drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78
  79 #include <linux/uaccess.h>
  80
  81 /* Uncomment to enable debugging */
  82 /* #define TUN_DEBUG 1 */
  83
  84 #ifdef TUN_DEBUG
  85 static int debug;
  86
  87 #define tun_debug(level, tun, fmt, args...)                     \
  88 do {                                                            \
  89         if (tun->debug)                                         \
  90                 netdev_printk(level, tun->dev, fmt, ##args);    \
  91 } while (0)
  92 #define DBG1(level, fmt, args...)                               \
  93 do {                                                            \
  94         if (debug == 2)                                         \
  95                 printk(level fmt, ##args);                      \
  96 } while (0)
  97 #else
  98 #define tun_debug(level, tun, fmt, args...)                     \
  99 do {                                                            \
 100         if (0)                                                  \
 101                 netdev_printk(level, tun->dev, fmt, ##args);    \
 102 } while (0)
 103 #define DBG1(level, fmt, args...)                               \
 104 do {                                                            \
 105         if (0)                                                  \
 106                 printk(level fmt, ##args);                      \
 107 } while (0)
 108 #endif
 109
 110 #define TUN_HEADROOM 256
 111 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD + TUN_HEADROOM)
 112
 113 /* TUN device flags */
 114
 115 /* IFF_ATTACH_QUEUE is never stored in device flags,
 116  * overload it to mean fasync when stored there.
 117  */
 118 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 119 /* High bits in flags field are unused. */
 120 #define TUN_VNET_LE     0x80000000
 121 #define TUN_VNET_BE     0x40000000
 122
 123 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 124                       IFF_MULTI_QUEUE)
 125 #define GOODCOPY_LEN 128
 126
 127 #define FLT_EXACT_COUNT 8
 128 struct tap_filter {
 129         unsigned int    count;    /* Number of addrs. Zero means disabled */
 130         u32             mask[2];  /* Mask of the hashed addrs */
 131         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 132 };
 133
 134 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 135  * to max number of VCPUs in guest. */
 136 #define MAX_TAP_QUEUES 256
 137 #define MAX_TAP_FLOWS  4096
 138
 139 #define TUN_FLOW_EXPIRE (3 * HZ)
 140
 141 struct tun_pcpu_stats {
 142         u64 rx_packets;
 143         u64 rx_bytes;
 144         u64 tx_packets;
 145         u64 tx_bytes;
 146         struct u64_stats_sync syncp;
 147         u32 rx_dropped;
 148         u32 tx_dropped;
 149         u32 rx_frame_errors;
 150 };
 151
 152 /* A tun_file connects an open character device to a tuntap netdevice. It
 153  * also contains all socket related structures (except sock_fprog and tap_filter)
 154  * to serve as one transmit queue for tuntap device. The sock_fprog and
 155  * tap_filter were kept in tun_struct since they were used for filtering for the
 156  * netdevice not for a specific queue (at least I didn't see the requirement for
 157  * this).
 158  *
 159  * RCU usage:
 160  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 161  * other can only be read while rcu_read_lock or rtnl_lock is held.
 162  */
 163 struct tun_file {
 164         struct sock sk;
 165         struct socket socket;
 166         struct socket_wq wq;
 167         struct tun_struct __rcu *tun;
 168         struct fasync_struct *fasync;
 169         /* only used for fasnyc */
 170         unsigned int flags;
 171         union {
 172                 u16 queue_index;
 173                 unsigned int ifindex;
 174         };
 175         struct list_head next;
 176         struct tun_struct *detached;
 177         struct skb_array tx_array;
 178         struct page_frag alloc_frag;
 179 };
 180
 181 struct tun_flow_entry {
 182         struct hlist_node hash_link;
 183         struct rcu_head rcu;
 184         struct tun_struct *tun;
 185
 186         u32 rxhash;
 187         u32 rps_rxhash;
 188         int queue_index;
 189         unsigned long updated;
 190 };
 191
 192 #define TUN_NUM_FLOW_ENTRIES 1024
 193
 194 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 195  * device, socket filter, sndbuf and vnet header size were restore when the
 196  * file were attached to a persist device.
 197  */
 198 struct tun_struct {
 199         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 200         unsigned int            numqueues;
 201         unsigned int            flags;
 202         kuid_t                  owner;
 203         kgid_t                  group;
 204
 205         struct net_device       *dev;
 206         netdev_features_t       set_features;
 207 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 208                           NETIF_F_TSO6)
 209
 210         int                     align;
 211         int                     vnet_hdr_sz;
 212         int                     sndbuf;
 213         struct tap_filter       txflt;
 214         struct sock_fprog       fprog;
 215         /* protected by rtnl lock */
 216         bool                    filter_attached;
 217 #ifdef TUN_DEBUG
 218         int debug;
 219 #endif
 220         spinlock_t lock;
 221         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 222         struct timer_list flow_gc_timer;
 223         unsigned long ageing_time;
 224         unsigned int numdisabled;
 225         struct list_head disabled;
 226         void *security;
 227         u32 flow_count;
 228         u32 rx_batched;
 229         struct tun_pcpu_stats __percpu *pcpu_stats;
 230         struct bpf_prog __rcu *xdp_prog;
 231 };
 232
 233 #ifdef CONFIG_TUN_VNET_CROSS_LE
 234 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 235 {
 236         return tun->flags & TUN_VNET_BE ? false :
 237                 virtio_legacy_is_little_endian();
 238 }
 239
 240 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 241 {
 242         int be = !!(tun->flags & TUN_VNET_BE);
 243
 244         if (put_user(be, argp))
 245                 return -EFAULT;
 246
 247         return 0;
 248 }
 249
 250 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 251 {
 252         int be;
 253
 254         if (get_user(be, argp))
 255                 return -EFAULT;
 256
 257         if (be)
 258                 tun->flags |= TUN_VNET_BE;
 259         else
 260                 tun->flags &= ~TUN_VNET_BE;
 261
 262         return 0;
 263 }
 264 #else
 265 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 266 {
 267         return virtio_legacy_is_little_endian();
 268 }
 269
 270 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 271 {
 272         return -EINVAL;
 273 }
 274
 275 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 276 {
 277         return -EINVAL;
 278 }
 279 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 280
 281 static inline bool tun_is_little_endian(struct tun_struct *tun)
 282 {
 283         return tun->flags & TUN_VNET_LE ||
 284                 tun_legacy_is_little_endian(tun);
 285 }
 286
 287 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 288 {
 289         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 290 }
 291
 292 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 293 {
 294         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 295 }
 296
 297 static inline u32 tun_hashfn(u32 rxhash)
 298 {
 299         return rxhash & 0x3ff;
 300 }
 301
 302 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 303 {
 304         struct tun_flow_entry *e;
 305
 306         hlist_for_each_entry_rcu(e, head, hash_link) {
 307                 if (e->rxhash == rxhash)
 308                         return e;
 309         }
 310         return NULL;
 311 }
 312
 313 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 314                                               struct hlist_head *head,
 315                                               u32 rxhash, u16 queue_index)
 316 {
 317         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 318
 319         if (e) {
 320                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 321                           rxhash, queue_index);
 322                 e->updated = jiffies;
 323                 e->rxhash = rxhash;
 324                 e->rps_rxhash = 0;
 325                 e->queue_index = queue_index;
 326                 e->tun = tun;
 327                 hlist_add_head_rcu(&e->hash_link, head);
 328                 ++tun->flow_count;
 329         }
 330         return e;
 331 }
 332
 333 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 334 {
 335         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 336                   e->rxhash, e->queue_index);
 337         hlist_del_rcu(&e->hash_link);
 338         kfree_rcu(e, rcu);
 339         --tun->flow_count;
 340 }
 341
 342 static void tun_flow_flush(struct tun_struct *tun)
 343 {
 344         int i;
 345
 346         spin_lock_bh(&tun->lock);
 347         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 348                 struct tun_flow_entry *e;
 349                 struct hlist_node *n;
 350
 351                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 352                         tun_flow_delete(tun, e);
 353         }
 354         spin_unlock_bh(&tun->lock);
 355 }
 356
 357 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 358 {
 359         int i;
 360
 361         spin_lock_bh(&tun->lock);
 362         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 363                 struct tun_flow_entry *e;
 364                 struct hlist_node *n;
 365
 366                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 367                         if (e->queue_index == queue_index)
 368                                 tun_flow_delete(tun, e);
 369                 }
 370         }
 371         spin_unlock_bh(&tun->lock);
 372 }
 373
 374 static void tun_flow_cleanup(unsigned long data)
 375 {
 376         struct tun_struct *tun = (struct tun_struct *)data;
 377         unsigned long delay = tun->ageing_time;
 378         unsigned long next_timer = jiffies + delay;
 379         unsigned long count = 0;
 380         int i;
 381
 382         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 383
 384         spin_lock_bh(&tun->lock);
 385         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 386                 struct tun_flow_entry *e;
 387                 struct hlist_node *n;
 388
 389                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 390                         unsigned long this_timer;
 391                         count++;
 392                         this_timer = e->updated + delay;
 393                         if (time_before_eq(this_timer, jiffies))
 394                                 tun_flow_delete(tun, e);
 395                         else if (time_before(this_timer, next_timer))
 396                                 next_timer = this_timer;
 397                 }
 398         }
 399
 400         if (count)
 401                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 402         spin_unlock_bh(&tun->lock);
 403 }
 404
 405 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 406                             struct tun_file *tfile)
 407 {
 408         struct hlist_head *head;
 409         struct tun_flow_entry *e;
 410         unsigned long delay = tun->ageing_time;
 411         u16 queue_index = tfile->queue_index;
 412
 413         if (!rxhash)
 414                 return;
 415         else
 416                 head = &tun->flows[tun_hashfn(rxhash)];
 417
 418         rcu_read_lock();
 419
 420         /* We may get a very small possibility of OOO during switching, not
 421          * worth to optimize.*/
 422         if (tun->numqueues == 1 || tfile->detached)
 423                 goto unlock;
 424
 425         e = tun_flow_find(head, rxhash);
 426         if (likely(e)) {
 427                 /* TODO: keep queueing to old queue until it's empty? */
 428                 e->queue_index = queue_index;
 429                 e->updated = jiffies;
 430                 sock_rps_record_flow_hash(e->rps_rxhash);
 431         } else {
 432                 spin_lock_bh(&tun->lock);
 433                 if (!tun_flow_find(head, rxhash) &&
 434                     tun->flow_count < MAX_TAP_FLOWS)
 435                         tun_flow_create(tun, head, rxhash, queue_index);
 436
 437                 if (!timer_pending(&tun->flow_gc_timer))
 438                         mod_timer(&tun->flow_gc_timer,
 439                                   round_jiffies_up(jiffies + delay));
 440                 spin_unlock_bh(&tun->lock);
 441         }
 442
 443 unlock:
 444         rcu_read_unlock();
 445 }
 446
 447 /**
 448  * Save the hash received in the stack receive path and update the
 449  * flow_hash table accordingly.
