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_page {
 192         struct page *page;
 193         int count;
 194 };
 195
 196 struct tun_flow_entry {
 197         struct hlist_node hash_link;
 198         struct rcu_head rcu;
 199         struct tun_struct *tun;
 200
 201         u32 rxhash;
 202         u32 rps_rxhash;
 203         int queue_index;
 204         unsigned long updated ____cacheline_aligned_in_smp;
 205 };
 206
 207 #define TUN_NUM_FLOW_ENTRIES 1024
 208 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 209
 210 struct tun_prog {
 211         struct rcu_head rcu;
 212         struct bpf_prog *prog;
 213 };
 214
 215 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 216  * device, socket filter, sndbuf and vnet header size were restore when the
 217  * file were attached to a persist device.
 218  */
 219 struct tun_struct {
 220         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 221         unsigned int            numqueues;
 222         unsigned int            flags;
 223         kuid_t                  owner;
 224         kgid_t                  group;
 225
 226         struct net_device       *dev;
 227         netdev_features_t       set_features;
 228 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 229                           NETIF_F_TSO6)
 230
 231         int                     align;
 232         int                     vnet_hdr_sz;
 233         int                     sndbuf;
 234         struct tap_filter       txflt;
 235         struct sock_fprog       fprog;
 236         /* protected by rtnl lock */
 237         bool                    filter_attached;
 238 #ifdef TUN_DEBUG
 239         int debug;
 240 #endif
 241         spinlock_t lock;
 242         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 243         struct timer_list flow_gc_timer;
 244         unsigned long ageing_time;
 245         unsigned int numdisabled;
 246         struct list_head disabled;
 247         void *security;
 248         u32 flow_count;
 249         u32 rx_batched;
 250         struct tun_pcpu_stats __percpu *pcpu_stats;
 251         struct bpf_prog __rcu *xdp_prog;
 252         struct tun_prog __rcu *steering_prog;
 253         struct tun_prog __rcu *filter_prog;
 254         struct ethtool_link_ksettings link_ksettings;
 255 };
 256
 257 struct veth {
 258         __be16 h_vlan_proto;
 259         __be16 h_vlan_TCI;
 260 };
 261
 262 bool tun_is_xdp_frame(void *ptr)
 263 {
 264         return (unsigned long)ptr & TUN_XDP_FLAG;
 265 }
 266 EXPORT_SYMBOL(tun_is_xdp_frame);
 267
 268 void *tun_xdp_to_ptr(void *ptr)
 269 {
 270         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 271 }
 272 EXPORT_SYMBOL(tun_xdp_to_ptr);
 273
 274 void *tun_ptr_to_xdp(void *ptr)
 275 {
 276         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 277 }
 278 EXPORT_SYMBOL(tun_ptr_to_xdp);
 279
 280 static int tun_napi_receive(struct napi_struct *napi, int budget)
 281 {
 282         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 283         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 284         struct sk_buff_head process_queue;
 285         struct sk_buff *skb;
 286         int received = 0;
 287
 288         __skb_queue_head_init(&process_queue);
 289
 290         spin_lock(&queue->lock);
 291         skb_queue_splice_tail_init(queue, &process_queue);
 292         spin_unlock(&queue->lock);
 293
 294         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 295                 napi_gro_receive(napi, skb);
 296                 ++received;
 297         }
 298
 299         if (!skb_queue_empty(&process_queue)) {
 300                 spin_lock(&queue->lock);
 301                 skb_queue_splice(&process_queue, queue);
 302                 spin_unlock(&queue->lock);
 303         }
 304
 305         return received;
 306 }
 307
 308 static int tun_napi_poll(struct napi_struct *napi, int budget)
 309 {
 310         unsigned int received;
 311
 312         received = tun_napi_receive(napi, budget);
 313
 314         if (received < budget)
 315                 napi_complete_done(napi, received);
 316
 317         return received;
 318 }
 319
 320 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 321                           bool napi_en, bool napi_frags)
 322 {
 323         tfile->napi_enabled = napi_en;
 324         tfile->napi_frags_enabled = napi_en && napi_frags;
 325         if (napi_en) {
 326                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 327                                NAPI_POLL_WEIGHT);
 328                 napi_enable(&tfile->napi);
 329         }
 330 }
 331
 332 static void tun_napi_disable(struct tun_file *tfile)
 333 {
 334         if (tfile->napi_enabled)
 335                 napi_disable(&tfile->napi);
 336 }
 337
 338 static void tun_napi_del(struct tun_file *tfile)
 339 {
 340         if (tfile->napi_enabled)
 341                 netif_napi_del(&tfile->napi);
 342 }
 343
 344 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 345 {
 346         return tfile->napi_frags_enabled;
 347 }
 348
 349 #ifdef CONFIG_TUN_VNET_CROSS_LE
 350 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 351 {
 352         return tun->flags & TUN_VNET_BE ? false :
 353                 virtio_legacy_is_little_endian();
 354 }
 355
 356 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 357 {
 358         int be = !!(tun->flags & TUN_VNET_BE);
 359
 360         if (put_user(be, argp))
 361                 return -EFAULT;
 362
 363         return 0;
 364 }
 365
 366 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 367 {
 368         int be;
 369
 370         if (get_user(be, argp))
 371                 return -EFAULT;
 372
 373         if (be)
 374                 tun->flags |= TUN_VNET_BE;
 375         else
 376                 tun->flags &= ~TUN_VNET_BE;
 377
 378         return 0;
 379 }
 380 #else
 381 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 382 {
 383         return virtio_legacy_is_little_endian();
 384 }
 385
 386 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 387 {
 388         return -EINVAL;
 389 }
 390
 391 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 392 {
 393         return -EINVAL;
 394 }
 395 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 396
 397 static inline bool tun_is_little_endian(struct tun_struct *tun)
 398 {
 399         return tun->flags & TUN_VNET_LE ||
 400                 tun_legacy_is_little_endian(tun);
 401 }
 402
 403 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 404 {
 405         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 406 }
 407
 408 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 409 {
 410         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 411 }
 412
 413 static inline u32 tun_hashfn(u32 rxhash)
 414 {
 415         return rxhash & TUN_MASK_FLOW_ENTRIES;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 419 {
 420         struct tun_flow_entry *e;
 421
 422         hlist_for_each_entry_rcu(e, head, hash_link) {
 423                 if (e->rxhash == rxhash)
 424                         return e;
 425         }
 426         return NULL;
 427 }
 428
 429 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 430                                               struct hlist_head *head,
 431                                               u32 rxhash, u16 queue_index)
 432 {
 433         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 434
 435         if (e) {
 436                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 437                           rxhash, queue_index);
 438                 e->updated = jiffies;
 439                 e->rxhash = rxhash;
 440                 e->rps_rxhash = 0;
 441                 e->queue_index = queue_index;
 442                 e->tun = tun;
 443                 hlist_add_head_rcu(&e->hash_link, head);
 444                 ++tun->flow_count;
 445         }
 446         return e;
 447 }
 448
 449 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 450 {
 451         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 452                   e->rxhash, e->queue_index);
 453         hlist_del_rcu(&e->hash_link);
 454         kfree_rcu(e, rcu);
 455         --tun->flow_count;
 456 }
 457
 458 static void tun_flow_flush(struct tun_struct *tun)
 459 {
 460         int i;
 461
 462         spin_lock_bh(&tun->lock);
 463         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 464                 struct tun_flow_entry *e;
 465                 struct hlist_node *n;
 466
 467                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 468                         tun_flow_delete(tun, e);
 469         }
 470         spin_unlock_bh(&tun->lock);
 471 }
 472
 473 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 474 {
 475         int i;
 476
 477         spin_lock_bh(&tun->lock);
 478         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 479                 struct tun_flow_entry *e;
 480                 struct hlist_node *n;
 481
 482                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 483                         if (e->queue_index == queue_index)
 484                                 tun_flow_delete(tun, e);
 485                 }
 486         }
 487         spin_unlock_bh(&tun->lock);
 488 }
 489
 490 static void tun_flow_cleanup(struct timer_list *t)
 491 {
 492         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 493         unsigned long delay = tun->ageing_time;
 494         unsigned long next_timer = jiffies + delay;
 495         unsigned long count = 0;
 496         int i;
 497
 498         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 499
 500         spin_lock(&tun->lock);
 501         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 502                 struct tun_flow_entry *e;
 503                 struct hlist_node *n;
 504
 505                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 506                         unsigned long this_timer;
 507
 508                         this_timer = e->updated + delay;
 509                         if (time_before_eq(this_timer, jiffies)) {
 510                                 tun_flow_delete(tun, e);
 511                                 continue;
 512                         }
 513                         count++;
 514                         if (time_before(this_timer, next_timer))
 515                                 next_timer = this_timer;
 516                 }
 517         }
 518
 519         if (count)
 520                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 521         spin_unlock(&tun->lock);
 522 }
 523
 524 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 525                             struct tun_file *tfile)
 526 {
 527         struct hlist_head *head;
 528         struct tun_flow_entry *e;
 529         unsigned long delay = tun->ageing_time;
 530         u16 queue_index = tfile->queue_index;
 531
 532         head = &tun->flows[tun_hashfn(rxhash)];
 533
 534         rcu_read_lock();
 535
 536         e = tun_flow_find(head, rxhash);
 537         if (likely(e)) {
 538                 /* TODO: keep queueing to old queue until it's empty? */
 539                 if (e->queue_index != queue_index)
 540                         e->queue_index = queue_index;
 541                 if (e->updated != jiffies)
 542                         e->updated = jiffies;
 543                 sock_rps_record_flow_hash(e->rps_rxhash);
 544         } else {
 545                 spin_lock_bh(&tun->lock);
 546                 if (!tun_flow_find(head, rxhash) &&
 547                     tun->flow_count < MAX_TAP_FLOWS)
 548                         tun_flow_create(tun, head, rxhash, queue_index);
 549
 550                 if (!timer_pending(&tun->flow_gc_timer))
 551                         mod_timer(&tun->flow_gc_timer,
 552                                   round_jiffies_up(jiffies + delay));
 553                 spin_unlock_bh(&tun->lock);
 554         }
 555
 556         rcu_read_unlock();
 557 }
 558
 559 /**
 560  * Save the hash received in the stack receive path and update the
 561  * flow_hash table accordingly.
 562  */
 563 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 564 {
 565         if (unlikely(e->rps_rxhash != hash))
 566                 e->rps_rxhash = hash;
 567 }
 568
 569 /* We try to identify a flow through its rxhash. The reason that
 570  * we do not check rxq no. is because some cards(e.g 82599), chooses
 571  * the rxq based on the txq where the last packet of the flow comes. As
 572  * the userspace application move between processors, we may get a
 573  * different rxq no. here.
