drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 117
 118 /* TUN device flags */
 119
 120 /* IFF_ATTACH_QUEUE is never stored in device flags,
 121  * overload it to mean fasync when stored there.
 122  */
 123 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 124 /* High bits in flags field are unused. */
 125 #define TUN_VNET_LE     0x80000000
 126 #define TUN_VNET_BE     0x40000000
 127
 128 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 129                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 130
 131 #define GOODCOPY_LEN 128
 132
 133 #define FLT_EXACT_COUNT 8
 134 struct tap_filter {
 135         unsigned int    count;    /* Number of addrs. Zero means disabled */
 136         u32             mask[2];  /* Mask of the hashed addrs */
 137         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 138 };
 139
 140 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 141  * to max number of VCPUs in guest. */
 142 #define MAX_TAP_QUEUES 256
 143 #define MAX_TAP_FLOWS  4096
 144
 145 #define TUN_FLOW_EXPIRE (3 * HZ)
 146
 147 struct tun_pcpu_stats {
 148         u64 rx_packets;
 149         u64 rx_bytes;
 150         u64 tx_packets;
 151         u64 tx_bytes;
 152         struct u64_stats_sync syncp;
 153         u32 rx_dropped;
 154         u32 tx_dropped;
 155         u32 rx_frame_errors;
 156 };
 157
 158 /* A tun_file connects an open character device to a tuntap netdevice. It
 159  * also contains all socket related structures (except sock_fprog and tap_filter)
 160  * to serve as one transmit queue for tuntap device. The sock_fprog and
 161  * tap_filter were kept in tun_struct since they were used for filtering for the
 162  * netdevice not for a specific queue (at least I didn't see the requirement for
 163  * this).
 164  *
 165  * RCU usage:
 166  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 167  * other can only be read while rcu_read_lock or rtnl_lock is held.
 168  */
 169 struct tun_file {
 170         struct sock sk;
 171         struct socket socket;
 172         struct socket_wq wq;
 173         struct tun_struct __rcu *tun;
 174         struct fasync_struct *fasync;
 175         /* only used for fasnyc */
 176         unsigned int flags;
 177         union {
 178                 u16 queue_index;
 179                 unsigned int ifindex;
 180         };
 181         struct napi_struct napi;
 182         bool napi_enabled;
 183         bool napi_frags_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_flow_entry {
 192         struct hlist_node hash_link;
 193         struct rcu_head rcu;
 194         struct tun_struct *tun;
 195
 196         u32 rxhash;
 197         u32 rps_rxhash;
 198         int queue_index;
 199         unsigned long updated;
 200 };
 201
 202 #define TUN_NUM_FLOW_ENTRIES 1024
 203 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 204
 205 struct tun_prog {
 206         struct rcu_head rcu;
 207         struct bpf_prog *prog;
 208 };
 209
 210 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 211  * device, socket filter, sndbuf and vnet header size were restore when the
 212  * file were attached to a persist device.
 213  */
 214 struct tun_struct {
 215         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 216         unsigned int            numqueues;
 217         unsigned int            flags;
 218         kuid_t                  owner;
 219         kgid_t                  group;
 220
 221         struct net_device       *dev;
 222         netdev_features_t       set_features;
 223 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 224                           NETIF_F_TSO6)
 225
 226         int                     align;
 227         int                     vnet_hdr_sz;
 228         int                     sndbuf;
 229         struct tap_filter       txflt;
 230         struct sock_fprog       fprog;
 231         /* protected by rtnl lock */
 232         bool                    filter_attached;
 233 #ifdef TUN_DEBUG
 234         int debug;
 235 #endif
 236         spinlock_t lock;
 237         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 238         struct timer_list flow_gc_timer;
 239         unsigned long ageing_time;
 240         unsigned int numdisabled;
 241         struct list_head disabled;
 242         void *security;
 243         u32 flow_count;
 244         u32 rx_batched;
 245         struct tun_pcpu_stats __percpu *pcpu_stats;
 246         struct bpf_prog __rcu *xdp_prog;
 247         struct tun_prog __rcu *steering_prog;
 248         struct tun_prog __rcu *filter_prog;
 249         struct ethtool_link_ksettings link_ksettings;
 250 };
 251
 252 struct veth {
 253         __be16 h_vlan_proto;
 254         __be16 h_vlan_TCI;
 255 };
 256
 257 bool tun_is_xdp_frame(void *ptr)
 258 {
 259         return (unsigned long)ptr & TUN_XDP_FLAG;
 260 }
 261 EXPORT_SYMBOL(tun_is_xdp_frame);
 262
 263 void *tun_xdp_to_ptr(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_xdp_to_ptr);
 268
 269 void *tun_ptr_to_xdp(void *ptr)
 270 {
 271         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 272 }
 273 EXPORT_SYMBOL(tun_ptr_to_xdp);
 274
 275 static int tun_napi_receive(struct napi_struct *napi, int budget)
 276 {
 277         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 278         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 279         struct sk_buff_head process_queue;
 280         struct sk_buff *skb;
 281         int received = 0;
 282
 283         __skb_queue_head_init(&process_queue);
 284
 285         spin_lock(&queue->lock);
 286         skb_queue_splice_tail_init(queue, &process_queue);
 287         spin_unlock(&queue->lock);
 288
 289         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 290                 napi_gro_receive(napi, skb);
 291                 ++received;
 292         }
 293
 294         if (!skb_queue_empty(&process_queue)) {
 295                 spin_lock(&queue->lock);
 296                 skb_queue_splice(&process_queue, queue);
 297                 spin_unlock(&queue->lock);
 298         }
 299
 300         return received;
 301 }
 302
 303 static int tun_napi_poll(struct napi_struct *napi, int budget)
 304 {
 305         unsigned int received;
 306
 307         received = tun_napi_receive(napi, budget);
 308
 309         if (received < budget)
 310                 napi_complete_done(napi, received);
 311
 312         return received;
 313 }
 314
 315 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 316                           bool napi_en, bool napi_frags)
 317 {
 318         tfile->napi_enabled = napi_en;
 319         tfile->napi_frags_enabled = napi_en && napi_frags;
 320         if (napi_en) {
 321                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 322                                NAPI_POLL_WEIGHT);
 323                 napi_enable(&tfile->napi);
 324         }
 325 }
 326
 327 static void tun_napi_disable(struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 napi_disable(&tfile->napi);
 331 }
 332
 333 static void tun_napi_del(struct tun_file *tfile)
 334 {
 335         if (tfile->napi_enabled)
 336                 netif_napi_del(&tfile->napi);
 337 }
 338
 339 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 340 {
 341         return tfile->napi_frags_enabled;
 342 }
 343
 344 #ifdef CONFIG_TUN_VNET_CROSS_LE
 345 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 346 {
 347         return tun->flags & TUN_VNET_BE ? false :
 348                 virtio_legacy_is_little_endian();
 349 }
 350
 351 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 352 {
 353         int be = !!(tun->flags & TUN_VNET_BE);
 354
 355         if (put_user(be, argp))
 356                 return -EFAULT;
 357
 358         return 0;
 359 }
 360
 361 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 362 {
 363         int be;
 364
 365         if (get_user(be, argp))
 366                 return -EFAULT;
 367
 368         if (be)
 369                 tun->flags |= TUN_VNET_BE;
 370         else
 371                 tun->flags &= ~TUN_VNET_BE;
 372
 373         return 0;
 374 }
 375 #else
 376 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 377 {
 378         return virtio_legacy_is_little_endian();
 379 }
 380
 381 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 382 {
 383         return -EINVAL;
 384 }
 385
 386 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 387 {
 388         return -EINVAL;
 389 }
 390 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 391
 392 static inline bool tun_is_little_endian(struct tun_struct *tun)
 393 {
 394         return tun->flags & TUN_VNET_LE ||
 395                 tun_legacy_is_little_endian(tun);
 396 }
 397
 398 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 399 {
 400         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 401 }
 402
 403 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 404 {
 405         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 406 }
 407
 408 static inline u32 tun_hashfn(u32 rxhash)
 409 {
 410         return rxhash & TUN_MASK_FLOW_ENTRIES;
 411 }
 412
 413 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 414 {
 415         struct tun_flow_entry *e;
 416
 417         hlist_for_each_entry_rcu(e, head, hash_link) {
 418                 if (e->rxhash == rxhash)
 419                         return e;
 420         }
 421         return NULL;
 422 }
 423
 424 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 425                                               struct hlist_head *head,
 426                                               u32 rxhash, u16 queue_index)
 427 {
 428         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 429
 430         if (e) {
 431                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 432                           rxhash, queue_index);
 433                 e->updated = jiffies;
 434                 e->rxhash = rxhash;
 435                 e->rps_rxhash = 0;
 436                 e->queue_index = queue_index;
 437                 e->tun = tun;
 438                 hlist_add_head_rcu(&e->hash_link, head);
 439                 ++tun->flow_count;
 440         }
 441         return e;
 442 }
 443
 444 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 445 {
 446         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 447                   e->rxhash, e->queue_index);
 448         hlist_del_rcu(&e->hash_link);
 449         kfree_rcu(e, rcu);
 450         --tun->flow_count;
 451 }
 452
 453 static void tun_flow_flush(struct tun_struct *tun)
 454 {
 455         int i;
 456
 457         spin_lock_bh(&tun->lock);
 458         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 459                 struct tun_flow_entry *e;
 460                 struct hlist_node *n;
 461
 462                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 463                         tun_flow_delete(tun, e);
 464         }
 465         spin_unlock_bh(&tun->lock);
 466 }
 467
 468 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 469 {
 470         int i;
 471
 472         spin_lock_bh(&tun->lock);
 473         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 474                 struct tun_flow_entry *e;
 475                 struct hlist_node *n;
 476
 477                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 478                         if (e->queue_index == queue_index)
 479                                 tun_flow_delete(tun, e);
 480                 }
 481         }
 482         spin_unlock_bh(&tun->lock);
 483 }
 484
 485 static void tun_flow_cleanup(struct timer_list *t)
 486 {
 487         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 488         unsigned long delay = tun->ageing_time;
 489         unsigned long next_timer = jiffies + delay;
 490         unsigned long count = 0;
 491         int i;
 492
 493         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 494
 495         spin_lock(&tun->lock);
 496         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 497                 struct tun_flow_entry *e;
 498                 struct hlist_node *n;
 499
 500                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 501                         unsigned long this_timer;
 502
 503                         this_timer = e->updated + delay;
 504                         if (time_before_eq(this_timer, jiffies)) {
 505                                 tun_flow_delete(tun, e);
 506                                 continue;
 507                         }
 508                         count++;
 509                         if (time_before(this_timer, next_timer))
 510                                 next_timer = this_timer;
 511                 }
 512         }
 513
 514         if (count)
 515                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 516         spin_unlock(&tun->lock);
 517 }
 518
 519 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 520                             struct tun_file *tfile)
 521 {
 522         struct hlist_head *head;
 523         struct tun_flow_entry *e;
 524         unsigned long delay = tun->ageing_time;
 525         u16 queue_index = tfile->queue_index;
 526
 527         if (!rxhash)
 528                 return;
 529         else
 530                 head = &tun->flows[tun_hashfn(rxhash)];
 531
 532         rcu_read_lock();
 533
 534         e = tun_flow_find(head, rxhash);
 535         if (likely(e)) {
 536                 /* TODO: keep queueing to old queue until it's empty? */
 537                 e->queue_index = queue_index;
 538                 e->updated = jiffies;
 539                 sock_rps_record_flow_hash(e->rps_rxhash);
 540         } else {
 541                 spin_lock_bh(&tun->lock);
 542                 if (!tun_flow_find(head, rxhash) &&
 543                     tun->flow_count < MAX_TAP_FLOWS)
 544                         tun_flow_create(tun, head, rxhash, queue_index);
 545
 546                 if (!timer_pending(&tun->flow_gc_timer))
 547                         mod_timer(&tun->flow_gc_timer,
 548                                   round_jiffies_up(jiffies + delay));
 549                 spin_unlock_bh(&tun->lock);
 550         }
 551
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here.