 450  */
 451 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 452 {
 453         if (unlikely(e->rps_rxhash != hash))
 454                 e->rps_rxhash = hash;
 455 }
 456
 457 /* We try to identify a flow through its rxhash first. The reason that
 458  * we do not check rxq no. is because some cards(e.g 82599), chooses
 459  * the rxq based on the txq where the last packet of the flow comes. As
 460  * the userspace application move between processors, we may get a
 461  * different rxq no. here. If we could not get rxhash, then we would
 462  * hope the rxq no. may help here.
 463  */
 464 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 465                             void *accel_priv, select_queue_fallback_t fallback)
 466 {
 467         struct tun_struct *tun = netdev_priv(dev);
 468         struct tun_flow_entry *e;
 469         u32 txq = 0;
 470         u32 numqueues = 0;
 471
 472         rcu_read_lock();
 473         numqueues = ACCESS_ONCE(tun->numqueues);
 474
 475         txq = __skb_get_hash_symmetric(skb);
 476         if (txq) {
 477                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 478                 if (e) {
 479                         tun_flow_save_rps_rxhash(e, txq);
 480                         txq = e->queue_index;
 481                 } else
 482                         /* use multiply and shift instead of expensive divide */
 483                         txq = ((u64)txq * numqueues) >> 32;
 484         } else if (likely(skb_rx_queue_recorded(skb))) {
 485                 txq = skb_get_rx_queue(skb);
 486                 while (unlikely(txq >= numqueues))
 487                         txq -= numqueues;
 488         }
 489
 490         rcu_read_unlock();
 491         return txq;
 492 }
 493
 494 static inline bool tun_not_capable(struct tun_struct *tun)
 495 {
 496         const struct cred *cred = current_cred();
 497         struct net *net = dev_net(tun->dev);
 498
 499         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 500                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 501                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 502 }
 503
 504 static void tun_set_real_num_queues(struct tun_struct *tun)
 505 {
 506         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 507         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 508 }
 509
 510 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 511 {
 512         tfile->detached = tun;
 513         list_add_tail(&tfile->next, &tun->disabled);
 514         ++tun->numdisabled;
 515 }
 516
 517 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 518 {
 519         struct tun_struct *tun = tfile->detached;
 520
 521         tfile->detached = NULL;
 522         list_del_init(&tfile->next);
 523         --tun->numdisabled;
 524         return tun;
 525 }
 526
 527 static void tun_queue_purge(struct tun_file *tfile)
 528 {
 529         struct sk_buff *skb;
 530
 531         while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
 532                 kfree_skb(skb);
 533
 534         skb_queue_purge(&tfile->sk.sk_write_queue);
 535         skb_queue_purge(&tfile->sk.sk_error_queue);
 536 }
 537
 538 static void __tun_detach(struct tun_file *tfile, bool clean)
 539 {
 540         struct tun_file *ntfile;
 541         struct tun_struct *tun;
 542
 543         tun = rtnl_dereference(tfile->tun);
 544
 545         if (tun && !tfile->detached) {
 546                 u16 index = tfile->queue_index;
 547                 BUG_ON(index >= tun->numqueues);
 548
 549                 rcu_assign_pointer(tun->tfiles[index],
 550                                    tun->tfiles[tun->numqueues - 1]);
 551                 ntfile = rtnl_dereference(tun->tfiles[index]);
 552                 ntfile->queue_index = index;
 553
 554                 --tun->numqueues;
 555                 if (clean) {
 556                         RCU_INIT_POINTER(tfile->tun, NULL);
 557                         sock_put(&tfile->sk);
 558                 } else
 559                         tun_disable_queue(tun, tfile);
 560
 561                 synchronize_net();
 562                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 563                 /* Drop read queue */
 564                 tun_queue_purge(tfile);
 565                 tun_set_real_num_queues(tun);
 566         } else if (tfile->detached && clean) {
 567                 tun = tun_enable_queue(tfile);
 568                 sock_put(&tfile->sk);
 569         }
 570
 571         if (clean) {
 572                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 573                         netif_carrier_off(tun->dev);
 574
 575                         if (!(tun->flags & IFF_PERSIST) &&
 576                             tun->dev->reg_state == NETREG_REGISTERED)
 577                                 unregister_netdevice(tun->dev);
 578                 }
 579                 if (tun)
 580                         skb_array_cleanup(&tfile->tx_array);
 581                 if (tfile->alloc_frag.page)
 582                         put_page(tfile->alloc_frag.page);
 583                 sock_put(&tfile->sk);
 584         }
 585 }
 586
 587 static void tun_detach(struct tun_file *tfile, bool clean)
 588 {
 589         rtnl_lock();
 590         __tun_detach(tfile, clean);
 591         rtnl_unlock();
 592 }
 593
 594 static void tun_detach_all(struct net_device *dev)
 595 {
 596         struct tun_struct *tun = netdev_priv(dev);
 597         struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
 598         struct tun_file *tfile, *tmp;
 599         int i, n = tun->numqueues;
 600
 601         for (i = 0; i < n; i++) {
 602                 tfile = rtnl_dereference(tun->tfiles[i]);
 603                 BUG_ON(!tfile);
 604                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 605                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 606                 RCU_INIT_POINTER(tfile->tun, NULL);
 607                 --tun->numqueues;
 608         }
 609         list_for_each_entry(tfile, &tun->disabled, next) {
 610                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 611                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 612                 RCU_INIT_POINTER(tfile->tun, NULL);
 613         }
 614         BUG_ON(tun->numqueues != 0);
 615
 616         synchronize_net();
 617         for (i = 0; i < n; i++) {
 618                 tfile = rtnl_dereference(tun->tfiles[i]);
 619                 /* Drop read queue */
 620                 tun_queue_purge(tfile);
 621                 sock_put(&tfile->sk);
 622         }
 623         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 624                 tun_enable_queue(tfile);
 625                 tun_queue_purge(tfile);
 626                 sock_put(&tfile->sk);
 627         }
 628         BUG_ON(tun->numdisabled != 0);
 629
 630         if (xdp_prog)
 631                 bpf_prog_put(xdp_prog);
 632
 633         if (tun->flags & IFF_PERSIST)
 634                 module_put(THIS_MODULE);
 635 }
 636
 637 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 638 {
 639         struct tun_file *tfile = file->private_data;
 640         struct net_device *dev = tun->dev;
 641         int err;
 642
 643         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 644         if (err < 0)
 645                 goto out;
 646
 647         err = -EINVAL;
 648         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 649                 goto out;
 650
 651         err = -EBUSY;
 652         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 653                 goto out;
 654
 655         err = -E2BIG;
 656         if (!tfile->detached &&
 657             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 658                 goto out;
 659
 660         err = 0;
 661
 662         /* Re-attach the filter to persist device */
 663         if (!skip_filter && (tun->filter_attached == true)) {
 664                 lock_sock(tfile->socket.sk);
 665                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 666                 release_sock(tfile->socket.sk);
 667                 if (!err)
 668                         goto out;
 669         }
 670
 671         if (!tfile->detached &&
 672             skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
 673                 err = -ENOMEM;
 674                 goto out;
 675         }
 676
 677         tfile->queue_index = tun->numqueues;
 678         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 679         rcu_assign_pointer(tfile->tun, tun);
 680         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 681         tun->numqueues++;
 682
 683         if (tfile->detached)
 684                 tun_enable_queue(tfile);
 685         else
 686                 sock_hold(&tfile->sk);
 687
 688         tun_set_real_num_queues(tun);
 689
 690         /* device is allowed to go away first, so no need to hold extra
 691          * refcnt.
 692          */
 693
 694 out:
 695         return err;
 696 }
 697
 698 static struct tun_struct *__tun_get(struct tun_file *tfile)
 699 {
 700         struct tun_struct *tun;
 701
 702         rcu_read_lock();
 703         tun = rcu_dereference(tfile->tun);
 704         if (tun)
 705                 dev_hold(tun->dev);
 706         rcu_read_unlock();
 707
 708         return tun;
 709 }
 710
 711 static struct tun_struct *tun_get(struct file *file)
 712 {
 713         return __tun_get(file->private_data);
 714 }
 715
 716 static void tun_put(struct tun_struct *tun)
 717 {
 718         dev_put(tun->dev);
 719 }
 720
 721 /* TAP filtering */
 722 static void addr_hash_set(u32 *mask, const u8 *addr)
 723 {
 724         int n = ether_crc(ETH_ALEN, addr) >> 26;
 725         mask[n >> 5] |= (1 << (n & 31));
 726 }
 727
 728 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 729 {
 730         int n = ether_crc(ETH_ALEN, addr) >> 26;
 731         return mask[n >> 5] & (1 << (n & 31));
 732 }
 733
 734 static int update_filter(struct tap_filter *filter, void __user *arg)
 735 {
 736         struct { u8 u[ETH_ALEN]; } *addr;
 737         struct tun_filter uf;
 738         int err, alen, n, nexact;
 739
 740         if (copy_from_user(&uf, arg, sizeof(uf)))
 741                 return -EFAULT;
 742
 743         if (!uf.count) {
 744                 /* Disabled */
 745                 filter->count = 0;
 746                 return 0;
 747         }
 748
 749         alen = ETH_ALEN * uf.count;
 750         addr = memdup_user(arg + sizeof(uf), alen);
 751         if (IS_ERR(addr))
 752                 return PTR_ERR(addr);
 753
 754         /* The filter is updated without holding any locks. Which is
 755          * perfectly safe. We disable it first and in the worst
 756          * case we'll accept a few undesired packets. */
 757         filter->count = 0;
 758         wmb();
 759
 760         /* Use first set of addresses as an exact filter */
 761         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 762                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 763
 764         nexact = n;
 765
 766         /* Remaining multicast addresses are hashed,
 767          * unicast will leave the filter disabled. */
 768         memset(filter->mask, 0, sizeof(filter->mask));
 769         for (; n < uf.count; n++) {
 770                 if (!is_multicast_ether_addr(addr[n].u)) {
 771                         err = 0; /* no filter */
 772                         goto free_addr;
 773                 }
 774                 addr_hash_set(filter->mask, addr[n].u);
 775         }
 776
 777         /* For ALLMULTI just set the mask to all ones.