 574  */
 575 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 576 {
 577         struct tun_flow_entry *e;
 578         u32 txq = 0;
 579         u32 numqueues = 0;
 580
 581         numqueues = READ_ONCE(tun->numqueues);
 582
 583         txq = __skb_get_hash_symmetric(skb);
 584         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 585         if (e) {
 586                 tun_flow_save_rps_rxhash(e, txq);
 587                 txq = e->queue_index;
 588         } else {
 589                 /* use multiply and shift instead of expensive divide */
 590                 txq = ((u64)txq * numqueues) >> 32;
 591         }
 592
 593         return txq;
 594 }
 595
 596 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 597 {
 598         struct tun_prog *prog;
 599         u16 ret = 0;
 600
 601         prog = rcu_dereference(tun->steering_prog);
 602         if (prog)
 603                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 604
 605         return ret % tun->numqueues;
 606 }
 607
 608 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 609                             struct net_device *sb_dev)
 610 {
 611         struct tun_struct *tun = netdev_priv(dev);
 612         u16 ret;
 613
 614         rcu_read_lock();
 615         if (rcu_dereference(tun->steering_prog))
 616                 ret = tun_ebpf_select_queue(tun, skb);
 617         else
 618                 ret = tun_automq_select_queue(tun, skb);
 619         rcu_read_unlock();
 620
 621         return ret;
 622 }
 623
 624 static inline bool tun_not_capable(struct tun_struct *tun)
 625 {
 626         const struct cred *cred = current_cred();
 627         struct net *net = dev_net(tun->dev);
 628
 629         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 630                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 631                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 632 }
 633
 634 static void tun_set_real_num_queues(struct tun_struct *tun)
 635 {
 636         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 637         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 638 }
 639
 640 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 641 {
 642         tfile->detached = tun;
 643         list_add_tail(&tfile->next, &tun->disabled);
 644         ++tun->numdisabled;
 645 }
 646
 647 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 648 {
 649         struct tun_struct *tun = tfile->detached;
 650
 651         tfile->detached = NULL;
 652         list_del_init(&tfile->next);
 653         --tun->numdisabled;
 654         return tun;
 655 }
 656
 657 void tun_ptr_free(void *ptr)
 658 {
 659         if (!ptr)
 660                 return;
 661         if (tun_is_xdp_frame(ptr)) {
 662                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 663
 664                 xdp_return_frame(xdpf);
 665         } else {
 666                 __skb_array_destroy_skb(ptr);
 667         }
 668 }
 669 EXPORT_SYMBOL_GPL(tun_ptr_free);
 670
 671 static void tun_queue_purge(struct tun_file *tfile)
 672 {
 673         void *ptr;
 674
 675         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 676                 tun_ptr_free(ptr);
 677
 678         skb_queue_purge(&tfile->sk.sk_write_queue);
 679         skb_queue_purge(&tfile->sk.sk_error_queue);
 680 }
 681
 682 static void __tun_detach(struct tun_file *tfile, bool clean)
 683 {
 684         struct tun_file *ntfile;
 685         struct tun_struct *tun;
 686
 687         tun = rtnl_dereference(tfile->tun);
 688
 689         if (tun && clean) {
 690                 tun_napi_disable(tfile);
 691                 tun_napi_del(tfile);
 692         }
 693
 694         if (tun && !tfile->detached) {
 695                 u16 index = tfile->queue_index;
 696                 BUG_ON(index >= tun->numqueues);
 697
 698                 rcu_assign_pointer(tun->tfiles[index],
 699                                    tun->tfiles[tun->numqueues - 1]);
 700                 ntfile = rtnl_dereference(tun->tfiles[index]);
 701                 ntfile->queue_index = index;
 702
 703                 --tun->numqueues;
 704                 if (clean) {
 705                         RCU_INIT_POINTER(tfile->tun, NULL);
 706                         sock_put(&tfile->sk);
 707                 } else
 708                         tun_disable_queue(tun, tfile);
 709
 710                 synchronize_net();
 711                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 712                 /* Drop read queue */
 713                 tun_queue_purge(tfile);
 714                 tun_set_real_num_queues(tun);
 715         } else if (tfile->detached && clean) {
 716                 tun = tun_enable_queue(tfile);
 717                 sock_put(&tfile->sk);
 718         }
 719
 720         if (clean) {
 721                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 722                         netif_carrier_off(tun->dev);
 723
 724                         if (!(tun->flags & IFF_PERSIST) &&
 725                             tun->dev->reg_state == NETREG_REGISTERED)
 726                                 unregister_netdevice(tun->dev);
 727                 }
 728                 if (tun)
 729                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 730                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 731                 sock_put(&tfile->sk);
 732         }
 733 }
 734
 735 static void tun_detach(struct tun_file *tfile, bool clean)
 736 {
 737         struct tun_struct *tun;
 738         struct net_device *dev;
 739
 740         rtnl_lock();
 741         tun = rtnl_dereference(tfile->tun);
 742         dev = tun ? tun->dev : NULL;
 743         __tun_detach(tfile, clean);
 744         if (dev)
 745                 netdev_state_change(dev);
 746         rtnl_unlock();
 747 }
 748
 749 static void tun_detach_all(struct net_device *dev)
 750 {
 751         struct tun_struct *tun = netdev_priv(dev);
 752         struct tun_file *tfile, *tmp;
 753         int i, n = tun->numqueues;
 754
 755         for (i = 0; i < n; i++) {
 756                 tfile = rtnl_dereference(tun->tfiles[i]);
 757                 BUG_ON(!tfile);
 758                 tun_napi_disable(tfile);
 759                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 760                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 761                 RCU_INIT_POINTER(tfile->tun, NULL);
 762                 --tun->numqueues;
 763         }
 764         list_for_each_entry(tfile, &tun->disabled, next) {
 765                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 766                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 767                 RCU_INIT_POINTER(tfile->tun, NULL);
 768         }
 769         BUG_ON(tun->numqueues != 0);
 770
 771         synchronize_net();
 772         for (i = 0; i < n; i++) {
 773                 tfile = rtnl_dereference(tun->tfiles[i]);
 774                 tun_napi_del(tfile);
 775                 /* Drop read queue */
 776                 tun_queue_purge(tfile);
 777                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 778                 sock_put(&tfile->sk);
 779         }
 780         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 781                 tun_enable_queue(tfile);
 782                 tun_queue_purge(tfile);
 783                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 784                 sock_put(&tfile->sk);
 785         }
 786         BUG_ON(tun->numdisabled != 0);
 787
 788         if (tun->flags & IFF_PERSIST)
 789                 module_put(THIS_MODULE);
 790 }
 791
 792 static int tun_attach(struct tun_struct *tun, struct file *file,
 793                       bool skip_filter, bool napi, bool napi_frags)
 794 {
 795         struct tun_file *tfile = file->private_data;
 796         struct net_device *dev = tun->dev;
 797         int err;
 798
 799         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 800         if (err < 0)
 801                 goto out;
 802
 803         err = -EINVAL;
 804         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 805                 goto out;
 806
 807         err = -EBUSY;
 808         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 809                 goto out;
 810
 811         err = -E2BIG;
 812         if (!tfile->detached &&
 813             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 814                 goto out;
 815
 816         err = 0;
 817
 818         /* Re-attach the filter to persist device */
 819         if (!skip_filter && (tun->filter_attached == true)) {
 820                 lock_sock(tfile->socket.sk);
 821                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 822                 release_sock(tfile->socket.sk);
 823                 if (!err)
 824                         goto out;
 825         }
 826
 827         if (!tfile->detached &&
 828             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 829                             GFP_KERNEL, tun_ptr_free)) {
 830                 err = -ENOMEM;
 831                 goto out;
 832         }
 833
 834         tfile->queue_index = tun->numqueues;
 835         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 836
 837         if (tfile->detached) {
 838                 /* Re-attach detached tfile, updating XDP queue_index */
 839                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 840
 841                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 842                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 843         } else {
 844                 /* Setup XDP RX-queue info, for new tfile getting attached */
 845                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 846                                        tun->dev, tfile->queue_index);
 847                 if (err < 0)
 848                         goto out;
 849                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 850                                                  MEM_TYPE_PAGE_SHARED, NULL);
 851                 if (err < 0) {
 852                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 853                         goto out;
 854                 }
 855                 err = 0;
 856         }
 857
 858         if (tfile->detached) {
 859                 tun_enable_queue(tfile);
 860         } else {
 861                 sock_hold(&tfile->sk);
 862                 tun_napi_init(tun, tfile, napi, napi_frags);
 863         }
 864
 865         if (rtnl_dereference(tun->xdp_prog))
 866                 sock_set_flag(&tfile->sk, SOCK_XDP);
 867
 868         /* device is allowed to go away first, so no need to hold extra
 869          * refcnt.
 870          */
 871
 872         /* Publish tfile->tun and tun->tfiles only after we've fully
 873          * initialized tfile; otherwise we risk using half-initialized
 874          * object.
 875          */
 876         rcu_assign_pointer(tfile->tun, tun);
 877         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 878         tun->numqueues++;
 879         tun_set_real_num_queues(tun);
 880 out:
 881         return err;
 882 }
 883
 884 static struct tun_struct *tun_get(struct tun_file *tfile)
 885 {
 886         struct tun_struct *tun;
 887
 888         rcu_read_lock();
 889         tun = rcu_dereference(tfile->tun);
 890         if (tun)
 891                 dev_hold(tun->dev);
 892         rcu_read_unlock();
 893
 894         return tun;
 895 }
 896
 897 static void tun_put(struct tun_struct *tun)
 898 {
 899         dev_put(tun->dev);
 900 }
 901
 902 /* TAP filtering */
 903 static void addr_hash_set(u32 *mask, const u8 *addr)
 904 {
 905         int n = ether_crc(ETH_ALEN, addr) >> 26;
 906         mask[n >> 5] |= (1 << (n & 31));
 907 }
 908
 909 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 910 {
 911         int n = ether_crc(ETH_ALEN, addr) >> 26;
 912         return mask[n >> 5] & (1 << (n & 31));
 913 }
 914
 915 static int update_filter(struct tap_filter *filter, void __user *arg)
 916 {
 917         struct { u8 u[ETH_ALEN]; } *addr;
 918         struct tun_filter uf;
 919         int err, alen, n, nexact;
 920
 921         if (copy_from_user(&uf, arg, sizeof(uf)))
 922                 return -EFAULT;
 923
 924         if (!uf.count) {
 925                 /* Disabled */
 926                 filter->count = 0;
 927                 return 0;
 928         }
 929
 930         alen = ETH_ALEN * uf.count;
 931         addr = memdup_user(arg + sizeof(uf), alen);
 932         if (IS_ERR(addr))
 933                 return PTR_ERR(addr);
 934
 935         /* The filter is updated without holding any locks. Which is
 936          * perfectly safe. We disable it first and in the worst
 937          * case we'll accept a few undesired packets. */
 938         filter->count = 0;
 939         wmb();
 940
 941         /* Use first set of addresses as an exact filter */
 942         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 943                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 944
 945         nexact = n;
 946
 947         /* Remaining multicast addresses are hashed,
 948          * unicast will leave the filter disabled. */
 949         memset(filter->mask, 0, sizeof(filter->mask));
 950         for (; n < uf.count; n++) {
 951                 if (!is_multicast_ether_addr(addr[n].u)) {
 952                         err = 0; /* no filter */
 953                         goto free_addr;
 954                 }
 955                 addr_hash_set(filter->mask, addr[n].u);
 956         }
 957
 958         /* For ALLMULTI just set the mask to all ones.
 959          * This overrides the mask populated above. */
 960         if ((uf.flags & TUN_FLT_ALLMULTI))
 961                 memset(filter->mask, ~0, sizeof(filter->mask));
 962
 963         /* Now enable the filter */
 964         wmb();
 965         filter->count = nexact;
 966
 967         /* Return the number of exact filters */
 968         err = nexact;
 969 free_addr:
 970         kfree(addr);
 971         return err;
 972 }
 973
 974 /* Returns: 0 - drop, !=0 - accept */
 975 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 976 {
 977         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 978          * at this point. */
 979         struct ethhdr *eh = (struct ethhdr *) skb->data;
 980         int i;
 981
 982         /* Exact match */
 983         for (i = 0; i < filter->count; i++)
 984                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 985                         return 1;
 986
 987         /* Inexact match (multicast only) */
 988         if (is_multicast_ether_addr(eh->h_dest))
 989                 return addr_hash_test(filter->mask, eh->h_dest);
 990
 991         return 0;
 992 }
 993
 994 /*
 995  * Checks whether the packet is accepted or not.
 996  * Returns: 0 - drop, !=0 - accept
 997  */
 998 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 999 {
1000         if (!filter->count)
1001                 return 1;
1002
1003         return run_filter(filter, skb);
1004 }
1005
1006 /* Network device part of the driver */
1007
1008 static const struct ethtool_ops tun_ethtool_ops;
1009
1010 /* Net device detach from fd. */
1011 static void tun_net_uninit(struct net_device *dev)
1012 {
1013         tun_detach_all(dev);
1014 }
1015
1016 /* Net device open. */
1017 static int tun_net_open(struct net_device *dev)
1018 {
1019         struct tun_struct *tun = netdev_priv(dev);
1020         int i;
1021
1022         netif_tx_start_all_queues(dev);
1023
1024         for (i = 0; i < tun->numqueues; i++) {
1025                 struct tun_file *tfile;
1026
1027                 tfile = rtnl_dereference(tun->tfiles[i]);
1028                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1029         }
1030
1031         return 0;
1032 }
1033
1034 /* Net device close. */
1035 static int tun_net_close(struct net_device *dev)
1036 {
1037         netif_tx_stop_all_queues(dev);
1038         return 0;
1039 }
1040
1041 /* Net device start xmit */
1042 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1043 {
1044 #ifdef CONFIG_RPS
1045         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1046                 /* Select queue was not called for the skbuff, so we extract the
1047                  * RPS hash and save it into the flow_table here.
1048                  */
1049                 struct tun_flow_entry *e;
1050                 __u32 rxhash;
1051
1052                 rxhash = __skb_get_hash_symmetric(skb);
1053                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1054                 if (e)
1055                         tun_flow_save_rps_rxhash(e, rxhash);
1056         }
1057 #endif
1058 }
1059
1060 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1061                                     struct sk_buff *skb,
1062                                     int len)
1063 {
1064         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1065
1066         if (prog)
1067                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1068
1069         return len;
1070 }
1071
1072 /* Net device start xmit */
1073 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1074 {
1075         struct tun_struct *tun = netdev_priv(dev);
1076         int txq = skb->queue_mapping;
1077         struct tun_file *tfile;
1078         int len = skb->len;
1079
1080         rcu_read_lock();
1081         tfile = rcu_dereference(tun->tfiles[txq]);
1082
1083         /* Drop packet if interface is not attached */
1084         if (txq >= tun->numqueues)
1085                 goto drop;
1086
1087         if (!rcu_dereference(tun->steering_prog))
1088                 tun_automq_xmit(tun, skb);
1089
1090         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1091
1092         BUG_ON(!tfile);
1093
1094         /* Drop if the filter does not like it.
1095          * This is a noop if the filter is disabled.