 570  */
 571 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 572 {
 573         struct tun_flow_entry *e;
 574         u32 txq = 0;
 575         u32 numqueues = 0;
 576
 577         numqueues = READ_ONCE(tun->numqueues);
 578
 579         txq = __skb_get_hash_symmetric(skb);
 580         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 581         if (e) {
 582                 tun_flow_save_rps_rxhash(e, txq);
 583                 txq = e->queue_index;
 584         } else {
 585                 /* use multiply and shift instead of expensive divide */
 586                 txq = ((u64)txq * numqueues) >> 32;
 587         }
 588
 589         return txq;
 590 }
 591
 592 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 593 {
 594         struct tun_prog *prog;
 595         u16 ret = 0;
 596
 597         prog = rcu_dereference(tun->steering_prog);
 598         if (prog)
 599                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 600
 601         return ret % tun->numqueues;
 602 }
 603
 604 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 605                             struct net_device *sb_dev,
 606                             select_queue_fallback_t fallback)
 607 {
 608         struct tun_struct *tun = netdev_priv(dev);
 609         u16 ret;
 610
 611         rcu_read_lock();
 612         if (rcu_dereference(tun->steering_prog))
 613                 ret = tun_ebpf_select_queue(tun, skb);
 614         else
 615                 ret = tun_automq_select_queue(tun, skb);
 616         rcu_read_unlock();
 617
 618         return ret;
 619 }
 620
 621 static inline bool tun_not_capable(struct tun_struct *tun)
 622 {
 623         const struct cred *cred = current_cred();
 624         struct net *net = dev_net(tun->dev);
 625
 626         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 627                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 628                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 629 }
 630
 631 static void tun_set_real_num_queues(struct tun_struct *tun)
 632 {
 633         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 634         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 635 }
 636
 637 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 638 {
 639         tfile->detached = tun;
 640         list_add_tail(&tfile->next, &tun->disabled);
 641         ++tun->numdisabled;
 642 }
 643
 644 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 645 {
 646         struct tun_struct *tun = tfile->detached;
 647
 648         tfile->detached = NULL;
 649         list_del_init(&tfile->next);
 650         --tun->numdisabled;
 651         return tun;
 652 }
 653
 654 void tun_ptr_free(void *ptr)
 655 {
 656         if (!ptr)
 657                 return;
 658         if (tun_is_xdp_frame(ptr)) {
 659                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 660
 661                 xdp_return_frame(xdpf);
 662         } else {
 663                 __skb_array_destroy_skb(ptr);
 664         }
 665 }
 666 EXPORT_SYMBOL_GPL(tun_ptr_free);
 667
 668 static void tun_queue_purge(struct tun_file *tfile)
 669 {
 670         void *ptr;
 671
 672         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 673                 tun_ptr_free(ptr);
 674
 675         skb_queue_purge(&tfile->sk.sk_write_queue);
 676         skb_queue_purge(&tfile->sk.sk_error_queue);
 677 }
 678
 679 static void __tun_detach(struct tun_file *tfile, bool clean)
 680 {
 681         struct tun_file *ntfile;
 682         struct tun_struct *tun;
 683
 684         tun = rtnl_dereference(tfile->tun);
 685
 686         if (tun && clean) {
 687                 tun_napi_disable(tfile);
 688                 tun_napi_del(tfile);
 689         }
 690
 691         if (tun && !tfile->detached) {
 692                 u16 index = tfile->queue_index;
 693                 BUG_ON(index >= tun->numqueues);
 694
 695                 rcu_assign_pointer(tun->tfiles[index],
 696                                    tun->tfiles[tun->numqueues - 1]);
 697                 ntfile = rtnl_dereference(tun->tfiles[index]);
 698                 ntfile->queue_index = index;
 699
 700                 --tun->numqueues;
 701                 if (clean) {
 702                         RCU_INIT_POINTER(tfile->tun, NULL);
 703                         sock_put(&tfile->sk);
 704                 } else
 705                         tun_disable_queue(tun, tfile);
 706
 707                 synchronize_net();
 708                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 709                 /* Drop read queue */
 710                 tun_queue_purge(tfile);
 711                 tun_set_real_num_queues(tun);
 712         } else if (tfile->detached && clean) {
 713                 tun = tun_enable_queue(tfile);
 714                 sock_put(&tfile->sk);
 715         }
 716
 717         if (clean) {
 718                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 719                         netif_carrier_off(tun->dev);
 720
 721                         if (!(tun->flags & IFF_PERSIST) &&
 722                             tun->dev->reg_state == NETREG_REGISTERED)
 723                                 unregister_netdevice(tun->dev);
 724                 }
 725                 if (tun)
 726                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 727                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 728                 sock_put(&tfile->sk);
 729         }
 730 }
 731
 732 static void tun_detach(struct tun_file *tfile, bool clean)
 733 {
 734         struct tun_struct *tun;
 735         struct net_device *dev;
 736
 737         rtnl_lock();
 738         tun = rtnl_dereference(tfile->tun);
 739         dev = tun ? tun->dev : NULL;
 740         __tun_detach(tfile, clean);
 741         if (dev)
 742                 netdev_state_change(dev);
 743         rtnl_unlock();
 744 }
 745
 746 static void tun_detach_all(struct net_device *dev)
 747 {
 748         struct tun_struct *tun = netdev_priv(dev);
 749         struct tun_file *tfile, *tmp;
 750         int i, n = tun->numqueues;
 751
 752         for (i = 0; i < n; i++) {
 753                 tfile = rtnl_dereference(tun->tfiles[i]);
 754                 BUG_ON(!tfile);
 755                 tun_napi_disable(tfile);
 756                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 757                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 758                 RCU_INIT_POINTER(tfile->tun, NULL);
 759                 --tun->numqueues;
 760         }
 761         list_for_each_entry(tfile, &tun->disabled, next) {
 762                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 763                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 764                 RCU_INIT_POINTER(tfile->tun, NULL);
 765         }
 766         BUG_ON(tun->numqueues != 0);
 767
 768         synchronize_net();
 769         for (i = 0; i < n; i++) {
 770                 tfile = rtnl_dereference(tun->tfiles[i]);
 771                 tun_napi_del(tfile);
 772                 /* Drop read queue */
 773                 tun_queue_purge(tfile);
 774                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 775                 sock_put(&tfile->sk);
 776         }
 777         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 778                 tun_enable_queue(tfile);
 779                 tun_queue_purge(tfile);
 780                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 781                 sock_put(&tfile->sk);
 782         }
 783         BUG_ON(tun->numdisabled != 0);
 784
 785         if (tun->flags & IFF_PERSIST)
 786                 module_put(THIS_MODULE);
 787 }
 788
 789 static int tun_attach(struct tun_struct *tun, struct file *file,
 790                       bool skip_filter, bool napi, bool napi_frags)
 791 {
 792         struct tun_file *tfile = file->private_data;
 793         struct net_device *dev = tun->dev;
 794         int err;
 795
 796         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 797         if (err < 0)
 798                 goto out;
 799
 800         err = -EINVAL;
 801         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 802                 goto out;
 803
 804         err = -EBUSY;
 805         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 806                 goto out;
 807
 808         err = -E2BIG;
 809         if (!tfile->detached &&
 810             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 811                 goto out;
 812
 813         err = 0;
 814
 815         /* Re-attach the filter to persist device */
 816         if (!skip_filter && (tun->filter_attached == true)) {
 817                 lock_sock(tfile->socket.sk);
 818                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 819                 release_sock(tfile->socket.sk);
 820                 if (!err)
 821                         goto out;
 822         }
 823
 824         if (!tfile->detached &&
 825             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 826                             GFP_KERNEL, tun_ptr_free)) {
 827                 err = -ENOMEM;
 828                 goto out;
 829         }
 830
 831         tfile->queue_index = tun->numqueues;
 832         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 833
 834         if (tfile->detached) {
 835                 /* Re-attach detached tfile, updating XDP queue_index */
 836                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 837
 838                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 839                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 840         } else {
 841                 /* Setup XDP RX-queue info, for new tfile getting attached */
 842                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 843                                        tun->dev, tfile->queue_index);
 844                 if (err < 0)
 845                         goto out;
 846                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 847                                                  MEM_TYPE_PAGE_SHARED, NULL);
 848                 if (err < 0) {
 849                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 850                         goto out;
 851                 }
 852                 err = 0;
 853         }
 854
 855         rcu_assign_pointer(tfile->tun, tun);
 856         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 857         tun->numqueues++;
 858
 859         if (tfile->detached) {
 860                 tun_enable_queue(tfile);
 861         } else {
 862                 sock_hold(&tfile->sk);
 863                 tun_napi_init(tun, tfile, napi, napi_frags);
 864         }
 865
 866         if (rtnl_dereference(tun->xdp_prog))
 867                 sock_set_flag(&tfile->sk, SOCK_XDP);
 868
 869         tun_set_real_num_queues(tun);
 870
 871         /* device is allowed to go away first, so no need to hold extra
 872          * refcnt.
 873          */
 874
 875 out:
 876         return err;
 877 }
 878
 879 static struct tun_struct *tun_get(struct tun_file *tfile)
 880 {
 881         struct tun_struct *tun;
 882
 883         rcu_read_lock();
 884         tun = rcu_dereference(tfile->tun);
 885         if (tun)
 886                 dev_hold(tun->dev);
 887         rcu_read_unlock();
 888
 889         return tun;
 890 }
 891
 892 static void tun_put(struct tun_struct *tun)
 893 {
 894         dev_put(tun->dev);
 895 }
 896
 897 /* TAP filtering */
 898 static void addr_hash_set(u32 *mask, const u8 *addr)
 899 {
 900         int n = ether_crc(ETH_ALEN, addr) >> 26;
 901         mask[n >> 5] |= (1 << (n & 31));
 902 }
 903
 904 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 905 {
 906         int n = ether_crc(ETH_ALEN, addr) >> 26;
 907         return mask[n >> 5] & (1 << (n & 31));
 908 }
 909
 910 static int update_filter(struct tap_filter *filter, void __user *arg)
 911 {
 912         struct { u8 u[ETH_ALEN]; } *addr;
 913         struct tun_filter uf;
 914         int err, alen, n, nexact;
 915
 916         if (copy_from_user(&uf, arg, sizeof(uf)))
 917                 return -EFAULT;
 918
 919         if (!uf.count) {
 920                 /* Disabled */
 921                 filter->count = 0;
 922                 return 0;
 923         }
 924
 925         alen = ETH_ALEN * uf.count;
 926         addr = memdup_user(arg + sizeof(uf), alen);
 927         if (IS_ERR(addr))
 928                 return PTR_ERR(addr);
 929
 930         /* The filter is updated without holding any locks. Which is
 931          * perfectly safe. We disable it first and in the worst
 932          * case we'll accept a few undesired packets. */
 933         filter->count = 0;
 934         wmb();
 935
 936         /* Use first set of addresses as an exact filter */
 937         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 938                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 939
 940         nexact = n;
 941
 942         /* Remaining multicast addresses are hashed,
 943          * unicast will leave the filter disabled. */
 944         memset(filter->mask, 0, sizeof(filter->mask));
 945         for (; n < uf.count; n++) {
 946                 if (!is_multicast_ether_addr(addr[n].u)) {
 947                         err = 0; /* no filter */
 948                         goto free_addr;
 949                 }
 950                 addr_hash_set(filter->mask, addr[n].u);
 951         }
 952
 953         /* For ALLMULTI just set the mask to all ones.
 954          * This overrides the mask populated above. */
 955         if ((uf.flags & TUN_FLT_ALLMULTI))
 956                 memset(filter->mask, ~0, sizeof(filter->mask));
 957
 958         /* Now enable the filter */
 959         wmb();
 960         filter->count = nexact;
 961
 962         /* Return the number of exact filters */
 963         err = nexact;
 964 free_addr:
 965         kfree(addr);
 966         return err;
 967 }
 968
 969 /* Returns: 0 - drop, !=0 - accept */
 970 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 971 {
 972         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 973          * at this point. */
 974         struct ethhdr *eh = (struct ethhdr *) skb->data;
 975         int i;
 976
 977         /* Exact match */
 978         for (i = 0; i < filter->count; i++)
 979                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 980                         return 1;
 981
 982         /* Inexact match (multicast only) */
 983         if (is_multicast_ether_addr(eh->h_dest))
 984                 return addr_hash_test(filter->mask, eh->h_dest);
 985
 986         return 0;
 987 }
 988
 989 /*
 990  * Checks whether the packet is accepted or not.
 991  * Returns: 0 - drop, !=0 - accept
 992  */
 993 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 994 {
 995         if (!filter->count)
 996                 return 1;
 997
 998         return run_filter(filter, skb);
 999 }
1000
1001 /* Network device part of the driver */
1002
1003 static const struct ethtool_ops tun_ethtool_ops;
1004
1005 /* Net device detach from fd. */
1006 static void tun_net_uninit(struct net_device *dev)
1007 {
1008         tun_detach_all(dev);
1009 }
1010
1011 /* Net device open. */
1012 static int tun_net_open(struct net_device *dev)
1013 {
1014         struct tun_struct *tun = netdev_priv(dev);
1015         int i;
1016
1017         netif_tx_start_all_queues(dev);
1018
1019         for (i = 0; i < tun->numqueues; i++) {
1020                 struct tun_file *tfile;
1021
1022                 tfile = rtnl_dereference(tun->tfiles[i]);
1023                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1024         }
1025
1026         return 0;
1027 }
1028
1029 /* Net device close. */
1030 static int tun_net_close(struct net_device *dev)
1031 {
1032         netif_tx_stop_all_queues(dev);
1033         return 0;
1034 }
1035
1036 /* Net device start xmit */
1037 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1038 {
1039 #ifdef CONFIG_RPS
1040         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1041                 /* Select queue was not called for the skbuff, so we extract the
1042                  * RPS hash and save it into the flow_table here.