 778          * This overrides the mask populated above. */
 779         if ((uf.flags & TUN_FLT_ALLMULTI))
 780                 memset(filter->mask, ~0, sizeof(filter->mask));
 781
 782         /* Now enable the filter */
 783         wmb();
 784         filter->count = nexact;
 785
 786         /* Return the number of exact filters */
 787         err = nexact;
 788 free_addr:
 789         kfree(addr);
 790         return err;
 791 }
 792
 793 /* Returns: 0 - drop, !=0 - accept */
 794 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 795 {
 796         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 797          * at this point. */
 798         struct ethhdr *eh = (struct ethhdr *) skb->data;
 799         int i;
 800
 801         /* Exact match */
 802         for (i = 0; i < filter->count; i++)
 803                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 804                         return 1;
 805
 806         /* Inexact match (multicast only) */
 807         if (is_multicast_ether_addr(eh->h_dest))
 808                 return addr_hash_test(filter->mask, eh->h_dest);
 809
 810         return 0;
 811 }
 812
 813 /*
 814  * Checks whether the packet is accepted or not.
 815  * Returns: 0 - drop, !=0 - accept
 816  */
 817 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 818 {
 819         if (!filter->count)
 820                 return 1;
 821
 822         return run_filter(filter, skb);
 823 }
 824
 825 /* Network device part of the driver */
 826
 827 static const struct ethtool_ops tun_ethtool_ops;
 828
 829 /* Net device detach from fd. */
 830 static void tun_net_uninit(struct net_device *dev)
 831 {
 832         tun_detach_all(dev);
 833 }
 834
 835 /* Net device open. */
 836 static int tun_net_open(struct net_device *dev)
 837 {
 838         struct tun_struct *tun = netdev_priv(dev);
 839         int i;
 840
 841         netif_tx_start_all_queues(dev);
 842
 843         for (i = 0; i < tun->numqueues; i++) {
 844                 struct tun_file *tfile;
 845
 846                 tfile = rtnl_dereference(tun->tfiles[i]);
 847                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
 848         }
 849
 850         return 0;
 851 }
 852
 853 /* Net device close. */
 854 static int tun_net_close(struct net_device *dev)
 855 {
 856         netif_tx_stop_all_queues(dev);
 857         return 0;
 858 }
 859
 860 /* Net device start xmit */
 861 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 862 {
 863         struct tun_struct *tun = netdev_priv(dev);
 864         int txq = skb->queue_mapping;
 865         struct tun_file *tfile;
 866         u32 numqueues = 0;
 867
 868         rcu_read_lock();
 869         tfile = rcu_dereference(tun->tfiles[txq]);
 870         numqueues = ACCESS_ONCE(tun->numqueues);
 871
 872         /* Drop packet if interface is not attached */
 873         if (txq >= numqueues)
 874                 goto drop;
 875
 876 #ifdef CONFIG_RPS
 877         if (numqueues == 1 && static_key_false(&rps_needed)) {
 878                 /* Select queue was not called for the skbuff, so we extract the
 879                  * RPS hash and save it into the flow_table here.
 880                  */
 881                 __u32 rxhash;
 882
 883                 rxhash = __skb_get_hash_symmetric(skb);
 884                 if (rxhash) {
 885                         struct tun_flow_entry *e;
 886                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
 887                                         rxhash);
 888                         if (e)
 889                                 tun_flow_save_rps_rxhash(e, rxhash);
 890                 }
 891         }
 892 #endif
 893
 894         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
 895
 896         BUG_ON(!tfile);
 897
 898         /* Drop if the filter does not like it.
 899          * This is a noop if the filter is disabled.
 900          * Filter can be enabled only for the TAP devices. */
 901         if (!check_filter(&tun->txflt, skb))
 902                 goto drop;
 903
 904         if (tfile->socket.sk->sk_filter &&
 905             sk_filter(tfile->socket.sk, skb))
 906                 goto drop;
 907
 908         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
 909                 goto drop;
 910
 911         skb_tx_timestamp(skb);
 912
 913         /* Orphan the skb - required as we might hang on to it
 914          * for indefinite time.
 915          */
 916         skb_orphan(skb);
 917
 918         nf_reset(skb);
 919
 920         if (skb_array_produce(&tfile->tx_array, skb))
 921                 goto drop;
 922
 923         /* Notify and wake up reader process */
 924         if (tfile->flags & TUN_FASYNC)
 925                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
 926         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 927
 928         rcu_read_unlock();
 929         return NETDEV_TX_OK;
 930
 931 drop:
 932         this_cpu_inc(tun->pcpu_stats->tx_dropped);
 933         skb_tx_error(skb);
 934         kfree_skb(skb);
 935         rcu_read_unlock();
 936         return NET_XMIT_DROP;
 937 }
 938
 939 static void tun_net_mclist(struct net_device *dev)
 940 {
 941         /*
 942          * This callback is supposed to deal with mc filter in
 943          * _rx_ path and has nothing to do with the _tx_ path.
 944          * In rx path we always accept everything userspace gives us.
 945          */
 946 }
 947
 948 static netdev_features_t tun_net_fix_features(struct net_device *dev,
 949         netdev_features_t features)
 950 {
 951         struct tun_struct *tun = netdev_priv(dev);
 952
 953         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
 954 }
 955 #ifdef CONFIG_NET_POLL_CONTROLLER
 956 static void tun_poll_controller(struct net_device *dev)
 957 {
 958         /*
 959          * Tun only receives frames when:
 960          * 1) the char device endpoint gets data from user space
 961          * 2) the tun socket gets a sendmsg call from user space
 962          * Since both of those are synchronous operations, we are guaranteed
 963          * never to have pending data when we poll for it
 964          * so there is nothing to do here but return.
 965          * We need this though so netpoll recognizes us as an interface that
 966          * supports polling, which enables bridge devices in virt setups to
 967          * still use netconsole
 968          */
 969         return;
 970 }
 971 #endif
 972
 973 static void tun_set_headroom(struct net_device *dev, int new_hr)
 974 {
 975         struct tun_struct *tun = netdev_priv(dev);
 976
 977         if (new_hr < NET_SKB_PAD)
 978                 new_hr = NET_SKB_PAD;
 979
 980         tun->align = new_hr;
 981 }
 982
 983 static void
 984 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
 985 {
 986         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
 987         struct tun_struct *tun = netdev_priv(dev);
 988         struct tun_pcpu_stats *p;
 989         int i;
 990
 991         for_each_possible_cpu(i) {
 992                 u64 rxpackets, rxbytes, txpackets, txbytes;
 993                 unsigned int start;
 994
 995                 p = per_cpu_ptr(tun->pcpu_stats, i);
 996                 do {
 997                         start = u64_stats_fetch_begin(&p->syncp);
 998                         rxpackets       = p->rx_packets;
 999                         rxbytes         = p->rx_bytes;
1000                         txpackets       = p->tx_packets;
1001                         txbytes         = p->tx_bytes;
1002                 } while (u64_stats_fetch_retry(&p->syncp, start));
1003
1004                 stats->rx_packets       += rxpackets;
1005                 stats->rx_bytes         += rxbytes;
1006                 stats->tx_packets       += txpackets;
1007                 stats->tx_bytes         += txbytes;
1008
1009                 /* u32 counters */
1010                 rx_dropped      += p->rx_dropped;
1011                 rx_frame_errors += p->rx_frame_errors;
1012                 tx_dropped      += p->tx_dropped;
1013         }
1014         stats->rx_dropped  = rx_dropped;
1015         stats->rx_frame_errors = rx_frame_errors;
1016         stats->tx_dropped = tx_dropped;
1017 }
1018
1019 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1020                        struct netlink_ext_ack *extack)
1021 {
1022         struct tun_struct *tun = netdev_priv(dev);
1023         struct bpf_prog *old_prog;
1024
1025         old_prog = rtnl_dereference(tun->xdp_prog);
1026         rcu_assign_pointer(tun->xdp_prog, prog);
1027         if (old_prog)
1028                 bpf_prog_put(old_prog);
1029
1030         return 0;
1031 }
1032
1033 static u32 tun_xdp_query(struct net_device *dev)
1034 {
1035         struct tun_struct *tun = netdev_priv(dev);
1036         const struct bpf_prog *xdp_prog;
1037
1038         xdp_prog = rtnl_dereference(tun->xdp_prog);
1039         if (xdp_prog)
1040                 return xdp_prog->aux->id;
1041
1042         return 0;
1043 }
1044
1045 static int tun_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1046 {
1047         switch (xdp->command) {
1048         case XDP_SETUP_PROG:
1049                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1050         case XDP_QUERY_PROG:
1051                 xdp->prog_id = tun_xdp_query(dev);
1052                 xdp->prog_attached = !!xdp->prog_id;
1053                 return 0;
1054         default:
1055                 return -EINVAL;
1056         }
1057 }
1058
1059 static const struct net_device_ops tun_netdev_ops = {
1060         .ndo_uninit             = tun_net_uninit,
1061         .ndo_open               = tun_net_open,
1062         .ndo_stop               = tun_net_close,
1063         .ndo_start_xmit         = tun_net_xmit,
1064         .ndo_fix_features       = tun_net_fix_features,
1065         .ndo_select_queue       = tun_select_queue,
1066 #ifdef CONFIG_NET_POLL_CONTROLLER
1067         .ndo_poll_controller    = tun_poll_controller,
1068 #endif
1069         .ndo_set_rx_headroom    = tun_set_headroom,
1070         .ndo_get_stats64        = tun_net_get_stats64,
1071 };
1072
1073 static const struct net_device_ops tap_netdev_ops = {
1074         .ndo_uninit             = tun_net_uninit,
1075         .ndo_open               = tun_net_open,
1076         .ndo_stop               = tun_net_close,
1077         .ndo_start_xmit         = tun_net_xmit,
1078         .ndo_fix_features       = tun_net_fix_features,
1079         .ndo_set_rx_mode        = tun_net_mclist,
1080         .ndo_set_mac_address    = eth_mac_addr,
1081         .ndo_validate_addr      = eth_validate_addr,
1082         .ndo_select_queue       = tun_select_queue,
1083 #ifdef CONFIG_NET_POLL_CONTROLLER
1084         .ndo_poll_controller    = tun_poll_controller,
1085 #endif
1086         .ndo_features_check     = passthru_features_check,
1087         .ndo_set_rx_headroom    = tun_set_headroom,
1088         .ndo_get_stats64        = tun_net_get_stats64,
1089         .ndo_xdp                = tun_xdp,
1090 };
1091
1092 static void tun_flow_init(struct tun_struct *tun)
1093 {
1094         int i;
1095
1096         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1097                 INIT_HLIST_HEAD(&tun->flows[i]);
1098
1099         tun->ageing_time = TUN_FLOW_EXPIRE;
1100         setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1101         mod_timer(&tun->flow_gc_timer,
1102                   round_jiffies_up(jiffies + tun->ageing_time));
1103 }
1104
1105 static void tun_flow_uninit(struct tun_struct *tun)
1106 {
1107         del_timer_sync(&tun->flow_gc_timer);
1108         tun_flow_flush(tun);
1109 }
1110
1111 #define MIN_MTU 68
1112 #define MAX_MTU 65535
1113
1114 /* Initialize net device. */
1115 static void tun_net_init(struct net_device *dev)
1116 {
1117         struct tun_struct *tun = netdev_priv(dev);
1118
1119         switch (tun->flags & TUN_TYPE_MASK) {
1120         case IFF_TUN:
1121                 dev->netdev_ops = &tun_netdev_ops;
1122
1123                 /* Point-to-Point TUN Device */
1124                 dev->hard_header_len = 0;
1125                 dev->addr_len = 0;
1126                 dev->mtu = 1500;
1127
1128                 /* Zero header length */
1129                 dev->type = ARPHRD_NONE;
1130                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1131                 break;
1132
1133         case IFF_TAP:
1134                 dev->netdev_ops = &tap_netdev_ops;
1135                 /* Ethernet TAP Device */
1136                 ether_setup(dev);
1137                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1138                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1139
1140                 eth_hw_addr_random(dev);
1141
1142                 break;
1143         }
1144
1145         dev->min_mtu = MIN_MTU;
1146         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1147 }
1148
1149 /* Character device part */
1150
1151 /* Poll */
1152 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1153 {
1154         struct tun_file *tfile = file->private_data;
1155         struct tun_struct *tun = __tun_get(tfile);
1156         struct sock *sk;
1157         unsigned int mask = 0;
1158
1159         if (!tun)
1160                 return POLLERR;
1161
1162         sk = tfile->socket.sk;
1163
1164         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1165
1166         poll_wait(file, sk_sleep(sk), wait);
1167
1168         if (!skb_array_empty(&tfile->tx_array))
1169                 mask |= POLLIN | POLLRDNORM;
1170
1171         if (tun->dev->flags & IFF_UP &&
1172             (sock_writeable(sk) ||
1173              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1174               sock_writeable(sk))))
1175                 mask |= POLLOUT | POLLWRNORM;
1176
1177         if (tun->dev->reg_state != NETREG_REGISTERED)
1178                 mask = POLLERR;
1179
1180         tun_put(tun);
1181         return mask;
1182 }
1183
1184 /* prepad is the amount to reserve at front.  len is length after that.
