drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_HEADROOM 256
 117 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 118
 119 /* TUN device flags */
 120
 121 /* IFF_ATTACH_QUEUE is never stored in device flags,
 122  * overload it to mean fasync when stored there.
 123  */
 124 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 125 /* High bits in flags field are unused. */
 126 #define TUN_VNET_LE     0x80000000
 127 #define TUN_VNET_BE     0x40000000
 128
 129 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 130                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 131
 132 #define GOODCOPY_LEN 128
 133
 134 #define FLT_EXACT_COUNT 8
 135 struct tap_filter {
 136         unsigned int    count;    /* Number of addrs. Zero means disabled */
 137         u32             mask[2];  /* Mask of the hashed addrs */
 138         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 139 };
 140
 141 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 142  * to max number of VCPUs in guest. */
 143 #define MAX_TAP_QUEUES 256
 144 #define MAX_TAP_FLOWS  4096
 145
 146 #define TUN_FLOW_EXPIRE (3 * HZ)
 147
 148 struct tun_pcpu_stats {
 149         u64 rx_packets;
 150         u64 rx_bytes;
 151         u64 tx_packets;
 152         u64 tx_bytes;
 153         struct u64_stats_sync syncp;
 154         u32 rx_dropped;
 155         u32 tx_dropped;
 156         u32 rx_frame_errors;
 157 };
 158
 159 /* A tun_file connects an open character device to a tuntap netdevice. It
 160  * also contains all socket related structures (except sock_fprog and tap_filter)
 161  * to serve as one transmit queue for tuntap device. The sock_fprog and
 162  * tap_filter were kept in tun_struct since they were used for filtering for the
 163  * netdevice not for a specific queue (at least I didn't see the requirement for
 164  * this).
 165  *
 166  * RCU usage:
 167  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 168  * other can only be read while rcu_read_lock or rtnl_lock is held.
 169  */
 170 struct tun_file {
 171         struct sock sk;
 172         struct socket socket;
 173         struct socket_wq wq;
 174         struct tun_struct __rcu *tun;
 175         struct fasync_struct *fasync;
 176         /* only used for fasnyc */
 177         unsigned int flags;
 178         union {
 179                 u16 queue_index;
 180                 unsigned int ifindex;
 181         };
 182         struct napi_struct napi;
 183         bool napi_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_flow_entry {
 192         struct hlist_node hash_link;
 193         struct rcu_head rcu;
 194         struct tun_struct *tun;
 195
 196         u32 rxhash;
 197         u32 rps_rxhash;
 198         int queue_index;
 199         unsigned long updated;
 200 };
 201
 202 #define TUN_NUM_FLOW_ENTRIES 1024
 203 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 204
 205 struct tun_prog {
 206         struct rcu_head rcu;
 207         struct bpf_prog *prog;
 208 };
 209
 210 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 211  * device, socket filter, sndbuf and vnet header size were restore when the
 212  * file were attached to a persist device.
 213  */
 214 struct tun_struct {
 215         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 216         unsigned int            numqueues;
 217         unsigned int            flags;
 218         kuid_t                  owner;
 219         kgid_t                  group;
 220
 221         struct net_device       *dev;
 222         netdev_features_t       set_features;
 223 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 224                           NETIF_F_TSO6)
 225
 226         int                     align;
 227         int                     vnet_hdr_sz;
 228         int                     sndbuf;
 229         struct tap_filter       txflt;
 230         struct sock_fprog       fprog;
 231         /* protected by rtnl lock */
 232         bool                    filter_attached;
 233 #ifdef TUN_DEBUG
 234         int debug;
 235 #endif
 236         spinlock_t lock;
 237         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 238         struct timer_list flow_gc_timer;
 239         unsigned long ageing_time;
 240         unsigned int numdisabled;
 241         struct list_head disabled;
 242         void *security;
 243         u32 flow_count;
 244         u32 rx_batched;
 245         struct tun_pcpu_stats __percpu *pcpu_stats;
 246         struct bpf_prog __rcu *xdp_prog;
 247         struct tun_prog __rcu *steering_prog;
 248         struct tun_prog __rcu *filter_prog;
 249         struct ethtool_link_ksettings link_ksettings;
 250 };
 251
 252 struct veth {
 253         __be16 h_vlan_proto;
 254         __be16 h_vlan_TCI;
 255 };
 256
 257 bool tun_is_xdp_frame(void *ptr)
 258 {
 259         return (unsigned long)ptr & TUN_XDP_FLAG;
 260 }
 261 EXPORT_SYMBOL(tun_is_xdp_frame);
 262
 263 void *tun_xdp_to_ptr(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_xdp_to_ptr);
 268
 269 void *tun_ptr_to_xdp(void *ptr)
 270 {
 271         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 272 }
 273 EXPORT_SYMBOL(tun_ptr_to_xdp);
 274
 275 static int tun_napi_receive(struct napi_struct *napi, int budget)
 276 {
 277         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 278         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 279         struct sk_buff_head process_queue;
 280         struct sk_buff *skb;
 281         int received = 0;
 282
 283         __skb_queue_head_init(&process_queue);
 284
 285         spin_lock(&queue->lock);
 286         skb_queue_splice_tail_init(queue, &process_queue);
 287         spin_unlock(&queue->lock);
 288
 289         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 290                 napi_gro_receive(napi, skb);
 291                 ++received;
 292         }
 293
 294         if (!skb_queue_empty(&process_queue)) {
 295                 spin_lock(&queue->lock);
 296                 skb_queue_splice(&process_queue, queue);
 297                 spin_unlock(&queue->lock);
 298         }
 299
 300         return received;
 301 }
 302
 303 static int tun_napi_poll(struct napi_struct *napi, int budget)
 304 {
 305         unsigned int received;
 306
 307         received = tun_napi_receive(napi, budget);
 308
 309         if (received < budget)
 310                 napi_complete_done(napi, received);
 311
 312         return received;
 313 }
 314
 315 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 316                           bool napi_en)
 317 {
 318         tfile->napi_enabled = napi_en;
 319         if (napi_en) {
 320                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 321                                NAPI_POLL_WEIGHT);
 322                 napi_enable(&tfile->napi);
 323         }
 324 }
 325
 326 static void tun_napi_disable(struct tun_file *tfile)
 327 {
 328         if (tfile->napi_enabled)
 329                 napi_disable(&tfile->napi);
 330 }
 331
 332 static void tun_napi_del(struct tun_file *tfile)
 333 {
 334         if (tfile->napi_enabled)
 335                 netif_napi_del(&tfile->napi);
 336 }
 337
 338 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 339 {
 340         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 341 }
 342
 343 #ifdef CONFIG_TUN_VNET_CROSS_LE
 344 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 345 {
 346         return tun->flags & TUN_VNET_BE ? false :
 347                 virtio_legacy_is_little_endian();
 348 }
 349
 350 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 351 {
 352         int be = !!(tun->flags & TUN_VNET_BE);
 353
 354         if (put_user(be, argp))
 355                 return -EFAULT;
 356
 357         return 0;
 358 }
 359
 360 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 361 {
 362         int be;
 363
 364         if (get_user(be, argp))
 365                 return -EFAULT;
 366
 367         if (be)
 368                 tun->flags |= TUN_VNET_BE;
 369         else
 370                 tun->flags &= ~TUN_VNET_BE;
 371
 372         return 0;
 373 }
 374 #else
 375 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 376 {
 377         return virtio_legacy_is_little_endian();
 378 }
 379
 380 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384
 385 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 386 {
 387         return -EINVAL;
 388 }
 389 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 390
 391 static inline bool tun_is_little_endian(struct tun_struct *tun)
 392 {
 393         return tun->flags & TUN_VNET_LE ||
 394                 tun_legacy_is_little_endian(tun);
 395 }
 396
 397 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 398 {
 399         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 403 {
 404         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 405 }
 406
 407 static inline u32 tun_hashfn(u32 rxhash)
 408 {
 409         return rxhash & TUN_MASK_FLOW_ENTRIES;
 410 }
 411
 412 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 413 {
 414         struct tun_flow_entry *e;
 415
 416         hlist_for_each_entry_rcu(e, head, hash_link) {
 417                 if (e->rxhash == rxhash)
 418                         return e;
 419         }
 420         return NULL;
 421 }
 422
 423 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 424                                               struct hlist_head *head,
 425                                               u32 rxhash, u16 queue_index)
 426 {
 427         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 428
 429         if (e) {
 430                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 431                           rxhash, queue_index);
 432                 e->updated = jiffies;
 433                 e->rxhash = rxhash;
 434                 e->rps_rxhash = 0;
 435                 e->queue_index = queue_index;
 436                 e->tun = tun;
 437                 hlist_add_head_rcu(&e->hash_link, head);
 438                 ++tun->flow_count;
 439         }
 440         return e;
 441 }
 442
 443 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 444 {
 445         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 446                   e->rxhash, e->queue_index);
 447         hlist_del_rcu(&e->hash_link);
 448         kfree_rcu(e, rcu);
 449         --tun->flow_count;
 450 }
 451
 452 static void tun_flow_flush(struct tun_struct *tun)
 453 {
 454         int i;
 455
 456         spin_lock_bh(&tun->lock);
 457         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 458                 struct tun_flow_entry *e;
 459                 struct hlist_node *n;
 460
 461                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 462                         tun_flow_delete(tun, e);
 463         }
 464         spin_unlock_bh(&tun->lock);
 465 }
 466
 467 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 468 {
 469         int i;
 470
 471         spin_lock_bh(&tun->lock);
 472         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 473                 struct tun_flow_entry *e;
 474                 struct hlist_node *n;
 475
 476                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 477                         if (e->queue_index == queue_index)
 478                                 tun_flow_delete(tun, e);
 479                 }
 480         }
 481         spin_unlock_bh(&tun->lock);
 482 }
 483
 484 static void tun_flow_cleanup(struct timer_list *t)
 485 {
 486         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 487         unsigned long delay = tun->ageing_time;
 488         unsigned long next_timer = jiffies + delay;
 489         unsigned long count = 0;
 490         int i;
 491
 492         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 493
 494         spin_lock(&tun->lock);
 495         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 496                 struct tun_flow_entry *e;
 497                 struct hlist_node *n;
 498
 499                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 500                         unsigned long this_timer;
 501
 502                         this_timer = e->updated + delay;
 503                         if (time_before_eq(this_timer, jiffies)) {
 504                                 tun_flow_delete(tun, e);
 505                                 continue;
 506                         }
 507                         count++;
 508                         if (time_before(this_timer, next_timer))
 509                                 next_timer = this_timer;
 510                 }
 511         }
 512
 513         if (count)
 514                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 515         spin_unlock(&tun->lock);
 516 }
 517
 518 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 519                             struct tun_file *tfile)
 520 {
 521         struct hlist_head *head;
 522         struct tun_flow_entry *e;
 523         unsigned long delay = tun->ageing_time;
 524         u16 queue_index = tfile->queue_index;
 525
 526         if (!rxhash)
 527                 return;
 528         else
 529                 head = &tun->flows[tun_hashfn(rxhash)];
 530
 531         rcu_read_lock();
 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         rcu_read_unlock();
 552 }
 553
 554 /**
 555  * Save the hash received in the stack receive path and update the
 556  * flow_hash table accordingly.
 557  */
 558 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 559 {
 560         if (unlikely(e->rps_rxhash != hash))
 561                 e->rps_rxhash = hash;
 562 }
 563
 564 /* We try to identify a flow through its rxhash first. The reason that
 565  * we do not check rxq no. is because some cards(e.g 82599), chooses
 566  * the rxq based on the txq where the last packet of the flow comes. As
 567  * the userspace application move between processors, we may get a
 568  * different rxq no. here. If we could not get rxhash, then we would
 569  * hope the rxq no. may help here.