1043                  */
1044                 struct tun_flow_entry *e;
1045                 __u32 rxhash;
1046
1047                 rxhash = __skb_get_hash_symmetric(skb);
1048                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1049                 if (e)
1050                         tun_flow_save_rps_rxhash(e, rxhash);
1051         }
1052 #endif
1053 }
1054
1055 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1056                                     struct sk_buff *skb,
1057                                     int len)
1058 {
1059         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1060
1061         if (prog)
1062                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1063
1064         return len;
1065 }
1066
1067 /* Net device start xmit */
1068 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1069 {
1070         struct tun_struct *tun = netdev_priv(dev);
1071         int txq = skb->queue_mapping;
1072         struct tun_file *tfile;
1073         int len = skb->len;
1074
1075         rcu_read_lock();
1076         tfile = rcu_dereference(tun->tfiles[txq]);
1077
1078         /* Drop packet if interface is not attached */
1079         if (txq >= tun->numqueues)
1080                 goto drop;
1081
1082         if (!rcu_dereference(tun->steering_prog))
1083                 tun_automq_xmit(tun, skb);
1084
1085         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1086
1087         BUG_ON(!tfile);
1088
1089         /* Drop if the filter does not like it.
1090          * This is a noop if the filter is disabled.
1091          * Filter can be enabled only for the TAP devices. */
1092         if (!check_filter(&tun->txflt, skb))
1093                 goto drop;
1094
1095         if (tfile->socket.sk->sk_filter &&
1096             sk_filter(tfile->socket.sk, skb))
1097                 goto drop;
1098
1099         len = run_ebpf_filter(tun, skb, len);
1100         if (len == 0 || pskb_trim(skb, len))
1101                 goto drop;
1102
1103         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1104                 goto drop;
1105
1106         skb_tx_timestamp(skb);
1107
1108         /* Orphan the skb - required as we might hang on to it
1109          * for indefinite time.
1110          */
1111         skb_orphan(skb);
1112
1113         nf_reset(skb);
1114
1115         if (ptr_ring_produce(&tfile->tx_ring, skb))
1116                 goto drop;
1117
1118         /* Notify and wake up reader process */
1119         if (tfile->flags & TUN_FASYNC)
1120                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1121         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1122
1123         rcu_read_unlock();
1124         return NETDEV_TX_OK;
1125
1126 drop:
1127         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1128         skb_tx_error(skb);
1129         kfree_skb(skb);
1130         rcu_read_unlock();
1131         return NET_XMIT_DROP;
1132 }
1133
1134 static void tun_net_mclist(struct net_device *dev)
1135 {
1136         /*
1137          * This callback is supposed to deal with mc filter in
1138          * _rx_ path and has nothing to do with the _tx_ path.
1139          * In rx path we always accept everything userspace gives us.
1140          */
1141 }
1142
1143 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1144         netdev_features_t features)
1145 {
1146         struct tun_struct *tun = netdev_priv(dev);
1147
1148         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1149 }
1150
1151 static void tun_set_headroom(struct net_device *dev, int new_hr)
1152 {
1153         struct tun_struct *tun = netdev_priv(dev);
1154
1155         if (new_hr < NET_SKB_PAD)
1156                 new_hr = NET_SKB_PAD;
1157
1158         tun->align = new_hr;
1159 }
1160
1161 static void
1162 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1163 {
1164         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1165         struct tun_struct *tun = netdev_priv(dev);
1166         struct tun_pcpu_stats *p;
1167         int i;
1168
1169         for_each_possible_cpu(i) {
1170                 u64 rxpackets, rxbytes, txpackets, txbytes;
1171                 unsigned int start;
1172
1173                 p = per_cpu_ptr(tun->pcpu_stats, i);
1174                 do {
1175                         start = u64_stats_fetch_begin(&p->syncp);
1176                         rxpackets       = p->rx_packets;
1177                         rxbytes         = p->rx_bytes;
1178                         txpackets       = p->tx_packets;
1179                         txbytes         = p->tx_bytes;
1180                 } while (u64_stats_fetch_retry(&p->syncp, start));
1181
1182                 stats->rx_packets       += rxpackets;
1183                 stats->rx_bytes         += rxbytes;
1184                 stats->tx_packets       += txpackets;
1185                 stats->tx_bytes         += txbytes;
1186
1187                 /* u32 counters */
1188                 rx_dropped      += p->rx_dropped;
1189                 rx_frame_errors += p->rx_frame_errors;
1190                 tx_dropped      += p->tx_dropped;
1191         }
1192         stats->rx_dropped  = rx_dropped;
1193         stats->rx_frame_errors = rx_frame_errors;
1194         stats->tx_dropped = tx_dropped;
1195 }
1196
1197 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1198                        struct netlink_ext_ack *extack)
1199 {
1200         struct tun_struct *tun = netdev_priv(dev);
1201         struct tun_file *tfile;
1202         struct bpf_prog *old_prog;
1203         int i;
1204
1205         old_prog = rtnl_dereference(tun->xdp_prog);
1206         rcu_assign_pointer(tun->xdp_prog, prog);
1207         if (old_prog)
1208                 bpf_prog_put(old_prog);
1209
1210         for (i = 0; i < tun->numqueues; i++) {
1211                 tfile = rtnl_dereference(tun->tfiles[i]);
1212                 if (prog)
1213                         sock_set_flag(&tfile->sk, SOCK_XDP);
1214                 else
1215                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1216         }
1217         list_for_each_entry(tfile, &tun->disabled, next) {
1218                 if (prog)
1219                         sock_set_flag(&tfile->sk, SOCK_XDP);
1220                 else
1221                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1222         }
1223
1224         return 0;
1225 }
1226
1227 static u32 tun_xdp_query(struct net_device *dev)
1228 {
1229         struct tun_struct *tun = netdev_priv(dev);
1230         const struct bpf_prog *xdp_prog;
1231
1232         xdp_prog = rtnl_dereference(tun->xdp_prog);
1233         if (xdp_prog)
1234                 return xdp_prog->aux->id;
1235
1236         return 0;
1237 }
1238
1239 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1240 {
1241         switch (xdp->command) {
1242         case XDP_SETUP_PROG:
1243                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1244         case XDP_QUERY_PROG:
1245                 xdp->prog_id = tun_xdp_query(dev);
1246                 return 0;
1247         default:
1248                 return -EINVAL;
1249         }
1250 }
1251
1252 static const struct net_device_ops tun_netdev_ops = {
1253         .ndo_uninit             = tun_net_uninit,
1254         .ndo_open               = tun_net_open,
1255         .ndo_stop               = tun_net_close,
1256         .ndo_start_xmit         = tun_net_xmit,
1257         .ndo_fix_features       = tun_net_fix_features,
1258         .ndo_select_queue       = tun_select_queue,
1259         .ndo_set_rx_headroom    = tun_set_headroom,
1260         .ndo_get_stats64        = tun_net_get_stats64,
1261 };
1262
1263 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1264 {
1265         /* Notify and wake up reader process */
1266         if (tfile->flags & TUN_FASYNC)
1267                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1268         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1269 }
1270
1271 static int tun_xdp_xmit(struct net_device *dev, int n,
1272                         struct xdp_frame **frames, u32 flags)
1273 {
1274         struct tun_struct *tun = netdev_priv(dev);
1275         struct tun_file *tfile;
1276         u32 numqueues;
1277         int drops = 0;
1278         int cnt = n;
1279         int i;
1280
1281         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1282                 return -EINVAL;
1283
1284         rcu_read_lock();
1285
1286         numqueues = READ_ONCE(tun->numqueues);
1287         if (!numqueues) {
1288                 rcu_read_unlock();
1289                 return -ENXIO; /* Caller will free/return all frames */
1290         }
1291
1292         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1293                                             numqueues]);
1294
1295         spin_lock(&tfile->tx_ring.producer_lock);
1296         for (i = 0; i < n; i++) {
1297                 struct xdp_frame *xdp = frames[i];
1298                 /* Encode the XDP flag into lowest bit for consumer to differ
1299                  * XDP buffer from sk_buff.
1300                  */
1301                 void *frame = tun_xdp_to_ptr(xdp);
1302
1303                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1304                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1305                         xdp_return_frame_rx_napi(xdp);
1306                         drops++;
1307                 }
1308         }
1309         spin_unlock(&tfile->tx_ring.producer_lock);
1310
1311         if (flags & XDP_XMIT_FLUSH)
1312                 __tun_xdp_flush_tfile(tfile);
1313
1314         rcu_read_unlock();
1315         return cnt - drops;
1316 }
1317
1318 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1319 {
1320         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1321
1322         if (unlikely(!frame))
1323                 return -EOVERFLOW;
1324
1325         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1326 }
1327
1328 static const struct net_device_ops tap_netdev_ops = {
1329         .ndo_uninit             = tun_net_uninit,
1330         .ndo_open               = tun_net_open,
1331         .ndo_stop               = tun_net_close,
1332         .ndo_start_xmit         = tun_net_xmit,
1333         .ndo_fix_features       = tun_net_fix_features,
1334         .ndo_set_rx_mode        = tun_net_mclist,
1335         .ndo_set_mac_address    = eth_mac_addr,
1336         .ndo_validate_addr      = eth_validate_addr,
1337         .ndo_select_queue       = tun_select_queue,
1338         .ndo_features_check     = passthru_features_check,
1339         .ndo_set_rx_headroom    = tun_set_headroom,
1340         .ndo_get_stats64        = tun_net_get_stats64,
1341         .ndo_bpf                = tun_xdp,
1342         .ndo_xdp_xmit           = tun_xdp_xmit,
1343 };
1344
1345 static void tun_flow_init(struct tun_struct *tun)
1346 {
1347         int i;
1348
1349         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1350                 INIT_HLIST_HEAD(&tun->flows[i]);
1351
1352         tun->ageing_time = TUN_FLOW_EXPIRE;
1353         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1354         mod_timer(&tun->flow_gc_timer,
1355                   round_jiffies_up(jiffies + tun->ageing_time));
1356 }
1357
1358 static void tun_flow_uninit(struct tun_struct *tun)
1359 {
1360         del_timer_sync(&tun->flow_gc_timer);
1361         tun_flow_flush(tun);
1362 }
1363
1364 #define MIN_MTU 68
1365 #define MAX_MTU 65535
1366
1367 /* Initialize net device. */
1368 static void tun_net_init(struct net_device *dev)
1369 {
1370         struct tun_struct *tun = netdev_priv(dev);
1371
1372         switch (tun->flags & TUN_TYPE_MASK) {
1373         case IFF_TUN:
1374                 dev->netdev_ops = &tun_netdev_ops;
1375
1376                 /* Point-to-Point TUN Device */
1377                 dev->hard_header_len = 0;
1378                 dev->addr_len = 0;
1379                 dev->mtu = 1500;
1380
1381                 /* Zero header length */
1382                 dev->type = ARPHRD_NONE;
1383                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1384                 break;
1385
1386         case IFF_TAP:
1387                 dev->netdev_ops = &tap_netdev_ops;
1388                 /* Ethernet TAP Device */
1389                 ether_setup(dev);
1390                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1391                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1392
1393                 eth_hw_addr_random(dev);
1394
1395                 break;
1396         }
1397
1398         dev->min_mtu = MIN_MTU;
1399         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1400 }
1401
1402 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1403 {
1404         struct sock *sk = tfile->socket.sk;
1405
1406         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1407 }
1408
1409 /* Character device part */
1410
1411 /* Poll */
1412 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1413 {
1414         struct tun_file *tfile = file->private_data;
1415         struct tun_struct *tun = tun_get(tfile);
1416         struct sock *sk;
1417         __poll_t mask = 0;
1418
1419         if (!tun)
1420                 return EPOLLERR;
1421
1422         sk = tfile->socket.sk;
1423
1424         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1425
1426         poll_wait(file, sk_sleep(sk), wait);
1427
1428         if (!ptr_ring_empty(&tfile->tx_ring))
1429                 mask |= EPOLLIN | EPOLLRDNORM;
1430
1431         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1432          * guarantee EPOLLOUT to be raised by either here or
1433          * tun_sock_write_space(). Then process could get notification
1434          * after it writes to a down device and meets -EIO.