1185  * linear is a hint as to how much to copy (usually headers). */
1186 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1187                                      size_t prepad, size_t len,
1188                                      size_t linear, int noblock)
1189 {
1190         struct sock *sk = tfile->socket.sk;
1191         struct sk_buff *skb;
1192         int err;
1193
1194         /* Under a page?  Don't bother with paged skb. */
1195         if (prepad + len < PAGE_SIZE || !linear)
1196                 linear = len;
1197
1198         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1199                                    &err, 0);
1200         if (!skb)
1201                 return ERR_PTR(err);
1202
1203         skb_reserve(skb, prepad);
1204         skb_put(skb, linear);
1205         skb->data_len = len - linear;
1206         skb->len += len - linear;
1207
1208         return skb;
1209 }
1210
1211 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1212                            struct sk_buff *skb, int more)
1213 {
1214         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1215         struct sk_buff_head process_queue;
1216         u32 rx_batched = tun->rx_batched;
1217         bool rcv = false;
1218
1219         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1220                 local_bh_disable();
1221                 netif_receive_skb(skb);
1222                 local_bh_enable();
1223                 return;
1224         }
1225
1226         spin_lock(&queue->lock);
1227         if (!more || skb_queue_len(queue) == rx_batched) {
1228                 __skb_queue_head_init(&process_queue);
1229                 skb_queue_splice_tail_init(queue, &process_queue);
1230                 rcv = true;
1231         } else {
1232                 __skb_queue_tail(queue, skb);
1233         }
1234         spin_unlock(&queue->lock);
1235
1236         if (rcv) {
1237                 struct sk_buff *nskb;
1238
1239                 local_bh_disable();
1240                 while ((nskb = __skb_dequeue(&process_queue)))
1241                         netif_receive_skb(nskb);
1242                 netif_receive_skb(skb);
1243                 local_bh_enable();
1244         }
1245 }
1246
1247 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1248                               int len, int noblock, bool zerocopy)
1249 {
1250         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1251                 return false;
1252
1253         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1254                 return false;
1255
1256         if (!noblock)
1257                 return false;
1258
1259         if (zerocopy)
1260                 return false;
1261
1262         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1263             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1264                 return false;
1265
1266         return true;
1267 }
1268
1269 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1270                                      struct tun_file *tfile,
1271                                      struct iov_iter *from,
1272                                      struct virtio_net_hdr *hdr,
1273                                      int len, int *generic_xdp)
1274 {
1275         struct page_frag *alloc_frag = &tfile->alloc_frag;
1276         struct sk_buff *skb;
1277         struct bpf_prog *xdp_prog;
1278         int buflen = SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1279                      SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1280         unsigned int delta = 0;
1281         char *buf;
1282         size_t copied;
1283         bool xdp_xmit = false;
1284         int err;
1285
1286         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1287                 return ERR_PTR(-ENOMEM);
1288
1289         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1290         copied = copy_page_from_iter(alloc_frag->page,
1291                                      alloc_frag->offset + TUN_RX_PAD,
1292                                      len, from);
1293         if (copied != len)
1294                 return ERR_PTR(-EFAULT);
1295
1296         if (hdr->gso_type)
1297                 *generic_xdp = 1;
1298         else
1299                 *generic_xdp = 0;
1300
1301         rcu_read_lock();
1302         xdp_prog = rcu_dereference(tun->xdp_prog);
1303         if (xdp_prog && !*generic_xdp) {
1304                 struct xdp_buff xdp;
1305                 void *orig_data;
1306                 u32 act;
1307
1308                 xdp.data_hard_start = buf;
1309                 xdp.data = buf + TUN_RX_PAD;
1310                 xdp.data_end = xdp.data + len;
1311                 orig_data = xdp.data;
1312                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1313
1314                 switch (act) {
1315                 case XDP_REDIRECT:
1316                         get_page(alloc_frag->page);
1317                         alloc_frag->offset += buflen;
1318                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1319                         if (err)
1320                                 goto err_redirect;
1321                         return NULL;
1322                 case XDP_TX:
1323                         xdp_xmit = true;
1324                         /* fall through */
1325                 case XDP_PASS:
1326                         delta = orig_data - xdp.data;
1327                         break;
1328                 default:
1329                         bpf_warn_invalid_xdp_action(act);
1330                         /* fall through */
1331                 case XDP_ABORTED:
1332                         trace_xdp_exception(tun->dev, xdp_prog, act);
1333                         /* fall through */
1334                 case XDP_DROP:
1335                         goto err_xdp;
1336                 }
1337         }
1338
1339         skb = build_skb(buf, buflen);
1340         if (!skb) {
1341                 rcu_read_unlock();
1342                 return ERR_PTR(-ENOMEM);
1343         }
1344
1345         skb_reserve(skb, TUN_RX_PAD - delta);
1346         skb_put(skb, len + delta);
1347         get_page(alloc_frag->page);
1348         alloc_frag->offset += buflen;
1349
1350         if (xdp_xmit) {
1351                 skb->dev = tun->dev;
1352                 generic_xdp_tx(skb, xdp_prog);
1353                 rcu_read_lock();
1354                 return NULL;
1355         }
1356
1357         rcu_read_unlock();
1358
1359         return skb;
1360
1361 err_redirect:
1362         put_page(alloc_frag->page);
1363 err_xdp:
1364         rcu_read_unlock();
1365         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1366         return NULL;
1367 }
1368
1369 /* Get packet from user space buffer */
1370 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1371                             void *msg_control, struct iov_iter *from,
1372                             int noblock, bool more)
1373 {
1374         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1375         struct sk_buff *skb;
1376         size_t total_len = iov_iter_count(from);
1377         size_t len = total_len, align = tun->align, linear;
1378         struct virtio_net_hdr gso = { 0 };
1379         struct tun_pcpu_stats *stats;
1380         int good_linear;
1381         int copylen;
1382         bool zerocopy = false;
1383         int err;
1384         u32 rxhash;
1385         int generic_xdp = 1;
1386
1387         if (!(tun->dev->flags & IFF_UP))
1388                 return -EIO;
1389
1390         if (!(tun->flags & IFF_NO_PI)) {
1391                 if (len < sizeof(pi))
1392                         return -EINVAL;
1393                 len -= sizeof(pi);
1394
1395                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1396                         return -EFAULT;
1397         }
1398
1399         if (tun->flags & IFF_VNET_HDR) {
1400                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1401
1402                 if (len < vnet_hdr_sz)
1403                         return -EINVAL;
1404                 len -= vnet_hdr_sz;
1405
1406                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1407                         return -EFAULT;
1408
1409                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1410                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1411                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1412
1413                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1414                         return -EINVAL;
1415                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1416         }
1417
1418         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1419                 align += NET_IP_ALIGN;
1420                 if (unlikely(len < ETH_HLEN ||
1421                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1422                         return -EINVAL;
1423         }
1424
1425         good_linear = SKB_MAX_HEAD(align);
1426
1427         if (msg_control) {
1428                 struct iov_iter i = *from;
1429
1430                 /* There are 256 bytes to be copied in skb, so there is
1431                  * enough room for skb expand head in case it is used.
1432                  * The rest of the buffer is mapped from userspace.