 570  */
 571 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 572 {
 573         struct tun_flow_entry *e;
 574         u32 txq = 0;
 575         u32 numqueues = 0;
 576
 577         numqueues = READ_ONCE(tun->numqueues);
 578
 579         txq = __skb_get_hash_symmetric(skb);
 580         if (txq) {
 581                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 582                 if (e) {
 583                         tun_flow_save_rps_rxhash(e, txq);
 584                         txq = e->queue_index;
 585                 } else
 586                         /* use multiply and shift instead of expensive divide */
 587                         txq = ((u64)txq * numqueues) >> 32;
 588         } else if (likely(skb_rx_queue_recorded(skb))) {
 589                 txq = skb_get_rx_queue(skb);
 590                 while (unlikely(txq >= numqueues))
 591                         txq -= numqueues;
 592         }
 593
 594         return txq;
 595 }
 596
 597 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 598 {
 599         struct tun_prog *prog;
 600         u16 ret = 0;
 601
 602         prog = rcu_dereference(tun->steering_prog);
 603         if (prog)
 604                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 605
 606         return ret % tun->numqueues;
 607 }
 608
 609 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 610                             struct net_device *sb_dev,
 611                             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_detach(struct tun_file *tfile, bool clean)
 685 {
 686         struct tun_file *ntfile;
 687         struct tun_struct *tun;
 688
 689         tun = rtnl_dereference(tfile->tun);
 690
 691         if (tun && clean) {
 692                 tun_napi_disable(tfile);
 693                 tun_napi_del(tfile);
 694         }
 695
 696         if (tun && !tfile->detached) {
 697                 u16 index = tfile->queue_index;
 698                 BUG_ON(index >= tun->numqueues);
 699
 700                 rcu_assign_pointer(tun->tfiles[index],
 701                                    tun->tfiles[tun->numqueues - 1]);
 702                 ntfile = rtnl_dereference(tun->tfiles[index]);
 703                 ntfile->queue_index = index;
 704
 705                 --tun->numqueues;
 706                 if (clean) {
 707                         RCU_INIT_POINTER(tfile->tun, NULL);
 708                         sock_put(&tfile->sk);
 709                 } else
 710                         tun_disable_queue(tun, tfile);
 711
 712                 synchronize_net();
 713                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 714                 /* Drop read queue */
 715                 tun_queue_purge(tfile);
 716                 tun_set_real_num_queues(tun);
 717         } else if (tfile->detached && clean) {
 718                 tun = tun_enable_queue(tfile);
 719                 sock_put(&tfile->sk);
 720         }
 721
 722         if (clean) {
 723                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 724                         netif_carrier_off(tun->dev);
 725
 726                         if (!(tun->flags & IFF_PERSIST) &&
 727                             tun->dev->reg_state == NETREG_REGISTERED)
 728                                 unregister_netdevice(tun->dev);
 729                 }
 730                 if (tun)
 731                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 732                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 733                 sock_put(&tfile->sk);
 734         }
 735 }
 736
 737 static void tun_detach(struct tun_file *tfile, bool clean)
 738 {
 739         struct tun_struct *tun;
 740         struct net_device *dev;
 741
 742         rtnl_lock();
 743         tun = rtnl_dereference(tfile->tun);
 744         dev = tun ? tun->dev : NULL;
 745         __tun_detach(tfile, clean);
 746         if (dev)
 747                 netdev_state_change(dev);
 748         rtnl_unlock();
 749 }
 750
 751 static void tun_detach_all(struct net_device *dev)
 752 {
 753         struct tun_struct *tun = netdev_priv(dev);
 754         struct tun_file *tfile, *tmp;
 755         int i, n = tun->numqueues;
 756
 757         for (i = 0; i < n; i++) {
 758                 tfile = rtnl_dereference(tun->tfiles[i]);
 759                 BUG_ON(!tfile);
 760                 tun_napi_disable(tfile);
 761                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 762                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 763                 RCU_INIT_POINTER(tfile->tun, NULL);
 764                 --tun->numqueues;
 765         }
 766         list_for_each_entry(tfile, &tun->disabled, next) {
 767                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 768                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 769                 RCU_INIT_POINTER(tfile->tun, NULL);
 770         }
 771         BUG_ON(tun->numqueues != 0);
 772
 773         synchronize_net();
 774         for (i = 0; i < n; i++) {
 775                 tfile = rtnl_dereference(tun->tfiles[i]);
 776                 tun_napi_del(tfile);
 777                 /* Drop read queue */
 778                 tun_queue_purge(tfile);
 779                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 780                 sock_put(&tfile->sk);
 781         }
 782         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 783                 tun_enable_queue(tfile);
 784                 tun_queue_purge(tfile);
 785                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 786                 sock_put(&tfile->sk);
 787         }
 788         BUG_ON(tun->numdisabled != 0);
 789
 790         if (tun->flags & IFF_PERSIST)
 791                 module_put(THIS_MODULE);
 792 }
 793
 794 static int tun_attach(struct tun_struct *tun, struct file *file,
 795                       bool skip_filter, bool napi)
 796 {
 797         struct tun_file *tfile = file->private_data;
 798         struct net_device *dev = tun->dev;
 799         int err;
 800
 801         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 802         if (err < 0)
 803                 goto out;
 804
 805         err = -EINVAL;
 806         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 807                 goto out;
 808
 809         err = -EBUSY;
 810         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 811                 goto out;
 812
 813         err = -E2BIG;
 814         if (!tfile->detached &&
 815             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 816                 goto out;
 817
 818         err = 0;
 819
 820         /* Re-attach the filter to persist device */
 821         if (!skip_filter && (tun->filter_attached == true)) {
 822                 lock_sock(tfile->socket.sk);
 823                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 824                 release_sock(tfile->socket.sk);
 825                 if (!err)
 826                         goto out;
 827         }
 828
 829         if (!tfile->detached &&
 830             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 831                             GFP_KERNEL, tun_ptr_free)) {
 832                 err = -ENOMEM;
 833                 goto out;
 834         }
 835
 836         tfile->queue_index = tun->numqueues;
 837         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 838
 839         if (tfile->detached) {
 840                 /* Re-attach detached tfile, updating XDP queue_index */
 841                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 842
 843                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 844                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 845         } else {
 846                 /* Setup XDP RX-queue info, for new tfile getting attached */
 847                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 848                                        tun->dev, tfile->queue_index);
 849                 if (err < 0)
 850                         goto out;
 851                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 852                                                  MEM_TYPE_PAGE_SHARED, NULL);
 853                 if (err < 0) {
 854                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 855                         goto out;
 856                 }
 857                 err = 0;
 858         }
 859
 860         rcu_assign_pointer(tfile->tun, tun);
 861         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 862         tun->numqueues++;
 863
 864         if (tfile->detached) {
 865                 tun_enable_queue(tfile);
 866         } else {
 867                 sock_hold(&tfile->sk);
 868                 tun_napi_init(tun, tfile, napi);
 869         }
 870
 871         tun_set_real_num_queues(tun);
 872
 873         /* device is allowed to go away first, so no need to hold extra
 874          * refcnt.
 875          */
 876
 877 out:
 878         return err;
 879 }
 880
 881 static struct tun_struct *tun_get(struct tun_file *tfile)
 882 {
 883         struct tun_struct *tun;
 884
 885         rcu_read_lock();
 886         tun = rcu_dereference(tfile->tun);
 887         if (tun)
 888                 dev_hold(tun->dev);
 889         rcu_read_unlock();
 890
 891         return tun;
 892 }
 893
 894 static void tun_put(struct tun_struct *tun)
 895 {
 896         dev_put(tun->dev);
 897 }
 898
 899 /* TAP filtering */
 900 static void addr_hash_set(u32 *mask, const u8 *addr)
 901 {
 902         int n = ether_crc(ETH_ALEN, addr) >> 26;
 903         mask[n >> 5] |= (1 << (n & 31));
 904 }
 905
 906 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 907 {
 908         int n = ether_crc(ETH_ALEN, addr) >> 26;
 909         return mask[n >> 5] & (1 << (n & 31));
 910 }
 911
 912 static int update_filter(struct tap_filter *filter, void __user *arg)
 913 {
 914         struct { u8 u[ETH_ALEN]; } *addr;
 915         struct tun_filter uf;
 916         int err, alen, n, nexact;
 917
 918         if (copy_from_user(&uf, arg, sizeof(uf)))
 919                 return -EFAULT;
 920
 921         if (!uf.count) {
 922                 /* Disabled */
 923                 filter->count = 0;
 924                 return 0;
 925         }
 926
 927         alen = ETH_ALEN * uf.count;
 928         addr = memdup_user(arg + sizeof(uf), alen);
 929         if (IS_ERR(addr))
 930                 return PTR_ERR(addr);
 931
 932         /* The filter is updated without holding any locks. Which is
 933          * perfectly safe. We disable it first and in the worst
 934          * case we'll accept a few undesired packets. */
 935         filter->count = 0;
 936         wmb();
 937
 938         /* Use first set of addresses as an exact filter */
 939         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 940                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 941
 942         nexact = n;
 943
 944         /* Remaining multicast addresses are hashed,
 945          * unicast will leave the filter disabled. */
 946         memset(filter->mask, 0, sizeof(filter->mask));
 947         for (; n < uf.count; n++) {
 948                 if (!is_multicast_ether_addr(addr[n].u)) {
 949                         err = 0; /* no filter */
 950                         goto free_addr;
 951                 }
 952                 addr_hash_set(filter->mask, addr[n].u);
 953         }
 954
 955         /* For ALLMULTI just set the mask to all ones.
 956          * This overrides the mask populated above. */
 957         if ((uf.flags & TUN_FLT_ALLMULTI))
 958                 memset(filter->mask, ~0, sizeof(filter->mask));
 959
 960         /* Now enable the filter */
 961         wmb();
 962         filter->count = nexact;
 963
 964         /* Return the number of exact filters */
 965         err = nexact;
 966 free_addr:
 967         kfree(addr);
 968         return err;
 969 }
 970
 971 /* Returns: 0 - drop, !=0 - accept */
 972 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 973 {
 974         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 975          * at this point. */
 976         struct ethhdr *eh = (struct ethhdr *) skb->data;
 977         int i;
 978
 979         /* Exact match */
 980         for (i = 0; i < filter->count; i++)
 981                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 982                         return 1;
 983
 984         /* Inexact match (multicast only) */
 985         if (is_multicast_ether_addr(eh->h_dest))
 986                 return addr_hash_test(filter->mask, eh->h_dest);
 987
 988         return 0;
 989 }
 990
 991 /*
 992  * Checks whether the packet is accepted or not.
 993  * Returns: 0 - drop, !=0 - accept
 994  */
 995 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 996 {
 997         if (!filter->count)
 998                 return 1;
 999
1000         return run_filter(filter, skb);
1001 }
1002
1003 /* Network device part of the driver */
1004
1005 static const struct ethtool_ops tun_ethtool_ops;
1006
1007 /* Net device detach from fd. */
1008 static void tun_net_uninit(struct net_device *dev)
1009 {
1010         tun_detach_all(dev);
1011 }
1012
1013 /* Net device open. */
1014 static int tun_net_open(struct net_device *dev)
1015 {
1016         struct tun_struct *tun = netdev_priv(dev);
1017         int i;
1018
1019         netif_tx_start_all_queues(dev);
1020
1021         for (i = 0; i < tun->numqueues; i++) {
1022                 struct tun_file *tfile;
1023
1024                 tfile = rtnl_dereference(tun->tfiles[i]);
1025                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1026         }
1027
1028         return 0;
1029 }
1030
1031 /* Net device close. */
1032 static int tun_net_close(struct net_device *dev)
1033 {
1034         netif_tx_stop_all_queues(dev);
1035         return 0;
1036 }
1037
1038 /* Net device start xmit */
1039 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 {
1041 #ifdef CONFIG_RPS
1042         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1043                 /* Select queue was not called for the skbuff, so we extract the
1044                  * RPS hash and save it into the flow_table here.