1435          */
1436         if (tun_sock_writeable(tun, tfile) ||
1437             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1438              tun_sock_writeable(tun, tfile)))
1439                 mask |= EPOLLOUT | EPOLLWRNORM;
1440
1441         if (tun->dev->reg_state != NETREG_REGISTERED)
1442                 mask = EPOLLERR;
1443
1444         tun_put(tun);
1445         return mask;
1446 }
1447
1448 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1449                                             size_t len,
1450                                             const struct iov_iter *it)
1451 {
1452         struct sk_buff *skb;
1453         size_t linear;
1454         int err;
1455         int i;
1456
1457         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1458                 return ERR_PTR(-ENOMEM);
1459
1460         local_bh_disable();
1461         skb = napi_get_frags(&tfile->napi);
1462         local_bh_enable();
1463         if (!skb)
1464                 return ERR_PTR(-ENOMEM);
1465
1466         linear = iov_iter_single_seg_count(it);
1467         err = __skb_grow(skb, linear);
1468         if (err)
1469                 goto free;
1470
1471         skb->len = len;
1472         skb->data_len = len - linear;
1473         skb->truesize += skb->data_len;
1474
1475         for (i = 1; i < it->nr_segs; i++) {
1476                 struct page_frag *pfrag = &current->task_frag;
1477                 size_t fragsz = it->iov[i].iov_len;
1478
1479                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1480                         err = -EINVAL;
1481                         goto free;
1482                 }
1483
1484                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1485                         err = -ENOMEM;
1486                         goto free;
1487                 }
1488
1489                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1490                                    pfrag->offset, fragsz);
1491                 page_ref_inc(pfrag->page);
1492                 pfrag->offset += fragsz;
1493         }
1494
1495         return skb;
1496 free:
1497         /* frees skb and all frags allocated with napi_alloc_frag() */
1498         napi_free_frags(&tfile->napi);
1499         return ERR_PTR(err);
1500 }
1501
1502 /* prepad is the amount to reserve at front.  len is length after that.
1503  * linear is a hint as to how much to copy (usually headers). */
1504 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1505                                      size_t prepad, size_t len,
1506                                      size_t linear, int noblock)
1507 {
1508         struct sock *sk = tfile->socket.sk;
1509         struct sk_buff *skb;
1510         int err;
1511
1512         /* Under a page?  Don't bother with paged skb. */
1513         if (prepad + len < PAGE_SIZE || !linear)
1514                 linear = len;
1515
1516         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1517                                    &err, 0);
1518         if (!skb)
1519                 return ERR_PTR(err);
1520
1521         skb_reserve(skb, prepad);
1522         skb_put(skb, linear);
1523         skb->data_len = len - linear;
1524         skb->len += len - linear;
1525
1526         return skb;
1527 }
1528
1529 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1530                            struct sk_buff *skb, int more)
1531 {
1532         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1533         struct sk_buff_head process_queue;
1534         u32 rx_batched = tun->rx_batched;
1535         bool rcv = false;
1536
1537         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1538                 local_bh_disable();
1539                 netif_receive_skb(skb);
1540                 local_bh_enable();
1541                 return;
1542         }
1543
1544         spin_lock(&queue->lock);
1545         if (!more || skb_queue_len(queue) == rx_batched) {
1546                 __skb_queue_head_init(&process_queue);
1547                 skb_queue_splice_tail_init(queue, &process_queue);
1548                 rcv = true;
1549         } else {
1550                 __skb_queue_tail(queue, skb);
1551         }
1552         spin_unlock(&queue->lock);
1553
1554         if (rcv) {
1555                 struct sk_buff *nskb;
1556
1557                 local_bh_disable();
1558                 while ((nskb = __skb_dequeue(&process_queue)))
1559                         netif_receive_skb(nskb);
1560                 netif_receive_skb(skb);
1561                 local_bh_enable();
1562         }
1563 }
1564
1565 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1566                               int len, int noblock, bool zerocopy)
1567 {
1568         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1569                 return false;
1570
1571         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1572                 return false;
1573
1574         if (!noblock)
1575                 return false;
1576
1577         if (zerocopy)
1578                 return false;
1579
1580         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1581             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1582                 return false;
1583
1584         return true;
1585 }
1586
1587 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1588                                        int buflen, int len, int pad)
1589 {
1590         struct sk_buff *skb = build_skb(buf, buflen);
1591
1592         if (!skb)
1593                 return ERR_PTR(-ENOMEM);
1594
1595         skb_reserve(skb, pad);
1596         skb_put(skb, len);
1597
1598         get_page(alloc_frag->page);
1599         alloc_frag->offset += buflen;
1600
1601         return skb;
1602 }
1603
1604 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1605                        struct xdp_buff *xdp, u32 act)
1606 {
1607         int err;
1608
1609         switch (act) {
1610         case XDP_REDIRECT:
1611                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1612                 if (err)
1613                         return err;
1614                 break;
1615         case XDP_TX:
1616                 err = tun_xdp_tx(tun->dev, xdp);
1617                 if (err < 0)
1618                         return err;
1619                 break;
1620         case XDP_PASS:
1621                 break;
1622         default:
1623                 bpf_warn_invalid_xdp_action(act);
1624                 /* fall through */
1625         case XDP_ABORTED:
1626                 trace_xdp_exception(tun->dev, xdp_prog, act);
1627                 /* fall through */
1628         case XDP_DROP:
1629                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1630                 break;
1631         }
1632
1633         return act;
1634 }
1635
1636 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1637                                      struct tun_file *tfile,
1638                                      struct iov_iter *from,
1639                                      struct virtio_net_hdr *hdr,
1640                                      int len, int *skb_xdp)
1641 {
1642         struct page_frag *alloc_frag = &current->task_frag;
1643         struct bpf_prog *xdp_prog;
1644         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1645         char *buf;
1646         size_t copied;
1647         int pad = TUN_RX_PAD;
1648         int err = 0;
1649
1650         rcu_read_lock();
1651         xdp_prog = rcu_dereference(tun->xdp_prog);
1652         if (xdp_prog)
1653                 pad += XDP_PACKET_HEADROOM;
1654         buflen += SKB_DATA_ALIGN(len + pad);
1655         rcu_read_unlock();
1656
1657         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1658         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1659                 return ERR_PTR(-ENOMEM);
1660
1661         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1662         copied = copy_page_from_iter(alloc_frag->page,
1663                                      alloc_frag->offset + pad,
1664                                      len, from);
1665         if (copied != len)
1666                 return ERR_PTR(-EFAULT);
1667
1668         /* There's a small window that XDP may be set after the check
1669          * of xdp_prog above, this should be rare and for simplicity
1670          * we do XDP on skb in case the headroom is not enough.
1671          */
1672         if (hdr->gso_type || !xdp_prog) {
1673                 *skb_xdp = 1;
1674                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1675         }
1676
1677         *skb_xdp = 0;
1678
1679         local_bh_disable();
1680         rcu_read_lock();
1681         xdp_prog = rcu_dereference(tun->xdp_prog);
1682         if (xdp_prog) {
1683                 struct xdp_buff xdp;
1684                 u32 act;
1685
1686                 xdp.data_hard_start = buf;
1687                 xdp.data = buf + pad;
1688                 xdp_set_data_meta_invalid(&xdp);
1689                 xdp.data_end = xdp.data + len;
1690                 xdp.rxq = &tfile->xdp_rxq;
1691
1692                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1693                 if (act == XDP_REDIRECT || act == XDP_TX) {
1694                         get_page(alloc_frag->page);
1695                         alloc_frag->offset += buflen;
1696                 }
1697                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1698                 if (err < 0)
1699                         goto err_xdp;
1700                 if (err == XDP_REDIRECT)
1701                         xdp_do_flush_map();
1702                 if (err != XDP_PASS)
1703                         goto out;
1704
1705                 pad = xdp.data - xdp.data_hard_start;
1706                 len = xdp.data_end - xdp.data;
1707         }
1708         rcu_read_unlock();
1709         local_bh_enable();
1710
1711         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1712
1713 err_xdp:
1714         put_page(alloc_frag->page);
1715 out:
1716         rcu_read_unlock();
1717         local_bh_enable();
1718         return NULL;
1719 }
1720
1721 /* Get packet from user space buffer */
1722 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1723                             void *msg_control, struct iov_iter *from,
1724                             int noblock, bool more)
1725 {
1726         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1727         struct sk_buff *skb;
1728         size_t total_len = iov_iter_count(from);
1729         size_t len = total_len, align = tun->align, linear;
1730         struct virtio_net_hdr gso = { 0 };
1731         struct tun_pcpu_stats *stats;
1732         int good_linear;
1733         int copylen;
1734         bool zerocopy = false;
1735         int err;
1736         u32 rxhash = 0;
1737         int skb_xdp = 1;
1738         bool frags = tun_napi_frags_enabled(tfile);
1739
1740         if (!(tun->dev->flags & IFF_UP))
1741                 return -EIO;
1742
1743         if (!(tun->flags & IFF_NO_PI)) {
1744                 if (len < sizeof(pi))
1745                         return -EINVAL;
1746                 len -= sizeof(pi);
1747
1748                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1749                         return -EFAULT;
1750         }
1751
1752         if (tun->flags & IFF_VNET_HDR) {
1753                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1754
1755                 if (len < vnet_hdr_sz)
1756                         return -EINVAL;
1757                 len -= vnet_hdr_sz;
1758
1759                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1760                         return -EFAULT;
1761
1762                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1763                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1764                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1765
1766                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1767                         return -EINVAL;
1768                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1769         }
1770
1771         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1772                 align += NET_IP_ALIGN;
1773                 if (unlikely(len < ETH_HLEN ||
1774                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1775                         return -EINVAL;
1776         }
1777
1778         good_linear = SKB_MAX_HEAD(align);
1779
1780         if (msg_control) {
1781                 struct iov_iter i = *from;
1782
1783                 /* There are 256 bytes to be copied in skb, so there is
1784                  * enough room for skb expand head in case it is used.
1785                  * The rest of the buffer is mapped from userspace.
1786                  */
1787                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1788                 if (copylen > good_linear)
1789                         copylen = good_linear;
1790                 linear = copylen;
1791                 iov_iter_advance(&i, copylen);
1792                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1793                         zerocopy = true;
1794         }
1795
1796         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1797                 /* For the packet that is not easy to be processed
1798                  * (e.g gso or jumbo packet), we will do it at after
1799                  * skb was created with generic XDP routine.
1800                  */
1801                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1802                 if (IS_ERR(skb)) {
1803                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1804                         return PTR_ERR(skb);
1805                 }
1806                 if (!skb)
1807                         return total_len;
1808         } else {
1809                 if (!zerocopy) {
1810                         copylen = len;
1811                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1812                                 linear = good_linear;
1813                         else
1814                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1815                 }
1816
1817                 if (frags) {
1818                         mutex_lock(&tfile->napi_mutex);
1819                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1820                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1821                          * If zerocopy is enabled, then this layout will be
1822                          * overwritten by zerocopy_sg_from_iter().
1823                          */
1824                         zerocopy = false;
1825                 } else {
1826                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1827                                             noblock);
1828                 }
1829
1830                 if (IS_ERR(skb)) {
1831                         if (PTR_ERR(skb) != -EAGAIN)
1832                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1833                         if (frags)
1834                                 mutex_unlock(&tfile->napi_mutex);
1835                         return PTR_ERR(skb);
1836                 }
1837
1838                 if (zerocopy)
1839                         err = zerocopy_sg_from_iter(skb, from);
1840                 else
1841                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1842
1843                 if (err) {
1844                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1845                         kfree_skb(skb);
1846                         if (frags) {
1847                                 tfile->napi.skb = NULL;
1848                                 mutex_unlock(&tfile->napi_mutex);
1849                         }
1850
1851                         return -EFAULT;
1852                 }
1853         }
1854
1855         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1856                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1857                 kfree_skb(skb);
1858                 if (frags) {
1859                         tfile->napi.skb = NULL;
1860                         mutex_unlock(&tfile->napi_mutex);
1861                 }
1862
1863                 return -EINVAL;
1864         }
1865
1866         switch (tun->flags & TUN_TYPE_MASK) {
1867         case IFF_TUN:
1868                 if (tun->flags & IFF_NO_PI) {
1869                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1870
1871                         switch (ip_version) {
1872                         case 4:
1873                                 pi.proto = htons(ETH_P_IP);
1874                                 break;
1875                         case 6:
1876                                 pi.proto = htons(ETH_P_IPV6);
1877                                 break;
1878                         default:
1879                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1880                                 kfree_skb(skb);
1881                                 return -EINVAL;
1882                         }
1883                 }
1884
1885                 skb_reset_mac_header(skb);
1886                 skb->protocol = pi.proto;
1887                 skb->dev = tun->dev;
1888                 break;
1889         case IFF_TAP:
1890                 if (!frags)
1891                         skb->protocol = eth_type_trans(skb, tun->dev);
1892                 break;
1893         }
1894
1895         /* copy skb_ubuf_info for callback when skb has no error */
1896         if (zerocopy) {
1897                 skb_shinfo(skb)->destructor_arg = msg_control;
1898                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1899                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1900         } else if (msg_control) {
1901                 struct ubuf_info *uarg = msg_control;
1902                 uarg->callback(uarg, false);
1903         }
1904
1905         skb_reset_network_header(skb);
1906         skb_probe_transport_header(skb, 0);
1907
1908         if (skb_xdp) {
1909                 struct bpf_prog *xdp_prog;
1910                 int ret;
1911
1912                 local_bh_disable();
1913                 rcu_read_lock();
1914                 xdp_prog = rcu_dereference(tun->xdp_prog);
1915                 if (xdp_prog) {
1916                         ret = do_xdp_generic(xdp_prog, skb);
1917                         if (ret != XDP_PASS) {
1918                                 rcu_read_unlock();
1919                                 local_bh_enable();
1920                                 return total_len;
1921                         }
1922                 }
1923                 rcu_read_unlock();
1924                 local_bh_enable();
1925         }
1926
1927         /* Compute the costly rx hash only if needed for flow updates.