1433                  */
1434                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1435                 if (copylen > good_linear)
1436                         copylen = good_linear;
1437                 linear = copylen;
1438                 iov_iter_advance(&i, copylen);
1439                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1440                         zerocopy = true;
1441         }
1442
1443         if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1444                 skb = tun_build_skb(tun, tfile, from, &gso, len, &generic_xdp);
1445                 if (IS_ERR(skb)) {
1446                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1447                         return PTR_ERR(skb);
1448                 }
1449                 if (!skb)
1450                         return total_len;
1451         } else {
1452                 if (!zerocopy) {
1453                         copylen = len;
1454                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1455                                 linear = good_linear;
1456                         else
1457                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1458                 }
1459
1460                 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1461                 if (IS_ERR(skb)) {
1462                         if (PTR_ERR(skb) != -EAGAIN)
1463                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1464                         return PTR_ERR(skb);
1465                 }
1466
1467                 if (zerocopy)
1468                         err = zerocopy_sg_from_iter(skb, from);
1469                 else
1470                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1471
1472                 if (err) {
1473                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1474                         kfree_skb(skb);
1475                         return -EFAULT;
1476                 }
1477         }
1478
1479         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1480                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1481                 kfree_skb(skb);
1482                 return -EINVAL;
1483         }
1484
1485         switch (tun->flags & TUN_TYPE_MASK) {
1486         case IFF_TUN:
1487                 if (tun->flags & IFF_NO_PI) {
1488                         switch (skb->data[0] & 0xf0) {
1489                         case 0x40:
1490                                 pi.proto = htons(ETH_P_IP);
1491                                 break;
1492                         case 0x60:
1493                                 pi.proto = htons(ETH_P_IPV6);
1494                                 break;
1495                         default:
1496                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1497                                 kfree_skb(skb);
1498                                 return -EINVAL;
1499                         }
1500                 }
1501
1502                 skb_reset_mac_header(skb);
1503                 skb->protocol = pi.proto;
1504                 skb->dev = tun->dev;
1505                 break;
1506         case IFF_TAP:
1507                 skb->protocol = eth_type_trans(skb, tun->dev);
1508                 break;
1509         }
1510
1511         /* copy skb_ubuf_info for callback when skb has no error */
1512         if (zerocopy) {
1513                 skb_shinfo(skb)->destructor_arg = msg_control;
1514                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1515                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1516         } else if (msg_control) {
1517                 struct ubuf_info *uarg = msg_control;
1518                 uarg->callback(uarg, false);
1519         }
1520
1521         skb_reset_network_header(skb);
1522         skb_probe_transport_header(skb, 0);
1523
1524         if (generic_xdp) {
1525                 struct bpf_prog *xdp_prog;
1526                 int ret;
1527
1528                 rcu_read_lock();
1529                 xdp_prog = rcu_dereference(tun->xdp_prog);
1530                 if (xdp_prog) {
1531                         ret = do_xdp_generic(xdp_prog, skb);
1532                         if (ret != XDP_PASS) {
1533                                 rcu_read_unlock();
1534                                 return total_len;
1535                         }
1536                 }
1537                 rcu_read_unlock();
1538         }
1539
1540         rxhash = __skb_get_hash_symmetric(skb);
1541 #ifndef CONFIG_4KSTACKS
1542         tun_rx_batched(tun, tfile, skb, more);
1543 #else
1544         netif_rx_ni(skb);
1545 #endif
1546
1547         stats = get_cpu_ptr(tun->pcpu_stats);
1548         u64_stats_update_begin(&stats->syncp);
1549         stats->rx_packets++;
1550         stats->rx_bytes += len;
1551         u64_stats_update_end(&stats->syncp);
1552         put_cpu_ptr(stats);
1553
1554         tun_flow_update(tun, rxhash, tfile);
1555         return total_len;
1556 }
1557
1558 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1559 {
1560         struct file *file = iocb->ki_filp;
1561         struct tun_struct *tun = tun_get(file);
1562         struct tun_file *tfile = file->private_data;
1563         ssize_t result;
1564
1565         if (!tun)
1566                 return -EBADFD;
1567
1568         result = tun_get_user(tun, tfile, NULL, from,
1569                               file->f_flags & O_NONBLOCK, false);
1570
1571         tun_put(tun);
1572         return result;
1573 }
1574
1575 /* Put packet to the user space buffer */
1576 static ssize_t tun_put_user(struct tun_struct *tun,
1577                             struct tun_file *tfile,
1578                             struct sk_buff *skb,
1579                             struct iov_iter *iter)
1580 {
1581         struct tun_pi pi = { 0, skb->protocol };
1582         struct tun_pcpu_stats *stats;
1583         ssize_t total;
1584         int vlan_offset = 0;
1585         int vlan_hlen = 0;
1586         int vnet_hdr_sz = 0;
1587
1588         if (skb_vlan_tag_present(skb))
1589                 vlan_hlen = VLAN_HLEN;
1590
1591         if (tun->flags & IFF_VNET_HDR)
1592                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1593
1594         total = skb->len + vlan_hlen + vnet_hdr_sz;
1595
1596         if (!(tun->flags & IFF_NO_PI)) {
1597                 if (iov_iter_count(iter) < sizeof(pi))
1598                         return -EINVAL;
1599
1600                 total += sizeof(pi);
1601                 if (iov_iter_count(iter) < total) {
1602                         /* Packet will be striped */
1603                         pi.flags |= TUN_PKT_STRIP;
1604                 }
1605
1606                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1607                         return -EFAULT;
1608         }
1609
1610         if (vnet_hdr_sz) {
1611                 struct virtio_net_hdr gso;
1612
1613                 if (iov_iter_count(iter) < vnet_hdr_sz)
1614                         return -EINVAL;
1615
1616                 if (virtio_net_hdr_from_skb(skb, &gso,
1617                                             tun_is_little_endian(tun), true)) {
1618                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1619                         pr_err("unexpected GSO type: "
1620                                "0x%x, gso_size %d, hdr_len %d\n",
1621                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1622                                tun16_to_cpu(tun, gso.hdr_len));
1623                         print_hex_dump(KERN_ERR, "tun: ",
1624                                        DUMP_PREFIX_NONE,
1625                                        16, 1, skb->head,
1626                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1627                         WARN_ON_ONCE(1);
1628                         return -EINVAL;
1629                 }
1630
1631                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1632                         return -EFAULT;
1633
1634                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1635         }
1636
1637         if (vlan_hlen) {
1638                 int ret;
1639                 struct {
1640                         __be16 h_vlan_proto;
1641                         __be16 h_vlan_TCI;
1642                 } veth;
1643
1644                 veth.h_vlan_proto = skb->vlan_proto;
1645                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1646
1647                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1648
1649                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1650                 if (ret || !iov_iter_count(iter))
1651                         goto done;
1652
1653                 ret = copy_to_iter(&veth, sizeof(veth), iter);
1654                 if (ret != sizeof(veth) || !iov_iter_count(iter))
1655                         goto done;
1656         }
1657
1658         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1659
1660 done:
1661         /* caller is in process context, */
1662         stats = get_cpu_ptr(tun->pcpu_stats);
1663         u64_stats_update_begin(&stats->syncp);
1664         stats->tx_packets++;
1665         stats->tx_bytes += skb->len + vlan_hlen;
1666         u64_stats_update_end(&stats->syncp);
1667         put_cpu_ptr(tun->pcpu_stats);
1668
1669         return total;
1670 }
1671
1672 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1673                                      int *err)
1674 {
1675         DECLARE_WAITQUEUE(wait, current);
1676         struct sk_buff *skb = NULL;
1677         int error = 0;
1678
1679         skb = skb_array_consume(&tfile->tx_array);
1680         if (skb)
1681                 goto out;
1682         if (noblock) {
1683                 error = -EAGAIN;
1684                 goto out;
1685         }
1686
1687         add_wait_queue(&tfile->wq.wait, &wait);
1688         current->state = TASK_INTERRUPTIBLE;
1689
1690         while (1) {
1691                 skb = skb_array_consume(&tfile->tx_array);
1692                 if (skb)
1693                         break;
1694                 if (signal_pending(current)) {
1695                         error = -ERESTARTSYS;
1696                         break;
1697                 }
1698                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1699                         error = -EFAULT;
1700                         break;
1701                 }
1702
1703                 schedule();
1704         }
1705
1706         current->state = TASK_RUNNING;
1707         remove_wait_queue(&tfile->wq.wait, &wait);
1708
1709 out:
1710         *err = error;
1711         return skb;
1712 }
1713
1714 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1715                            struct iov_iter *to,
1716                            int noblock, struct sk_buff *skb)
1717 {
1718         ssize_t ret;
1719         int err;
1720
1721         tun_debug(KERN_INFO, tun, "tun_do_read\n");
1722
1723         if (!iov_iter_count(to))
1724                 return 0;
1725
1726         if (!skb) {
1727                 /* Read frames from ring */
1728                 skb = tun_ring_recv(tfile, noblock, &err);
1729                 if (!skb)
1730                         return err;
1731         }
1732
1733         ret = tun_put_user(tun, tfile, skb, to);
1734         if (unlikely(ret < 0))
1735                 kfree_skb(skb);
1736         else
1737                 consume_skb(skb);
1738
1739         return ret;
1740 }
1741
1742 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1743 {
1744         struct file *file = iocb->ki_filp;
1745         struct tun_file *tfile = file->private_data;
1746         struct tun_struct *tun = __tun_get(tfile);
1747         ssize_t len = iov_iter_count(to), ret;
1748
1749         if (!tun)
1750                 return -EBADFD;
1751         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1752         ret = min_t(ssize_t, ret, len);
1753         if (ret > 0)
1754                 iocb->ki_pos = ret;
1755         tun_put(tun);
1756         return ret;
1757 }
1758
1759 static void tun_free_netdev(struct net_device *dev)
1760 {
1761         struct tun_struct *tun = netdev_priv(dev);
1762
1763         BUG_ON(!(list_empty(&tun->disabled)));
1764         free_percpu(tun->pcpu_stats);
1765         tun_flow_uninit(tun);
1766         security_tun_dev_free_security(tun->security);
1767 }
1768
1769 static void tun_setup(struct net_device *dev)
1770 {
1771         struct tun_struct *tun = netdev_priv(dev);
1772
1773         tun->owner = INVALID_UID;
1774         tun->group = INVALID_GID;
1775
1776         dev->ethtool_ops = &tun_ethtool_ops;
1777         dev->needs_free_netdev = true;
1778         dev->priv_destructor = tun_free_netdev;
1779         /* We prefer our own queue length */
1780         dev->tx_queue_len = TUN_READQ_SIZE;
1781 }
1782
1783 /* Trivial set of netlink ops to allow deleting tun or tap
1784  * device with netlink.