1096          * Filter can be enabled only for the TAP devices. */
1097         if (!check_filter(&tun->txflt, skb))
1098                 goto drop;
1099
1100         if (tfile->socket.sk->sk_filter &&
1101             sk_filter(tfile->socket.sk, skb))
1102                 goto drop;
1103
1104         len = run_ebpf_filter(tun, skb, len);
1105         if (len == 0 || pskb_trim(skb, len))
1106                 goto drop;
1107
1108         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1109                 goto drop;
1110
1111         skb_tx_timestamp(skb);
1112
1113         /* Orphan the skb - required as we might hang on to it
1114          * for indefinite time.
1115          */
1116         skb_orphan(skb);
1117
1118         nf_reset(skb);
1119
1120         if (ptr_ring_produce(&tfile->tx_ring, skb))
1121                 goto drop;
1122
1123         /* Notify and wake up reader process */
1124         if (tfile->flags & TUN_FASYNC)
1125                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1126         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1127
1128         rcu_read_unlock();
1129         return NETDEV_TX_OK;
1130
1131 drop:
1132         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1133         skb_tx_error(skb);
1134         kfree_skb(skb);
1135         rcu_read_unlock();
1136         return NET_XMIT_DROP;
1137 }
1138
1139 static void tun_net_mclist(struct net_device *dev)
1140 {
1141         /*
1142          * This callback is supposed to deal with mc filter in
1143          * _rx_ path and has nothing to do with the _tx_ path.
1144          * In rx path we always accept everything userspace gives us.
1145          */
1146 }
1147
1148 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1149         netdev_features_t features)
1150 {
1151         struct tun_struct *tun = netdev_priv(dev);
1152
1153         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1154 }
1155
1156 static void tun_set_headroom(struct net_device *dev, int new_hr)
1157 {
1158         struct tun_struct *tun = netdev_priv(dev);
1159
1160         if (new_hr < NET_SKB_PAD)
1161                 new_hr = NET_SKB_PAD;
1162
1163         tun->align = new_hr;
1164 }
1165
1166 static void
1167 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1168 {
1169         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1170         struct tun_struct *tun = netdev_priv(dev);
1171         struct tun_pcpu_stats *p;
1172         int i;
1173
1174         for_each_possible_cpu(i) {
1175                 u64 rxpackets, rxbytes, txpackets, txbytes;
1176                 unsigned int start;
1177
1178                 p = per_cpu_ptr(tun->pcpu_stats, i);
1179                 do {
1180                         start = u64_stats_fetch_begin(&p->syncp);
1181                         rxpackets       = p->rx_packets;
1182                         rxbytes         = p->rx_bytes;
1183                         txpackets       = p->tx_packets;
1184                         txbytes         = p->tx_bytes;
1185                 } while (u64_stats_fetch_retry(&p->syncp, start));
1186
1187                 stats->rx_packets       += rxpackets;
1188                 stats->rx_bytes         += rxbytes;
1189                 stats->tx_packets       += txpackets;
1190                 stats->tx_bytes         += txbytes;
1191
1192                 /* u32 counters */
1193                 rx_dropped      += p->rx_dropped;
1194                 rx_frame_errors += p->rx_frame_errors;
1195                 tx_dropped      += p->tx_dropped;
1196         }
1197         stats->rx_dropped  = rx_dropped;
1198         stats->rx_frame_errors = rx_frame_errors;
1199         stats->tx_dropped = tx_dropped;
1200 }
1201
1202 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1203                        struct netlink_ext_ack *extack)
1204 {
1205         struct tun_struct *tun = netdev_priv(dev);
1206         struct tun_file *tfile;
1207         struct bpf_prog *old_prog;
1208         int i;
1209
1210         old_prog = rtnl_dereference(tun->xdp_prog);
1211         rcu_assign_pointer(tun->xdp_prog, prog);
1212         if (old_prog)
1213                 bpf_prog_put(old_prog);
1214
1215         for (i = 0; i < tun->numqueues; i++) {
1216                 tfile = rtnl_dereference(tun->tfiles[i]);
1217                 if (prog)
1218                         sock_set_flag(&tfile->sk, SOCK_XDP);
1219                 else
1220                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1221         }
1222         list_for_each_entry(tfile, &tun->disabled, next) {
1223                 if (prog)
1224                         sock_set_flag(&tfile->sk, SOCK_XDP);
1225                 else
1226                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1227         }
1228
1229         return 0;
1230 }
1231
1232 static u32 tun_xdp_query(struct net_device *dev)
1233 {
1234         struct tun_struct *tun = netdev_priv(dev);
1235         const struct bpf_prog *xdp_prog;
1236
1237         xdp_prog = rtnl_dereference(tun->xdp_prog);
1238         if (xdp_prog)
1239                 return xdp_prog->aux->id;
1240
1241         return 0;
1242 }
1243
1244 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1245 {
1246         switch (xdp->command) {
1247         case XDP_SETUP_PROG:
1248                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1249         case XDP_QUERY_PROG:
1250                 xdp->prog_id = tun_xdp_query(dev);
1251                 return 0;
1252         default:
1253                 return -EINVAL;
1254         }
1255 }
1256
1257 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1258 {
1259         if (new_carrier) {
1260                 struct tun_struct *tun = netdev_priv(dev);
1261
1262                 if (!tun->numqueues)
1263                         return -EPERM;
1264
1265                 netif_carrier_on(dev);
1266         } else {
1267                 netif_carrier_off(dev);
1268         }
1269         return 0;
1270 }
1271
1272 static const struct net_device_ops tun_netdev_ops = {
1273         .ndo_uninit             = tun_net_uninit,
1274         .ndo_open               = tun_net_open,
1275         .ndo_stop               = tun_net_close,
1276         .ndo_start_xmit         = tun_net_xmit,
1277         .ndo_fix_features       = tun_net_fix_features,
1278         .ndo_select_queue       = tun_select_queue,
1279         .ndo_set_rx_headroom    = tun_set_headroom,
1280         .ndo_get_stats64        = tun_net_get_stats64,
1281         .ndo_change_carrier     = tun_net_change_carrier,
1282 };
1283
1284 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1285 {
1286         /* Notify and wake up reader process */
1287         if (tfile->flags & TUN_FASYNC)
1288                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1289         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1290 }
1291
1292 static int tun_xdp_xmit(struct net_device *dev, int n,
1293                         struct xdp_frame **frames, u32 flags)
1294 {
1295         struct tun_struct *tun = netdev_priv(dev);
1296         struct tun_file *tfile;
1297         u32 numqueues;
1298         int drops = 0;
1299         int cnt = n;
1300         int i;
1301
1302         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1303                 return -EINVAL;
1304
1305         rcu_read_lock();
1306
1307         numqueues = READ_ONCE(tun->numqueues);
1308         if (!numqueues) {
1309                 rcu_read_unlock();
1310                 return -ENXIO; /* Caller will free/return all frames */
1311         }
1312
1313         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1314                                             numqueues]);
1315
1316         spin_lock(&tfile->tx_ring.producer_lock);
1317         for (i = 0; i < n; i++) {
1318                 struct xdp_frame *xdp = frames[i];
1319                 /* Encode the XDP flag into lowest bit for consumer to differ
1320                  * XDP buffer from sk_buff.
1321                  */
1322                 void *frame = tun_xdp_to_ptr(xdp);
1323
1324                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1325                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1326                         xdp_return_frame_rx_napi(xdp);
1327                         drops++;
1328                 }
1329         }
1330         spin_unlock(&tfile->tx_ring.producer_lock);
1331
1332         if (flags & XDP_XMIT_FLUSH)
1333                 __tun_xdp_flush_tfile(tfile);
1334
1335         rcu_read_unlock();
1336         return cnt - drops;
1337 }
1338
1339 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1340 {
1341         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1342
1343         if (unlikely(!frame))
1344                 return -EOVERFLOW;
1345
1346         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1347 }
1348
1349 static const struct net_device_ops tap_netdev_ops = {
1350         .ndo_uninit             = tun_net_uninit,
1351         .ndo_open               = tun_net_open,
1352         .ndo_stop               = tun_net_close,
1353         .ndo_start_xmit         = tun_net_xmit,
1354         .ndo_fix_features       = tun_net_fix_features,
1355         .ndo_set_rx_mode        = tun_net_mclist,
1356         .ndo_set_mac_address    = eth_mac_addr,
1357         .ndo_validate_addr      = eth_validate_addr,
1358         .ndo_select_queue       = tun_select_queue,
1359         .ndo_features_check     = passthru_features_check,
1360         .ndo_set_rx_headroom    = tun_set_headroom,
1361         .ndo_get_stats64        = tun_net_get_stats64,
1362         .ndo_bpf                = tun_xdp,
1363         .ndo_xdp_xmit           = tun_xdp_xmit,
1364         .ndo_change_carrier     = tun_net_change_carrier,
1365 };
1366
1367 static void tun_flow_init(struct tun_struct *tun)
1368 {
1369         int i;
1370
1371         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1372                 INIT_HLIST_HEAD(&tun->flows[i]);
1373
1374         tun->ageing_time = TUN_FLOW_EXPIRE;
1375         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1376         mod_timer(&tun->flow_gc_timer,
1377                   round_jiffies_up(jiffies + tun->ageing_time));
1378 }
1379
1380 static void tun_flow_uninit(struct tun_struct *tun)
1381 {
1382         del_timer_sync(&tun->flow_gc_timer);
1383         tun_flow_flush(tun);
1384 }
1385
1386 #define MIN_MTU 68
1387 #define MAX_MTU 65535
1388
1389 /* Initialize net device. */
1390 static void tun_net_init(struct net_device *dev)
1391 {
1392         struct tun_struct *tun = netdev_priv(dev);
1393
1394         switch (tun->flags & TUN_TYPE_MASK) {
1395         case IFF_TUN:
1396                 dev->netdev_ops = &tun_netdev_ops;
1397
1398                 /* Point-to-Point TUN Device */
1399                 dev->hard_header_len = 0;
1400                 dev->addr_len = 0;
1401                 dev->mtu = 1500;
1402
1403                 /* Zero header length */
1404                 dev->type = ARPHRD_NONE;
1405                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1406                 break;
1407
1408         case IFF_TAP:
1409                 dev->netdev_ops = &tap_netdev_ops;
1410                 /* Ethernet TAP Device */
1411                 ether_setup(dev);
1412                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1413                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1414
1415                 eth_hw_addr_random(dev);
1416
1417                 break;
1418         }
1419
1420         dev->min_mtu = MIN_MTU;
1421         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1422 }
1423
1424 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1425 {
1426         struct sock *sk = tfile->socket.sk;
1427
1428         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1429 }
1430
1431 /* Character device part */
1432
1433 /* Poll */
1434 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1435 {
1436         struct tun_file *tfile = file->private_data;
1437         struct tun_struct *tun = tun_get(tfile);
1438         struct sock *sk;
1439         __poll_t mask = 0;
1440
1441         if (!tun)
1442                 return EPOLLERR;
1443
1444         sk = tfile->socket.sk;
1445
1446         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1447
1448         poll_wait(file, sk_sleep(sk), wait);
1449
1450         if (!ptr_ring_empty(&tfile->tx_ring))
1451                 mask |= EPOLLIN | EPOLLRDNORM;
1452
1453         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1454          * guarantee EPOLLOUT to be raised by either here or
1455          * tun_sock_write_space(). Then process could get notification
1456          * after it writes to a down device and meets -EIO.
1457          */
1458         if (tun_sock_writeable(tun, tfile) ||
1459             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1460              tun_sock_writeable(tun, tfile)))
1461                 mask |= EPOLLOUT | EPOLLWRNORM;
1462
1463         if (tun->dev->reg_state != NETREG_REGISTERED)
1464                 mask = EPOLLERR;
1465
1466         tun_put(tun);
1467         return mask;
1468 }
1469
1470 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1471                                             size_t len,
1472                                             const struct iov_iter *it)
1473 {
1474         struct sk_buff *skb;
1475         size_t linear;
1476         int err;
1477         int i;
1478
1479         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1480                 return ERR_PTR(-ENOMEM);
1481
1482         local_bh_disable();
1483         skb = napi_get_frags(&tfile->napi);
1484         local_bh_enable();
1485         if (!skb)
1486                 return ERR_PTR(-ENOMEM);
1487
1488         linear = iov_iter_single_seg_count(it);
1489         err = __skb_grow(skb, linear);
1490         if (err)
1491                 goto free;
1492
1493         skb->len = len;
1494         skb->data_len = len - linear;
1495         skb->truesize += skb->data_len;
1496
1497         for (i = 1; i < it->nr_segs; i++) {
1498                 size_t fragsz = it->iov[i].iov_len;
1499                 struct page *page;
1500                 void *frag;
1501
1502                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1503                         err = -EINVAL;
1504                         goto free;
1505                 }
1506                 frag = netdev_alloc_frag(fragsz);
1507                 if (!frag) {
1508                         err = -ENOMEM;
1509                         goto free;
1510                 }
1511                 page = virt_to_head_page(frag);
1512                 skb_fill_page_desc(skb, i - 1, page,
1513                                    frag - page_address(page), fragsz);
1514         }
1515
1516         return skb;
1517 free:
1518         /* frees skb and all frags allocated with napi_alloc_frag() */
1519         napi_free_frags(&tfile->napi);
1520         return ERR_PTR(err);
1521 }
1522
1523 /* prepad is the amount to reserve at front.  len is length after that.