1928          * We may get a very small possibility of OOO during switching, not
1929          * worth to optimize.
1930          */
1931         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1932             !tfile->detached)
1933                 rxhash = __skb_get_hash_symmetric(skb);
1934
1935         if (frags) {
1936                 /* Exercise flow dissector code path. */
1937                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1938
1939                 if (unlikely(headlen > skb_headlen(skb))) {
1940                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1941                         napi_free_frags(&tfile->napi);
1942                         mutex_unlock(&tfile->napi_mutex);
1943                         WARN_ON(1);
1944                         return -ENOMEM;
1945                 }
1946
1947                 local_bh_disable();
1948                 napi_gro_frags(&tfile->napi);
1949                 local_bh_enable();
1950                 mutex_unlock(&tfile->napi_mutex);
1951         } else if (tfile->napi_enabled) {
1952                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1953                 int queue_len;
1954
1955                 spin_lock_bh(&queue->lock);
1956                 __skb_queue_tail(queue, skb);
1957                 queue_len = skb_queue_len(queue);
1958                 spin_unlock(&queue->lock);
1959
1960                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1961                         napi_schedule(&tfile->napi);
1962
1963                 local_bh_enable();
1964         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1965                 tun_rx_batched(tun, tfile, skb, more);
1966         } else {
1967                 netif_rx_ni(skb);
1968         }
1969
1970         stats = get_cpu_ptr(tun->pcpu_stats);
1971         u64_stats_update_begin(&stats->syncp);
1972         stats->rx_packets++;
1973         stats->rx_bytes += len;
1974         u64_stats_update_end(&stats->syncp);
1975         put_cpu_ptr(stats);
1976
1977         if (rxhash)
1978                 tun_flow_update(tun, rxhash, tfile);
1979
1980         return total_len;
1981 }
1982
1983 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1984 {
1985         struct file *file = iocb->ki_filp;
1986         struct tun_file *tfile = file->private_data;
1987         struct tun_struct *tun = tun_get(tfile);
1988         ssize_t result;
1989
1990         if (!tun)
1991                 return -EBADFD;
1992
1993         result = tun_get_user(tun, tfile, NULL, from,
1994                               file->f_flags & O_NONBLOCK, false);
1995
1996         tun_put(tun);
1997         return result;
1998 }
1999
2000 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2001                                 struct tun_file *tfile,
2002                                 struct xdp_frame *xdp_frame,
2003                                 struct iov_iter *iter)
2004 {
2005         int vnet_hdr_sz = 0;
2006         size_t size = xdp_frame->len;
2007         struct tun_pcpu_stats *stats;
2008         size_t ret;
2009
2010         if (tun->flags & IFF_VNET_HDR) {
2011                 struct virtio_net_hdr gso = { 0 };
2012
2013                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2014                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2015                         return -EINVAL;
2016                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2017                              sizeof(gso)))
2018                         return -EFAULT;
2019                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2020         }
2021
2022         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2023
2024         stats = get_cpu_ptr(tun->pcpu_stats);
2025         u64_stats_update_begin(&stats->syncp);
2026         stats->tx_packets++;
2027         stats->tx_bytes += ret;
2028         u64_stats_update_end(&stats->syncp);
2029         put_cpu_ptr(tun->pcpu_stats);
2030
2031         return ret;
2032 }
2033
2034 /* Put packet to the user space buffer */
2035 static ssize_t tun_put_user(struct tun_struct *tun,
2036                             struct tun_file *tfile,
2037                             struct sk_buff *skb,
2038                             struct iov_iter *iter)
2039 {
2040         struct tun_pi pi = { 0, skb->protocol };
2041         struct tun_pcpu_stats *stats;
2042         ssize_t total;
2043         int vlan_offset = 0;
2044         int vlan_hlen = 0;
2045         int vnet_hdr_sz = 0;
2046
2047         if (skb_vlan_tag_present(skb))
2048                 vlan_hlen = VLAN_HLEN;
2049
2050         if (tun->flags & IFF_VNET_HDR)
2051                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2052
2053         total = skb->len + vlan_hlen + vnet_hdr_sz;
2054
2055         if (!(tun->flags & IFF_NO_PI)) {
2056                 if (iov_iter_count(iter) < sizeof(pi))
2057                         return -EINVAL;
2058
2059                 total += sizeof(pi);
2060                 if (iov_iter_count(iter) < total) {
2061                         /* Packet will be striped */
2062                         pi.flags |= TUN_PKT_STRIP;
2063                 }
2064
2065                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2066                         return -EFAULT;
2067         }
2068
2069         if (vnet_hdr_sz) {
2070                 struct virtio_net_hdr gso;
2071
2072                 if (iov_iter_count(iter) < vnet_hdr_sz)
2073                         return -EINVAL;
2074
2075                 if (virtio_net_hdr_from_skb(skb, &gso,
2076                                             tun_is_little_endian(tun), true,
2077                                             vlan_hlen)) {
2078                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2079                         pr_err("unexpected GSO type: "
2080                                "0x%x, gso_size %d, hdr_len %d\n",
2081                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2082                                tun16_to_cpu(tun, gso.hdr_len));
2083                         print_hex_dump(KERN_ERR, "tun: ",
2084                                        DUMP_PREFIX_NONE,
2085                                        16, 1, skb->head,
2086                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2087                         WARN_ON_ONCE(1);
2088                         return -EINVAL;
2089                 }
2090
2091                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2092                         return -EFAULT;
2093
2094                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2095         }
2096
2097         if (vlan_hlen) {
2098                 int ret;
2099                 struct veth veth;
2100
2101                 veth.h_vlan_proto = skb->vlan_proto;
2102                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2103
2104                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2105
2106                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2107                 if (ret || !iov_iter_count(iter))
2108                         goto done;
2109
2110                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2111                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2112                         goto done;
2113         }
2114
2115         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2116
2117 done:
2118         /* caller is in process context, */
2119         stats = get_cpu_ptr(tun->pcpu_stats);
2120         u64_stats_update_begin(&stats->syncp);
2121         stats->tx_packets++;
2122         stats->tx_bytes += skb->len + vlan_hlen;
2123         u64_stats_update_end(&stats->syncp);
2124         put_cpu_ptr(tun->pcpu_stats);
2125
2126         return total;
2127 }
2128
2129 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2130 {
2131         DECLARE_WAITQUEUE(wait, current);
2132         void *ptr = NULL;
2133         int error = 0;
2134
2135         ptr = ptr_ring_consume(&tfile->tx_ring);
2136         if (ptr)
2137                 goto out;
2138         if (noblock) {
2139                 error = -EAGAIN;
2140                 goto out;
2141         }
2142
2143         add_wait_queue(&tfile->wq.wait, &wait);
2144         current->state = TASK_INTERRUPTIBLE;
2145
2146         while (1) {
2147                 ptr = ptr_ring_consume(&tfile->tx_ring);
2148                 if (ptr)
2149                         break;
2150                 if (signal_pending(current)) {
2151                         error = -ERESTARTSYS;
2152                         break;
2153                 }
2154                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2155                         error = -EFAULT;
2156                         break;
2157                 }
2158
2159                 schedule();
2160         }
2161
2162         current->state = TASK_RUNNING;
2163         remove_wait_queue(&tfile->wq.wait, &wait);
2164
2165 out:
2166         *err = error;
2167         return ptr;
2168 }
2169
2170 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2171                            struct iov_iter *to,
2172                            int noblock, void *ptr)
2173 {
2174         ssize_t ret;
2175         int err;
2176
2177         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2178
2179         if (!iov_iter_count(to)) {
2180                 tun_ptr_free(ptr);
2181                 return 0;
2182         }
2183
2184         if (!ptr) {
2185                 /* Read frames from ring */
2186                 ptr = tun_ring_recv(tfile, noblock, &err);
2187                 if (!ptr)
2188                         return err;
2189         }
2190
2191         if (tun_is_xdp_frame(ptr)) {
2192                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2193
2194                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2195                 xdp_return_frame(xdpf);
2196         } else {
2197                 struct sk_buff *skb = ptr;
2198
2199                 ret = tun_put_user(tun, tfile, skb, to);
2200                 if (unlikely(ret < 0))
2201                         kfree_skb(skb);
2202                 else
2203                         consume_skb(skb);
2204         }
2205
2206         return ret;
2207 }
2208
2209 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2210 {
2211         struct file *file = iocb->ki_filp;
2212         struct tun_file *tfile = file->private_data;
2213         struct tun_struct *tun = tun_get(tfile);
2214         ssize_t len = iov_iter_count(to), ret;
2215
2216         if (!tun)
2217                 return -EBADFD;
2218         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2219         ret = min_t(ssize_t, ret, len);
2220         if (ret > 0)
2221                 iocb->ki_pos = ret;
2222         tun_put(tun);
2223         return ret;
2224 }
2225
2226 static void tun_prog_free(struct rcu_head *rcu)
2227 {
2228         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2229
2230         bpf_prog_destroy(prog->prog);
2231         kfree(prog);
2232 }
2233
2234 static int __tun_set_ebpf(struct tun_struct *tun,
2235                           struct tun_prog __rcu **prog_p,
2236                           struct bpf_prog *prog)
2237 {
2238         struct tun_prog *old, *new = NULL;
2239
2240         if (prog) {
2241                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2242                 if (!new)
2243                         return -ENOMEM;
2244                 new->prog = prog;
2245         }
2246
2247         spin_lock_bh(&tun->lock);
2248         old = rcu_dereference_protected(*prog_p,
2249                                         lockdep_is_held(&tun->lock));
2250         rcu_assign_pointer(*prog_p, new);
2251         spin_unlock_bh(&tun->lock);
2252
2253         if (old)
2254                 call_rcu(&old->rcu, tun_prog_free);
2255
2256         return 0;
2257 }
2258
2259 static void tun_free_netdev(struct net_device *dev)
2260 {
2261         struct tun_struct *tun = netdev_priv(dev);
2262
2263         BUG_ON(!(list_empty(&tun->disabled)));
2264         free_percpu(tun->pcpu_stats);
2265         tun_flow_uninit(tun);
2266         security_tun_dev_free_security(tun->security);
2267         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2268         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2269 }
2270
2271 static void tun_setup(struct net_device *dev)
2272 {
2273         struct tun_struct *tun = netdev_priv(dev);
2274
2275         tun->owner = INVALID_UID;
2276         tun->group = INVALID_GID;
2277         tun_default_link_ksettings(dev, &tun->link_ksettings);
2278
2279         dev->ethtool_ops = &tun_ethtool_ops;
2280         dev->needs_free_netdev = true;
2281         dev->priv_destructor = tun_free_netdev;
2282         /* We prefer our own queue length */
2283         dev->tx_queue_len = TUN_READQ_SIZE;
2284 }
2285
2286 /* Trivial set of netlink ops to allow deleting tun or tap
2287  * device with netlink.