1785  */
1786 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1787                         struct netlink_ext_ack *extack)
1788 {
1789         return -EINVAL;
1790 }
1791
1792 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1793         .kind           = DRV_NAME,
1794         .priv_size      = sizeof(struct tun_struct),
1795         .setup          = tun_setup,
1796         .validate       = tun_validate,
1797 };
1798
1799 static void tun_sock_write_space(struct sock *sk)
1800 {
1801         struct tun_file *tfile;
1802         wait_queue_head_t *wqueue;
1803
1804         if (!sock_writeable(sk))
1805                 return;
1806
1807         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1808                 return;
1809
1810         wqueue = sk_sleep(sk);
1811         if (wqueue && waitqueue_active(wqueue))
1812                 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1813                                                 POLLWRNORM | POLLWRBAND);
1814
1815         tfile = container_of(sk, struct tun_file, sk);
1816         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1817 }
1818
1819 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1820 {
1821         int ret;
1822         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1823         struct tun_struct *tun = __tun_get(tfile);
1824
1825         if (!tun)
1826                 return -EBADFD;
1827
1828         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1829                            m->msg_flags & MSG_DONTWAIT,
1830                            m->msg_flags & MSG_MORE);
1831         tun_put(tun);
1832         return ret;
1833 }
1834
1835 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1836                        int flags)
1837 {
1838         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1839         struct tun_struct *tun = __tun_get(tfile);
1840         int ret;
1841
1842         if (!tun)
1843                 return -EBADFD;
1844
1845         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1846                 ret = -EINVAL;
1847                 goto out;
1848         }
1849         if (flags & MSG_ERRQUEUE) {
1850                 ret = sock_recv_errqueue(sock->sk, m, total_len,
1851                                          SOL_PACKET, TUN_TX_TIMESTAMP);
1852                 goto out;
1853         }
1854         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT,
1855                           m->msg_control);
1856         if (ret > (ssize_t)total_len) {
1857                 m->msg_flags |= MSG_TRUNC;
1858                 ret = flags & MSG_TRUNC ? ret : total_len;
1859         }
1860 out:
1861         tun_put(tun);
1862         return ret;
1863 }
1864
1865 static int tun_peek_len(struct socket *sock)
1866 {
1867         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1868         struct tun_struct *tun;
1869         int ret = 0;
1870
1871         tun = __tun_get(tfile);
1872         if (!tun)
1873                 return 0;
1874
1875         ret = skb_array_peek_len(&tfile->tx_array);
1876         tun_put(tun);
1877
1878         return ret;
1879 }
1880
1881 /* Ops structure to mimic raw sockets with tun */
1882 static const struct proto_ops tun_socket_ops = {
1883         .peek_len = tun_peek_len,
1884         .sendmsg = tun_sendmsg,
1885         .recvmsg = tun_recvmsg,
1886 };
1887
1888 static struct proto tun_proto = {
1889         .name           = "tun",
1890         .owner          = THIS_MODULE,
1891         .obj_size       = sizeof(struct tun_file),
1892 };
1893
1894 static int tun_flags(struct tun_struct *tun)
1895 {
1896         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1897 }
1898
1899 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1900                               char *buf)
1901 {
1902         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1903         return sprintf(buf, "0x%x\n", tun_flags(tun));
1904 }
1905
1906 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1907                               char *buf)
1908 {
1909         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1910         return uid_valid(tun->owner)?
1911                 sprintf(buf, "%u\n",
1912                         from_kuid_munged(current_user_ns(), tun->owner)):
1913                 sprintf(buf, "-1\n");
1914 }
1915
1916 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1917                               char *buf)
1918 {
1919         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1920         return gid_valid(tun->group) ?
1921                 sprintf(buf, "%u\n",
1922                         from_kgid_munged(current_user_ns(), tun->group)):
1923                 sprintf(buf, "-1\n");
1924 }
1925
1926 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1927 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1928 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1929
1930 static struct attribute *tun_dev_attrs[] = {
1931         &dev_attr_tun_flags.attr,
1932         &dev_attr_owner.attr,
1933         &dev_attr_group.attr,
1934         NULL
1935 };
1936
1937 static const struct attribute_group tun_attr_group = {
1938         .attrs = tun_dev_attrs
1939 };
1940
1941 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1942 {
1943         struct tun_struct *tun;
1944         struct tun_file *tfile = file->private_data;
1945         struct net_device *dev;
1946         int err;
1947
1948         if (tfile->detached)
1949                 return -EINVAL;
1950
1951         dev = __dev_get_by_name(net, ifr->ifr_name);
1952         if (dev) {
1953                 if (ifr->ifr_flags & IFF_TUN_EXCL)
1954                         return -EBUSY;
1955                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1956                         tun = netdev_priv(dev);
1957                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1958                         tun = netdev_priv(dev);
1959                 else
1960                         return -EINVAL;
1961
1962                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1963                     !!(tun->flags & IFF_MULTI_QUEUE))
1964                         return -EINVAL;
1965
1966                 if (tun_not_capable(tun))
1967                         return -EPERM;
1968                 err = security_tun_dev_open(tun->security);
1969                 if (err < 0)
1970                         return err;
1971
1972                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1973                 if (err < 0)
1974                         return err;
1975
1976                 if (tun->flags & IFF_MULTI_QUEUE &&
1977                     (tun->numqueues + tun->numdisabled > 1)) {
1978                         /* One or more queue has already been attached, no need
1979                          * to initialize the device again.
1980                          */
1981                         return 0;
1982                 }
1983         }
1984         else {
1985                 char *name;
1986                 unsigned long flags = 0;
1987                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1988                              MAX_TAP_QUEUES : 1;
1989
1990                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1991                         return -EPERM;
1992                 err = security_tun_dev_create();
1993                 if (err < 0)
1994                         return err;
1995
1996                 /* Set dev type */
1997                 if (ifr->ifr_flags & IFF_TUN) {
1998                         /* TUN device */
1999                         flags |= IFF_TUN;
2000                         name = "tun%d";
2001                 } else if (ifr->ifr_flags & IFF_TAP) {
2002                         /* TAP device */
2003                         flags |= IFF_TAP;
2004                         name = "tap%d";
2005                 } else
2006                         return -EINVAL;
2007
2008                 if (*ifr->ifr_name)
2009                         name = ifr->ifr_name;
2010
2011                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2012                                        NET_NAME_UNKNOWN, tun_setup, queues,
2013                                        queues);
2014
2015                 if (!dev)
2016                         return -ENOMEM;
2017
2018                 dev_net_set(dev, net);
2019                 dev->rtnl_link_ops = &tun_link_ops;
2020                 dev->ifindex = tfile->ifindex;
2021                 dev->sysfs_groups[0] = &tun_attr_group;
2022
2023                 tun = netdev_priv(dev);
2024                 tun->dev = dev;
2025                 tun->flags = flags;
2026                 tun->txflt.count = 0;
2027                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2028
2029                 tun->align = NET_SKB_PAD;
2030                 tun->filter_attached = false;
2031                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2032                 tun->rx_batched = 0;
2033
2034                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2035                 if (!tun->pcpu_stats) {
2036                         err = -ENOMEM;
2037                         goto err_free_dev;
2038                 }
2039
2040                 spin_lock_init(&tun->lock);
2041
2042                 err = security_tun_dev_alloc_security(&tun->security);
2043                 if (err < 0)
2044                         goto err_free_stat;
2045
2046                 tun_net_init(dev);
2047                 tun_flow_init(tun);
2048
2049                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2050                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2051                                    NETIF_F_HW_VLAN_STAG_TX;
2052                 dev->features = dev->hw_features | NETIF_F_LLTX;
2053                 dev->vlan_features = dev->features &
2054                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2055                                        NETIF_F_HW_VLAN_STAG_TX);
2056
2057                 INIT_LIST_HEAD(&tun->disabled);
2058                 err = tun_attach(tun, file, false);
2059                 if (err < 0)
2060                         goto err_free_flow;
2061
2062                 err = register_netdevice(tun->dev);
2063                 if (err < 0)
2064                         goto err_detach;
2065         }
2066
2067         netif_carrier_on(tun->dev);
2068
2069         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2070
2071         tun->flags = (tun->flags & ~TUN_FEATURES) |
2072                 (ifr->ifr_flags & TUN_FEATURES);
2073
2074         /* Make sure persistent devices do not get stuck in
2075          * xoff state.
2076          */
2077         if (netif_running(tun->dev))
2078                 netif_tx_wake_all_queues(tun->dev);
2079
2080         strcpy(ifr->ifr_name, tun->dev->name);
2081         return 0;
2082
2083 err_detach:
2084         tun_detach_all(dev);
2085 err_free_flow:
2086         tun_flow_uninit(tun);
2087         security_tun_dev_free_security(tun->security);
2088 err_free_stat:
2089         free_percpu(tun->pcpu_stats);
2090 err_free_dev:
2091         free_netdev(dev);
2092         return err;
2093 }
2094
2095 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2096                        struct ifreq *ifr)
2097 {
2098         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2099
2100         strcpy(ifr->ifr_name, tun->dev->name);
2101
2102         ifr->ifr_flags = tun_flags(tun);
2103
2104 }
2105
2106 /* This is like a cut-down ethtool ops, except done via tun fd so no
2107  * privs required. */
2108 static int set_offload(struct tun_struct *tun, unsigned long arg)
2109 {
2110         netdev_features_t features = 0;
2111
2112         if (arg & TUN_F_CSUM) {
2113                 features |= NETIF_F_HW_CSUM;
2114                 arg &= ~TUN_F_CSUM;
2115
2116                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2117                         if (arg & TUN_F_TSO_ECN) {
2118                                 features |= NETIF_F_TSO_ECN;
2119                                 arg &= ~TUN_F_TSO_ECN;
2120                         }
2121                         if (arg & TUN_F_TSO4)
2122                                 features |= NETIF_F_TSO;
2123                         if (arg & TUN_F_TSO6)
2124                                 features |= NETIF_F_TSO6;
2125                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2126                 }
2127         }
2128
2129         /* This gives the user a way to test for new features in future by
2130          * trying to set them. */
2131         if (arg)
2132                 return -EINVAL;
2133
2134         tun->set_features = features;
2135         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2136         tun->dev->wanted_features |= features;
2137         netdev_update_features(tun->dev);
2138
2139         return 0;
2140 }
2141
2142 static void tun_detach_filter(struct tun_struct *tun, int n)
2143 {
2144         int i;
2145         struct tun_file *tfile;
2146
2147         for (i = 0; i < n; i++) {
2148                 tfile = rtnl_dereference(tun->tfiles[i]);
2149                 lock_sock(tfile->socket.sk);
2150                 sk_detach_filter(tfile->socket.sk);
2151                 release_sock(tfile->socket.sk);
2152         }
2153
2154         tun->filter_attached = false;
2155 }
2156
2157 static int tun_attach_filter(struct tun_struct *tun)
2158 {
2159         int i, ret = 0;
2160         struct tun_file *tfile;
2161
2162         for (i = 0; i < tun->numqueues; i++) {
2163                 tfile = rtnl_dereference(tun->tfiles[i]);
2164                 lock_sock(tfile->socket.sk);
2165                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2166                 release_sock(tfile->socket.sk);
2167                 if (ret) {
2168                         tun_detach_filter(tun, i);
2169                         return ret;
2170                 }
2171         }
2172
2173         tun->filter_attached = true;
2174         return ret;
2175 }
2176
2177 static void tun_set_sndbuf(struct tun_struct *tun)
2178 {
2179         struct tun_file *tfile;
2180         int i;
2181
2182         for (i = 0; i < tun->numqueues; i++) {
2183                 tfile = rtnl_dereference(tun->tfiles[i]);
2184                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2185         }
2186 }
2187
2188 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2189 {
2190         struct tun_file *tfile = file->private_data;
2191         struct tun_struct *tun;
2192         int ret = 0;
2193
2194         rtnl_lock();
2195
2196         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2197                 tun = tfile->detached;
2198                 if (!tun) {
2199                         ret = -EINVAL;
2200                         goto unlock;
2201                 }
2202                 ret = security_tun_dev_attach_queue(tun->security);
2203                 if (ret < 0)
2204                         goto unlock;
2205                 ret = tun_attach(tun, file, false);
2206         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2207                 tun = rtnl_dereference(tfile->tun);
2208                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2209                         ret = -EINVAL;
2210                 else
2211                         __tun_detach(tfile, false);
2212         } else
2213                 ret = -EINVAL;
2214
2215 unlock:
2216         rtnl_unlock();
2217         return ret;
2218 }
2219
2220 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2221                             unsigned long arg, int ifreq_len)
2222 {
2223         struct tun_file *tfile = file->private_data;
2224         struct tun_struct *tun;
2225         void __user* argp = (void __user*)arg;
2226         struct ifreq ifr;
2227         kuid_t owner;
2228         kgid_t group;
2229         int sndbuf;
2230         int vnet_hdr_sz;
2231         unsigned int ifindex;
2232         int le;
2233         int ret;
2234
2235         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2236                 if (copy_from_user(&ifr, argp, ifreq_len))
2237                         return -EFAULT;
2238         } else {
2239                 memset(&ifr, 0, sizeof(ifr));
2240         }
2241         if (cmd == TUNGETFEATURES) {
2242                 /* Currently this just means: "what IFF flags are valid?".