1524  * linear is a hint as to how much to copy (usually headers). */
1525 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1526                                      size_t prepad, size_t len,
1527                                      size_t linear, int noblock)
1528 {
1529         struct sock *sk = tfile->socket.sk;
1530         struct sk_buff *skb;
1531         int err;
1532
1533         /* Under a page?  Don't bother with paged skb. */
1534         if (prepad + len < PAGE_SIZE || !linear)
1535                 linear = len;
1536
1537         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1538                                    &err, 0);
1539         if (!skb)
1540                 return ERR_PTR(err);
1541
1542         skb_reserve(skb, prepad);
1543         skb_put(skb, linear);
1544         skb->data_len = len - linear;
1545         skb->len += len - linear;
1546
1547         return skb;
1548 }
1549
1550 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1551                            struct sk_buff *skb, int more)
1552 {
1553         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1554         struct sk_buff_head process_queue;
1555         u32 rx_batched = tun->rx_batched;
1556         bool rcv = false;
1557
1558         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1559                 local_bh_disable();
1560                 skb_record_rx_queue(skb, tfile->queue_index);
1561                 netif_receive_skb(skb);
1562                 local_bh_enable();
1563                 return;
1564         }
1565
1566         spin_lock(&queue->lock);
1567         if (!more || skb_queue_len(queue) == rx_batched) {
1568                 __skb_queue_head_init(&process_queue);
1569                 skb_queue_splice_tail_init(queue, &process_queue);
1570                 rcv = true;
1571         } else {
1572                 __skb_queue_tail(queue, skb);
1573         }
1574         spin_unlock(&queue->lock);
1575
1576         if (rcv) {
1577                 struct sk_buff *nskb;
1578
1579                 local_bh_disable();
1580                 while ((nskb = __skb_dequeue(&process_queue))) {
1581                         skb_record_rx_queue(nskb, tfile->queue_index);
1582                         netif_receive_skb(nskb);
1583                 }
1584                 skb_record_rx_queue(skb, tfile->queue_index);
1585                 netif_receive_skb(skb);
1586                 local_bh_enable();
1587         }
1588 }
1589
1590 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1591                               int len, int noblock, bool zerocopy)
1592 {
1593         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1594                 return false;
1595
1596         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1597                 return false;
1598
1599         if (!noblock)
1600                 return false;
1601
1602         if (zerocopy)
1603                 return false;
1604
1605         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1606             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1607                 return false;
1608
1609         return true;
1610 }
1611
1612 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1613                                        int buflen, int len, int pad)
1614 {
1615         struct sk_buff *skb = build_skb(buf, buflen);
1616
1617         if (!skb)
1618                 return ERR_PTR(-ENOMEM);
1619
1620         skb_reserve(skb, pad);
1621         skb_put(skb, len);
1622
1623         get_page(alloc_frag->page);
1624         alloc_frag->offset += buflen;
1625
1626         return skb;
1627 }
1628
1629 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1630                        struct xdp_buff *xdp, u32 act)
1631 {
1632         int err;
1633
1634         switch (act) {
1635         case XDP_REDIRECT:
1636                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1637                 if (err)
1638                         return err;
1639                 break;
1640         case XDP_TX:
1641                 err = tun_xdp_tx(tun->dev, xdp);
1642                 if (err < 0)
1643                         return err;
1644                 break;
1645         case XDP_PASS:
1646                 break;
1647         default:
1648                 bpf_warn_invalid_xdp_action(act);
1649                 /* fall through */
1650         case XDP_ABORTED:
1651                 trace_xdp_exception(tun->dev, xdp_prog, act);
1652                 /* fall through */
1653         case XDP_DROP:
1654                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1655                 break;
1656         }
1657
1658         return act;
1659 }
1660
1661 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1662                                      struct tun_file *tfile,
1663                                      struct iov_iter *from,
1664                                      struct virtio_net_hdr *hdr,
1665                                      int len, int *skb_xdp)
1666 {
1667         struct page_frag *alloc_frag = &current->task_frag;
1668         struct bpf_prog *xdp_prog;
1669         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1670         char *buf;
1671         size_t copied;
1672         int pad = TUN_RX_PAD;
1673         int err = 0;
1674
1675         rcu_read_lock();
1676         xdp_prog = rcu_dereference(tun->xdp_prog);
1677         if (xdp_prog)
1678                 pad += XDP_PACKET_HEADROOM;
1679         buflen += SKB_DATA_ALIGN(len + pad);
1680         rcu_read_unlock();
1681
1682         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1683         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1684                 return ERR_PTR(-ENOMEM);
1685
1686         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1687         copied = copy_page_from_iter(alloc_frag->page,
1688                                      alloc_frag->offset + pad,
1689                                      len, from);
1690         if (copied != len)
1691                 return ERR_PTR(-EFAULT);
1692
1693         /* There's a small window that XDP may be set after the check
1694          * of xdp_prog above, this should be rare and for simplicity
1695          * we do XDP on skb in case the headroom is not enough.
1696          */
1697         if (hdr->gso_type || !xdp_prog) {
1698                 *skb_xdp = 1;
1699                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1700         }
1701
1702         *skb_xdp = 0;
1703
1704         local_bh_disable();
1705         rcu_read_lock();
1706         xdp_prog = rcu_dereference(tun->xdp_prog);
1707         if (xdp_prog) {
1708                 struct xdp_buff xdp;
1709                 u32 act;
1710
1711                 xdp.data_hard_start = buf;
1712                 xdp.data = buf + pad;
1713                 xdp_set_data_meta_invalid(&xdp);
1714                 xdp.data_end = xdp.data + len;
1715                 xdp.rxq = &tfile->xdp_rxq;
1716
1717                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1718                 if (act == XDP_REDIRECT || act == XDP_TX) {
1719                         get_page(alloc_frag->page);
1720                         alloc_frag->offset += buflen;
1721                 }
1722                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1723                 if (err < 0)
1724                         goto err_xdp;
1725                 if (err == XDP_REDIRECT)
1726                         xdp_do_flush_map();
1727                 if (err != XDP_PASS)
1728                         goto out;
1729
1730                 pad = xdp.data - xdp.data_hard_start;
1731                 len = xdp.data_end - xdp.data;
1732         }
1733         rcu_read_unlock();
1734         local_bh_enable();
1735
1736         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1737
1738 err_xdp:
1739         put_page(alloc_frag->page);
1740 out:
1741         rcu_read_unlock();
1742         local_bh_enable();
1743         return NULL;
1744 }
1745
1746 /* Get packet from user space buffer */
1747 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1748                             void *msg_control, struct iov_iter *from,
1749                             int noblock, bool more)
1750 {
1751         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1752         struct sk_buff *skb;
1753         size_t total_len = iov_iter_count(from);
1754         size_t len = total_len, align = tun->align, linear;
1755         struct virtio_net_hdr gso = { 0 };
1756         struct tun_pcpu_stats *stats;
1757         int good_linear;
1758         int copylen;
1759         bool zerocopy = false;
1760         int err;
1761         u32 rxhash = 0;
1762         int skb_xdp = 1;
1763         bool frags = tun_napi_frags_enabled(tfile);
1764
1765         if (!(tun->flags & IFF_NO_PI)) {
1766                 if (len < sizeof(pi))
1767                         return -EINVAL;
1768                 len -= sizeof(pi);
1769
1770                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1771                         return -EFAULT;
1772         }
1773
1774         if (tun->flags & IFF_VNET_HDR) {
1775                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1776
1777                 if (len < vnet_hdr_sz)
1778                         return -EINVAL;
1779                 len -= vnet_hdr_sz;
1780
1781                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1782                         return -EFAULT;
1783
1784                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1785                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1786                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1787
1788                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1789                         return -EINVAL;
1790                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1791         }
1792
1793         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1794                 align += NET_IP_ALIGN;
1795                 if (unlikely(len < ETH_HLEN ||
1796                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1797                         return -EINVAL;
1798         }
1799
1800         good_linear = SKB_MAX_HEAD(align);
1801
1802         if (msg_control) {
1803                 struct iov_iter i = *from;
1804
1805                 /* There are 256 bytes to be copied in skb, so there is
1806                  * enough room for skb expand head in case it is used.
1807                  * The rest of the buffer is mapped from userspace.
1808                  */
1809                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1810                 if (copylen > good_linear)
1811                         copylen = good_linear;
1812                 linear = copylen;
1813                 iov_iter_advance(&i, copylen);
1814                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1815                         zerocopy = true;
1816         }
1817
1818         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1819                 /* For the packet that is not easy to be processed
1820                  * (e.g gso or jumbo packet), we will do it at after
1821                  * skb was created with generic XDP routine.
1822                  */
1823                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1824                 if (IS_ERR(skb)) {
1825                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1826                         return PTR_ERR(skb);
1827                 }
1828                 if (!skb)
1829                         return total_len;
1830         } else {
1831                 if (!zerocopy) {
1832                         copylen = len;
1833                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1834                                 linear = good_linear;
1835                         else
1836                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1837                 }
1838
1839                 if (frags) {
1840                         mutex_lock(&tfile->napi_mutex);
1841                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1842                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1843                          * If zerocopy is enabled, then this layout will be
1844                          * overwritten by zerocopy_sg_from_iter().
1845                          */
1846                         zerocopy = false;
1847                 } else {
1848                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1849                                             noblock);
1850                 }
1851
1852                 if (IS_ERR(skb)) {
1853                         if (PTR_ERR(skb) != -EAGAIN)
1854                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1855                         if (frags)
1856                                 mutex_unlock(&tfile->napi_mutex);
1857                         return PTR_ERR(skb);
1858                 }
1859
1860                 if (zerocopy)
1861                         err = zerocopy_sg_from_iter(skb, from);
1862                 else
1863                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1864
1865                 if (err) {
1866                         err = -EFAULT;
1867 drop:
1868                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1869                         kfree_skb(skb);
1870                         if (frags) {
1871                                 tfile->napi.skb = NULL;
1872                                 mutex_unlock(&tfile->napi_mutex);
1873                         }
1874
1875                         return err;
1876                 }
1877         }
1878
1879         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1880                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1881                 kfree_skb(skb);
1882                 if (frags) {
1883                         tfile->napi.skb = NULL;
1884                         mutex_unlock(&tfile->napi_mutex);
1885                 }
1886
1887                 return -EINVAL;
1888         }
1889
1890         switch (tun->flags & TUN_TYPE_MASK) {
1891         case IFF_TUN:
1892                 if (tun->flags & IFF_NO_PI) {
1893                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1894
1895                         switch (ip_version) {
1896                         case 4:
1897                                 pi.proto = htons(ETH_P_IP);
1898                                 break;
1899                         case 6:
1900                                 pi.proto = htons(ETH_P_IPV6);
1901                                 break;
1902                         default:
1903                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1904                                 kfree_skb(skb);
1905                                 return -EINVAL;
1906                         }
1907                 }
1908
1909                 skb_reset_mac_header(skb);
1910                 skb->protocol = pi.proto;
1911                 skb->dev = tun->dev;
1912                 break;
1913         case IFF_TAP:
1914                 if (!frags)
1915                         skb->protocol = eth_type_trans(skb, tun->dev);
1916                 break;
1917         }
1918
1919         /* copy skb_ubuf_info for callback when skb has no error */
1920         if (zerocopy) {
1921                 skb_shinfo(skb)->destructor_arg = msg_control;
1922                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1923                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1924         } else if (msg_control) {
1925                 struct ubuf_info *uarg = msg_control;
1926                 uarg->callback(uarg, false);
1927         }
1928
1929         skb_reset_network_header(skb);
1930         skb_probe_transport_header(skb);
1931
1932         if (skb_xdp) {
1933                 struct bpf_prog *xdp_prog;
1934                 int ret;
1935
1936                 local_bh_disable();
1937                 rcu_read_lock();
1938                 xdp_prog = rcu_dereference(tun->xdp_prog);
1939                 if (xdp_prog) {
1940                         ret = do_xdp_generic(xdp_prog, skb);
1941                         if (ret != XDP_PASS) {
1942                                 rcu_read_unlock();
1943                                 local_bh_enable();
1944                                 return total_len;
1945                         }
1946                 }
1947                 rcu_read_unlock();
1948                 local_bh_enable();
1949         }
1950
1951         /* Compute the costly rx hash only if needed for flow updates.
1952          * We may get a very small possibility of OOO during switching, not
1953          * worth to optimize.