1045                  */
1046                 __u32 rxhash;
1047
1048                 rxhash = __skb_get_hash_symmetric(skb);
1049                 if (rxhash) {
1050                         struct tun_flow_entry *e;
1051                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1052                                         rxhash);
1053                         if (e)
1054                                 tun_flow_save_rps_rxhash(e, rxhash);
1055                 }
1056         }
1057 #endif
1058 }
1059
1060 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1061                                     struct sk_buff *skb,
1062                                     int len)
1063 {
1064         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1065
1066         if (prog)
1067                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1068
1069         return len;
1070 }
1071
1072 /* Net device start xmit */
1073 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1074 {
1075         struct tun_struct *tun = netdev_priv(dev);
1076         int txq = skb->queue_mapping;
1077         struct tun_file *tfile;
1078         int len = skb->len;
1079
1080         rcu_read_lock();
1081         tfile = rcu_dereference(tun->tfiles[txq]);
1082
1083         /* Drop packet if interface is not attached */
1084         if (txq >= tun->numqueues)
1085                 goto drop;
1086
1087         if (!rcu_dereference(tun->steering_prog))
1088                 tun_automq_xmit(tun, skb);
1089
1090         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1091
1092         BUG_ON(!tfile);
1093
1094         /* Drop if the filter does not like it.
1095          * This is a noop if the filter is disabled.
1096          * Filter can be enabled only for the TAP devices. */
1097         if (!check_filter(&tun->txflt, skb))
1098                 goto drop;
1099
1100         if (tfile->socket.sk->sk_filter &&
1101             sk_filter(tfile->socket.sk, skb))
1102                 goto drop;
1103
1104         len = run_ebpf_filter(tun, skb, len);
1105         if (len == 0 || pskb_trim(skb, len))
1106                 goto drop;
1107
1108         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1109                 goto drop;
1110
1111         skb_tx_timestamp(skb);
1112
1113         /* Orphan the skb - required as we might hang on to it
1114          * for indefinite time.
1115          */
1116         skb_orphan(skb);
1117
1118         nf_reset(skb);
1119
1120         if (ptr_ring_produce(&tfile->tx_ring, skb))
1121                 goto drop;
1122
1123         /* Notify and wake up reader process */
1124         if (tfile->flags & TUN_FASYNC)
1125                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1126         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1127
1128         rcu_read_unlock();
1129         return NETDEV_TX_OK;
1130
1131 drop:
1132         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1133         skb_tx_error(skb);
1134         kfree_skb(skb);
1135         rcu_read_unlock();
1136         return NET_XMIT_DROP;
1137 }
1138
1139 static void tun_net_mclist(struct net_device *dev)
1140 {
1141         /*
1142          * This callback is supposed to deal with mc filter in
1143          * _rx_ path and has nothing to do with the _tx_ path.
1144          * In rx path we always accept everything userspace gives us.
1145          */
1146 }
1147
1148 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1149         netdev_features_t features)
1150 {
1151         struct tun_struct *tun = netdev_priv(dev);
1152
1153         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1154 }
1155
1156 static void tun_set_headroom(struct net_device *dev, int new_hr)
1157 {
1158         struct tun_struct *tun = netdev_priv(dev);
1159
1160         if (new_hr < NET_SKB_PAD)
1161                 new_hr = NET_SKB_PAD;
1162
1163         tun->align = new_hr;
1164 }
1165
1166 static void
1167 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1168 {
1169         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1170         struct tun_struct *tun = netdev_priv(dev);
1171         struct tun_pcpu_stats *p;
1172         int i;
1173
1174         for_each_possible_cpu(i) {
1175                 u64 rxpackets, rxbytes, txpackets, txbytes;
1176                 unsigned int start;
1177
1178                 p = per_cpu_ptr(tun->pcpu_stats, i);
1179                 do {
1180                         start = u64_stats_fetch_begin(&p->syncp);
1181                         rxpackets       = p->rx_packets;
1182                         rxbytes         = p->rx_bytes;
1183                         txpackets       = p->tx_packets;
1184                         txbytes         = p->tx_bytes;
1185                 } while (u64_stats_fetch_retry(&p->syncp, start));
1186
1187                 stats->rx_packets       += rxpackets;
1188                 stats->rx_bytes         += rxbytes;
1189                 stats->tx_packets       += txpackets;
1190                 stats->tx_bytes         += txbytes;
1191
1192                 /* u32 counters */
1193                 rx_dropped      += p->rx_dropped;
1194                 rx_frame_errors += p->rx_frame_errors;
1195                 tx_dropped      += p->tx_dropped;
1196         }
1197         stats->rx_dropped  = rx_dropped;
1198         stats->rx_frame_errors = rx_frame_errors;
1199         stats->tx_dropped = tx_dropped;
1200 }
1201
1202 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1203                        struct netlink_ext_ack *extack)
1204 {
1205         struct tun_struct *tun = netdev_priv(dev);
1206         struct bpf_prog *old_prog;
1207
1208         old_prog = rtnl_dereference(tun->xdp_prog);
1209         rcu_assign_pointer(tun->xdp_prog, prog);
1210         if (old_prog)
1211                 bpf_prog_put(old_prog);
1212
1213         return 0;
1214 }
1215
1216 static u32 tun_xdp_query(struct net_device *dev)
1217 {
1218         struct tun_struct *tun = netdev_priv(dev);
1219         const struct bpf_prog *xdp_prog;
1220
1221         xdp_prog = rtnl_dereference(tun->xdp_prog);
1222         if (xdp_prog)
1223                 return xdp_prog->aux->id;
1224
1225         return 0;
1226 }
1227
1228 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1229 {
1230         switch (xdp->command) {
1231         case XDP_SETUP_PROG:
1232                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1233         case XDP_QUERY_PROG:
1234                 xdp->prog_id = tun_xdp_query(dev);
1235                 return 0;
1236         default:
1237                 return -EINVAL;
1238         }
1239 }
1240
1241 static const struct net_device_ops tun_netdev_ops = {
1242         .ndo_uninit             = tun_net_uninit,
1243         .ndo_open               = tun_net_open,
1244         .ndo_stop               = tun_net_close,
1245         .ndo_start_xmit         = tun_net_xmit,
1246         .ndo_fix_features       = tun_net_fix_features,
1247         .ndo_select_queue       = tun_select_queue,
1248         .ndo_set_rx_headroom    = tun_set_headroom,
1249         .ndo_get_stats64        = tun_net_get_stats64,
1250 };
1251
1252 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1253 {
1254         /* Notify and wake up reader process */
1255         if (tfile->flags & TUN_FASYNC)
1256                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1257         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1258 }
1259
1260 static int tun_xdp_xmit(struct net_device *dev, int n,
1261                         struct xdp_frame **frames, u32 flags)
1262 {
1263         struct tun_struct *tun = netdev_priv(dev);
1264         struct tun_file *tfile;
1265         u32 numqueues;
1266         int drops = 0;
1267         int cnt = n;
1268         int i;
1269
1270         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1271                 return -EINVAL;
1272
1273         rcu_read_lock();
1274
1275         numqueues = READ_ONCE(tun->numqueues);
1276         if (!numqueues) {
1277                 rcu_read_unlock();
1278                 return -ENXIO; /* Caller will free/return all frames */
1279         }
1280
1281         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1282                                             numqueues]);
1283
1284         spin_lock(&tfile->tx_ring.producer_lock);
1285         for (i = 0; i < n; i++) {
1286                 struct xdp_frame *xdp = frames[i];
1287                 /* Encode the XDP flag into lowest bit for consumer to differ
1288                  * XDP buffer from sk_buff.
1289                  */
1290                 void *frame = tun_xdp_to_ptr(xdp);
1291
1292                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1293                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1294                         xdp_return_frame_rx_napi(xdp);
1295                         drops++;
1296                 }
1297         }
1298         spin_unlock(&tfile->tx_ring.producer_lock);
1299
1300         if (flags & XDP_XMIT_FLUSH)
1301                 __tun_xdp_flush_tfile(tfile);
1302
1303         rcu_read_unlock();
1304         return cnt - drops;
1305 }
1306
1307 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1308 {
1309         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1310
1311         if (unlikely(!frame))
1312                 return -EOVERFLOW;
1313
1314         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1315 }
1316
1317 static const struct net_device_ops tap_netdev_ops = {
1318         .ndo_uninit             = tun_net_uninit,
1319         .ndo_open               = tun_net_open,
1320         .ndo_stop               = tun_net_close,
1321         .ndo_start_xmit         = tun_net_xmit,
1322         .ndo_fix_features       = tun_net_fix_features,
1323         .ndo_set_rx_mode        = tun_net_mclist,
1324         .ndo_set_mac_address    = eth_mac_addr,
1325         .ndo_validate_addr      = eth_validate_addr,
1326         .ndo_select_queue       = tun_select_queue,
1327         .ndo_features_check     = passthru_features_check,
1328         .ndo_set_rx_headroom    = tun_set_headroom,
1329         .ndo_get_stats64        = tun_net_get_stats64,
1330         .ndo_bpf                = tun_xdp,
1331         .ndo_xdp_xmit           = tun_xdp_xmit,
1332 };
1333
1334 static void tun_flow_init(struct tun_struct *tun)
1335 {
1336         int i;
1337
1338         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1339                 INIT_HLIST_HEAD(&tun->flows[i]);
1340
1341         tun->ageing_time = TUN_FLOW_EXPIRE;
1342         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1343         mod_timer(&tun->flow_gc_timer,
1344                   round_jiffies_up(jiffies + tun->ageing_time));
1345 }
1346
1347 static void tun_flow_uninit(struct tun_struct *tun)
1348 {
1349         del_timer_sync(&tun->flow_gc_timer);
1350         tun_flow_flush(tun);
1351 }
1352
1353 #define MIN_MTU 68
1354 #define MAX_MTU 65535
1355
1356 /* Initialize net device. */
1357 static void tun_net_init(struct net_device *dev)
1358 {
1359         struct tun_struct *tun = netdev_priv(dev);
1360
1361         switch (tun->flags & TUN_TYPE_MASK) {
1362         case IFF_TUN:
1363                 dev->netdev_ops = &tun_netdev_ops;
1364
1365                 /* Point-to-Point TUN Device */
1366                 dev->hard_header_len = 0;
1367                 dev->addr_len = 0;
1368                 dev->mtu = 1500;
1369
1370                 /* Zero header length */
1371                 dev->type = ARPHRD_NONE;
1372                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1373                 break;
1374
1375         case IFF_TAP:
1376                 dev->netdev_ops = &tap_netdev_ops;
1377                 /* Ethernet TAP Device */
1378                 ether_setup(dev);
1379                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1380                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1381
1382                 eth_hw_addr_random(dev);
1383
1384                 break;
1385         }
1386
1387         dev->min_mtu = MIN_MTU;
1388         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1389 }
1390
1391 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1392 {
1393         struct sock *sk = tfile->socket.sk;
1394
1395         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1396 }
1397
1398 /* Character device part */
1399
1400 /* Poll */
1401 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1402 {
1403         struct tun_file *tfile = file->private_data;
1404         struct tun_struct *tun = tun_get(tfile);
1405         struct sock *sk;
1406         __poll_t mask = 0;
1407
1408         if (!tun)
1409                 return EPOLLERR;
1410
1411         sk = tfile->socket.sk;
1412
1413         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1414
1415         poll_wait(file, sk_sleep(sk), wait);
1416
1417         if (!ptr_ring_empty(&tfile->tx_ring))
1418                 mask |= EPOLLIN | EPOLLRDNORM;
1419
1420         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1421          * guarantee EPOLLOUT to be raised by either here or
1422          * tun_sock_write_space(). Then process could get notification
1423          * after it writes to a down device and meets -EIO.