2243                  * This is needed because we never checked for invalid flags on
2244                  * TUNSETIFF.
2245                  */
2246                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2247                                 (unsigned int __user*)argp);
2248         } else if (cmd == TUNSETQUEUE)
2249                 return tun_set_queue(file, &ifr);
2250
2251         ret = 0;
2252         rtnl_lock();
2253
2254         tun = __tun_get(tfile);
2255         if (cmd == TUNSETIFF) {
2256                 ret = -EEXIST;
2257                 if (tun)
2258                         goto unlock;
2259
2260                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2261
2262                 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2263
2264                 if (ret)
2265                         goto unlock;
2266
2267                 if (copy_to_user(argp, &ifr, ifreq_len))
2268                         ret = -EFAULT;
2269                 goto unlock;
2270         }
2271         if (cmd == TUNSETIFINDEX) {
2272                 ret = -EPERM;
2273                 if (tun)
2274                         goto unlock;
2275
2276                 ret = -EFAULT;
2277                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2278                         goto unlock;
2279
2280                 ret = 0;
2281                 tfile->ifindex = ifindex;
2282                 goto unlock;
2283         }
2284
2285         ret = -EBADFD;
2286         if (!tun)
2287                 goto unlock;
2288
2289         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2290
2291         ret = 0;
2292         switch (cmd) {
2293         case TUNGETIFF:
2294                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2295
2296                 if (tfile->detached)
2297                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2298                 if (!tfile->socket.sk->sk_filter)
2299                         ifr.ifr_flags |= IFF_NOFILTER;
2300
2301                 if (copy_to_user(argp, &ifr, ifreq_len))
2302                         ret = -EFAULT;
2303                 break;
2304
2305         case TUNSETNOCSUM:
2306                 /* Disable/Enable checksum */
2307
2308                 /* [unimplemented] */
2309                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2310                           arg ? "disabled" : "enabled");
2311                 break;
2312
2313         case TUNSETPERSIST:
2314                 /* Disable/Enable persist mode. Keep an extra reference to the
2315                  * module to prevent the module being unprobed.
2316                  */
2317                 if (arg && !(tun->flags & IFF_PERSIST)) {
2318                         tun->flags |= IFF_PERSIST;
2319                         __module_get(THIS_MODULE);
2320                 }
2321                 if (!arg && (tun->flags & IFF_PERSIST)) {
2322                         tun->flags &= ~IFF_PERSIST;
2323                         module_put(THIS_MODULE);
2324                 }
2325
2326                 tun_debug(KERN_INFO, tun, "persist %s\n",
2327                           arg ? "enabled" : "disabled");
2328                 break;
2329
2330         case TUNSETOWNER:
2331                 /* Set owner of the device */
2332                 owner = make_kuid(current_user_ns(), arg);
2333                 if (!uid_valid(owner)) {
2334                         ret = -EINVAL;
2335                         break;
2336                 }
2337                 tun->owner = owner;
2338                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2339                           from_kuid(&init_user_ns, tun->owner));
2340                 break;
2341
2342         case TUNSETGROUP:
2343                 /* Set group of the device */
2344                 group = make_kgid(current_user_ns(), arg);
2345                 if (!gid_valid(group)) {
2346                         ret = -EINVAL;
2347                         break;
2348                 }
2349                 tun->group = group;
2350                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2351                           from_kgid(&init_user_ns, tun->group));
2352                 break;
2353
2354         case TUNSETLINK:
2355                 /* Only allow setting the type when the interface is down */
2356                 if (tun->dev->flags & IFF_UP) {
2357                         tun_debug(KERN_INFO, tun,
2358                                   "Linktype set failed because interface is up\n");
2359                         ret = -EBUSY;
2360                 } else {
2361                         tun->dev->type = (int) arg;
2362                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2363                                   tun->dev->type);
2364                         ret = 0;
2365                 }
2366                 break;
2367
2368 #ifdef TUN_DEBUG
2369         case TUNSETDEBUG:
2370                 tun->debug = arg;
2371                 break;
2372 #endif
2373         case TUNSETOFFLOAD:
2374                 ret = set_offload(tun, arg);
2375                 break;
2376
2377         case TUNSETTXFILTER:
2378                 /* Can be set only for TAPs */
2379                 ret = -EINVAL;
2380                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2381                         break;
2382                 ret = update_filter(&tun->txflt, (void __user *)arg);
2383                 break;
2384
2385         case SIOCGIFHWADDR:
2386                 /* Get hw address */
2387                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2388                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2389                 if (copy_to_user(argp, &ifr, ifreq_len))
2390                         ret = -EFAULT;
2391                 break;
2392
2393         case SIOCSIFHWADDR:
2394                 /* Set hw address */
2395                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2396                           ifr.ifr_hwaddr.sa_data);
2397
2398                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2399                 break;
2400
2401         case TUNGETSNDBUF:
2402                 sndbuf = tfile->socket.sk->sk_sndbuf;
2403                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2404                         ret = -EFAULT;
2405                 break;
2406
2407         case TUNSETSNDBUF:
2408                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2409                         ret = -EFAULT;
2410                         break;
2411                 }
2412
2413                 tun->sndbuf = sndbuf;
2414                 tun_set_sndbuf(tun);
2415                 break;
2416
2417         case TUNGETVNETHDRSZ:
2418                 vnet_hdr_sz = tun->vnet_hdr_sz;
2419                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2420                         ret = -EFAULT;
2421                 break;
2422
2423         case TUNSETVNETHDRSZ:
2424                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2425                         ret = -EFAULT;
2426                         break;
2427                 }
2428                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2429                         ret = -EINVAL;
2430                         break;
2431                 }
2432
2433                 tun->vnet_hdr_sz = vnet_hdr_sz;
2434                 break;
2435
2436         case TUNGETVNETLE:
2437                 le = !!(tun->flags & TUN_VNET_LE);
2438                 if (put_user(le, (int __user *)argp))
2439                         ret = -EFAULT;
2440                 break;
2441
2442         case TUNSETVNETLE:
2443                 if (get_user(le, (int __user *)argp)) {
2444                         ret = -EFAULT;
2445                         break;
2446                 }
2447                 if (le)
2448                         tun->flags |= TUN_VNET_LE;
2449                 else
2450                         tun->flags &= ~TUN_VNET_LE;
2451                 break;
2452
2453         case TUNGETVNETBE:
2454                 ret = tun_get_vnet_be(tun, argp);
2455                 break;
2456
2457         case TUNSETVNETBE:
2458                 ret = tun_set_vnet_be(tun, argp);
2459                 break;
2460
2461         case TUNATTACHFILTER:
2462                 /* Can be set only for TAPs */
2463                 ret = -EINVAL;
2464                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2465                         break;
2466                 ret = -EFAULT;
2467                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2468                         break;
2469
2470                 ret = tun_attach_filter(tun);
2471                 break;
2472
2473         case TUNDETACHFILTER:
2474                 /* Can be set only for TAPs */
2475                 ret = -EINVAL;
2476                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2477                         break;
2478                 ret = 0;
2479                 tun_detach_filter(tun, tun->numqueues);
2480                 break;
2481
2482         case TUNGETFILTER:
2483                 ret = -EINVAL;
2484                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2485                         break;
2486                 ret = -EFAULT;
2487                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2488                         break;
2489                 ret = 0;
2490                 break;
2491
2492         default:
2493                 ret = -EINVAL;
2494                 break;
2495         }
2496
2497 unlock:
2498         rtnl_unlock();
2499         if (tun)
2500                 tun_put(tun);
2501         return ret;
2502 }
2503
2504 static long tun_chr_ioctl(struct file *file,
2505                           unsigned int cmd, unsigned long arg)
2506 {
2507         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2508 }
2509
2510 #ifdef CONFIG_COMPAT
2511 static long tun_chr_compat_ioctl(struct file *file,
2512                          unsigned int cmd, unsigned long arg)
2513 {
2514         switch (cmd) {
2515         case TUNSETIFF:
2516         case TUNGETIFF:
2517         case TUNSETTXFILTER:
2518         case TUNGETSNDBUF:
2519         case TUNSETSNDBUF:
2520         case SIOCGIFHWADDR:
2521         case SIOCSIFHWADDR:
2522                 arg = (unsigned long)compat_ptr(arg);
2523                 break;
2524         default:
2525                 arg = (compat_ulong_t)arg;
2526                 break;
2527         }
2528
2529         /*
2530          * compat_ifreq is shorter than ifreq, so we must not access beyond
2531          * the end of that structure. All fields that are used in this
2532          * driver are compatible though, we don't need to convert the
2533          * contents.