2288  */
2289 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2290                         struct netlink_ext_ack *extack)
2291 {
2292         if (!data)
2293                 return 0;
2294         return -EINVAL;
2295 }
2296
2297 static size_t tun_get_size(const struct net_device *dev)
2298 {
2299         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2300         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2301
2302         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2303                nla_total_size(sizeof(gid_t)) + /* GROUP */
2304                nla_total_size(sizeof(u8)) + /* TYPE */
2305                nla_total_size(sizeof(u8)) + /* PI */
2306                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2307                nla_total_size(sizeof(u8)) + /* PERSIST */
2308                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2309                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2310                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2311                0;
2312 }
2313
2314 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2315 {
2316         struct tun_struct *tun = netdev_priv(dev);
2317
2318         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2319                 goto nla_put_failure;
2320         if (uid_valid(tun->owner) &&
2321             nla_put_u32(skb, IFLA_TUN_OWNER,
2322                         from_kuid_munged(current_user_ns(), tun->owner)))
2323                 goto nla_put_failure;
2324         if (gid_valid(tun->group) &&
2325             nla_put_u32(skb, IFLA_TUN_GROUP,
2326                         from_kgid_munged(current_user_ns(), tun->group)))
2327                 goto nla_put_failure;
2328         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2329                 goto nla_put_failure;
2330         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2331                 goto nla_put_failure;
2332         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2333                 goto nla_put_failure;
2334         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2335                        !!(tun->flags & IFF_MULTI_QUEUE)))
2336                 goto nla_put_failure;
2337         if (tun->flags & IFF_MULTI_QUEUE) {
2338                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2339                         goto nla_put_failure;
2340                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2341                                 tun->numdisabled))
2342                         goto nla_put_failure;
2343         }
2344
2345         return 0;
2346
2347 nla_put_failure:
2348         return -EMSGSIZE;
2349 }
2350
2351 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2352         .kind           = DRV_NAME,
2353         .priv_size      = sizeof(struct tun_struct),
2354         .setup          = tun_setup,
2355         .validate       = tun_validate,
2356         .get_size       = tun_get_size,
2357         .fill_info      = tun_fill_info,
2358 };
2359
2360 static void tun_sock_write_space(struct sock *sk)
2361 {
2362         struct tun_file *tfile;
2363         wait_queue_head_t *wqueue;
2364
2365         if (!sock_writeable(sk))
2366                 return;
2367
2368         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2369                 return;
2370
2371         wqueue = sk_sleep(sk);
2372         if (wqueue && waitqueue_active(wqueue))
2373                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2374                                                 EPOLLWRNORM | EPOLLWRBAND);
2375
2376         tfile = container_of(sk, struct tun_file, sk);
2377         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2378 }
2379
2380 static int tun_xdp_one(struct tun_struct *tun,
2381                        struct tun_file *tfile,
2382                        struct xdp_buff *xdp, int *flush)
2383 {
2384         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2385         struct virtio_net_hdr *gso = &hdr->gso;
2386         struct tun_pcpu_stats *stats;
2387         struct bpf_prog *xdp_prog;
2388         struct sk_buff *skb = NULL;
2389         u32 rxhash = 0, act;
2390         int buflen = hdr->buflen;
2391         int err = 0;
2392         bool skb_xdp = false;
2393
2394         xdp_prog = rcu_dereference(tun->xdp_prog);
2395         if (xdp_prog) {
2396                 if (gso->gso_type) {
2397                         skb_xdp = true;
2398                         goto build;
2399                 }
2400                 xdp_set_data_meta_invalid(xdp);
2401                 xdp->rxq = &tfile->xdp_rxq;
2402
2403                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2404                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2405                 if (err < 0) {
2406                         put_page(virt_to_head_page(xdp->data));
2407                         return err;
2408                 }
2409
2410                 switch (err) {
2411                 case XDP_REDIRECT:
2412                         *flush = true;
2413                         /* fall through */
2414                 case XDP_TX:
2415                         return 0;
2416                 case XDP_PASS:
2417                         break;
2418                 default:
2419                         put_page(virt_to_head_page(xdp->data));
2420                         return 0;
2421                 }
2422         }
2423
2424 build:
2425         skb = build_skb(xdp->data_hard_start, buflen);
2426         if (!skb) {
2427                 err = -ENOMEM;
2428                 goto out;
2429         }
2430
2431         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2432         skb_put(skb, xdp->data_end - xdp->data);
2433
2434         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2435                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2436                 kfree_skb(skb);
2437                 err = -EINVAL;
2438                 goto out;
2439         }
2440
2441         skb->protocol = eth_type_trans(skb, tun->dev);
2442         skb_reset_network_header(skb);
2443         skb_probe_transport_header(skb, 0);
2444
2445         if (skb_xdp) {
2446                 err = do_xdp_generic(xdp_prog, skb);
2447                 if (err != XDP_PASS)
2448                         goto out;
2449         }
2450
2451         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2452             !tfile->detached)
2453                 rxhash = __skb_get_hash_symmetric(skb);
2454
2455         netif_receive_skb(skb);
2456
2457         stats = get_cpu_ptr(tun->pcpu_stats);
2458         u64_stats_update_begin(&stats->syncp);
2459         stats->rx_packets++;
2460         stats->rx_bytes += skb->len;
2461         u64_stats_update_end(&stats->syncp);
2462         put_cpu_ptr(stats);
2463
2464         if (rxhash)
2465                 tun_flow_update(tun, rxhash, tfile);
2466
2467 out:
2468         return err;
2469 }
2470
2471 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2472 {
2473         int ret, i;
2474         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2475         struct tun_struct *tun = tun_get(tfile);
2476         struct tun_msg_ctl *ctl = m->msg_control;
2477         struct xdp_buff *xdp;
2478
2479         if (!tun)
2480                 return -EBADFD;
2481
2482         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2483                 int n = ctl->num;
2484                 int flush = 0;
2485
2486                 local_bh_disable();
2487                 rcu_read_lock();
2488
2489                 for (i = 0; i < n; i++) {
2490                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2491                         tun_xdp_one(tun, tfile, xdp, &flush);
2492                 }
2493
2494                 if (flush)
2495                         xdp_do_flush_map();
2496
2497                 rcu_read_unlock();
2498                 local_bh_enable();
2499
2500                 ret = total_len;
2501                 goto out;
2502         }
2503
2504         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2505                            m->msg_flags & MSG_DONTWAIT,
2506                            m->msg_flags & MSG_MORE);
2507 out:
2508         tun_put(tun);
2509         return ret;
2510 }
2511
2512 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2513                        int flags)
2514 {
2515         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2516         struct tun_struct *tun = tun_get(tfile);
2517         void *ptr = m->msg_control;
2518         int ret;
2519
2520         if (!tun) {
2521                 ret = -EBADFD;
2522                 goto out_free;
2523         }
2524
2525         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2526                 ret = -EINVAL;
2527                 goto out_put_tun;
2528         }
2529         if (flags & MSG_ERRQUEUE) {
2530                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2531                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2532                 goto out;
2533         }
2534         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2535         if (ret > (ssize_t)total_len) {
2536                 m->msg_flags |= MSG_TRUNC;
2537                 ret = flags & MSG_TRUNC ? ret : total_len;
2538         }
2539 out:
2540         tun_put(tun);
2541         return ret;
2542
2543 out_put_tun:
2544         tun_put(tun);
2545 out_free:
2546         tun_ptr_free(ptr);
2547         return ret;
2548 }
2549
2550 static int tun_ptr_peek_len(void *ptr)
2551 {
2552         if (likely(ptr)) {
2553                 if (tun_is_xdp_frame(ptr)) {
2554                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2555
2556                         return xdpf->len;
2557                 }
2558                 return __skb_array_len_with_tag(ptr);
2559         } else {
2560                 return 0;
2561         }
2562 }
2563
2564 static int tun_peek_len(struct socket *sock)
2565 {
2566         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2567         struct tun_struct *tun;
2568         int ret = 0;
2569
2570         tun = tun_get(tfile);
2571         if (!tun)
2572                 return 0;
2573
2574         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2575         tun_put(tun);
2576
2577         return ret;
2578 }
2579
2580 /* Ops structure to mimic raw sockets with tun */
2581 static const struct proto_ops tun_socket_ops = {
2582         .peek_len = tun_peek_len,
2583         .sendmsg = tun_sendmsg,
2584         .recvmsg = tun_recvmsg,
2585 };
2586
2587 static struct proto tun_proto = {
2588         .name           = "tun",
2589         .owner          = THIS_MODULE,
2590         .obj_size       = sizeof(struct tun_file),
2591 };
2592
2593 static int tun_flags(struct tun_struct *tun)
2594 {
2595         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2596 }
2597
2598 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2599                               char *buf)
2600 {
2601         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2602         return sprintf(buf, "0x%x\n", tun_flags(tun));
2603 }
2604
2605 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2606                               char *buf)
2607 {
2608         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2609         return uid_valid(tun->owner)?
2610                 sprintf(buf, "%u\n",
2611                         from_kuid_munged(current_user_ns(), tun->owner)):
2612                 sprintf(buf, "-1\n");
2613 }
2614
2615 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2616                               char *buf)
2617 {
2618         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2619         return gid_valid(tun->group) ?
2620                 sprintf(buf, "%u\n",
2621                         from_kgid_munged(current_user_ns(), tun->group)):
2622                 sprintf(buf, "-1\n");
2623 }
2624
2625 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2626 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2627 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2628
2629 static struct attribute *tun_dev_attrs[] = {
2630         &dev_attr_tun_flags.attr,
2631         &dev_attr_owner.attr,
2632         &dev_attr_group.attr,
2633         NULL
2634 };
2635
2636 static const struct attribute_group tun_attr_group = {
2637         .attrs = tun_dev_attrs
2638 };
2639
2640 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2641 {
2642         struct tun_struct *tun;
2643         struct tun_file *tfile = file->private_data;
2644         struct net_device *dev;
2645         int err;
2646
2647         if (tfile->detached)
2648                 return -EINVAL;
2649
2650         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2651                 if (!capable(CAP_NET_ADMIN))
2652                         return -EPERM;
2653
2654                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2655                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2656                         return -EINVAL;
2657         }
2658
2659         dev = __dev_get_by_name(net, ifr->ifr_name);
2660         if (dev) {
2661                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2662                         return -EBUSY;
2663                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2664                         tun = netdev_priv(dev);
2665                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2666                         tun = netdev_priv(dev);
2667                 else
2668                         return -EINVAL;
2669
2670                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2671                     !!(tun->flags & IFF_MULTI_QUEUE))
2672                         return -EINVAL;
2673
2674                 if (tun_not_capable(tun))
2675                         return -EPERM;
2676                 err = security_tun_dev_open(tun->security);
2677                 if (err < 0)
2678                         return err;
2679
2680                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2681                                  ifr->ifr_flags & IFF_NAPI,
2682                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2683                 if (err < 0)
2684                         return err;
2685
2686                 if (tun->flags & IFF_MULTI_QUEUE &&
2687                     (tun->numqueues + tun->numdisabled > 1)) {
2688                         /* One or more queue has already been attached, no need
2689                          * to initialize the device again.
2690                          */
2691                         netdev_state_change(dev);
2692                         return 0;
2693                 }
2694
2695                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2696                               (ifr->ifr_flags & TUN_FEATURES);
2697
2698                 netdev_state_change(dev);
2699         } else {
2700                 char *name;
2701                 unsigned long flags = 0;
2702                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2703                              MAX_TAP_QUEUES : 1;
2704
2705                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2706                         return -EPERM;
2707                 err = security_tun_dev_create();
2708                 if (err < 0)
2709                         return err;
2710
2711                 /* Set dev type */
2712                 if (ifr->ifr_flags & IFF_TUN) {
2713                         /* TUN device */
2714                         flags |= IFF_TUN;
2715                         name = "tun%d";
2716                 } else if (ifr->ifr_flags & IFF_TAP) {
2717                         /* TAP device */
2718                         flags |= IFF_TAP;
2719                         name = "tap%d";
2720                 } else
2721                         return -EINVAL;
2722
2723                 if (*ifr->ifr_name)
2724                         name = ifr->ifr_name;
2725
2726                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2727                                        NET_NAME_UNKNOWN, tun_setup, queues,
2728                                        queues);
2729
2730                 if (!dev)
2731                         return -ENOMEM;
2732                 err = dev_get_valid_name(net, dev, name);
2733                 if (err < 0)
2734                         goto err_free_dev;
2735
2736                 dev_net_set(dev, net);
2737                 dev->rtnl_link_ops = &tun_link_ops;
2738                 dev->ifindex = tfile->ifindex;
2739                 dev->sysfs_groups[0] = &tun_attr_group;
2740
2741                 tun = netdev_priv(dev);
2742                 tun->dev = dev;
2743                 tun->flags = flags;
2744                 tun->txflt.count = 0;
2745                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2746
2747                 tun->align = NET_SKB_PAD;
2748                 tun->filter_attached = false;
2749                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2750                 tun->rx_batched = 0;
2751                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2752
2753                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2754                 if (!tun->pcpu_stats) {
2755                         err = -ENOMEM;
2756                         goto err_free_dev;
2757                 }
2758
2759                 spin_lock_init(&tun->lock);
2760
2761                 err = security_tun_dev_alloc_security(&tun->security);
2762                 if (err < 0)
2763                         goto err_free_stat;
2764
2765                 tun_net_init(dev);
2766                 tun_flow_init(tun);
2767
2768                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2769                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2770                                    NETIF_F_HW_VLAN_STAG_TX;
2771                 dev->features = dev->hw_features | NETIF_F_LLTX;
2772                 dev->vlan_features = dev->features &
2773                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2774                                        NETIF_F_HW_VLAN_STAG_TX);
2775
2776                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2777                               (ifr->ifr_flags & TUN_FEATURES);
2778
2779                 INIT_LIST_HEAD(&tun->disabled);
2780                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2781                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2782                 if (err < 0)
2783                         goto err_free_flow;
2784
2785                 err = register_netdevice(tun->dev);
2786                 if (err < 0)
2787                         goto err_detach;
2788         }
2789
2790         netif_carrier_on(tun->dev);
2791
2792         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2793
2794         /* Make sure persistent devices do not get stuck in
2795          * xoff state.