1424          */
1425         if (tun_sock_writeable(tun, tfile) ||
1426             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1427              tun_sock_writeable(tun, tfile)))
1428                 mask |= EPOLLOUT | EPOLLWRNORM;
1429
1430         if (tun->dev->reg_state != NETREG_REGISTERED)
1431                 mask = EPOLLERR;
1432
1433         tun_put(tun);
1434         return mask;
1435 }
1436
1437 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1438                                             size_t len,
1439                                             const struct iov_iter *it)
1440 {
1441         struct sk_buff *skb;
1442         size_t linear;
1443         int err;
1444         int i;
1445
1446         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1447                 return ERR_PTR(-ENOMEM);
1448
1449         local_bh_disable();
1450         skb = napi_get_frags(&tfile->napi);
1451         local_bh_enable();
1452         if (!skb)
1453                 return ERR_PTR(-ENOMEM);
1454
1455         linear = iov_iter_single_seg_count(it);
1456         err = __skb_grow(skb, linear);
1457         if (err)
1458                 goto free;
1459
1460         skb->len = len;
1461         skb->data_len = len - linear;
1462         skb->truesize += skb->data_len;
1463
1464         for (i = 1; i < it->nr_segs; i++) {
1465                 struct page_frag *pfrag = &current->task_frag;
1466                 size_t fragsz = it->iov[i].iov_len;
1467
1468                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1469                         err = -EINVAL;
1470                         goto free;
1471                 }
1472
1473                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1474                         err = -ENOMEM;
1475                         goto free;
1476                 }
1477
1478                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1479                                    pfrag->offset, fragsz);
1480                 page_ref_inc(pfrag->page);
1481                 pfrag->offset += fragsz;
1482         }
1483
1484         return skb;
1485 free:
1486         /* frees skb and all frags allocated with napi_alloc_frag() */
1487         napi_free_frags(&tfile->napi);
1488         return ERR_PTR(err);
1489 }
1490
1491 /* prepad is the amount to reserve at front.  len is length after that.
1492  * linear is a hint as to how much to copy (usually headers). */
1493 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1494                                      size_t prepad, size_t len,
1495                                      size_t linear, int noblock)
1496 {
1497         struct sock *sk = tfile->socket.sk;
1498         struct sk_buff *skb;
1499         int err;
1500
1501         /* Under a page?  Don't bother with paged skb. */
1502         if (prepad + len < PAGE_SIZE || !linear)
1503                 linear = len;
1504
1505         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1506                                    &err, 0);
1507         if (!skb)
1508                 return ERR_PTR(err);
1509
1510         skb_reserve(skb, prepad);
1511         skb_put(skb, linear);
1512         skb->data_len = len - linear;
1513         skb->len += len - linear;
1514
1515         return skb;
1516 }
1517
1518 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1519                            struct sk_buff *skb, int more)
1520 {
1521         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1522         struct sk_buff_head process_queue;
1523         u32 rx_batched = tun->rx_batched;
1524         bool rcv = false;
1525
1526         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1527                 local_bh_disable();
1528                 netif_receive_skb(skb);
1529                 local_bh_enable();
1530                 return;
1531         }
1532
1533         spin_lock(&queue->lock);
1534         if (!more || skb_queue_len(queue) == rx_batched) {
1535                 __skb_queue_head_init(&process_queue);
1536                 skb_queue_splice_tail_init(queue, &process_queue);
1537                 rcv = true;
1538         } else {
1539                 __skb_queue_tail(queue, skb);
1540         }
1541         spin_unlock(&queue->lock);
1542
1543         if (rcv) {
1544                 struct sk_buff *nskb;
1545
1546                 local_bh_disable();
1547                 while ((nskb = __skb_dequeue(&process_queue)))
1548                         netif_receive_skb(nskb);
1549                 netif_receive_skb(skb);
1550                 local_bh_enable();
1551         }
1552 }
1553
1554 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1555                               int len, int noblock, bool zerocopy)
1556 {
1557         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1558                 return false;
1559
1560         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1561                 return false;
1562
1563         if (!noblock)
1564                 return false;
1565
1566         if (zerocopy)
1567                 return false;
1568
1569         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1570             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1571                 return false;
1572
1573         return true;
1574 }
1575
1576 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1577                                      struct tun_file *tfile,
1578                                      struct iov_iter *from,
1579                                      struct virtio_net_hdr *hdr,
1580                                      int len, int *skb_xdp)
1581 {
1582         struct page_frag *alloc_frag = &current->task_frag;
1583         struct sk_buff *skb;
1584         struct bpf_prog *xdp_prog;
1585         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1586         unsigned int delta = 0;
1587         char *buf;
1588         size_t copied;
1589         int err, pad = TUN_RX_PAD;
1590
1591         rcu_read_lock();
1592         xdp_prog = rcu_dereference(tun->xdp_prog);
1593         if (xdp_prog)
1594                 pad += TUN_HEADROOM;
1595         buflen += SKB_DATA_ALIGN(len + pad);
1596         rcu_read_unlock();
1597
1598         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1599         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1600                 return ERR_PTR(-ENOMEM);
1601
1602         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1603         copied = copy_page_from_iter(alloc_frag->page,
1604                                      alloc_frag->offset + pad,
1605                                      len, from);
1606         if (copied != len)
1607                 return ERR_PTR(-EFAULT);
1608
1609         /* There's a small window that XDP may be set after the check
1610          * of xdp_prog above, this should be rare and for simplicity
1611          * we do XDP on skb in case the headroom is not enough.
1612          */
1613         if (hdr->gso_type || !xdp_prog)
1614                 *skb_xdp = 1;
1615         else
1616                 *skb_xdp = 0;
1617
1618         local_bh_disable();
1619         rcu_read_lock();
1620         xdp_prog = rcu_dereference(tun->xdp_prog);
1621         if (xdp_prog && !*skb_xdp) {
1622                 struct xdp_buff xdp;
1623                 void *orig_data;
1624                 u32 act;
1625
1626                 xdp.data_hard_start = buf;
1627                 xdp.data = buf + pad;
1628                 xdp_set_data_meta_invalid(&xdp);
1629                 xdp.data_end = xdp.data + len;
1630                 xdp.rxq = &tfile->xdp_rxq;
1631                 orig_data = xdp.data;
1632                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1633
1634                 switch (act) {
1635                 case XDP_REDIRECT:
1636                         get_page(alloc_frag->page);
1637                         alloc_frag->offset += buflen;
1638                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1639                         xdp_do_flush_map();
1640                         if (err)
1641                                 goto err_redirect;
1642                         rcu_read_unlock();
1643                         local_bh_enable();
1644                         return NULL;
1645                 case XDP_TX:
1646                         get_page(alloc_frag->page);
1647                         alloc_frag->offset += buflen;
1648                         if (tun_xdp_tx(tun->dev, &xdp) < 0)
1649                                 goto err_redirect;
1650                         rcu_read_unlock();
1651                         local_bh_enable();
1652                         return NULL;
1653                 case XDP_PASS:
1654                         delta = orig_data - xdp.data;
1655                         len = xdp.data_end - xdp.data;
1656                         break;
1657                 default:
1658                         bpf_warn_invalid_xdp_action(act);
1659                         /* fall through */
1660                 case XDP_ABORTED:
1661                         trace_xdp_exception(tun->dev, xdp_prog, act);
1662                         /* fall through */
1663                 case XDP_DROP:
1664                         goto err_xdp;
1665                 }
1666         }
1667
1668         skb = build_skb(buf, buflen);
1669         if (!skb) {
1670                 rcu_read_unlock();
1671                 local_bh_enable();
1672                 return ERR_PTR(-ENOMEM);
1673         }
1674
1675         skb_reserve(skb, pad - delta);
1676         skb_put(skb, len);
1677         get_page(alloc_frag->page);
1678         alloc_frag->offset += buflen;
1679
1680         rcu_read_unlock();
1681         local_bh_enable();
1682
1683         return skb;
1684
1685 err_redirect:
1686         put_page(alloc_frag->page);
1687 err_xdp:
1688         rcu_read_unlock();
1689         local_bh_enable();
1690         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1691         return NULL;
1692 }
1693
1694 /* Get packet from user space buffer */
1695 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1696                             void *msg_control, struct iov_iter *from,
1697                             int noblock, bool more)
1698 {
1699         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1700         struct sk_buff *skb;
1701         size_t total_len = iov_iter_count(from);
1702         size_t len = total_len, align = tun->align, linear;
1703         struct virtio_net_hdr gso = { 0 };
1704         struct tun_pcpu_stats *stats;
1705         int good_linear;
1706         int copylen;
1707         bool zerocopy = false;
1708         int err;
1709         u32 rxhash = 0;
1710         int skb_xdp = 1;
1711         bool frags = tun_napi_frags_enabled(tun);
1712
1713         if (!(tun->dev->flags & IFF_UP))
1714                 return -EIO;
1715
1716         if (!(tun->flags & IFF_NO_PI)) {
1717                 if (len < sizeof(pi))
1718                         return -EINVAL;
1719                 len -= sizeof(pi);
1720
1721                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1722                         return -EFAULT;
1723         }
1724
1725         if (tun->flags & IFF_VNET_HDR) {
1726                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1727
1728                 if (len < vnet_hdr_sz)
1729                         return -EINVAL;
1730                 len -= vnet_hdr_sz;
1731
1732                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1733                         return -EFAULT;
1734
1735                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1736                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1737                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1738
1739                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1740                         return -EINVAL;
1741                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1742         }
1743
1744         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1745                 align += NET_IP_ALIGN;
1746                 if (unlikely(len < ETH_HLEN ||
1747                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1748                         return -EINVAL;
1749         }
1750
1751         good_linear = SKB_MAX_HEAD(align);
1752
1753         if (msg_control) {
1754                 struct iov_iter i = *from;
1755
1756                 /* There are 256 bytes to be copied in skb, so there is
1757                  * enough room for skb expand head in case it is used.
1758                  * The rest of the buffer is mapped from userspace.
1759                  */
1760                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1761                 if (copylen > good_linear)
1762                         copylen = good_linear;
1763                 linear = copylen;
1764                 iov_iter_advance(&i, copylen);
1765                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1766                         zerocopy = true;
1767         }
1768
1769         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1770                 /* For the packet that is not easy to be processed
1771                  * (e.g gso or jumbo packet), we will do it at after
1772                  * skb was created with generic XDP routine.
1773                  */
1774                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1775                 if (IS_ERR(skb)) {
1776                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1777                         return PTR_ERR(skb);
1778                 }
1779                 if (!skb)
1780                         return total_len;
1781         } else {
1782                 if (!zerocopy) {
1783                         copylen = len;
1784                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1785                                 linear = good_linear;
1786                         else
1787                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1788                 }
1789
1790                 if (frags) {
1791                         mutex_lock(&tfile->napi_mutex);
1792                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1793                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1794                          * If zerocopy is enabled, then this layout will be
1795                          * overwritten by zerocopy_sg_from_iter().
1796                          */
1797                         zerocopy = false;
1798                 } else {
1799                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1800                                             noblock);
1801                 }
1802
1803                 if (IS_ERR(skb)) {
1804                         if (PTR_ERR(skb) != -EAGAIN)
1805                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1806                         if (frags)
1807                                 mutex_unlock(&tfile->napi_mutex);
1808                         return PTR_ERR(skb);
1809                 }
1810
1811                 if (zerocopy)
1812                         err = zerocopy_sg_from_iter(skb, from);
1813                 else
1814                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1815
1816                 if (err) {
1817                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1818                         kfree_skb(skb);
1819                         if (frags) {
1820                                 tfile->napi.skb = NULL;
1821                                 mutex_unlock(&tfile->napi_mutex);
1822                         }
1823
1824                         return -EFAULT;
1825                 }
1826         }
1827
1828         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1829                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1830                 kfree_skb(skb);
1831                 if (frags) {
1832                         tfile->napi.skb = NULL;
1833                         mutex_unlock(&tfile->napi_mutex);
1834                 }
1835
1836                 return -EINVAL;
1837         }
1838
1839         switch (tun->flags & TUN_TYPE_MASK) {
1840         case IFF_TUN:
1841                 if (tun->flags & IFF_NO_PI) {
1842                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1843
1844                         switch (ip_version) {
1845                         case 4:
1846                                 pi.proto = htons(ETH_P_IP);
1847                                 break;
1848                         case 6:
1849                                 pi.proto = htons(ETH_P_IPV6);
1850                                 break;
1851                         default:
1852                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1853                                 kfree_skb(skb);
1854                                 return -EINVAL;
1855                         }
1856                 }
1857
1858                 skb_reset_mac_header(skb);
1859                 skb->protocol = pi.proto;
1860                 skb->dev = tun->dev;
1861                 break;
1862         case IFF_TAP:
1863                 if (!frags)
1864                         skb->protocol = eth_type_trans(skb, tun->dev);
1865                 break;
1866         }
1867
1868         /* copy skb_ubuf_info for callback when skb has no error */
1869         if (zerocopy) {
1870                 skb_shinfo(skb)->destructor_arg = msg_control;
1871                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1872                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1873         } else if (msg_control) {
1874                 struct ubuf_info *uarg = msg_control;
1875                 uarg->callback(uarg, false);
1876         }
1877
1878         skb_reset_network_header(skb);
1879         skb_probe_transport_header(skb, 0);
1880
1881         if (skb_xdp) {
1882                 struct bpf_prog *xdp_prog;
1883                 int ret;
1884
1885                 local_bh_disable();
1886                 rcu_read_lock();
1887                 xdp_prog = rcu_dereference(tun->xdp_prog);
1888                 if (xdp_prog) {
1889                         ret = do_xdp_generic(xdp_prog, skb);
1890                         if (ret != XDP_PASS) {
1891                                 rcu_read_unlock();
1892                                 local_bh_enable();
1893                                 return total_len;
1894                         }
1895                 }
1896                 rcu_read_unlock();
1897                 local_bh_enable();
1898         }
1899
1900         /* Compute the costly rx hash only if needed for flow updates.