1954          */
1955         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1956             !tfile->detached)
1957                 rxhash = __skb_get_hash_symmetric(skb);
1958
1959         rcu_read_lock();
1960         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1961                 err = -EIO;
1962                 rcu_read_unlock();
1963                 goto drop;
1964         }
1965
1966         if (frags) {
1967                 /* Exercise flow dissector code path. */
1968                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1969
1970                 if (unlikely(headlen > skb_headlen(skb))) {
1971                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1972                         napi_free_frags(&tfile->napi);
1973                         rcu_read_unlock();
1974                         mutex_unlock(&tfile->napi_mutex);
1975                         WARN_ON(1);
1976                         return -ENOMEM;
1977                 }
1978
1979                 local_bh_disable();
1980                 napi_gro_frags(&tfile->napi);
1981                 local_bh_enable();
1982                 mutex_unlock(&tfile->napi_mutex);
1983         } else if (tfile->napi_enabled) {
1984                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1985                 int queue_len;
1986
1987                 spin_lock_bh(&queue->lock);
1988                 __skb_queue_tail(queue, skb);
1989                 queue_len = skb_queue_len(queue);
1990                 spin_unlock(&queue->lock);
1991
1992                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1993                         napi_schedule(&tfile->napi);
1994
1995                 local_bh_enable();
1996         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1997                 tun_rx_batched(tun, tfile, skb, more);
1998         } else {
1999                 netif_rx_ni(skb);
2000         }
2001         rcu_read_unlock();
2002
2003         stats = get_cpu_ptr(tun->pcpu_stats);
2004         u64_stats_update_begin(&stats->syncp);
2005         stats->rx_packets++;
2006         stats->rx_bytes += len;
2007         u64_stats_update_end(&stats->syncp);
2008         put_cpu_ptr(stats);
2009
2010         if (rxhash)
2011                 tun_flow_update(tun, rxhash, tfile);
2012
2013         return total_len;
2014 }
2015
2016 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2017 {
2018         struct file *file = iocb->ki_filp;
2019         struct tun_file *tfile = file->private_data;
2020         struct tun_struct *tun = tun_get(tfile);
2021         ssize_t result;
2022
2023         if (!tun)
2024                 return -EBADFD;
2025
2026         result = tun_get_user(tun, tfile, NULL, from,
2027                               file->f_flags & O_NONBLOCK, false);
2028
2029         tun_put(tun);
2030         return result;
2031 }
2032
2033 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2034                                 struct tun_file *tfile,
2035                                 struct xdp_frame *xdp_frame,
2036                                 struct iov_iter *iter)
2037 {
2038         int vnet_hdr_sz = 0;
2039         size_t size = xdp_frame->len;
2040         struct tun_pcpu_stats *stats;
2041         size_t ret;
2042
2043         if (tun->flags & IFF_VNET_HDR) {
2044                 struct virtio_net_hdr gso = { 0 };
2045
2046                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2047                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2048                         return -EINVAL;
2049                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2050                              sizeof(gso)))
2051                         return -EFAULT;
2052                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2053         }
2054
2055         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2056
2057         stats = get_cpu_ptr(tun->pcpu_stats);
2058         u64_stats_update_begin(&stats->syncp);
2059         stats->tx_packets++;
2060         stats->tx_bytes += ret;
2061         u64_stats_update_end(&stats->syncp);
2062         put_cpu_ptr(tun->pcpu_stats);
2063
2064         return ret;
2065 }
2066
2067 /* Put packet to the user space buffer */
2068 static ssize_t tun_put_user(struct tun_struct *tun,
2069                             struct tun_file *tfile,
2070                             struct sk_buff *skb,
2071                             struct iov_iter *iter)
2072 {
2073         struct tun_pi pi = { 0, skb->protocol };
2074         struct tun_pcpu_stats *stats;
2075         ssize_t total;
2076         int vlan_offset = 0;
2077         int vlan_hlen = 0;
2078         int vnet_hdr_sz = 0;
2079
2080         if (skb_vlan_tag_present(skb))
2081                 vlan_hlen = VLAN_HLEN;
2082
2083         if (tun->flags & IFF_VNET_HDR)
2084                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2085
2086         total = skb->len + vlan_hlen + vnet_hdr_sz;
2087
2088         if (!(tun->flags & IFF_NO_PI)) {
2089                 if (iov_iter_count(iter) < sizeof(pi))
2090                         return -EINVAL;
2091
2092                 total += sizeof(pi);
2093                 if (iov_iter_count(iter) < total) {
2094                         /* Packet will be striped */
2095                         pi.flags |= TUN_PKT_STRIP;
2096                 }
2097
2098                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2099                         return -EFAULT;
2100         }
2101
2102         if (vnet_hdr_sz) {
2103                 struct virtio_net_hdr gso;
2104
2105                 if (iov_iter_count(iter) < vnet_hdr_sz)
2106                         return -EINVAL;
2107
2108                 if (virtio_net_hdr_from_skb(skb, &gso,
2109                                             tun_is_little_endian(tun), true,
2110                                             vlan_hlen)) {
2111                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2112                         pr_err("unexpected GSO type: "
2113                                "0x%x, gso_size %d, hdr_len %d\n",
2114                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2115                                tun16_to_cpu(tun, gso.hdr_len));
2116                         print_hex_dump(KERN_ERR, "tun: ",
2117                                        DUMP_PREFIX_NONE,
2118                                        16, 1, skb->head,
2119                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2120                         WARN_ON_ONCE(1);
2121                         return -EINVAL;
2122                 }
2123
2124                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2125                         return -EFAULT;
2126
2127                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2128         }
2129
2130         if (vlan_hlen) {
2131                 int ret;
2132                 struct veth veth;
2133
2134                 veth.h_vlan_proto = skb->vlan_proto;
2135                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2136
2137                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2138
2139                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2140                 if (ret || !iov_iter_count(iter))
2141                         goto done;
2142
2143                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2144                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2145                         goto done;
2146         }
2147
2148         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2149
2150 done:
2151         /* caller is in process context, */
2152         stats = get_cpu_ptr(tun->pcpu_stats);
2153         u64_stats_update_begin(&stats->syncp);
2154         stats->tx_packets++;
2155         stats->tx_bytes += skb->len + vlan_hlen;
2156         u64_stats_update_end(&stats->syncp);
2157         put_cpu_ptr(tun->pcpu_stats);
2158
2159         return total;
2160 }
2161
2162 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2163 {
2164         DECLARE_WAITQUEUE(wait, current);
2165         void *ptr = NULL;
2166         int error = 0;
2167
2168         ptr = ptr_ring_consume(&tfile->tx_ring);
2169         if (ptr)
2170                 goto out;
2171         if (noblock) {
2172                 error = -EAGAIN;
2173                 goto out;
2174         }
2175
2176         add_wait_queue(&tfile->wq.wait, &wait);
2177
2178         while (1) {
2179                 set_current_state(TASK_INTERRUPTIBLE);
2180                 ptr = ptr_ring_consume(&tfile->tx_ring);
2181                 if (ptr)
2182                         break;
2183                 if (signal_pending(current)) {
2184                         error = -ERESTARTSYS;
2185                         break;
2186                 }
2187                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2188                         error = -EFAULT;
2189                         break;
2190                 }
2191
2192                 schedule();
2193         }
2194
2195         __set_current_state(TASK_RUNNING);
2196         remove_wait_queue(&tfile->wq.wait, &wait);
2197
2198 out:
2199         *err = error;
2200         return ptr;
2201 }
2202
2203 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2204                            struct iov_iter *to,
2205                            int noblock, void *ptr)
2206 {
2207         ssize_t ret;
2208         int err;
2209
2210         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2211
2212         if (!iov_iter_count(to)) {
2213                 tun_ptr_free(ptr);
2214                 return 0;
2215         }
2216
2217         if (!ptr) {
2218                 /* Read frames from ring */
2219                 ptr = tun_ring_recv(tfile, noblock, &err);
2220                 if (!ptr)
2221                         return err;
2222         }
2223
2224         if (tun_is_xdp_frame(ptr)) {
2225                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2226
2227                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2228                 xdp_return_frame(xdpf);
2229         } else {
2230                 struct sk_buff *skb = ptr;
2231
2232                 ret = tun_put_user(tun, tfile, skb, to);
2233                 if (unlikely(ret < 0))
2234                         kfree_skb(skb);
2235                 else
2236                         consume_skb(skb);
2237         }
2238
2239         return ret;
2240 }
2241
2242 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2243 {
2244         struct file *file = iocb->ki_filp;
2245         struct tun_file *tfile = file->private_data;
2246         struct tun_struct *tun = tun_get(tfile);
2247         ssize_t len = iov_iter_count(to), ret;
2248
2249         if (!tun)
2250                 return -EBADFD;
2251         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2252         ret = min_t(ssize_t, ret, len);
2253         if (ret > 0)
2254                 iocb->ki_pos = ret;
2255         tun_put(tun);
2256         return ret;
2257 }
2258
2259 static void tun_prog_free(struct rcu_head *rcu)
2260 {
2261         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2262
2263         bpf_prog_destroy(prog->prog);
2264         kfree(prog);
2265 }
2266
2267 static int __tun_set_ebpf(struct tun_struct *tun,
2268                           struct tun_prog __rcu **prog_p,
2269                           struct bpf_prog *prog)
2270 {
2271         struct tun_prog *old, *new = NULL;
2272
2273         if (prog) {
2274                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2275                 if (!new)
2276                         return -ENOMEM;
2277                 new->prog = prog;
2278         }
2279
2280         spin_lock_bh(&tun->lock);
2281         old = rcu_dereference_protected(*prog_p,
2282                                         lockdep_is_held(&tun->lock));
2283         rcu_assign_pointer(*prog_p, new);
2284         spin_unlock_bh(&tun->lock);
2285
2286         if (old)
2287                 call_rcu(&old->rcu, tun_prog_free);
2288
2289         return 0;
2290 }
2291
2292 static void tun_free_netdev(struct net_device *dev)
2293 {
2294         struct tun_struct *tun = netdev_priv(dev);
2295
2296         BUG_ON(!(list_empty(&tun->disabled)));
2297         free_percpu(tun->pcpu_stats);
2298         tun_flow_uninit(tun);
2299         security_tun_dev_free_security(tun->security);
2300         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2301         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2302 }
2303
2304 static void tun_setup(struct net_device *dev)
2305 {
2306         struct tun_struct *tun = netdev_priv(dev);
2307
2308         tun->owner = INVALID_UID;
2309         tun->group = INVALID_GID;
2310         tun_default_link_ksettings(dev, &tun->link_ksettings);
2311
2312         dev->ethtool_ops = &tun_ethtool_ops;
2313         dev->needs_free_netdev = true;
2314         dev->priv_destructor = tun_free_netdev;
2315         /* We prefer our own queue length */
2316         dev->tx_queue_len = TUN_READQ_SIZE;
2317 }
2318
2319 /* Trivial set of netlink ops to allow deleting tun or tap
2320  * device with netlink.