2796          */
2797         if (netif_running(tun->dev))
2798                 netif_tx_wake_all_queues(tun->dev);
2799
2800         strcpy(ifr->ifr_name, tun->dev->name);
2801         return 0;
2802
2803 err_detach:
2804         tun_detach_all(dev);
2805         /* register_netdevice() already called tun_free_netdev() */
2806         goto err_free_dev;
2807
2808 err_free_flow:
2809         tun_flow_uninit(tun);
2810         security_tun_dev_free_security(tun->security);
2811 err_free_stat:
2812         free_percpu(tun->pcpu_stats);
2813 err_free_dev:
2814         free_netdev(dev);
2815         return err;
2816 }
2817
2818 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2819                        struct ifreq *ifr)
2820 {
2821         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2822
2823         strcpy(ifr->ifr_name, tun->dev->name);
2824
2825         ifr->ifr_flags = tun_flags(tun);
2826
2827 }
2828
2829 /* This is like a cut-down ethtool ops, except done via tun fd so no
2830  * privs required. */
2831 static int set_offload(struct tun_struct *tun, unsigned long arg)
2832 {
2833         netdev_features_t features = 0;
2834
2835         if (arg & TUN_F_CSUM) {
2836                 features |= NETIF_F_HW_CSUM;
2837                 arg &= ~TUN_F_CSUM;
2838
2839                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2840                         if (arg & TUN_F_TSO_ECN) {
2841                                 features |= NETIF_F_TSO_ECN;
2842                                 arg &= ~TUN_F_TSO_ECN;
2843                         }
2844                         if (arg & TUN_F_TSO4)
2845                                 features |= NETIF_F_TSO;
2846                         if (arg & TUN_F_TSO6)
2847                                 features |= NETIF_F_TSO6;
2848                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2849                 }
2850
2851                 arg &= ~TUN_F_UFO;
2852         }
2853
2854         /* This gives the user a way to test for new features in future by
2855          * trying to set them. */
2856         if (arg)
2857                 return -EINVAL;
2858
2859         tun->set_features = features;
2860         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2861         tun->dev->wanted_features |= features;
2862         netdev_update_features(tun->dev);
2863
2864         return 0;
2865 }
2866
2867 static void tun_detach_filter(struct tun_struct *tun, int n)
2868 {
2869         int i;
2870         struct tun_file *tfile;
2871
2872         for (i = 0; i < n; i++) {
2873                 tfile = rtnl_dereference(tun->tfiles[i]);
2874                 lock_sock(tfile->socket.sk);
2875                 sk_detach_filter(tfile->socket.sk);
2876                 release_sock(tfile->socket.sk);
2877         }
2878
2879         tun->filter_attached = false;
2880 }
2881
2882 static int tun_attach_filter(struct tun_struct *tun)
2883 {
2884         int i, ret = 0;
2885         struct tun_file *tfile;
2886
2887         for (i = 0; i < tun->numqueues; i++) {
2888                 tfile = rtnl_dereference(tun->tfiles[i]);
2889                 lock_sock(tfile->socket.sk);
2890                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2891                 release_sock(tfile->socket.sk);
2892                 if (ret) {
2893                         tun_detach_filter(tun, i);
2894                         return ret;
2895                 }
2896         }
2897
2898         tun->filter_attached = true;
2899         return ret;
2900 }
2901
2902 static void tun_set_sndbuf(struct tun_struct *tun)
2903 {
2904         struct tun_file *tfile;
2905         int i;
2906
2907         for (i = 0; i < tun->numqueues; i++) {
2908                 tfile = rtnl_dereference(tun->tfiles[i]);
2909                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2910         }
2911 }
2912
2913 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2914 {
2915         struct tun_file *tfile = file->private_data;
2916         struct tun_struct *tun;
2917         int ret = 0;
2918
2919         rtnl_lock();
2920
2921         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2922                 tun = tfile->detached;
2923                 if (!tun) {
2924                         ret = -EINVAL;
2925                         goto unlock;
2926                 }
2927                 ret = security_tun_dev_attach_queue(tun->security);
2928                 if (ret < 0)
2929                         goto unlock;
2930                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2931                                  tun->flags & IFF_NAPI_FRAGS);
2932         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2933                 tun = rtnl_dereference(tfile->tun);
2934                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2935                         ret = -EINVAL;
2936                 else
2937                         __tun_detach(tfile, false);
2938         } else
2939                 ret = -EINVAL;
2940
2941         if (ret >= 0)
2942                 netdev_state_change(tun->dev);
2943
2944 unlock:
2945         rtnl_unlock();
2946         return ret;
2947 }
2948
2949 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2950                         void __user *data)
2951 {
2952         struct bpf_prog *prog;
2953         int fd;
2954
2955         if (copy_from_user(&fd, data, sizeof(fd)))
2956                 return -EFAULT;
2957
2958         if (fd == -1) {
2959                 prog = NULL;
2960         } else {
2961                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2962                 if (IS_ERR(prog))
2963                         return PTR_ERR(prog);
2964         }
2965
2966         return __tun_set_ebpf(tun, prog_p, prog);
2967 }
2968
2969 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2970                             unsigned long arg, int ifreq_len)
2971 {
2972         struct tun_file *tfile = file->private_data;
2973         struct net *net = sock_net(&tfile->sk);
2974         struct tun_struct *tun;
2975         void __user* argp = (void __user*)arg;
2976         struct ifreq ifr;
2977         kuid_t owner;
2978         kgid_t group;
2979         int sndbuf;
2980         int vnet_hdr_sz;
2981         unsigned int ifindex;
2982         int le;
2983         int ret;
2984         bool do_notify = false;
2985
2986         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2987             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2988                 if (copy_from_user(&ifr, argp, ifreq_len))
2989                         return -EFAULT;
2990         } else {
2991                 memset(&ifr, 0, sizeof(ifr));
2992         }
2993         if (cmd == TUNGETFEATURES) {
2994                 /* Currently this just means: "what IFF flags are valid?".
2995                  * This is needed because we never checked for invalid flags on
2996                  * TUNSETIFF.
2997                  */
2998                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2999                                 (unsigned int __user*)argp);
3000         } else if (cmd == TUNSETQUEUE) {
3001                 return tun_set_queue(file, &ifr);
3002         } else if (cmd == SIOCGSKNS) {
3003                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3004                         return -EPERM;
3005                 return open_related_ns(&net->ns, get_net_ns);
3006         }
3007
3008         ret = 0;
3009         rtnl_lock();
3010
3011         tun = tun_get(tfile);
3012         if (cmd == TUNSETIFF) {
3013                 ret = -EEXIST;
3014                 if (tun)
3015                         goto unlock;
3016
3017                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3018
3019                 ret = tun_set_iff(net, file, &ifr);
3020
3021                 if (ret)
3022                         goto unlock;
3023
3024                 if (copy_to_user(argp, &ifr, ifreq_len))
3025                         ret = -EFAULT;
3026                 goto unlock;
3027         }
3028         if (cmd == TUNSETIFINDEX) {
3029                 ret = -EPERM;
3030                 if (tun)
3031                         goto unlock;
3032
3033                 ret = -EFAULT;
3034                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3035                         goto unlock;
3036
3037                 ret = 0;
3038                 tfile->ifindex = ifindex;
3039                 goto unlock;
3040         }
3041
3042         ret = -EBADFD;
3043         if (!tun)
3044                 goto unlock;
3045
3046         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3047
3048         ret = 0;
3049         switch (cmd) {
3050         case TUNGETIFF:
3051                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3052
3053                 if (tfile->detached)
3054                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3055                 if (!tfile->socket.sk->sk_filter)
3056                         ifr.ifr_flags |= IFF_NOFILTER;
3057
3058                 if (copy_to_user(argp, &ifr, ifreq_len))
3059                         ret = -EFAULT;
3060                 break;
3061
3062         case TUNSETNOCSUM:
3063                 /* Disable/Enable checksum */
3064
3065                 /* [unimplemented] */
3066                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3067                           arg ? "disabled" : "enabled");
3068                 break;
3069
3070         case TUNSETPERSIST:
3071                 /* Disable/Enable persist mode. Keep an extra reference to the
3072                  * module to prevent the module being unprobed.
3073                  */
3074                 if (arg && !(tun->flags & IFF_PERSIST)) {
3075                         tun->flags |= IFF_PERSIST;
3076                         __module_get(THIS_MODULE);
3077                         do_notify = true;
3078                 }
3079                 if (!arg && (tun->flags & IFF_PERSIST)) {
3080                         tun->flags &= ~IFF_PERSIST;
3081                         module_put(THIS_MODULE);
3082                         do_notify = true;
3083                 }
3084
3085                 tun_debug(KERN_INFO, tun, "persist %s\n",
3086                           arg ? "enabled" : "disabled");
3087                 break;
3088
3089         case TUNSETOWNER:
3090                 /* Set owner of the device */
3091                 owner = make_kuid(current_user_ns(), arg);
3092                 if (!uid_valid(owner)) {
3093                         ret = -EINVAL;
3094                         break;
3095                 }
3096                 tun->owner = owner;
3097                 do_notify = true;
3098                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3099                           from_kuid(&init_user_ns, tun->owner));
3100                 break;
3101
3102         case TUNSETGROUP:
3103                 /* Set group of the device */
3104                 group = make_kgid(current_user_ns(), arg);
3105                 if (!gid_valid(group)) {
3106                         ret = -EINVAL;
3107                         break;
3108                 }
3109                 tun->group = group;
3110                 do_notify = true;
3111                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3112                           from_kgid(&init_user_ns, tun->group));
3113                 break;
3114
3115         case TUNSETLINK:
3116                 /* Only allow setting the type when the interface is down */
3117                 if (tun->dev->flags & IFF_UP) {
3118                         tun_debug(KERN_INFO, tun,
3119                                   "Linktype set failed because interface is up\n");
3120                         ret = -EBUSY;
3121                 } else {
3122                         tun->dev->type = (int) arg;
3123                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3124                                   tun->dev->type);
3125                         ret = 0;
3126                 }
3127                 break;
3128
3129 #ifdef TUN_DEBUG
3130         case TUNSETDEBUG:
3131                 tun->debug = arg;
3132                 break;
3133 #endif
3134         case TUNSETOFFLOAD:
3135                 ret = set_offload(tun, arg);
3136                 break;
3137
3138         case TUNSETTXFILTER:
3139                 /* Can be set only for TAPs */
3140                 ret = -EINVAL;
3141                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3142                         break;
3143                 ret = update_filter(&tun->txflt, (void __user *)arg);
3144                 break;
3145
3146         case SIOCGIFHWADDR:
3147                 /* Get hw address */
3148                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3149                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3150                 if (copy_to_user(argp, &ifr, ifreq_len))
3151                         ret = -EFAULT;
3152                 break;
3153
3154         case SIOCSIFHWADDR:
3155                 /* Set hw address */
3156                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3157                           ifr.ifr_hwaddr.sa_data);
3158
3159                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3160                 break;
3161
3162         case TUNGETSNDBUF:
3163                 sndbuf = tfile->socket.sk->sk_sndbuf;
3164                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3165                         ret = -EFAULT;
3166                 break;
3167
3168         case TUNSETSNDBUF:
3169                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3170                         ret = -EFAULT;
3171                         break;
3172                 }
3173                 if (sndbuf <= 0) {
3174                         ret = -EINVAL;
3175                         break;
3176                 }
3177
3178                 tun->sndbuf = sndbuf;
3179                 tun_set_sndbuf(tun);
3180                 break;
3181
3182         case TUNGETVNETHDRSZ:
3183                 vnet_hdr_sz = tun->vnet_hdr_sz;
3184                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3185                         ret = -EFAULT;
3186                 break;
3187
3188         case TUNSETVNETHDRSZ:
3189                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3190                         ret = -EFAULT;
3191                         break;
3192                 }
3193                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3194                         ret = -EINVAL;
3195                         break;
3196                 }
3197
3198                 tun->vnet_hdr_sz = vnet_hdr_sz;
3199                 break;
3200
3201         case TUNGETVNETLE:
3202                 le = !!(tun->flags & TUN_VNET_LE);
3203                 if (put_user(le, (int __user *)argp))
3204                         ret = -EFAULT;
3205                 break;
3206
3207         case TUNSETVNETLE:
3208                 if (get_user(le, (int __user *)argp)) {
3209                         ret = -EFAULT;
3210                         break;
3211                 }
3212                 if (le)
3213                         tun->flags |= TUN_VNET_LE;
3214                 else
3215                         tun->flags &= ~TUN_VNET_LE;
3216                 break;
3217
3218         case TUNGETVNETBE:
3219                 ret = tun_get_vnet_be(tun, argp);
3220                 break;
3221
3222         case TUNSETVNETBE:
3223                 ret = tun_set_vnet_be(tun, argp);
3224                 break;
3225
3226         case TUNATTACHFILTER:
3227                 /* Can be set only for TAPs */
3228                 ret = -EINVAL;
3229                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3230                         break;
3231                 ret = -EFAULT;
3232                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3233                         break;
3234
3235                 ret = tun_attach_filter(tun);
3236                 break;
3237
3238         case TUNDETACHFILTER:
3239                 /* Can be set only for TAPs */
3240                 ret = -EINVAL;
3241                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3242                         break;
3243                 ret = 0;
3244                 tun_detach_filter(tun, tun->numqueues);
3245                 break;
3246
3247         case TUNGETFILTER:
3248                 ret = -EINVAL;
3249                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3250                         break;
3251                 ret = -EFAULT;
3252                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3253                         break;
3254                 ret = 0;
3255                 break;
3256
3257         case TUNSETSTEERINGEBPF:
3258                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3259                 break;
3260
3261         case TUNSETFILTEREBPF:
3262                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3263                 break;
3264
3265         default:
3266                 ret = -EINVAL;
3267                 break;
3268         }
3269
3270         if (do_notify)
3271                 netdev_state_change(tun->dev);
3272
3273 unlock:
3274         rtnl_unlock();
3275         if (tun)
3276                 tun_put(tun);
3277         return ret;
3278 }
3279
3280 static long tun_chr_ioctl(struct file *file,
3281                           unsigned int cmd, unsigned long arg)
3282 {
3283         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3284 }
3285
3286 #ifdef CONFIG_COMPAT
3287 static long tun_chr_compat_ioctl(struct file *file,
3288                          unsigned int cmd, unsigned long arg)
3289 {
3290         switch (cmd) {
3291         case TUNSETIFF:
3292         case TUNGETIFF:
3293         case TUNSETTXFILTER:
3294         case TUNGETSNDBUF:
3295         case TUNSETSNDBUF:
3296         case SIOCGIFHWADDR:
3297         case SIOCSIFHWADDR:
3298                 arg = (unsigned long)compat_ptr(arg);
3299                 break;
3300         default:
3301                 arg = (compat_ulong_t)arg;
3302                 break;
3303         }
3304
3305         /*
3306          * compat_ifreq is shorter than ifreq, so we must not access beyond
3307          * the end of that structure. All fields that are used in this
3308          * driver are compatible though, we don't need to convert the
3309          * contents.