1901          * We may get a very small possibility of OOO during switching, not
1902          * worth to optimize.
1903          */
1904         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1905             !tfile->detached)
1906                 rxhash = __skb_get_hash_symmetric(skb);
1907
1908         if (frags) {
1909                 /* Exercise flow dissector code path. */
1910                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1911
1912                 if (unlikely(headlen > skb_headlen(skb))) {
1913                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1914                         napi_free_frags(&tfile->napi);
1915                         mutex_unlock(&tfile->napi_mutex);
1916                         WARN_ON(1);
1917                         return -ENOMEM;
1918                 }
1919
1920                 local_bh_disable();
1921                 napi_gro_frags(&tfile->napi);
1922                 local_bh_enable();
1923                 mutex_unlock(&tfile->napi_mutex);
1924         } else if (tfile->napi_enabled) {
1925                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1926                 int queue_len;
1927
1928                 spin_lock_bh(&queue->lock);
1929                 __skb_queue_tail(queue, skb);
1930                 queue_len = skb_queue_len(queue);
1931                 spin_unlock(&queue->lock);
1932
1933                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1934                         napi_schedule(&tfile->napi);
1935
1936                 local_bh_enable();
1937         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1938                 tun_rx_batched(tun, tfile, skb, more);
1939         } else {
1940                 netif_rx_ni(skb);
1941         }
1942
1943         stats = get_cpu_ptr(tun->pcpu_stats);
1944         u64_stats_update_begin(&stats->syncp);
1945         stats->rx_packets++;
1946         stats->rx_bytes += len;
1947         u64_stats_update_end(&stats->syncp);
1948         put_cpu_ptr(stats);
1949
1950         if (rxhash)
1951                 tun_flow_update(tun, rxhash, tfile);
1952
1953         return total_len;
1954 }
1955
1956 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1957 {
1958         struct file *file = iocb->ki_filp;
1959         struct tun_file *tfile = file->private_data;
1960         struct tun_struct *tun = tun_get(tfile);
1961         ssize_t result;
1962
1963         if (!tun)
1964                 return -EBADFD;
1965
1966         result = tun_get_user(tun, tfile, NULL, from,
1967                               file->f_flags & O_NONBLOCK, false);
1968
1969         tun_put(tun);
1970         return result;
1971 }
1972
1973 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1974                                 struct tun_file *tfile,
1975                                 struct xdp_frame *xdp_frame,
1976                                 struct iov_iter *iter)
1977 {
1978         int vnet_hdr_sz = 0;
1979         size_t size = xdp_frame->len;
1980         struct tun_pcpu_stats *stats;
1981         size_t ret;
1982
1983         if (tun->flags & IFF_VNET_HDR) {
1984                 struct virtio_net_hdr gso = { 0 };
1985
1986                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1987                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1988                         return -EINVAL;
1989                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1990                              sizeof(gso)))
1991                         return -EFAULT;
1992                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1993         }
1994
1995         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
1996
1997         stats = get_cpu_ptr(tun->pcpu_stats);
1998         u64_stats_update_begin(&stats->syncp);
1999         stats->tx_packets++;
2000         stats->tx_bytes += ret;
2001         u64_stats_update_end(&stats->syncp);
2002         put_cpu_ptr(tun->pcpu_stats);
2003
2004         return ret;
2005 }
2006
2007 /* Put packet to the user space buffer */
2008 static ssize_t tun_put_user(struct tun_struct *tun,
2009                             struct tun_file *tfile,
2010                             struct sk_buff *skb,
2011                             struct iov_iter *iter)
2012 {
2013         struct tun_pi pi = { 0, skb->protocol };
2014         struct tun_pcpu_stats *stats;
2015         ssize_t total;
2016         int vlan_offset = 0;
2017         int vlan_hlen = 0;
2018         int vnet_hdr_sz = 0;
2019
2020         if (skb_vlan_tag_present(skb))
2021                 vlan_hlen = VLAN_HLEN;
2022
2023         if (tun->flags & IFF_VNET_HDR)
2024                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2025
2026         total = skb->len + vlan_hlen + vnet_hdr_sz;
2027
2028         if (!(tun->flags & IFF_NO_PI)) {
2029                 if (iov_iter_count(iter) < sizeof(pi))
2030                         return -EINVAL;
2031
2032                 total += sizeof(pi);
2033                 if (iov_iter_count(iter) < total) {
2034                         /* Packet will be striped */
2035                         pi.flags |= TUN_PKT_STRIP;
2036                 }
2037
2038                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2039                         return -EFAULT;
2040         }
2041
2042         if (vnet_hdr_sz) {
2043                 struct virtio_net_hdr gso;
2044
2045                 if (iov_iter_count(iter) < vnet_hdr_sz)
2046                         return -EINVAL;
2047
2048                 if (virtio_net_hdr_from_skb(skb, &gso,
2049                                             tun_is_little_endian(tun), true,
2050                                             vlan_hlen)) {
2051                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2052                         pr_err("unexpected GSO type: "
2053                                "0x%x, gso_size %d, hdr_len %d\n",
2054                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2055                                tun16_to_cpu(tun, gso.hdr_len));
2056                         print_hex_dump(KERN_ERR, "tun: ",
2057                                        DUMP_PREFIX_NONE,
2058                                        16, 1, skb->head,
2059                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2060                         WARN_ON_ONCE(1);
2061                         return -EINVAL;
2062                 }
2063
2064                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2065                         return -EFAULT;
2066
2067                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2068         }
2069
2070         if (vlan_hlen) {
2071                 int ret;
2072                 struct veth veth;
2073
2074                 veth.h_vlan_proto = skb->vlan_proto;
2075                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2076
2077                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2078
2079                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2080                 if (ret || !iov_iter_count(iter))
2081                         goto done;
2082
2083                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2084                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2085                         goto done;
2086         }
2087
2088         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2089
2090 done:
2091         /* caller is in process context, */
2092         stats = get_cpu_ptr(tun->pcpu_stats);
2093         u64_stats_update_begin(&stats->syncp);
2094         stats->tx_packets++;
2095         stats->tx_bytes += skb->len + vlan_hlen;
2096         u64_stats_update_end(&stats->syncp);
2097         put_cpu_ptr(tun->pcpu_stats);
2098
2099         return total;
2100 }
2101
2102 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2103 {
2104         DECLARE_WAITQUEUE(wait, current);
2105         void *ptr = NULL;
2106         int error = 0;
2107
2108         ptr = ptr_ring_consume(&tfile->tx_ring);
2109         if (ptr)
2110                 goto out;
2111         if (noblock) {
2112                 error = -EAGAIN;
2113                 goto out;
2114         }
2115
2116         add_wait_queue(&tfile->wq.wait, &wait);
2117         current->state = TASK_INTERRUPTIBLE;
2118
2119         while (1) {
2120                 ptr = ptr_ring_consume(&tfile->tx_ring);
2121                 if (ptr)
2122                         break;
2123                 if (signal_pending(current)) {
2124                         error = -ERESTARTSYS;
2125                         break;
2126                 }
2127                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2128                         error = -EFAULT;
2129                         break;
2130                 }
2131
2132                 schedule();
2133         }
2134
2135         current->state = TASK_RUNNING;
2136         remove_wait_queue(&tfile->wq.wait, &wait);
2137
2138 out:
2139         *err = error;
2140         return ptr;
2141 }
2142
2143 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2144                            struct iov_iter *to,
2145                            int noblock, void *ptr)
2146 {
2147         ssize_t ret;
2148         int err;
2149
2150         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2151
2152         if (!iov_iter_count(to)) {
2153                 tun_ptr_free(ptr);
2154                 return 0;
2155         }
2156
2157         if (!ptr) {
2158                 /* Read frames from ring */
2159                 ptr = tun_ring_recv(tfile, noblock, &err);
2160                 if (!ptr)
2161                         return err;
2162         }
2163
2164         if (tun_is_xdp_frame(ptr)) {
2165                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2166
2167                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2168                 xdp_return_frame(xdpf);
2169         } else {
2170                 struct sk_buff *skb = ptr;
2171
2172                 ret = tun_put_user(tun, tfile, skb, to);
2173                 if (unlikely(ret < 0))
2174                         kfree_skb(skb);
2175                 else
2176                         consume_skb(skb);
2177         }
2178
2179         return ret;
2180 }
2181
2182 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2183 {
2184         struct file *file = iocb->ki_filp;
2185         struct tun_file *tfile = file->private_data;
2186         struct tun_struct *tun = tun_get(tfile);
2187         ssize_t len = iov_iter_count(to), ret;
2188
2189         if (!tun)
2190                 return -EBADFD;
2191         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2192         ret = min_t(ssize_t, ret, len);
2193         if (ret > 0)
2194                 iocb->ki_pos = ret;
2195         tun_put(tun);
2196         return ret;
2197 }
2198
2199 static void tun_prog_free(struct rcu_head *rcu)
2200 {
2201         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2202
2203         bpf_prog_destroy(prog->prog);
2204         kfree(prog);
2205 }
2206
2207 static int __tun_set_ebpf(struct tun_struct *tun,
2208                           struct tun_prog __rcu **prog_p,
2209                           struct bpf_prog *prog)
2210 {
2211         struct tun_prog *old, *new = NULL;
2212
2213         if (prog) {
2214                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2215                 if (!new)
2216                         return -ENOMEM;
2217                 new->prog = prog;
2218         }
2219
2220         spin_lock_bh(&tun->lock);
2221         old = rcu_dereference_protected(*prog_p,
2222                                         lockdep_is_held(&tun->lock));
2223         rcu_assign_pointer(*prog_p, new);
2224         spin_unlock_bh(&tun->lock);
2225
2226         if (old)
2227                 call_rcu(&old->rcu, tun_prog_free);
2228
2229         return 0;
2230 }
2231
2232 static void tun_free_netdev(struct net_device *dev)
2233 {
2234         struct tun_struct *tun = netdev_priv(dev);
2235
2236         BUG_ON(!(list_empty(&tun->disabled)));
2237         free_percpu(tun->pcpu_stats);
2238         tun_flow_uninit(tun);
2239         security_tun_dev_free_security(tun->security);
2240         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2241         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2242 }
2243
2244 static void tun_setup(struct net_device *dev)
2245 {
2246         struct tun_struct *tun = netdev_priv(dev);
2247
2248         tun->owner = INVALID_UID;
2249         tun->group = INVALID_GID;
2250         tun_default_link_ksettings(dev, &tun->link_ksettings);
2251
2252         dev->ethtool_ops = &tun_ethtool_ops;
2253         dev->needs_free_netdev = true;
2254         dev->priv_destructor = tun_free_netdev;
2255         /* We prefer our own queue length */
2256         dev->tx_queue_len = TUN_READQ_SIZE;
2257 }
2258
2259 /* Trivial set of netlink ops to allow deleting tun or tap
2260  * device with netlink.