2321  */
2322 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2323                         struct netlink_ext_ack *extack)
2324 {
2325         NL_SET_ERR_MSG(extack,
2326                        "tun/tap creation via rtnetlink is not supported.");
2327         return -EOPNOTSUPP;
2328 }
2329
2330 static size_t tun_get_size(const struct net_device *dev)
2331 {
2332         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2333         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2334
2335         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2336                nla_total_size(sizeof(gid_t)) + /* GROUP */
2337                nla_total_size(sizeof(u8)) + /* TYPE */
2338                nla_total_size(sizeof(u8)) + /* PI */
2339                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2340                nla_total_size(sizeof(u8)) + /* PERSIST */
2341                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2342                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2343                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2344                0;
2345 }
2346
2347 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2348 {
2349         struct tun_struct *tun = netdev_priv(dev);
2350
2351         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2352                 goto nla_put_failure;
2353         if (uid_valid(tun->owner) &&
2354             nla_put_u32(skb, IFLA_TUN_OWNER,
2355                         from_kuid_munged(current_user_ns(), tun->owner)))
2356                 goto nla_put_failure;
2357         if (gid_valid(tun->group) &&
2358             nla_put_u32(skb, IFLA_TUN_GROUP,
2359                         from_kgid_munged(current_user_ns(), tun->group)))
2360                 goto nla_put_failure;
2361         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2362                 goto nla_put_failure;
2363         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2364                 goto nla_put_failure;
2365         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2366                 goto nla_put_failure;
2367         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2368                        !!(tun->flags & IFF_MULTI_QUEUE)))
2369                 goto nla_put_failure;
2370         if (tun->flags & IFF_MULTI_QUEUE) {
2371                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2372                         goto nla_put_failure;
2373                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2374                                 tun->numdisabled))
2375                         goto nla_put_failure;
2376         }
2377
2378         return 0;
2379
2380 nla_put_failure:
2381         return -EMSGSIZE;
2382 }
2383
2384 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2385         .kind           = DRV_NAME,
2386         .priv_size      = sizeof(struct tun_struct),
2387         .setup          = tun_setup,
2388         .validate       = tun_validate,
2389         .get_size       = tun_get_size,
2390         .fill_info      = tun_fill_info,
2391 };
2392
2393 static void tun_sock_write_space(struct sock *sk)
2394 {
2395         struct tun_file *tfile;
2396         wait_queue_head_t *wqueue;
2397
2398         if (!sock_writeable(sk))
2399                 return;
2400
2401         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2402                 return;
2403
2404         wqueue = sk_sleep(sk);
2405         if (wqueue && waitqueue_active(wqueue))
2406                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2407                                                 EPOLLWRNORM | EPOLLWRBAND);
2408
2409         tfile = container_of(sk, struct tun_file, sk);
2410         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2411 }
2412
2413 static void tun_put_page(struct tun_page *tpage)
2414 {
2415         if (tpage->page)
2416                 __page_frag_cache_drain(tpage->page, tpage->count);
2417 }
2418
2419 static int tun_xdp_one(struct tun_struct *tun,
2420                        struct tun_file *tfile,
2421                        struct xdp_buff *xdp, int *flush,
2422                        struct tun_page *tpage)
2423 {
2424         unsigned int datasize = xdp->data_end - xdp->data;
2425         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2426         struct virtio_net_hdr *gso = &hdr->gso;
2427         struct tun_pcpu_stats *stats;
2428         struct bpf_prog *xdp_prog;
2429         struct sk_buff *skb = NULL;
2430         u32 rxhash = 0, act;
2431         int buflen = hdr->buflen;
2432         int err = 0;
2433         bool skb_xdp = false;
2434         struct page *page;
2435
2436         xdp_prog = rcu_dereference(tun->xdp_prog);
2437         if (xdp_prog) {
2438                 if (gso->gso_type) {
2439                         skb_xdp = true;
2440                         goto build;
2441                 }
2442                 xdp_set_data_meta_invalid(xdp);
2443                 xdp->rxq = &tfile->xdp_rxq;
2444
2445                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2446                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2447                 if (err < 0) {
2448                         put_page(virt_to_head_page(xdp->data));
2449                         return err;
2450                 }
2451
2452                 switch (err) {
2453                 case XDP_REDIRECT:
2454                         *flush = true;
2455                         /* fall through */
2456                 case XDP_TX:
2457                         return 0;
2458                 case XDP_PASS:
2459                         break;
2460                 default:
2461                         page = virt_to_head_page(xdp->data);
2462                         if (tpage->page == page) {
2463                                 ++tpage->count;
2464                         } else {
2465                                 tun_put_page(tpage);
2466                                 tpage->page = page;
2467                                 tpage->count = 1;
2468                         }
2469                         return 0;
2470                 }
2471         }
2472
2473 build:
2474         skb = build_skb(xdp->data_hard_start, buflen);
2475         if (!skb) {
2476                 err = -ENOMEM;
2477                 goto out;
2478         }
2479
2480         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2481         skb_put(skb, xdp->data_end - xdp->data);
2482
2483         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2484                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2485                 kfree_skb(skb);
2486                 err = -EINVAL;
2487                 goto out;
2488         }
2489
2490         skb->protocol = eth_type_trans(skb, tun->dev);
2491         skb_reset_network_header(skb);
2492         skb_probe_transport_header(skb);
2493
2494         if (skb_xdp) {
2495                 err = do_xdp_generic(xdp_prog, skb);
2496                 if (err != XDP_PASS)
2497                         goto out;
2498         }
2499
2500         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2501             !tfile->detached)
2502                 rxhash = __skb_get_hash_symmetric(skb);
2503
2504         skb_record_rx_queue(skb, tfile->queue_index);
2505         netif_receive_skb(skb);
2506
2507         /* No need for get_cpu_ptr() here since this function is
2508          * always called with bh disabled
2509          */
2510         stats = this_cpu_ptr(tun->pcpu_stats);
2511         u64_stats_update_begin(&stats->syncp);
2512         stats->rx_packets++;
2513         stats->rx_bytes += datasize;
2514         u64_stats_update_end(&stats->syncp);
2515
2516         if (rxhash)
2517                 tun_flow_update(tun, rxhash, tfile);
2518
2519 out:
2520         return err;
2521 }
2522
2523 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2524 {
2525         int ret, i;
2526         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2527         struct tun_struct *tun = tun_get(tfile);
2528         struct tun_msg_ctl *ctl = m->msg_control;
2529         struct xdp_buff *xdp;
2530
2531         if (!tun)
2532                 return -EBADFD;
2533
2534         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2535                 struct tun_page tpage;
2536                 int n = ctl->num;
2537                 int flush = 0;
2538
2539                 memset(&tpage, 0, sizeof(tpage));
2540
2541                 local_bh_disable();
2542                 rcu_read_lock();
2543
2544                 for (i = 0; i < n; i++) {
2545                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2546                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2547                 }
2548
2549                 if (flush)
2550                         xdp_do_flush_map();
2551
2552                 rcu_read_unlock();
2553                 local_bh_enable();
2554
2555                 tun_put_page(&tpage);
2556
2557                 ret = total_len;
2558                 goto out;
2559         }
2560
2561         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2562                            m->msg_flags & MSG_DONTWAIT,
2563                            m->msg_flags & MSG_MORE);
2564 out:
2565         tun_put(tun);
2566         return ret;
2567 }
2568
2569 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2570                        int flags)
2571 {
2572         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2573         struct tun_struct *tun = tun_get(tfile);
2574         void *ptr = m->msg_control;
2575         int ret;
2576
2577         if (!tun) {
2578                 ret = -EBADFD;
2579                 goto out_free;
2580         }
2581
2582         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2583                 ret = -EINVAL;
2584                 goto out_put_tun;
2585         }
2586         if (flags & MSG_ERRQUEUE) {
2587                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2588                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2589                 goto out;
2590         }
2591         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2592         if (ret > (ssize_t)total_len) {
2593                 m->msg_flags |= MSG_TRUNC;
2594                 ret = flags & MSG_TRUNC ? ret : total_len;
2595         }
2596 out:
2597         tun_put(tun);
2598         return ret;
2599
2600 out_put_tun:
2601         tun_put(tun);
2602 out_free:
2603         tun_ptr_free(ptr);
2604         return ret;
2605 }
2606
2607 static int tun_ptr_peek_len(void *ptr)
2608 {
2609         if (likely(ptr)) {
2610                 if (tun_is_xdp_frame(ptr)) {
2611                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2612
2613                         return xdpf->len;
2614                 }
2615                 return __skb_array_len_with_tag(ptr);
2616         } else {
2617                 return 0;
2618         }
2619 }
2620
2621 static int tun_peek_len(struct socket *sock)
2622 {
2623         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2624         struct tun_struct *tun;
2625         int ret = 0;
2626
2627         tun = tun_get(tfile);
2628         if (!tun)
2629                 return 0;
2630
2631         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2632         tun_put(tun);
2633
2634         return ret;
2635 }
2636
2637 /* Ops structure to mimic raw sockets with tun */
2638 static const struct proto_ops tun_socket_ops = {
2639         .peek_len = tun_peek_len,
2640         .sendmsg = tun_sendmsg,
2641         .recvmsg = tun_recvmsg,
2642 };
2643
2644 static struct proto tun_proto = {
2645         .name           = "tun",
2646         .owner          = THIS_MODULE,
2647         .obj_size       = sizeof(struct tun_file),
2648 };
2649
2650 static int tun_flags(struct tun_struct *tun)
2651 {
2652         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2653 }
2654
2655 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2656                               char *buf)
2657 {
2658         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2659         return sprintf(buf, "0x%x\n", tun_flags(tun));
2660 }
2661
2662 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2663                               char *buf)
2664 {
2665         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2666         return uid_valid(tun->owner)?
2667                 sprintf(buf, "%u\n",
2668                         from_kuid_munged(current_user_ns(), tun->owner)):
2669                 sprintf(buf, "-1\n");
2670 }
2671
2672 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2673                               char *buf)
2674 {
2675         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2676         return gid_valid(tun->group) ?
2677                 sprintf(buf, "%u\n",
2678                         from_kgid_munged(current_user_ns(), tun->group)):
2679                 sprintf(buf, "-1\n");
2680 }
2681
2682 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2683 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2684 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2685
2686 static struct attribute *tun_dev_attrs[] = {
2687         &dev_attr_tun_flags.attr,
2688         &dev_attr_owner.attr,
2689         &dev_attr_group.attr,
2690         NULL
2691 };
2692
2693 static const struct attribute_group tun_attr_group = {
2694         .attrs = tun_dev_attrs
2695 };
2696
2697 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2698 {
2699         struct tun_struct *tun;
2700         struct tun_file *tfile = file->private_data;
2701         struct net_device *dev;
2702         int err;
2703
2704         if (tfile->detached)
2705                 return -EINVAL;
2706
2707         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2708                 if (!capable(CAP_NET_ADMIN))
2709                         return -EPERM;
2710
2711                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2712                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2713                         return -EINVAL;
2714         }
2715
2716         dev = __dev_get_by_name(net, ifr->ifr_name);
2717         if (dev) {
2718                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2719                         return -EBUSY;
2720                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2721                         tun = netdev_priv(dev);
2722                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2723                         tun = netdev_priv(dev);
2724                 else
2725                         return -EINVAL;
2726
2727                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2728                     !!(tun->flags & IFF_MULTI_QUEUE))
2729                         return -EINVAL;
2730
2731                 if (tun_not_capable(tun))
2732                         return -EPERM;
2733                 err = security_tun_dev_open(tun->security);
2734                 if (err < 0)
2735                         return err;
2736
2737                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2738                                  ifr->ifr_flags & IFF_NAPI,
2739                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2740                 if (err < 0)
2741                         return err;
2742
2743                 if (tun->flags & IFF_MULTI_QUEUE &&
2744                     (tun->numqueues + tun->numdisabled > 1)) {
2745                         /* One or more queue has already been attached, no need
2746                          * to initialize the device again.
2747                          */
2748                         netdev_state_change(dev);
2749                         return 0;
2750                 }
2751
2752                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2753                               (ifr->ifr_flags & TUN_FEATURES);
2754
2755                 netdev_state_change(dev);
2756         } else {
2757                 char *name;
2758                 unsigned long flags = 0;
2759                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2760                              MAX_TAP_QUEUES : 1;
2761
2762                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2763                         return -EPERM;
2764                 err = security_tun_dev_create();
2765                 if (err < 0)
2766                         return err;
2767
2768                 /* Set dev type */
2769                 if (ifr->ifr_flags & IFF_TUN) {
2770                         /* TUN device */
2771                         flags |= IFF_TUN;
2772                         name = "tun%d";
2773                 } else if (ifr->ifr_flags & IFF_TAP) {
2774                         /* TAP device */
2775                         flags |= IFF_TAP;
2776                         name = "tap%d";
2777                 } else
2778                         return -EINVAL;
2779
2780                 if (*ifr->ifr_name)
2781                         name = ifr->ifr_name;
2782
2783                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2784                                        NET_NAME_UNKNOWN, tun_setup, queues,
2785                                        queues);
2786
2787                 if (!dev)
2788                         return -ENOMEM;
2789                 err = dev_get_valid_name(net, dev, name);
2790                 if (err < 0)
2791                         goto err_free_dev;
2792
2793                 dev_net_set(dev, net);
2794                 dev->rtnl_link_ops = &tun_link_ops;
2795                 dev->ifindex = tfile->ifindex;
2796                 dev->sysfs_groups[0] = &tun_attr_group;
2797
2798                 tun = netdev_priv(dev);
2799                 tun->dev = dev;
2800                 tun->flags = flags;
2801                 tun->txflt.count = 0;
2802                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2803
2804                 tun->align = NET_SKB_PAD;
2805                 tun->filter_attached = false;
2806                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2807                 tun->rx_batched = 0;
2808                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2809
2810                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2811                 if (!tun->pcpu_stats) {
2812                         err = -ENOMEM;
2813                         goto err_free_dev;
2814                 }
2815
2816                 spin_lock_init(&tun->lock);
2817
2818                 err = security_tun_dev_alloc_security(&tun->security);
2819                 if (err < 0)
2820                         goto err_free_stat;
2821
2822                 tun_net_init(dev);
2823                 tun_flow_init(tun);
2824
2825                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2826                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2827                                    NETIF_F_HW_VLAN_STAG_TX;
2828                 dev->features = dev->hw_features | NETIF_F_LLTX;
2829                 dev->vlan_features = dev->features &
2830                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2831                                        NETIF_F_HW_VLAN_STAG_TX);
2832
2833                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2834                               (ifr->ifr_flags & TUN_FEATURES);
2835
2836                 INIT_LIST_HEAD(&tun->disabled);
2837                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2838                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2839                 if (err < 0)
2840                         goto err_free_flow;
2841
2842                 err = register_netdevice(tun->dev);
2843                 if (err < 0)
2844                         goto err_detach;
2845         }
2846
2847         netif_carrier_on(tun->dev);
2848
2849         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2850
2851         /* Make sure persistent devices do not get stuck in
2852          * xoff state.
