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