2261  */
2262 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2263                         struct netlink_ext_ack *extack)
2264 {
2265         return -EINVAL;
2266 }
2267
2268 static size_t tun_get_size(const struct net_device *dev)
2269 {
2270         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2271         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2272
2273         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2274                nla_total_size(sizeof(gid_t)) + /* GROUP */
2275                nla_total_size(sizeof(u8)) + /* TYPE */
2276                nla_total_size(sizeof(u8)) + /* PI */
2277                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2278                nla_total_size(sizeof(u8)) + /* PERSIST */
2279                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2280                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2281                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2282                0;
2283 }
2284
2285 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2286 {
2287         struct tun_struct *tun = netdev_priv(dev);
2288
2289         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2290                 goto nla_put_failure;
2291         if (uid_valid(tun->owner) &&
2292             nla_put_u32(skb, IFLA_TUN_OWNER,
2293                         from_kuid_munged(current_user_ns(), tun->owner)))
2294                 goto nla_put_failure;
2295         if (gid_valid(tun->group) &&
2296             nla_put_u32(skb, IFLA_TUN_GROUP,
2297                         from_kgid_munged(current_user_ns(), tun->group)))
2298                 goto nla_put_failure;
2299         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2300                 goto nla_put_failure;
2301         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2302                 goto nla_put_failure;
2303         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2304                 goto nla_put_failure;
2305         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2306                        !!(tun->flags & IFF_MULTI_QUEUE)))
2307                 goto nla_put_failure;
2308         if (tun->flags & IFF_MULTI_QUEUE) {
2309                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2310                         goto nla_put_failure;
2311                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2312                                 tun->numdisabled))
2313                         goto nla_put_failure;
2314         }
2315
2316         return 0;
2317
2318 nla_put_failure:
2319         return -EMSGSIZE;
2320 }
2321
2322 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2323         .kind           = DRV_NAME,
2324         .priv_size      = sizeof(struct tun_struct),
2325         .setup          = tun_setup,
2326         .validate       = tun_validate,
2327         .get_size       = tun_get_size,
2328         .fill_info      = tun_fill_info,
2329 };
2330
2331 static void tun_sock_write_space(struct sock *sk)
2332 {
2333         struct tun_file *tfile;
2334         wait_queue_head_t *wqueue;
2335
2336         if (!sock_writeable(sk))
2337                 return;
2338
2339         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2340                 return;
2341
2342         wqueue = sk_sleep(sk);
2343         if (wqueue && waitqueue_active(wqueue))
2344                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2345                                                 EPOLLWRNORM | EPOLLWRBAND);
2346
2347         tfile = container_of(sk, struct tun_file, sk);
2348         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2349 }
2350
2351 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2352 {
2353         int ret;
2354         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2355         struct tun_struct *tun = tun_get(tfile);
2356
2357         if (!tun)
2358                 return -EBADFD;
2359
2360         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2361                            m->msg_flags & MSG_DONTWAIT,
2362                            m->msg_flags & MSG_MORE);
2363         tun_put(tun);
2364         return ret;
2365 }
2366
2367 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2368                        int flags)
2369 {
2370         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2371         struct tun_struct *tun = tun_get(tfile);
2372         void *ptr = m->msg_control;
2373         int ret;
2374
2375         if (!tun) {
2376                 ret = -EBADFD;
2377                 goto out_free;
2378         }
2379
2380         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2381                 ret = -EINVAL;
2382                 goto out_put_tun;
2383         }
2384         if (flags & MSG_ERRQUEUE) {
2385                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2386                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2387                 goto out;
2388         }
2389         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2390         if (ret > (ssize_t)total_len) {
2391                 m->msg_flags |= MSG_TRUNC;
2392                 ret = flags & MSG_TRUNC ? ret : total_len;
2393         }
2394 out:
2395         tun_put(tun);
2396         return ret;
2397
2398 out_put_tun:
2399         tun_put(tun);
2400 out_free:
2401         tun_ptr_free(ptr);
2402         return ret;
2403 }
2404
2405 static int tun_ptr_peek_len(void *ptr)
2406 {
2407         if (likely(ptr)) {
2408                 if (tun_is_xdp_frame(ptr)) {
2409                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2410
2411                         return xdpf->len;
2412                 }
2413                 return __skb_array_len_with_tag(ptr);
2414         } else {
2415                 return 0;
2416         }
2417 }
2418
2419 static int tun_peek_len(struct socket *sock)
2420 {
2421         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2422         struct tun_struct *tun;
2423         int ret = 0;
2424
2425         tun = tun_get(tfile);
2426         if (!tun)
2427                 return 0;
2428
2429         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2430         tun_put(tun);
2431
2432         return ret;
2433 }
2434
2435 /* Ops structure to mimic raw sockets with tun */
2436 static const struct proto_ops tun_socket_ops = {
2437         .peek_len = tun_peek_len,
2438         .sendmsg = tun_sendmsg,
2439         .recvmsg = tun_recvmsg,
2440 };
2441
2442 static struct proto tun_proto = {
2443         .name           = "tun",
2444         .owner          = THIS_MODULE,
2445         .obj_size       = sizeof(struct tun_file),
2446 };
2447
2448 static int tun_flags(struct tun_struct *tun)
2449 {
2450         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2451 }
2452
2453 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2454                               char *buf)
2455 {
2456         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2457         return sprintf(buf, "0x%x\n", tun_flags(tun));
2458 }
2459
2460 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2461                               char *buf)
2462 {
2463         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2464         return uid_valid(tun->owner)?
2465                 sprintf(buf, "%u\n",
2466                         from_kuid_munged(current_user_ns(), tun->owner)):
2467                 sprintf(buf, "-1\n");
2468 }
2469
2470 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2471                               char *buf)
2472 {
2473         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2474         return gid_valid(tun->group) ?
2475                 sprintf(buf, "%u\n",
2476                         from_kgid_munged(current_user_ns(), tun->group)):
2477                 sprintf(buf, "-1\n");
2478 }
2479
2480 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2481 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2482 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2483
2484 static struct attribute *tun_dev_attrs[] = {
2485         &dev_attr_tun_flags.attr,
2486         &dev_attr_owner.attr,
2487         &dev_attr_group.attr,
2488         NULL
2489 };
2490
2491 static const struct attribute_group tun_attr_group = {
2492         .attrs = tun_dev_attrs
2493 };
2494
2495 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2496 {
2497         struct tun_struct *tun;
2498         struct tun_file *tfile = file->private_data;
2499         struct net_device *dev;
2500         int err;
2501
2502         if (tfile->detached)
2503                 return -EINVAL;
2504
2505         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2506                 if (!capable(CAP_NET_ADMIN))
2507                         return -EPERM;
2508
2509                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2510                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2511                         return -EINVAL;
2512         }
2513
2514         dev = __dev_get_by_name(net, ifr->ifr_name);
2515         if (dev) {
2516                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2517                         return -EBUSY;
2518                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2519                         tun = netdev_priv(dev);
2520                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2521                         tun = netdev_priv(dev);
2522                 else
2523                         return -EINVAL;
2524
2525                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2526                     !!(tun->flags & IFF_MULTI_QUEUE))
2527                         return -EINVAL;
2528
2529                 if (tun_not_capable(tun))
2530                         return -EPERM;
2531                 err = security_tun_dev_open(tun->security);
2532                 if (err < 0)
2533                         return err;
2534
2535                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2536                                  ifr->ifr_flags & IFF_NAPI);
2537                 if (err < 0)
2538                         return err;
2539
2540                 if (tun->flags & IFF_MULTI_QUEUE &&
2541                     (tun->numqueues + tun->numdisabled > 1)) {
2542                         /* One or more queue has already been attached, no need
2543                          * to initialize the device again.
2544                          */
2545                         netdev_state_change(dev);
2546                         return 0;
2547                 }
2548
2549                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2550                               (ifr->ifr_flags & TUN_FEATURES);
2551
2552                 netdev_state_change(dev);
2553         } else {
2554                 char *name;
2555                 unsigned long flags = 0;
2556                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2557                              MAX_TAP_QUEUES : 1;
2558
2559                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2560                         return -EPERM;
2561                 err = security_tun_dev_create();
2562                 if (err < 0)
2563                         return err;
2564
2565                 /* Set dev type */
2566                 if (ifr->ifr_flags & IFF_TUN) {
2567                         /* TUN device */
2568                         flags |= IFF_TUN;
2569                         name = "tun%d";
2570                 } else if (ifr->ifr_flags & IFF_TAP) {
2571                         /* TAP device */
2572                         flags |= IFF_TAP;
2573                         name = "tap%d";
2574                 } else
2575                         return -EINVAL;
2576
2577                 if (*ifr->ifr_name)
2578                         name = ifr->ifr_name;
2579
2580                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2581                                        NET_NAME_UNKNOWN, tun_setup, queues,
2582                                        queues);
2583
2584                 if (!dev)
2585                         return -ENOMEM;
2586                 err = dev_get_valid_name(net, dev, name);
2587                 if (err < 0)
2588                         goto err_free_dev;
2589
2590                 dev_net_set(dev, net);
2591                 dev->rtnl_link_ops = &tun_link_ops;
2592                 dev->ifindex = tfile->ifindex;
2593                 dev->sysfs_groups[0] = &tun_attr_group;
2594
2595                 tun = netdev_priv(dev);
2596                 tun->dev = dev;
2597                 tun->flags = flags;
2598                 tun->txflt.count = 0;
2599                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2600
2601                 tun->align = NET_SKB_PAD;
2602                 tun->filter_attached = false;
2603                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2604                 tun->rx_batched = 0;
2605                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2606
2607                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2608                 if (!tun->pcpu_stats) {
2609                         err = -ENOMEM;
2610                         goto err_free_dev;
2611                 }
2612
2613                 spin_lock_init(&tun->lock);
2614
2615                 err = security_tun_dev_alloc_security(&tun->security);
2616                 if (err < 0)
2617                         goto err_free_stat;
2618
2619                 tun_net_init(dev);
2620                 tun_flow_init(tun);
2621
2622                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2623                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2624                                    NETIF_F_HW_VLAN_STAG_TX;
2625                 dev->features = dev->hw_features | NETIF_F_LLTX;
2626                 dev->vlan_features = dev->features &
2627                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2628                                        NETIF_F_HW_VLAN_STAG_TX);
2629
2630                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2631                               (ifr->ifr_flags & TUN_FEATURES);
2632
2633                 INIT_LIST_HEAD(&tun->disabled);
2634                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2635                 if (err < 0)
2636                         goto err_free_flow;
2637
2638                 err = register_netdevice(tun->dev);
2639                 if (err < 0)
2640                         goto err_detach;
2641         }
2642
2643         netif_carrier_on(tun->dev);
2644
2645         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2646
2647         /* Make sure persistent devices do not get stuck in
2648          * xoff state.