2534          */
2535         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2536 }
2537 #endif /* CONFIG_COMPAT */
2538
2539 static int tun_chr_fasync(int fd, struct file *file, int on)
2540 {
2541         struct tun_file *tfile = file->private_data;
2542         int ret;
2543
2544         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2545                 goto out;
2546
2547         if (on) {
2548                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2549                 tfile->flags |= TUN_FASYNC;
2550         } else
2551                 tfile->flags &= ~TUN_FASYNC;
2552         ret = 0;
2553 out:
2554         return ret;
2555 }
2556
2557 static int tun_chr_open(struct inode *inode, struct file * file)
2558 {
2559         struct net *net = current->nsproxy->net_ns;
2560         struct tun_file *tfile;
2561
2562         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2563
2564         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2565                                             &tun_proto, 0);
2566         if (!tfile)
2567                 return -ENOMEM;
2568         RCU_INIT_POINTER(tfile->tun, NULL);
2569         tfile->flags = 0;
2570         tfile->ifindex = 0;
2571
2572         init_waitqueue_head(&tfile->wq.wait);
2573         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2574
2575         tfile->socket.file = file;
2576         tfile->socket.ops = &tun_socket_ops;
2577
2578         sock_init_data(&tfile->socket, &tfile->sk);
2579
2580         tfile->sk.sk_write_space = tun_sock_write_space;
2581         tfile->sk.sk_sndbuf = INT_MAX;
2582
2583         tfile->alloc_frag.page = NULL;
2584
2585         file->private_data = tfile;
2586         INIT_LIST_HEAD(&tfile->next);
2587
2588         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2589
2590         return 0;
2591 }
2592
2593 static int tun_chr_close(struct inode *inode, struct file *file)
2594 {
2595         struct tun_file *tfile = file->private_data;
2596
2597         tun_detach(tfile, true);
2598
2599         return 0;
2600 }
2601
2602 #ifdef CONFIG_PROC_FS
2603 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2604 {
2605         struct tun_struct *tun;
2606         struct ifreq ifr;
2607
2608         memset(&ifr, 0, sizeof(ifr));
2609
2610         rtnl_lock();
2611         tun = tun_get(f);
2612         if (tun)
2613                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2614         rtnl_unlock();
2615
2616         if (tun)
2617                 tun_put(tun);
2618
2619         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2620 }
2621 #endif
2622
2623 static const struct file_operations tun_fops = {
2624         .owner  = THIS_MODULE,
2625         .llseek = no_llseek,
2626         .read_iter  = tun_chr_read_iter,
2627         .write_iter = tun_chr_write_iter,
2628         .poll   = tun_chr_poll,
2629         .unlocked_ioctl = tun_chr_ioctl,
2630 #ifdef CONFIG_COMPAT
2631         .compat_ioctl = tun_chr_compat_ioctl,
2632 #endif
2633         .open   = tun_chr_open,
2634         .release = tun_chr_close,
2635         .fasync = tun_chr_fasync,
2636 #ifdef CONFIG_PROC_FS
2637         .show_fdinfo = tun_chr_show_fdinfo,
2638 #endif
2639 };
2640
2641 static struct miscdevice tun_miscdev = {
2642         .minor = TUN_MINOR,
2643         .name = "tun",
2644         .nodename = "net/tun",
2645         .fops = &tun_fops,
2646 };
2647
2648 /* ethtool interface */
2649
2650 static int tun_get_link_ksettings(struct net_device *dev,
2651                                   struct ethtool_link_ksettings *cmd)
2652 {
2653         ethtool_link_ksettings_zero_link_mode(cmd, supported);
2654         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2655         cmd->base.speed         = SPEED_10;
2656         cmd->base.duplex        = DUPLEX_FULL;
2657         cmd->base.port          = PORT_TP;
2658         cmd->base.phy_address   = 0;
2659         cmd->base.autoneg       = AUTONEG_DISABLE;
2660         return 0;
2661 }
2662
2663 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2664 {
2665         struct tun_struct *tun = netdev_priv(dev);
2666
2667         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2668         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2669
2670         switch (tun->flags & TUN_TYPE_MASK) {
2671         case IFF_TUN:
2672                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2673                 break;
2674         case IFF_TAP:
2675                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2676                 break;
2677         }
2678 }
2679
2680 static u32 tun_get_msglevel(struct net_device *dev)
2681 {
2682 #ifdef TUN_DEBUG
2683         struct tun_struct *tun = netdev_priv(dev);
2684         return tun->debug;
2685 #else
2686         return -EOPNOTSUPP;
2687 #endif
2688 }
2689
2690 static void tun_set_msglevel(struct net_device *dev, u32 value)
2691 {
2692 #ifdef TUN_DEBUG
2693         struct tun_struct *tun = netdev_priv(dev);
2694         tun->debug = value;
2695 #endif
2696 }
2697
2698 static int tun_get_coalesce(struct net_device *dev,
2699                             struct ethtool_coalesce *ec)
2700 {
2701         struct tun_struct *tun = netdev_priv(dev);
2702
2703         ec->rx_max_coalesced_frames = tun->rx_batched;
2704
2705         return 0;
2706 }
2707
2708 static int tun_set_coalesce(struct net_device *dev,
2709                             struct ethtool_coalesce *ec)
2710 {
2711         struct tun_struct *tun = netdev_priv(dev);
2712
2713         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2714                 tun->rx_batched = NAPI_POLL_WEIGHT;
2715         else
2716                 tun->rx_batched = ec->rx_max_coalesced_frames;
2717
2718         return 0;
2719 }
2720
2721 static const struct ethtool_ops tun_ethtool_ops = {
2722         .get_drvinfo    = tun_get_drvinfo,
2723         .get_msglevel   = tun_get_msglevel,
2724         .set_msglevel   = tun_set_msglevel,
2725         .get_link       = ethtool_op_get_link,
2726         .get_ts_info    = ethtool_op_get_ts_info,
2727         .get_coalesce   = tun_get_coalesce,
2728         .set_coalesce   = tun_set_coalesce,
2729         .get_link_ksettings = tun_get_link_ksettings,
2730 };
2731
2732 static int tun_queue_resize(struct tun_struct *tun)
2733 {
2734         struct net_device *dev = tun->dev;
2735         struct tun_file *tfile;
2736         struct skb_array **arrays;
2737         int n = tun->numqueues + tun->numdisabled;
2738         int ret, i;
2739
2740         arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2741         if (!arrays)
2742                 return -ENOMEM;
2743
2744         for (i = 0; i < tun->numqueues; i++) {
2745                 tfile = rtnl_dereference(tun->tfiles[i]);
2746                 arrays[i] = &tfile->tx_array;
2747         }
2748         list_for_each_entry(tfile, &tun->disabled, next)
2749                 arrays[i++] = &tfile->tx_array;
2750
2751         ret = skb_array_resize_multiple(arrays, n,
2752                                         dev->tx_queue_len, GFP_KERNEL);
2753
2754         kfree(arrays);
2755         return ret;
2756 }
2757
2758 static int tun_device_event(struct notifier_block *unused,
2759                             unsigned long event, void *ptr)
2760 {
2761         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2762         struct tun_struct *tun = netdev_priv(dev);
2763
2764         if (dev->rtnl_link_ops != &tun_link_ops)
2765                 return NOTIFY_DONE;
2766
2767         switch (event) {
2768         case NETDEV_CHANGE_TX_QUEUE_LEN:
2769                 if (tun_queue_resize(tun))
2770                         return NOTIFY_BAD;
2771                 break;
2772         default:
2773                 break;
2774         }
2775
2776         return NOTIFY_DONE;
2777 }
2778
2779 static struct notifier_block tun_notifier_block __read_mostly = {
2780         .notifier_call  = tun_device_event,
2781 };
2782
2783 static int __init tun_init(void)
2784 {
2785         int ret = 0;
2786
2787         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2788
2789         ret = rtnl_link_register(&tun_link_ops);
2790         if (ret) {
2791                 pr_err("Can't register link_ops\n");
2792                 goto err_linkops;
2793         }
2794
2795         ret = misc_register(&tun_miscdev);
2796         if (ret) {
2797                 pr_err("Can't register misc device %d\n", TUN_MINOR);
2798                 goto err_misc;
2799         }
2800
2801         ret = register_netdevice_notifier(&tun_notifier_block);
2802         if (ret) {
2803                 pr_err("Can't register netdevice notifier\n");
2804                 goto err_notifier;
2805         }
2806
2807         return  0;
2808
2809 err_notifier:
2810         misc_deregister(&tun_miscdev);
2811 err_misc:
2812         rtnl_link_unregister(&tun_link_ops);
2813 err_linkops:
2814         return ret;
2815 }
2816
2817 static void tun_cleanup(void)
2818 {
2819         misc_deregister(&tun_miscdev);
2820         rtnl_link_unregister(&tun_link_ops);
2821         unregister_netdevice_notifier(&tun_notifier_block);
2822 }
2823
2824 /* Get an underlying socket object from tun file.  Returns error unless file is
2825  * attached to a device.  The returned object works like a packet socket, it
2826  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
2827  * holding a reference to the file for as long as the socket is in use. */
2828 struct socket *tun_get_socket(struct file *file)
2829 {
2830         struct tun_file *tfile;
2831         if (file->f_op != &tun_fops)
2832                 return ERR_PTR(-EINVAL);
2833         tfile = file->private_data;
2834         if (!tfile)
2835                 return ERR_PTR(-EBADFD);
2836         return &tfile->socket;
2837 }
2838 EXPORT_SYMBOL_GPL(tun_get_socket);
2839
2840 struct skb_array *tun_get_skb_array(struct file *file)
2841 {
2842         struct tun_file *tfile;
2843
2844         if (file->f_op != &tun_fops)
2845                 return ERR_PTR(-EINVAL);
2846         tfile = file->private_data;
2847         if (!tfile)
2848                 return ERR_PTR(-EBADFD);
2849         return &tfile->tx_array;
2850 }
2851 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2852
2853 module_init(tun_init);
2854 module_exit(tun_cleanup);
2855 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2856 MODULE_AUTHOR(DRV_COPYRIGHT);
2857 MODULE_LICENSE("GPL");
2858 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2859 MODULE_ALIAS("devname:net/tun");