2853          */
2854         if (netif_running(tun->dev))
2855                 netif_tx_wake_all_queues(tun->dev);
2856
2857         strcpy(ifr->ifr_name, tun->dev->name);
2858         return 0;
2859
2860 err_detach:
2861         tun_detach_all(dev);
2862         /* register_netdevice() already called tun_free_netdev() */
2863         goto err_free_dev;
2864
2865 err_free_flow:
2866         tun_flow_uninit(tun);
2867         security_tun_dev_free_security(tun->security);
2868 err_free_stat:
2869         free_percpu(tun->pcpu_stats);
2870 err_free_dev:
2871         free_netdev(dev);
2872         return err;
2873 }
2874
2875 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2876 {
2877         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2878
2879         strcpy(ifr->ifr_name, tun->dev->name);
2880
2881         ifr->ifr_flags = tun_flags(tun);
2882
2883 }
2884
2885 /* This is like a cut-down ethtool ops, except done via tun fd so no
2886  * privs required. */
2887 static int set_offload(struct tun_struct *tun, unsigned long arg)
2888 {
2889         netdev_features_t features = 0;
2890
2891         if (arg & TUN_F_CSUM) {
2892                 features |= NETIF_F_HW_CSUM;
2893                 arg &= ~TUN_F_CSUM;
2894
2895                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2896                         if (arg & TUN_F_TSO_ECN) {
2897                                 features |= NETIF_F_TSO_ECN;
2898                                 arg &= ~TUN_F_TSO_ECN;
2899                         }
2900                         if (arg & TUN_F_TSO4)
2901                                 features |= NETIF_F_TSO;
2902                         if (arg & TUN_F_TSO6)
2903                                 features |= NETIF_F_TSO6;
2904                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2905                 }
2906
2907                 arg &= ~TUN_F_UFO;
2908         }
2909
2910         /* This gives the user a way to test for new features in future by
2911          * trying to set them. */
2912         if (arg)
2913                 return -EINVAL;
2914
2915         tun->set_features = features;
2916         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2917         tun->dev->wanted_features |= features;
2918         netdev_update_features(tun->dev);
2919
2920         return 0;
2921 }
2922
2923 static void tun_detach_filter(struct tun_struct *tun, int n)
2924 {
2925         int i;
2926         struct tun_file *tfile;
2927
2928         for (i = 0; i < n; i++) {
2929                 tfile = rtnl_dereference(tun->tfiles[i]);
2930                 lock_sock(tfile->socket.sk);
2931                 sk_detach_filter(tfile->socket.sk);
2932                 release_sock(tfile->socket.sk);
2933         }
2934
2935         tun->filter_attached = false;
2936 }
2937
2938 static int tun_attach_filter(struct tun_struct *tun)
2939 {
2940         int i, ret = 0;
2941         struct tun_file *tfile;
2942
2943         for (i = 0; i < tun->numqueues; i++) {
2944                 tfile = rtnl_dereference(tun->tfiles[i]);
2945                 lock_sock(tfile->socket.sk);
2946                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2947                 release_sock(tfile->socket.sk);
2948                 if (ret) {
2949                         tun_detach_filter(tun, i);
2950                         return ret;
2951                 }
2952         }
2953
2954         tun->filter_attached = true;
2955         return ret;
2956 }
2957
2958 static void tun_set_sndbuf(struct tun_struct *tun)
2959 {
2960         struct tun_file *tfile;
2961         int i;
2962
2963         for (i = 0; i < tun->numqueues; i++) {
2964                 tfile = rtnl_dereference(tun->tfiles[i]);
2965                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2966         }
2967 }
2968
2969 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2970 {
2971         struct tun_file *tfile = file->private_data;
2972         struct tun_struct *tun;
2973         int ret = 0;
2974
2975         rtnl_lock();
2976
2977         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2978                 tun = tfile->detached;
2979                 if (!tun) {
2980                         ret = -EINVAL;
2981                         goto unlock;
2982                 }
2983                 ret = security_tun_dev_attach_queue(tun->security);
2984                 if (ret < 0)
2985                         goto unlock;
2986                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2987                                  tun->flags & IFF_NAPI_FRAGS);
2988         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2989                 tun = rtnl_dereference(tfile->tun);
2990                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2991                         ret = -EINVAL;
2992                 else
2993                         __tun_detach(tfile, false);
2994         } else
2995                 ret = -EINVAL;
2996
2997         if (ret >= 0)
2998                 netdev_state_change(tun->dev);
2999
3000 unlock:
3001         rtnl_unlock();
3002         return ret;
3003 }
3004
3005 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3006                         void __user *data)
3007 {
3008         struct bpf_prog *prog;
3009         int fd;
3010
3011         if (copy_from_user(&fd, data, sizeof(fd)))
3012                 return -EFAULT;
3013
3014         if (fd == -1) {
3015                 prog = NULL;
3016         } else {
3017                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3018                 if (IS_ERR(prog))
3019                         return PTR_ERR(prog);
3020         }
3021
3022         return __tun_set_ebpf(tun, prog_p, prog);
3023 }
3024
3025 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3026                             unsigned long arg, int ifreq_len)
3027 {
3028         struct tun_file *tfile = file->private_data;
3029         struct net *net = sock_net(&tfile->sk);
3030         struct tun_struct *tun;
3031         void __user* argp = (void __user*)arg;
3032         unsigned int ifindex, carrier;
3033         struct ifreq ifr;
3034         kuid_t owner;
3035         kgid_t group;
3036         int sndbuf;
3037         int vnet_hdr_sz;
3038         int le;
3039         int ret;
3040         bool do_notify = false;
3041
3042         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3043             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3044                 if (copy_from_user(&ifr, argp, ifreq_len))
3045                         return -EFAULT;
3046         } else {
3047                 memset(&ifr, 0, sizeof(ifr));
3048         }
3049         if (cmd == TUNGETFEATURES) {
3050                 /* Currently this just means: "what IFF flags are valid?".
3051                  * This is needed because we never checked for invalid flags on
3052                  * TUNSETIFF.
3053                  */
3054                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3055                                 (unsigned int __user*)argp);
3056         } else if (cmd == TUNSETQUEUE) {
3057                 return tun_set_queue(file, &ifr);
3058         } else if (cmd == SIOCGSKNS) {
3059                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3060                         return -EPERM;
3061                 return open_related_ns(&net->ns, get_net_ns);
3062         }
3063
3064         ret = 0;
3065         rtnl_lock();
3066
3067         tun = tun_get(tfile);
3068         if (cmd == TUNSETIFF) {
3069                 ret = -EEXIST;
3070                 if (tun)
3071                         goto unlock;
3072
3073                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3074
3075                 ret = tun_set_iff(net, file, &ifr);
3076
3077                 if (ret)
3078                         goto unlock;
3079
3080                 if (copy_to_user(argp, &ifr, ifreq_len))
3081                         ret = -EFAULT;
3082                 goto unlock;
3083         }
3084         if (cmd == TUNSETIFINDEX) {
3085                 ret = -EPERM;
3086                 if (tun)
3087                         goto unlock;
3088
3089                 ret = -EFAULT;
3090                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3091                         goto unlock;
3092
3093                 ret = 0;
3094                 tfile->ifindex = ifindex;
3095                 goto unlock;
3096         }
3097
3098         ret = -EBADFD;
3099         if (!tun)
3100                 goto unlock;
3101
3102         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3103
3104         net = dev_net(tun->dev);
3105         ret = 0;
3106         switch (cmd) {
3107         case TUNGETIFF:
3108                 tun_get_iff(tun, &ifr);
3109
3110                 if (tfile->detached)
3111                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3112                 if (!tfile->socket.sk->sk_filter)
3113                         ifr.ifr_flags |= IFF_NOFILTER;
3114
3115                 if (copy_to_user(argp, &ifr, ifreq_len))
3116                         ret = -EFAULT;
3117                 break;
3118
3119         case TUNSETNOCSUM:
3120                 /* Disable/Enable checksum */
3121
3122                 /* [unimplemented] */
3123                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3124                           arg ? "disabled" : "enabled");
3125                 break;
3126
3127         case TUNSETPERSIST:
3128                 /* Disable/Enable persist mode. Keep an extra reference to the
3129                  * module to prevent the module being unprobed.
3130                  */
3131                 if (arg && !(tun->flags & IFF_PERSIST)) {
3132                         tun->flags |= IFF_PERSIST;
3133                         __module_get(THIS_MODULE);
3134                         do_notify = true;
3135                 }
3136                 if (!arg && (tun->flags & IFF_PERSIST)) {
3137                         tun->flags &= ~IFF_PERSIST;
3138                         module_put(THIS_MODULE);
3139                         do_notify = true;
3140                 }
3141
3142                 tun_debug(KERN_INFO, tun, "persist %s\n",
3143                           arg ? "enabled" : "disabled");
3144                 break;
3145
3146         case TUNSETOWNER:
3147                 /* Set owner of the device */
3148                 owner = make_kuid(current_user_ns(), arg);
3149                 if (!uid_valid(owner)) {
3150                         ret = -EINVAL;
3151                         break;
3152                 }
3153                 tun->owner = owner;
3154                 do_notify = true;
3155                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3156                           from_kuid(&init_user_ns, tun->owner));
3157                 break;
3158
3159         case TUNSETGROUP:
3160                 /* Set group of the device */
3161                 group = make_kgid(current_user_ns(), arg);
3162                 if (!gid_valid(group)) {
3163                         ret = -EINVAL;
3164                         break;
3165                 }
3166                 tun->group = group;
3167                 do_notify = true;
3168                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3169                           from_kgid(&init_user_ns, tun->group));
3170                 break;
3171
3172         case TUNSETLINK:
3173                 /* Only allow setting the type when the interface is down */
3174                 if (tun->dev->flags & IFF_UP) {
3175                         tun_debug(KERN_INFO, tun,
3176                                   "Linktype set failed because interface is up\n");
3177                         ret = -EBUSY;
3178                 } else {
3179                         tun->dev->type = (int) arg;
3180                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3181                                   tun->dev->type);
3182                         ret = 0;
3183                 }
3184                 break;
3185
3186 #ifdef TUN_DEBUG
3187         case TUNSETDEBUG:
3188                 tun->debug = arg;
3189                 break;
3190 #endif
3191         case TUNSETOFFLOAD:
3192                 ret = set_offload(tun, arg);
3193                 break;
3194
3195         case TUNSETTXFILTER:
3196                 /* Can be set only for TAPs */
3197                 ret = -EINVAL;
3198                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3199                         break;
3200                 ret = update_filter(&tun->txflt, (void __user *)arg);
3201                 break;
3202
3203         case SIOCGIFHWADDR:
3204                 /* Get hw address */
3205                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3206                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3207                 if (copy_to_user(argp, &ifr, ifreq_len))
3208                         ret = -EFAULT;
3209                 break;
3210
3211         case SIOCSIFHWADDR:
3212                 /* Set hw address */
3213                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3214                           ifr.ifr_hwaddr.sa_data);
3215
3216                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3217                 break;
3218
3219         case TUNGETSNDBUF:
3220                 sndbuf = tfile->socket.sk->sk_sndbuf;
3221                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3222                         ret = -EFAULT;
3223                 break;
3224
3225         case TUNSETSNDBUF:
3226                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3227                         ret = -EFAULT;
3228                         break;
3229                 }
3230                 if (sndbuf <= 0) {
3231                         ret = -EINVAL;
3232                         break;
3233                 }
3234
3235                 tun->sndbuf = sndbuf;
3236                 tun_set_sndbuf(tun);
3237                 break;
3238
3239         case TUNGETVNETHDRSZ:
3240                 vnet_hdr_sz = tun->vnet_hdr_sz;
3241                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3242                         ret = -EFAULT;
3243                 break;
3244
3245         case TUNSETVNETHDRSZ:
3246                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3247                         ret = -EFAULT;
3248                         break;
3249                 }
3250                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3251                         ret = -EINVAL;
3252                         break;
3253                 }
3254
3255                 tun->vnet_hdr_sz = vnet_hdr_sz;
3256                 break;
3257
3258         case TUNGETVNETLE:
3259                 le = !!(tun->flags & TUN_VNET_LE);
3260                 if (put_user(le, (int __user *)argp))
3261                         ret = -EFAULT;
3262                 break;
3263
3264         case TUNSETVNETLE:
3265                 if (get_user(le, (int __user *)argp)) {
3266                         ret = -EFAULT;
3267                         break;
3268                 }
3269                 if (le)
3270                         tun->flags |= TUN_VNET_LE;
3271                 else
3272                         tun->flags &= ~TUN_VNET_LE;
3273                 break;
3274
3275         case TUNGETVNETBE:
3276                 ret = tun_get_vnet_be(tun, argp);
3277                 break;
3278
3279         case TUNSETVNETBE:
3280                 ret = tun_set_vnet_be(tun, argp);
3281                 break;
3282
3283         case TUNATTACHFILTER:
3284                 /* Can be set only for TAPs */
3285                 ret = -EINVAL;
3286                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3287                         break;
3288                 ret = -EFAULT;
3289                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3290                         break;
3291
3292                 ret = tun_attach_filter(tun);
3293                 break;
3294
3295         case TUNDETACHFILTER:
3296                 /* Can be set only for TAPs */
3297                 ret = -EINVAL;
3298                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3299                         break;
3300                 ret = 0;
3301                 tun_detach_filter(tun, tun->numqueues);
3302                 break;
3303
3304         case TUNGETFILTER:
3305                 ret = -EINVAL;
3306                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3307                         break;
3308                 ret = -EFAULT;
3309                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3310                         break;
3311                 ret = 0;
3312                 break;
3313
3314         case TUNSETSTEERINGEBPF:
3315                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3316                 break;
3317
3318         case TUNSETFILTEREBPF:
3319                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3320                 break;
3321
3322         case TUNSETCARRIER:
3323                 ret = -EFAULT;
3324                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3325                         goto unlock;
3326
3327                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3328                 break;
3329
3330         case TUNGETDEVNETNS:
3331                 ret = -EPERM;
3332                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3333                         goto unlock;
3334                 ret = open_related_ns(&net->ns, get_net_ns);
3335                 break;
3336
3337         default:
3338                 ret = -EINVAL;
3339                 break;
3340         }
3341
3342         if (do_notify)
3343                 netdev_state_change(tun->dev);
3344
3345 unlock:
3346         rtnl_unlock();
3347         if (tun)
3348                 tun_put(tun);
3349         return ret;
3350 }
3351
3352 static long tun_chr_ioctl(struct file *file,
3353                           unsigned int cmd, unsigned long arg)
3354 {
3355         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3356 }
3357
3358 #ifdef CONFIG_COMPAT
3359 static long tun_chr_compat_ioctl(struct file *file,
3360                          unsigned int cmd, unsigned long arg)
3361 {
3362         switch (cmd) {
3363         case TUNSETIFF:
3364         case TUNGETIFF:
3365         case TUNSETTXFILTER:
3366         case TUNGETSNDBUF:
3367         case TUNSETSNDBUF:
3368         case SIOCGIFHWADDR:
3369         case SIOCSIFHWADDR:
3370                 arg = (unsigned long)compat_ptr(arg);
3371                 break;
3372         default:
3373                 arg = (compat_ulong_t)arg;
3374                 break;
3375         }
3376
3377         /*
3378          * compat_ifreq is shorter than ifreq, so we must not access beyond
3379          * the end of that structure. All fields that are used in this
3380          * driver are compatible though, we don't need to convert the
3381          * contents.