2649          */
2650         if (netif_running(tun->dev))
2651                 netif_tx_wake_all_queues(tun->dev);
2652
2653         strcpy(ifr->ifr_name, tun->dev->name);
2654         return 0;
2655
2656 err_detach:
2657         tun_detach_all(dev);
2658         /* register_netdevice() already called tun_free_netdev() */
2659         goto err_free_dev;
2660
2661 err_free_flow:
2662         tun_flow_uninit(tun);
2663         security_tun_dev_free_security(tun->security);
2664 err_free_stat:
2665         free_percpu(tun->pcpu_stats);
2666 err_free_dev:
2667         free_netdev(dev);
2668         return err;
2669 }
2670
2671 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2672                        struct ifreq *ifr)
2673 {
2674         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2675
2676         strcpy(ifr->ifr_name, tun->dev->name);
2677
2678         ifr->ifr_flags = tun_flags(tun);
2679
2680 }
2681
2682 /* This is like a cut-down ethtool ops, except done via tun fd so no
2683  * privs required. */
2684 static int set_offload(struct tun_struct *tun, unsigned long arg)
2685 {
2686         netdev_features_t features = 0;
2687
2688         if (arg & TUN_F_CSUM) {
2689                 features |= NETIF_F_HW_CSUM;
2690                 arg &= ~TUN_F_CSUM;
2691
2692                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2693                         if (arg & TUN_F_TSO_ECN) {
2694                                 features |= NETIF_F_TSO_ECN;
2695                                 arg &= ~TUN_F_TSO_ECN;
2696                         }
2697                         if (arg & TUN_F_TSO4)
2698                                 features |= NETIF_F_TSO;
2699                         if (arg & TUN_F_TSO6)
2700                                 features |= NETIF_F_TSO6;
2701                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2702                 }
2703
2704                 arg &= ~TUN_F_UFO;
2705         }
2706
2707         /* This gives the user a way to test for new features in future by
2708          * trying to set them. */
2709         if (arg)
2710                 return -EINVAL;
2711
2712         tun->set_features = features;
2713         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2714         tun->dev->wanted_features |= features;
2715         netdev_update_features(tun->dev);
2716
2717         return 0;
2718 }
2719
2720 static void tun_detach_filter(struct tun_struct *tun, int n)
2721 {
2722         int i;
2723         struct tun_file *tfile;
2724
2725         for (i = 0; i < n; i++) {
2726                 tfile = rtnl_dereference(tun->tfiles[i]);
2727                 lock_sock(tfile->socket.sk);
2728                 sk_detach_filter(tfile->socket.sk);
2729                 release_sock(tfile->socket.sk);
2730         }
2731
2732         tun->filter_attached = false;
2733 }
2734
2735 static int tun_attach_filter(struct tun_struct *tun)
2736 {
2737         int i, ret = 0;
2738         struct tun_file *tfile;
2739
2740         for (i = 0; i < tun->numqueues; i++) {
2741                 tfile = rtnl_dereference(tun->tfiles[i]);
2742                 lock_sock(tfile->socket.sk);
2743                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2744                 release_sock(tfile->socket.sk);
2745                 if (ret) {
2746                         tun_detach_filter(tun, i);
2747                         return ret;
2748                 }
2749         }
2750
2751         tun->filter_attached = true;
2752         return ret;
2753 }
2754
2755 static void tun_set_sndbuf(struct tun_struct *tun)
2756 {
2757         struct tun_file *tfile;
2758         int i;
2759
2760         for (i = 0; i < tun->numqueues; i++) {
2761                 tfile = rtnl_dereference(tun->tfiles[i]);
2762                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2763         }
2764 }
2765
2766 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2767 {
2768         struct tun_file *tfile = file->private_data;
2769         struct tun_struct *tun;
2770         int ret = 0;
2771
2772         rtnl_lock();
2773
2774         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2775                 tun = tfile->detached;
2776                 if (!tun) {
2777                         ret = -EINVAL;
2778                         goto unlock;
2779                 }
2780                 ret = security_tun_dev_attach_queue(tun->security);
2781                 if (ret < 0)
2782                         goto unlock;
2783                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2784         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2785                 tun = rtnl_dereference(tfile->tun);
2786                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2787                         ret = -EINVAL;
2788                 else
2789                         __tun_detach(tfile, false);
2790         } else
2791                 ret = -EINVAL;
2792
2793         if (ret >= 0)
2794                 netdev_state_change(tun->dev);
2795
2796 unlock:
2797         rtnl_unlock();
2798         return ret;
2799 }
2800
2801 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2802                         void __user *data)
2803 {
2804         struct bpf_prog *prog;
2805         int fd;
2806
2807         if (copy_from_user(&fd, data, sizeof(fd)))
2808                 return -EFAULT;
2809
2810         if (fd == -1) {
2811                 prog = NULL;
2812         } else {
2813                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2814                 if (IS_ERR(prog))
2815                         return PTR_ERR(prog);
2816         }
2817
2818         return __tun_set_ebpf(tun, prog_p, prog);
2819 }
2820
2821 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2822                             unsigned long arg, int ifreq_len)
2823 {
2824         struct tun_file *tfile = file->private_data;
2825         struct net *net = sock_net(&tfile->sk);
2826         struct tun_struct *tun;
2827         void __user* argp = (void __user*)arg;
2828         struct ifreq ifr;
2829         kuid_t owner;
2830         kgid_t group;
2831         int sndbuf;
2832         int vnet_hdr_sz;
2833         unsigned int ifindex;
2834         int le;
2835         int ret;
2836         bool do_notify = false;
2837
2838         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2839             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2840                 if (copy_from_user(&ifr, argp, ifreq_len))
2841                         return -EFAULT;
2842         } else {
2843                 memset(&ifr, 0, sizeof(ifr));
2844         }
2845         if (cmd == TUNGETFEATURES) {
2846                 /* Currently this just means: "what IFF flags are valid?".
2847                  * This is needed because we never checked for invalid flags on
2848                  * TUNSETIFF.
2849                  */
2850                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2851                                 (unsigned int __user*)argp);
2852         } else if (cmd == TUNSETQUEUE) {
2853                 return tun_set_queue(file, &ifr);
2854         } else if (cmd == SIOCGSKNS) {
2855                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2856                         return -EPERM;
2857                 return open_related_ns(&net->ns, get_net_ns);
2858         }
2859
2860         ret = 0;
2861         rtnl_lock();
2862
2863         tun = tun_get(tfile);
2864         if (cmd == TUNSETIFF) {
2865                 ret = -EEXIST;
2866                 if (tun)
2867                         goto unlock;
2868
2869                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2870
2871                 ret = tun_set_iff(net, file, &ifr);
2872
2873                 if (ret)
2874                         goto unlock;
2875
2876                 if (copy_to_user(argp, &ifr, ifreq_len))
2877                         ret = -EFAULT;
2878                 goto unlock;
2879         }
2880         if (cmd == TUNSETIFINDEX) {
2881                 ret = -EPERM;
2882                 if (tun)
2883                         goto unlock;
2884
2885                 ret = -EFAULT;
2886                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2887                         goto unlock;
2888
2889                 ret = 0;
2890                 tfile->ifindex = ifindex;
2891                 goto unlock;
2892         }
2893
2894         ret = -EBADFD;
2895         if (!tun)
2896                 goto unlock;
2897
2898         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2899
2900         ret = 0;
2901         switch (cmd) {
2902         case TUNGETIFF:
2903                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2904
2905                 if (tfile->detached)
2906                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2907                 if (!tfile->socket.sk->sk_filter)
2908                         ifr.ifr_flags |= IFF_NOFILTER;
2909
2910                 if (copy_to_user(argp, &ifr, ifreq_len))
2911                         ret = -EFAULT;
2912                 break;
2913
2914         case TUNSETNOCSUM:
2915                 /* Disable/Enable checksum */
2916
2917                 /* [unimplemented] */
2918                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2919                           arg ? "disabled" : "enabled");
2920                 break;
2921
2922         case TUNSETPERSIST:
2923                 /* Disable/Enable persist mode. Keep an extra reference to the
2924                  * module to prevent the module being unprobed.
2925                  */
2926                 if (arg && !(tun->flags & IFF_PERSIST)) {
2927                         tun->flags |= IFF_PERSIST;
2928                         __module_get(THIS_MODULE);
2929                         do_notify = true;
2930                 }
2931                 if (!arg && (tun->flags & IFF_PERSIST)) {
2932                         tun->flags &= ~IFF_PERSIST;
2933                         module_put(THIS_MODULE);
2934                         do_notify = true;
2935                 }
2936
2937                 tun_debug(KERN_INFO, tun, "persist %s\n",
2938                           arg ? "enabled" : "disabled");
2939                 break;
2940
2941         case TUNSETOWNER:
2942                 /* Set owner of the device */
2943                 owner = make_kuid(current_user_ns(), arg);
2944                 if (!uid_valid(owner)) {
2945                         ret = -EINVAL;
2946                         break;
2947                 }
2948                 tun->owner = owner;
2949                 do_notify = true;
2950                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2951                           from_kuid(&init_user_ns, tun->owner));
2952                 break;
2953
2954         case TUNSETGROUP:
2955                 /* Set group of the device */
2956                 group = make_kgid(current_user_ns(), arg);
2957                 if (!gid_valid(group)) {
2958                         ret = -EINVAL;
2959                         break;
2960                 }
2961                 tun->group = group;
2962                 do_notify = true;
2963                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2964                           from_kgid(&init_user_ns, tun->group));
2965                 break;
2966
2967         case TUNSETLINK:
2968                 /* Only allow setting the type when the interface is down */
2969                 if (tun->dev->flags & IFF_UP) {
2970                         tun_debug(KERN_INFO, tun,
2971                                   "Linktype set failed because interface is up\n");
2972                         ret = -EBUSY;
2973                 } else {
2974                         tun->dev->type = (int) arg;
2975                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2976                                   tun->dev->type);
2977                         ret = 0;
2978                 }
2979                 break;
2980
2981 #ifdef TUN_DEBUG
2982         case TUNSETDEBUG:
2983                 tun->debug = arg;
2984                 break;
2985 #endif
2986         case TUNSETOFFLOAD:
2987                 ret = set_offload(tun, arg);
2988                 break;
2989
2990         case TUNSETTXFILTER:
2991                 /* Can be set only for TAPs */
2992                 ret = -EINVAL;
2993                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2994                         break;
2995                 ret = update_filter(&tun->txflt, (void __user *)arg);
2996                 break;
2997
2998         case SIOCGIFHWADDR:
2999                 /* Get hw address */
3000                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3001                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3002                 if (copy_to_user(argp, &ifr, ifreq_len))
3003                         ret = -EFAULT;
3004                 break;
3005
3006         case SIOCSIFHWADDR:
3007                 /* Set hw address */
3008                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3009                           ifr.ifr_hwaddr.sa_data);
3010
3011                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3012                 break;
3013
3014         case TUNGETSNDBUF:
3015                 sndbuf = tfile->socket.sk->sk_sndbuf;
3016                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3017                         ret = -EFAULT;
3018                 break;
3019
3020         case TUNSETSNDBUF:
3021                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3022                         ret = -EFAULT;
3023                         break;
3024                 }
3025                 if (sndbuf <= 0) {
3026                         ret = -EINVAL;
3027                         break;
3028                 }
3029
3030                 tun->sndbuf = sndbuf;
3031                 tun_set_sndbuf(tun);
3032                 break;
3033
3034         case TUNGETVNETHDRSZ:
3035                 vnet_hdr_sz = tun->vnet_hdr_sz;
3036                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3037                         ret = -EFAULT;
3038                 break;
3039
3040         case TUNSETVNETHDRSZ:
3041                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3042                         ret = -EFAULT;
3043                         break;
3044                 }
3045                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3046                         ret = -EINVAL;
3047                         break;
3048                 }
3049
3050                 tun->vnet_hdr_sz = vnet_hdr_sz;
3051                 break;
3052
3053         case TUNGETVNETLE:
3054                 le = !!(tun->flags & TUN_VNET_LE);
3055                 if (put_user(le, (int __user *)argp))
3056                         ret = -EFAULT;
3057                 break;
3058
3059         case TUNSETVNETLE:
3060                 if (get_user(le, (int __user *)argp)) {
3061                         ret = -EFAULT;
3062                         break;
3063                 }
3064                 if (le)
3065                         tun->flags |= TUN_VNET_LE;
3066                 else
3067                         tun->flags &= ~TUN_VNET_LE;
3068                 break;
3069
3070         case TUNGETVNETBE:
3071                 ret = tun_get_vnet_be(tun, argp);
3072                 break;
3073
3074         case TUNSETVNETBE:
3075                 ret = tun_set_vnet_be(tun, argp);
3076                 break;
3077
3078         case TUNATTACHFILTER:
3079                 /* Can be set only for TAPs */
3080                 ret = -EINVAL;
3081                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3082                         break;
3083                 ret = -EFAULT;
3084                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3085                         break;
3086
3087                 ret = tun_attach_filter(tun);
3088                 break;
3089
3090         case TUNDETACHFILTER:
3091                 /* Can be set only for TAPs */
3092                 ret = -EINVAL;
3093                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3094                         break;
3095                 ret = 0;
3096                 tun_detach_filter(tun, tun->numqueues);
3097                 break;
3098
3099         case TUNGETFILTER:
3100                 ret = -EINVAL;
3101                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3102                         break;
3103                 ret = -EFAULT;
3104                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3105                         break;
3106                 ret = 0;
3107                 break;
3108
3109         case TUNSETSTEERINGEBPF:
3110                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3111                 break;
3112
3113         case TUNSETFILTEREBPF:
3114                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3115                 break;
3116
3117         default:
3118                 ret = -EINVAL;
3119                 break;
3120         }
3121
3122         if (do_notify)
3123                 netdev_state_change(tun->dev);
3124
3125 unlock:
3126         rtnl_unlock();
3127         if (tun)
3128                 tun_put(tun);
3129         return ret;
3130 }
3131
3132 static long tun_chr_ioctl(struct file *file,
3133                           unsigned int cmd, unsigned long arg)
3134 {
3135         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3136 }
3137
3138 #ifdef CONFIG_COMPAT
3139 static long tun_chr_compat_ioctl(struct file *file,
3140                          unsigned int cmd, unsigned long arg)
3141 {
3142         switch (cmd) {
3143         case TUNSETIFF:
3144         case TUNGETIFF:
3145         case TUNSETTXFILTER:
3146         case TUNGETSNDBUF:
3147         case TUNSETSNDBUF:
3148         case SIOCGIFHWADDR:
3149         case SIOCSIFHWADDR:
3150                 arg = (unsigned long)compat_ptr(arg);
3151                 break;
3152         default:
3153                 arg = (compat_ulong_t)arg;
3154                 break;
3155         }
3156
3157         /*
3158          * compat_ifreq is shorter than ifreq, so we must not access beyond
3159          * the end of that structure. All fields that are used in this
3160          * driver are compatible though, we don't need to convert the
3161          * contents.