3310          */
3311         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3312 }
3313 #endif /* CONFIG_COMPAT */
3314
3315 static int tun_chr_fasync(int fd, struct file *file, int on)
3316 {
3317         struct tun_file *tfile = file->private_data;
3318         int ret;
3319
3320         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3321                 goto out;
3322
3323         if (on) {
3324                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3325                 tfile->flags |= TUN_FASYNC;
3326         } else
3327                 tfile->flags &= ~TUN_FASYNC;
3328         ret = 0;
3329 out:
3330         return ret;
3331 }
3332
3333 static int tun_chr_open(struct inode *inode, struct file * file)
3334 {
3335         struct net *net = current->nsproxy->net_ns;
3336         struct tun_file *tfile;
3337
3338         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3339
3340         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3341                                             &tun_proto, 0);
3342         if (!tfile)
3343                 return -ENOMEM;
3344         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3345                 sk_free(&tfile->sk);
3346                 return -ENOMEM;
3347         }
3348
3349         mutex_init(&tfile->napi_mutex);
3350         RCU_INIT_POINTER(tfile->tun, NULL);
3351         tfile->flags = 0;
3352         tfile->ifindex = 0;
3353
3354         init_waitqueue_head(&tfile->wq.wait);
3355         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3356
3357         tfile->socket.file = file;
3358         tfile->socket.ops = &tun_socket_ops;
3359
3360         sock_init_data(&tfile->socket, &tfile->sk);
3361
3362         tfile->sk.sk_write_space = tun_sock_write_space;
3363         tfile->sk.sk_sndbuf = INT_MAX;
3364
3365         file->private_data = tfile;
3366         INIT_LIST_HEAD(&tfile->next);
3367
3368         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3369
3370         return 0;
3371 }
3372
3373 static int tun_chr_close(struct inode *inode, struct file *file)
3374 {
3375         struct tun_file *tfile = file->private_data;
3376
3377         tun_detach(tfile, true);
3378
3379         return 0;
3380 }
3381
3382 #ifdef CONFIG_PROC_FS
3383 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3384 {
3385         struct tun_file *tfile = file->private_data;
3386         struct tun_struct *tun;
3387         struct ifreq ifr;
3388
3389         memset(&ifr, 0, sizeof(ifr));
3390
3391         rtnl_lock();
3392         tun = tun_get(tfile);
3393         if (tun)
3394                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3395         rtnl_unlock();
3396
3397         if (tun)
3398                 tun_put(tun);
3399
3400         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3401 }
3402 #endif
3403
3404 static const struct file_operations tun_fops = {
3405         .owner  = THIS_MODULE,
3406         .llseek = no_llseek,
3407         .read_iter  = tun_chr_read_iter,
3408         .write_iter = tun_chr_write_iter,
3409         .poll   = tun_chr_poll,
3410         .unlocked_ioctl = tun_chr_ioctl,
3411 #ifdef CONFIG_COMPAT
3412         .compat_ioctl = tun_chr_compat_ioctl,
3413 #endif
3414         .open   = tun_chr_open,
3415         .release = tun_chr_close,
3416         .fasync = tun_chr_fasync,
3417 #ifdef CONFIG_PROC_FS
3418         .show_fdinfo = tun_chr_show_fdinfo,
3419 #endif
3420 };
3421
3422 static struct miscdevice tun_miscdev = {
3423         .minor = TUN_MINOR,
3424         .name = "tun",
3425         .nodename = "net/tun",
3426         .fops = &tun_fops,
3427 };
3428
3429 /* ethtool interface */
3430
3431 static void tun_default_link_ksettings(struct net_device *dev,
3432                                        struct ethtool_link_ksettings *cmd)
3433 {
3434         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3435         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3436         cmd->base.speed         = SPEED_10;
3437         cmd->base.duplex        = DUPLEX_FULL;
3438         cmd->base.port          = PORT_TP;
3439         cmd->base.phy_address   = 0;
3440         cmd->base.autoneg       = AUTONEG_DISABLE;
3441 }
3442
3443 static int tun_get_link_ksettings(struct net_device *dev,
3444                                   struct ethtool_link_ksettings *cmd)
3445 {
3446         struct tun_struct *tun = netdev_priv(dev);
3447
3448         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3449         return 0;
3450 }
3451
3452 static int tun_set_link_ksettings(struct net_device *dev,
3453                                   const struct ethtool_link_ksettings *cmd)
3454 {
3455         struct tun_struct *tun = netdev_priv(dev);
3456
3457         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3458         return 0;
3459 }
3460
3461 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3462 {
3463         struct tun_struct *tun = netdev_priv(dev);
3464
3465         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3466         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3467
3468         switch (tun->flags & TUN_TYPE_MASK) {
3469         case IFF_TUN:
3470                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3471                 break;
3472         case IFF_TAP:
3473                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3474                 break;
3475         }
3476 }
3477
3478 static u32 tun_get_msglevel(struct net_device *dev)
3479 {
3480 #ifdef TUN_DEBUG
3481         struct tun_struct *tun = netdev_priv(dev);
3482         return tun->debug;
3483 #else
3484         return -EOPNOTSUPP;
3485 #endif
3486 }
3487
3488 static void tun_set_msglevel(struct net_device *dev, u32 value)
3489 {
3490 #ifdef TUN_DEBUG
3491         struct tun_struct *tun = netdev_priv(dev);
3492         tun->debug = value;
3493 #endif
3494 }
3495
3496 static int tun_get_coalesce(struct net_device *dev,
3497                             struct ethtool_coalesce *ec)
3498 {
3499         struct tun_struct *tun = netdev_priv(dev);
3500
3501         ec->rx_max_coalesced_frames = tun->rx_batched;
3502
3503         return 0;
3504 }
3505
3506 static int tun_set_coalesce(struct net_device *dev,
3507                             struct ethtool_coalesce *ec)
3508 {
3509         struct tun_struct *tun = netdev_priv(dev);
3510
3511         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3512                 tun->rx_batched = NAPI_POLL_WEIGHT;
3513         else
3514                 tun->rx_batched = ec->rx_max_coalesced_frames;
3515
3516         return 0;
3517 }
3518
3519 static const struct ethtool_ops tun_ethtool_ops = {
3520         .get_drvinfo    = tun_get_drvinfo,
3521         .get_msglevel   = tun_get_msglevel,
3522         .set_msglevel   = tun_set_msglevel,
3523         .get_link       = ethtool_op_get_link,
3524         .get_ts_info    = ethtool_op_get_ts_info,
3525         .get_coalesce   = tun_get_coalesce,
3526         .set_coalesce   = tun_set_coalesce,
3527         .get_link_ksettings = tun_get_link_ksettings,
3528         .set_link_ksettings = tun_set_link_ksettings,
3529 };
3530
3531 static int tun_queue_resize(struct tun_struct *tun)
3532 {
3533         struct net_device *dev = tun->dev;
3534         struct tun_file *tfile;
3535         struct ptr_ring **rings;
3536         int n = tun->numqueues + tun->numdisabled;
3537         int ret, i;
3538
3539         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3540         if (!rings)
3541                 return -ENOMEM;
3542
3543         for (i = 0; i < tun->numqueues; i++) {
3544                 tfile = rtnl_dereference(tun->tfiles[i]);
3545                 rings[i] = &tfile->tx_ring;
3546         }
3547         list_for_each_entry(tfile, &tun->disabled, next)
3548                 rings[i++] = &tfile->tx_ring;
3549
3550         ret = ptr_ring_resize_multiple(rings, n,
3551                                        dev->tx_queue_len, GFP_KERNEL,
3552                                        tun_ptr_free);
3553
3554         kfree(rings);
3555         return ret;
3556 }
3557
3558 static int tun_device_event(struct notifier_block *unused,
3559                             unsigned long event, void *ptr)
3560 {
3561         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3562         struct tun_struct *tun = netdev_priv(dev);
3563
3564         if (dev->rtnl_link_ops != &tun_link_ops)
3565                 return NOTIFY_DONE;
3566
3567         switch (event) {
3568         case NETDEV_CHANGE_TX_QUEUE_LEN:
3569                 if (tun_queue_resize(tun))
3570                         return NOTIFY_BAD;
3571                 break;
3572         default:
3573                 break;
3574         }
3575
3576         return NOTIFY_DONE;
3577 }
3578
3579 static struct notifier_block tun_notifier_block __read_mostly = {
3580         .notifier_call  = tun_device_event,
3581 };
3582
3583 static int __init tun_init(void)
3584 {
3585         int ret = 0;
3586
3587         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3588
3589         ret = rtnl_link_register(&tun_link_ops);
3590         if (ret) {
3591                 pr_err("Can't register link_ops\n");
3592                 goto err_linkops;
3593         }
3594
3595         ret = misc_register(&tun_miscdev);
3596         if (ret) {
3597                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3598                 goto err_misc;
3599         }
3600
3601         ret = register_netdevice_notifier(&tun_notifier_block);
3602         if (ret) {
3603                 pr_err("Can't register netdevice notifier\n");
3604                 goto err_notifier;
3605         }
3606
3607         return  0;
3608
3609 err_notifier:
3610         misc_deregister(&tun_miscdev);
3611 err_misc:
3612         rtnl_link_unregister(&tun_link_ops);
3613 err_linkops:
3614         return ret;
3615 }
3616
3617 static void tun_cleanup(void)
3618 {
3619         misc_deregister(&tun_miscdev);
3620         rtnl_link_unregister(&tun_link_ops);
3621         unregister_netdevice_notifier(&tun_notifier_block);
3622 }
3623
3624 /* Get an underlying socket object from tun file.  Returns error unless file is
3625  * attached to a device.  The returned object works like a packet socket, it
3626  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3627  * holding a reference to the file for as long as the socket is in use. */
3628 struct socket *tun_get_socket(struct file *file)
3629 {
3630         struct tun_file *tfile;
3631         if (file->f_op != &tun_fops)
3632                 return ERR_PTR(-EINVAL);
3633         tfile = file->private_data;
3634         if (!tfile)
3635                 return ERR_PTR(-EBADFD);
3636         return &tfile->socket;
3637 }
3638 EXPORT_SYMBOL_GPL(tun_get_socket);
3639
3640 struct ptr_ring *tun_get_tx_ring(struct file *file)
3641 {
3642         struct tun_file *tfile;
3643
3644         if (file->f_op != &tun_fops)
3645                 return ERR_PTR(-EINVAL);
3646         tfile = file->private_data;
3647         if (!tfile)
3648                 return ERR_PTR(-EBADFD);
3649         return &tfile->tx_ring;
3650 }
3651 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3652
3653 module_init(tun_init);
3654 module_exit(tun_cleanup);
3655 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3656 MODULE_AUTHOR(DRV_COPYRIGHT);
3657 MODULE_LICENSE("GPL");
3658 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3659 MODULE_ALIAS("devname:net/tun");