3382          */
3383         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3384 }
3385 #endif /* CONFIG_COMPAT */
3386
3387 static int tun_chr_fasync(int fd, struct file *file, int on)
3388 {
3389         struct tun_file *tfile = file->private_data;
3390         int ret;
3391
3392         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3393                 goto out;
3394
3395         if (on) {
3396                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3397                 tfile->flags |= TUN_FASYNC;
3398         } else
3399                 tfile->flags &= ~TUN_FASYNC;
3400         ret = 0;
3401 out:
3402         return ret;
3403 }
3404
3405 static int tun_chr_open(struct inode *inode, struct file * file)
3406 {
3407         struct net *net = current->nsproxy->net_ns;
3408         struct tun_file *tfile;
3409
3410         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3411
3412         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3413                                             &tun_proto, 0);
3414         if (!tfile)
3415                 return -ENOMEM;
3416         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3417                 sk_free(&tfile->sk);
3418                 return -ENOMEM;
3419         }
3420
3421         mutex_init(&tfile->napi_mutex);
3422         RCU_INIT_POINTER(tfile->tun, NULL);
3423         tfile->flags = 0;
3424         tfile->ifindex = 0;
3425
3426         init_waitqueue_head(&tfile->wq.wait);
3427         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3428
3429         tfile->socket.file = file;
3430         tfile->socket.ops = &tun_socket_ops;
3431
3432         sock_init_data(&tfile->socket, &tfile->sk);
3433
3434         tfile->sk.sk_write_space = tun_sock_write_space;
3435         tfile->sk.sk_sndbuf = INT_MAX;
3436
3437         file->private_data = tfile;
3438         INIT_LIST_HEAD(&tfile->next);
3439
3440         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3441
3442         return 0;
3443 }
3444
3445 static int tun_chr_close(struct inode *inode, struct file *file)
3446 {
3447         struct tun_file *tfile = file->private_data;
3448
3449         tun_detach(tfile, true);
3450
3451         return 0;
3452 }
3453
3454 #ifdef CONFIG_PROC_FS
3455 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3456 {
3457         struct tun_file *tfile = file->private_data;
3458         struct tun_struct *tun;
3459         struct ifreq ifr;
3460
3461         memset(&ifr, 0, sizeof(ifr));
3462
3463         rtnl_lock();
3464         tun = tun_get(tfile);
3465         if (tun)
3466                 tun_get_iff(tun, &ifr);
3467         rtnl_unlock();
3468
3469         if (tun)
3470                 tun_put(tun);
3471
3472         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3473 }
3474 #endif
3475
3476 static const struct file_operations tun_fops = {
3477         .owner  = THIS_MODULE,
3478         .llseek = no_llseek,
3479         .read_iter  = tun_chr_read_iter,
3480         .write_iter = tun_chr_write_iter,
3481         .poll   = tun_chr_poll,
3482         .unlocked_ioctl = tun_chr_ioctl,
3483 #ifdef CONFIG_COMPAT
3484         .compat_ioctl = tun_chr_compat_ioctl,
3485 #endif
3486         .open   = tun_chr_open,
3487         .release = tun_chr_close,
3488         .fasync = tun_chr_fasync,
3489 #ifdef CONFIG_PROC_FS
3490         .show_fdinfo = tun_chr_show_fdinfo,
3491 #endif
3492 };
3493
3494 static struct miscdevice tun_miscdev = {
3495         .minor = TUN_MINOR,
3496         .name = "tun",
3497         .nodename = "net/tun",
3498         .fops = &tun_fops,
3499 };
3500
3501 /* ethtool interface */
3502
3503 static void tun_default_link_ksettings(struct net_device *dev,
3504                                        struct ethtool_link_ksettings *cmd)
3505 {
3506         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3507         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3508         cmd->base.speed         = SPEED_10;
3509         cmd->base.duplex        = DUPLEX_FULL;
3510         cmd->base.port          = PORT_TP;
3511         cmd->base.phy_address   = 0;
3512         cmd->base.autoneg       = AUTONEG_DISABLE;
3513 }
3514
3515 static int tun_get_link_ksettings(struct net_device *dev,
3516                                   struct ethtool_link_ksettings *cmd)
3517 {
3518         struct tun_struct *tun = netdev_priv(dev);
3519
3520         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3521         return 0;
3522 }
3523
3524 static int tun_set_link_ksettings(struct net_device *dev,
3525                                   const struct ethtool_link_ksettings *cmd)
3526 {
3527         struct tun_struct *tun = netdev_priv(dev);
3528
3529         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3530         return 0;
3531 }
3532
3533 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3534 {
3535         struct tun_struct *tun = netdev_priv(dev);
3536
3537         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3538         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3539
3540         switch (tun->flags & TUN_TYPE_MASK) {
3541         case IFF_TUN:
3542                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3543                 break;
3544         case IFF_TAP:
3545                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3546                 break;
3547         }
3548 }
3549
3550 static u32 tun_get_msglevel(struct net_device *dev)
3551 {
3552 #ifdef TUN_DEBUG
3553         struct tun_struct *tun = netdev_priv(dev);
3554         return tun->debug;
3555 #else
3556         return -EOPNOTSUPP;
3557 #endif
3558 }
3559
3560 static void tun_set_msglevel(struct net_device *dev, u32 value)
3561 {
3562 #ifdef TUN_DEBUG
3563         struct tun_struct *tun = netdev_priv(dev);
3564         tun->debug = value;
3565 #endif
3566 }
3567
3568 static int tun_get_coalesce(struct net_device *dev,
3569                             struct ethtool_coalesce *ec)
3570 {
3571         struct tun_struct *tun = netdev_priv(dev);
3572
3573         ec->rx_max_coalesced_frames = tun->rx_batched;
3574
3575         return 0;
3576 }
3577
3578 static int tun_set_coalesce(struct net_device *dev,
3579                             struct ethtool_coalesce *ec)
3580 {
3581         struct tun_struct *tun = netdev_priv(dev);
3582
3583         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3584                 tun->rx_batched = NAPI_POLL_WEIGHT;
3585         else
3586                 tun->rx_batched = ec->rx_max_coalesced_frames;
3587
3588         return 0;
3589 }
3590
3591 static const struct ethtool_ops tun_ethtool_ops = {
3592         .get_drvinfo    = tun_get_drvinfo,
3593         .get_msglevel   = tun_get_msglevel,
3594         .set_msglevel   = tun_set_msglevel,
3595         .get_link       = ethtool_op_get_link,
3596         .get_ts_info    = ethtool_op_get_ts_info,
3597         .get_coalesce   = tun_get_coalesce,
3598         .set_coalesce   = tun_set_coalesce,
3599         .get_link_ksettings = tun_get_link_ksettings,
3600         .set_link_ksettings = tun_set_link_ksettings,
3601 };
3602
3603 static int tun_queue_resize(struct tun_struct *tun)
3604 {
3605         struct net_device *dev = tun->dev;
3606         struct tun_file *tfile;
3607         struct ptr_ring **rings;
3608         int n = tun->numqueues + tun->numdisabled;
3609         int ret, i;
3610
3611         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3612         if (!rings)
3613                 return -ENOMEM;
3614
3615         for (i = 0; i < tun->numqueues; i++) {
3616                 tfile = rtnl_dereference(tun->tfiles[i]);
3617                 rings[i] = &tfile->tx_ring;
3618         }
3619         list_for_each_entry(tfile, &tun->disabled, next)
3620                 rings[i++] = &tfile->tx_ring;
3621
3622         ret = ptr_ring_resize_multiple(rings, n,
3623                                        dev->tx_queue_len, GFP_KERNEL,
3624                                        tun_ptr_free);
3625
3626         kfree(rings);
3627         return ret;
3628 }
3629
3630 static int tun_device_event(struct notifier_block *unused,
3631                             unsigned long event, void *ptr)
3632 {
3633         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3634         struct tun_struct *tun = netdev_priv(dev);
3635
3636         if (dev->rtnl_link_ops != &tun_link_ops)
3637                 return NOTIFY_DONE;
3638
3639         switch (event) {
3640         case NETDEV_CHANGE_TX_QUEUE_LEN:
3641                 if (tun_queue_resize(tun))
3642                         return NOTIFY_BAD;
3643                 break;
3644         default:
3645                 break;
3646         }
3647
3648         return NOTIFY_DONE;
3649 }
3650
3651 static struct notifier_block tun_notifier_block __read_mostly = {
3652         .notifier_call  = tun_device_event,
3653 };
3654
3655 static int __init tun_init(void)
3656 {
3657         int ret = 0;
3658
3659         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3660
3661         ret = rtnl_link_register(&tun_link_ops);
3662         if (ret) {
3663                 pr_err("Can't register link_ops\n");
3664                 goto err_linkops;
3665         }
3666
3667         ret = misc_register(&tun_miscdev);
3668         if (ret) {
3669                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3670                 goto err_misc;
3671         }
3672
3673         ret = register_netdevice_notifier(&tun_notifier_block);
3674         if (ret) {
3675                 pr_err("Can't register netdevice notifier\n");
3676                 goto err_notifier;
3677         }
3678
3679         return  0;
3680
3681 err_notifier:
3682         misc_deregister(&tun_miscdev);
3683 err_misc:
3684         rtnl_link_unregister(&tun_link_ops);
3685 err_linkops:
3686         return ret;
3687 }
3688
3689 static void tun_cleanup(void)
3690 {
3691         misc_deregister(&tun_miscdev);
3692         rtnl_link_unregister(&tun_link_ops);
3693         unregister_netdevice_notifier(&tun_notifier_block);
3694 }
3695
3696 /* Get an underlying socket object from tun file.  Returns error unless file is
3697  * attached to a device.  The returned object works like a packet socket, it
3698  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3699  * holding a reference to the file for as long as the socket is in use. */
3700 struct socket *tun_get_socket(struct file *file)
3701 {
3702         struct tun_file *tfile;
3703         if (file->f_op != &tun_fops)
3704                 return ERR_PTR(-EINVAL);
3705         tfile = file->private_data;
3706         if (!tfile)
3707                 return ERR_PTR(-EBADFD);
3708         return &tfile->socket;
3709 }
3710 EXPORT_SYMBOL_GPL(tun_get_socket);
3711
3712 struct ptr_ring *tun_get_tx_ring(struct file *file)
3713 {
3714         struct tun_file *tfile;
3715
3716         if (file->f_op != &tun_fops)
3717                 return ERR_PTR(-EINVAL);
3718         tfile = file->private_data;
3719         if (!tfile)
3720                 return ERR_PTR(-EBADFD);
3721         return &tfile->tx_ring;
3722 }
3723 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3724
3725 module_init(tun_init);
3726 module_exit(tun_cleanup);
3727 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3728 MODULE_AUTHOR(DRV_COPYRIGHT);
3729 MODULE_LICENSE("GPL");
3730 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3731 MODULE_ALIAS("devname:net/tun");