3162          */
3163         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3164 }
3165 #endif /* CONFIG_COMPAT */
3166
3167 static int tun_chr_fasync(int fd, struct file *file, int on)
3168 {
3169         struct tun_file *tfile = file->private_data;
3170         int ret;
3171
3172         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3173                 goto out;
3174
3175         if (on) {
3176                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3177                 tfile->flags |= TUN_FASYNC;
3178         } else
3179                 tfile->flags &= ~TUN_FASYNC;
3180         ret = 0;
3181 out:
3182         return ret;
3183 }
3184
3185 static int tun_chr_open(struct inode *inode, struct file * file)
3186 {
3187         struct net *net = current->nsproxy->net_ns;
3188         struct tun_file *tfile;
3189
3190         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3191
3192         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3193                                             &tun_proto, 0);
3194         if (!tfile)
3195                 return -ENOMEM;
3196         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3197                 sk_free(&tfile->sk);
3198                 return -ENOMEM;
3199         }
3200
3201         mutex_init(&tfile->napi_mutex);
3202         RCU_INIT_POINTER(tfile->tun, NULL);
3203         tfile->flags = 0;
3204         tfile->ifindex = 0;
3205
3206         init_waitqueue_head(&tfile->wq.wait);
3207         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3208
3209         tfile->socket.file = file;
3210         tfile->socket.ops = &tun_socket_ops;
3211
3212         sock_init_data(&tfile->socket, &tfile->sk);
3213
3214         tfile->sk.sk_write_space = tun_sock_write_space;
3215         tfile->sk.sk_sndbuf = INT_MAX;
3216
3217         file->private_data = tfile;
3218         INIT_LIST_HEAD(&tfile->next);
3219
3220         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3221
3222         return 0;
3223 }
3224
3225 static int tun_chr_close(struct inode *inode, struct file *file)
3226 {
3227         struct tun_file *tfile = file->private_data;
3228
3229         tun_detach(tfile, true);
3230
3231         return 0;
3232 }
3233
3234 #ifdef CONFIG_PROC_FS
3235 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3236 {
3237         struct tun_file *tfile = file->private_data;
3238         struct tun_struct *tun;
3239         struct ifreq ifr;
3240
3241         memset(&ifr, 0, sizeof(ifr));
3242
3243         rtnl_lock();
3244         tun = tun_get(tfile);
3245         if (tun)
3246                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3247         rtnl_unlock();
3248
3249         if (tun)
3250                 tun_put(tun);
3251
3252         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3253 }
3254 #endif
3255
3256 static const struct file_operations tun_fops = {
3257         .owner  = THIS_MODULE,
3258         .llseek = no_llseek,
3259         .read_iter  = tun_chr_read_iter,
3260         .write_iter = tun_chr_write_iter,
3261         .poll   = tun_chr_poll,
3262         .unlocked_ioctl = tun_chr_ioctl,
3263 #ifdef CONFIG_COMPAT
3264         .compat_ioctl = tun_chr_compat_ioctl,
3265 #endif
3266         .open   = tun_chr_open,
3267         .release = tun_chr_close,
3268         .fasync = tun_chr_fasync,
3269 #ifdef CONFIG_PROC_FS
3270         .show_fdinfo = tun_chr_show_fdinfo,
3271 #endif
3272 };
3273
3274 static struct miscdevice tun_miscdev = {
3275         .minor = TUN_MINOR,
3276         .name = "tun",
3277         .nodename = "net/tun",
3278         .fops = &tun_fops,
3279 };
3280
3281 /* ethtool interface */
3282
3283 static void tun_default_link_ksettings(struct net_device *dev,
3284                                        struct ethtool_link_ksettings *cmd)
3285 {
3286         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3287         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3288         cmd->base.speed         = SPEED_10;
3289         cmd->base.duplex        = DUPLEX_FULL;
3290         cmd->base.port          = PORT_TP;
3291         cmd->base.phy_address   = 0;
3292         cmd->base.autoneg       = AUTONEG_DISABLE;
3293 }
3294
3295 static int tun_get_link_ksettings(struct net_device *dev,
3296                                   struct ethtool_link_ksettings *cmd)
3297 {
3298         struct tun_struct *tun = netdev_priv(dev);
3299
3300         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3301         return 0;
3302 }
3303
3304 static int tun_set_link_ksettings(struct net_device *dev,
3305                                   const struct ethtool_link_ksettings *cmd)
3306 {
3307         struct tun_struct *tun = netdev_priv(dev);
3308
3309         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3310         return 0;
3311 }
3312
3313 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3314 {
3315         struct tun_struct *tun = netdev_priv(dev);
3316
3317         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3318         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3319
3320         switch (tun->flags & TUN_TYPE_MASK) {
3321         case IFF_TUN:
3322                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3323                 break;
3324         case IFF_TAP:
3325                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3326                 break;
3327         }
3328 }
3329
3330 static u32 tun_get_msglevel(struct net_device *dev)
3331 {
3332 #ifdef TUN_DEBUG
3333         struct tun_struct *tun = netdev_priv(dev);
3334         return tun->debug;
3335 #else
3336         return -EOPNOTSUPP;
3337 #endif
3338 }
3339
3340 static void tun_set_msglevel(struct net_device *dev, u32 value)
3341 {
3342 #ifdef TUN_DEBUG
3343         struct tun_struct *tun = netdev_priv(dev);
3344         tun->debug = value;
3345 #endif
3346 }
3347
3348 static int tun_get_coalesce(struct net_device *dev,
3349                             struct ethtool_coalesce *ec)
3350 {
3351         struct tun_struct *tun = netdev_priv(dev);
3352
3353         ec->rx_max_coalesced_frames = tun->rx_batched;
3354
3355         return 0;
3356 }
3357
3358 static int tun_set_coalesce(struct net_device *dev,
3359                             struct ethtool_coalesce *ec)
3360 {
3361         struct tun_struct *tun = netdev_priv(dev);
3362
3363         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3364                 tun->rx_batched = NAPI_POLL_WEIGHT;
3365         else
3366                 tun->rx_batched = ec->rx_max_coalesced_frames;
3367
3368         return 0;
3369 }
3370
3371 static const struct ethtool_ops tun_ethtool_ops = {
3372         .get_drvinfo    = tun_get_drvinfo,
3373         .get_msglevel   = tun_get_msglevel,
3374         .set_msglevel   = tun_set_msglevel,
3375         .get_link       = ethtool_op_get_link,
3376         .get_ts_info    = ethtool_op_get_ts_info,
3377         .get_coalesce   = tun_get_coalesce,
3378         .set_coalesce   = tun_set_coalesce,
3379         .get_link_ksettings = tun_get_link_ksettings,
3380         .set_link_ksettings = tun_set_link_ksettings,
3381 };
3382
3383 static int tun_queue_resize(struct tun_struct *tun)
3384 {
3385         struct net_device *dev = tun->dev;
3386         struct tun_file *tfile;
3387         struct ptr_ring **rings;
3388         int n = tun->numqueues + tun->numdisabled;
3389         int ret, i;
3390
3391         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3392         if (!rings)
3393                 return -ENOMEM;
3394
3395         for (i = 0; i < tun->numqueues; i++) {
3396                 tfile = rtnl_dereference(tun->tfiles[i]);
3397                 rings[i] = &tfile->tx_ring;
3398         }
3399         list_for_each_entry(tfile, &tun->disabled, next)
3400                 rings[i++] = &tfile->tx_ring;
3401
3402         ret = ptr_ring_resize_multiple(rings, n,
3403                                        dev->tx_queue_len, GFP_KERNEL,
3404                                        tun_ptr_free);
3405
3406         kfree(rings);
3407         return ret;
3408 }
3409
3410 static int tun_device_event(struct notifier_block *unused,
3411                             unsigned long event, void *ptr)
3412 {
3413         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3414         struct tun_struct *tun = netdev_priv(dev);
3415
3416         if (dev->rtnl_link_ops != &tun_link_ops)
3417                 return NOTIFY_DONE;
3418
3419         switch (event) {
3420         case NETDEV_CHANGE_TX_QUEUE_LEN:
3421                 if (tun_queue_resize(tun))
3422                         return NOTIFY_BAD;
3423                 break;
3424         default:
3425                 break;
3426         }
3427
3428         return NOTIFY_DONE;
3429 }
3430
3431 static struct notifier_block tun_notifier_block __read_mostly = {
3432         .notifier_call  = tun_device_event,
3433 };
3434
3435 static int __init tun_init(void)
3436 {
3437         int ret = 0;
3438
3439         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3440
3441         ret = rtnl_link_register(&tun_link_ops);
3442         if (ret) {
3443                 pr_err("Can't register link_ops\n");
3444                 goto err_linkops;
3445         }
3446
3447         ret = misc_register(&tun_miscdev);
3448         if (ret) {
3449                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3450                 goto err_misc;
3451         }
3452
3453         ret = register_netdevice_notifier(&tun_notifier_block);
3454         if (ret) {
3455                 pr_err("Can't register netdevice notifier\n");
3456                 goto err_notifier;
3457         }
3458
3459         return  0;
3460
3461 err_notifier:
3462         misc_deregister(&tun_miscdev);
3463 err_misc:
3464         rtnl_link_unregister(&tun_link_ops);
3465 err_linkops:
3466         return ret;
3467 }
3468
3469 static void tun_cleanup(void)
3470 {
3471         misc_deregister(&tun_miscdev);
3472         rtnl_link_unregister(&tun_link_ops);
3473         unregister_netdevice_notifier(&tun_notifier_block);
3474 }
3475
3476 /* Get an underlying socket object from tun file.  Returns error unless file is
3477  * attached to a device.  The returned object works like a packet socket, it
3478  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3479  * holding a reference to the file for as long as the socket is in use. */
3480 struct socket *tun_get_socket(struct file *file)
3481 {
3482         struct tun_file *tfile;
3483         if (file->f_op != &tun_fops)
3484                 return ERR_PTR(-EINVAL);
3485         tfile = file->private_data;
3486         if (!tfile)
3487                 return ERR_PTR(-EBADFD);
3488         return &tfile->socket;
3489 }
3490 EXPORT_SYMBOL_GPL(tun_get_socket);
3491
3492 struct ptr_ring *tun_get_tx_ring(struct file *file)
3493 {
3494         struct tun_file *tfile;
3495
3496         if (file->f_op != &tun_fops)
3497                 return ERR_PTR(-EINVAL);
3498         tfile = file->private_data;
3499         if (!tfile)
3500                 return ERR_PTR(-EBADFD);
3501         return &tfile->tx_ring;
3502 }
3503 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3504
3505 module_init(tun_init);
3506 module_exit(tun_cleanup);
3507 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3508 MODULE_AUTHOR(DRV_COPYRIGHT);
3509 MODULE_LICENSE("GPL");
3510 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3511 MODULE_ALIAS("devname:net/tun");