue_len, GFP_KERNEL,
3421                                        tun_ptr_free);
3422
3423         kfree(rings);
3424         return ret;
3425 }
3426
3427 static int tun_device_event(struct notifier_block *unused,
3428                             unsigned long event, void *ptr)
3429 {
3430         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3431         struct tun_struct *tun = netdev_priv(dev);
3432
3433         if (dev->rtnl_link_ops != &tun_link_ops)
3434                 return NOTIFY_DONE;
3435
3436         switch (event) {
3437         case NETDEV_CHANGE_TX_QUEUE_LEN:
3438                 if (tun_queue_resize(tun))
3439                         return NOTIFY_BAD;
3440                 break;
3441         default:
3442                 break;
3443         }
3444
3445         return NOTIFY_DONE;
3446 }
3447
3448 static struct notifier_block tun_notifier_block __read_mostly = {
3449         .notifier_call  = tun_device_event,
3450 };
3451
3452 static int __init tun_init(void)
3453 {
3454         int ret = 0;
3455
3456         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3457
3458         ret = rtnl_link_register(&tun_link_ops);
3459         if (ret) {
3460                 pr_err("Can't register link_ops\n");
3461                 goto err_linkops;
3462         }
3463
3464         ret = misc_register(&tun_miscdev);
3465         if (ret) {
3466                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3467                 goto err_misc;
3468         }
3469
3470         ret = register_netdevice_notifier(&tun_notifier_block);
3471         if (ret) {
3472                 pr_err("Can't register netdevice notifier\n");
3473                 goto err_notifier;
3474         }
3475
3476         return  0;
3477
3478 err_notifier:
3479         misc_deregister(&tun_miscdev);
3480 err_misc:
3481         rtnl_link_unregister(&tun_link_ops);
3482 err_linkops:
3483         return ret;
3484 }
3485
3486 static void tun_cleanup(void)
3487 {
3488         misc_deregister(&tun_miscdev);
3489         rtnl_link_unregister(&tun_link_ops);
3490         unregister_netdevice_notifier(&tun_notifier_block);
3491 }
3492
3493 /* Get an underlying socket object from tun file.  Returns error unless file is
3494  * attached to a device.  The returned object works like a packet socket, it
3495  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3496  * holding a reference to the file for as long as the socket is in use. */
3497 struct socket *tun_get_socket(struct file *file)
3498 {
3499         struct tun_file *tfile;
3500         if (file->f_op != &tun_fops)
3501                 return ERR_PTR(-EINVAL);
3502         tfile = file->private_data;
3503         if (!tfile)
3504                 return ERR_PTR(-EBADFD);
3505         return &tfile->socket;
3506 }
3507 EXPORT_SYMBOL_GPL(tun_get_socket);
3508
3509 struct ptr_ring *tun_get_tx_ring(struct file *file)
3510 {
3511         struct tun_file *tfile;
3512
3513         if (file->f_op != &tun_fops)
3514                 return ERR_PTR(-EINVAL);
3515         tfile = file->private_data;
3516         if (!tfile)
3517                 return ERR_PTR(-EBADFD);
3518         return &tfile->tx_ring;
3519 }
3520 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3521
3522 module_init(tun_init);
3523 module_exit(tun_cleanup);
3524 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3525 MODULE_AUTHOR(DRV_COPYRIGHT);
3526 MODULE_LICENSE("GPL");
3527 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3528 MODULE_ALIAS("devname:net/tun");