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         u32 numqueues;
 600         u16 ret = 0;
 601
 602         numqueues = READ_ONCE(tun->numqueues);
 603         if (!numqueues)
 604                 return 0;
 605
 606         prog = rcu_dereference(tun->steering_prog);
 607         if (prog)
 608                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 609
 610         return ret % numqueues;
 611 }
 612
 613 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 614                             struct net_device *sb_dev)
 615 {
 616         struct tun_struct *tun = netdev_priv(dev);
 617         u16 ret;
 618
 619         rcu_read_lock();
 620         if (rcu_dereference(tun->steering_prog))
 621                 ret = tun_ebpf_select_queue(tun, skb);
 622         else
 623                 ret = tun_automq_select_queue(tun, skb);
 624         rcu_read_unlock();
 625
 626         return ret;
 627 }
 628
 629 static inline bool tun_not_capable(struct tun_struct *tun)
 630 {
 631         const struct cred *cred = current_cred();
 632         struct net *net = dev_net(tun->dev);
 633
 634         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 635                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 636                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 637 }
 638
 639 static void tun_set_real_num_queues(struct tun_struct *tun)
 640 {
 641         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 642         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 643 }
 644
 645 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 646 {
 647         tfile->detached = tun;
 648         list_add_tail(&tfile->next, &tun->disabled);
 649         ++tun->numdisabled;
 650 }
 651
 652 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 653 {
 654         struct tun_struct *tun = tfile->detached;
 655
 656         tfile->detached = NULL;
 657         list_del_init(&tfile->next);
 658         --tun->numdisabled;
 659         return tun;
 660 }
 661
 662 void tun_ptr_free(void *ptr)
 663 {
 664         if (!ptr)
 665                 return;
 666         if (tun_is_xdp_frame(ptr)) {
 667                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 668
 669                 xdp_return_frame(xdpf);
 670         } else {
 671                 __skb_array_destroy_skb(ptr);
 672         }
 673 }
 674 EXPORT_SYMBOL_GPL(tun_ptr_free);
 675
 676 static void tun_queue_purge(struct tun_file *tfile)
 677 {
 678         void *ptr;
 679
 680         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 681                 tun_ptr_free(ptr);
 682
 683         skb_queue_purge(&tfile->sk.sk_write_queue);
 684         skb_queue_purge(&tfile->sk.sk_error_queue);
 685 }
 686
 687 static void __tun_detach(struct tun_file *tfile, bool clean)
 688 {
 689         struct tun_file *ntfile;
 690         struct tun_struct *tun;
 691
 692         tun = rtnl_dereference(tfile->tun);
 693
 694         if (tun && clean) {
 695                 tun_napi_disable(tfile);
 696                 tun_napi_del(tfile);
 697         }
 698
 699         if (tun && !tfile->detached) {
 700                 u16 index = tfile->queue_index;
 701                 BUG_ON(index >= tun->numqueues);
 702
 703                 rcu_assign_pointer(tun->tfiles[index],
 704                                    tun->tfiles[tun->numqueues - 1]);
 705                 ntfile = rtnl_dereference(tun->tfiles[index]);
 706                 ntfile->queue_index = index;
 707
 708                 --tun->numqueues;
 709                 if (clean) {
 710                         RCU_INIT_POINTER(tfile->tun, NULL);
 711                         sock_put(&tfile->sk);
 712                 } else
 713                         tun_disable_queue(tun, tfile);
 714
 715                 synchronize_net();
 716                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 717                 /* Drop read queue */
 718                 tun_queue_purge(tfile);
 719                 tun_set_real_num_queues(tun);
 720         } else if (tfile->detached && clean) {
 721                 tun = tun_enable_queue(tfile);
 722                 sock_put(&tfile->sk);
 723         }
 724
 725         if (clean) {
 726                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 727                         netif_carrier_off(tun->dev);
 728
 729                         if (!(tun->flags & IFF_PERSIST) &&
 730                             tun->dev->reg_state == NETREG_REGISTERED)
 731                                 unregister_netdevice(tun->dev);
 732                 }
 733                 if (tun)
 734                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 735                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 736                 sock_put(&tfile->sk);
 737         }
 738 }
 739
 740 static void tun_detach(struct tun_file *tfile, bool clean)
 741 {
 742         struct tun_struct *tun;
 743         struct net_device *dev;
 744
 745         rtnl_lock();
 746         tun = rtnl_dereference(tfile->tun);
 747         dev = tun ? tun->dev : NULL;
 748         __tun_detach(tfile, clean);
 749         if (dev)
 750                 netdev_state_change(dev);
 751         rtnl_unlock();
 752 }
 753
 754 static void tun_detach_all(struct net_device *dev)
 755 {
 756         struct tun_struct *tun = netdev_priv(dev);
 757         struct tun_file *tfile, *tmp;
 758         int i, n = tun->numqueues;
 759
 760         for (i = 0; i < n; i++) {
 761                 tfile = rtnl_dereference(tun->tfiles[i]);
 762                 BUG_ON(!tfile);
 763                 tun_napi_disable(tfile);
 764                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 765                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 766                 RCU_INIT_POINTER(tfile->tun, NULL);
 767                 --tun->numqueues;
 768         }
 769         list_for_each_entry(tfile, &tun->disabled, next) {
 770                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 771                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 772                 RCU_INIT_POINTER(tfile->tun, NULL);
 773         }
 774         BUG_ON(tun->numqueues != 0);
 775
 776         synchronize_net();
 777         for (i = 0; i < n; i++) {
 778                 tfile = rtnl_dereference(tun->tfiles[i]);
 779                 tun_napi_del(tfile);
 780                 /* Drop read queue */
 781                 tun_queue_purge(tfile);
 782                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 783                 sock_put(&tfile->sk);
 784         }
 785         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 786                 tun_enable_queue(tfile);
 787                 tun_queue_purge(tfile);
 788                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 789                 sock_put(&tfile->sk);
 790         }
 791         BUG_ON(tun->numdisabled != 0);
 792
 793         if (tun->flags & IFF_PERSIST)
 794                 module_put(THIS_MODULE);
 795 }
 796
 797 static int tun_attach(struct tun_struct *tun, struct file *file,
 798                       bool skip_filter, bool napi, bool napi_frags)
 799 {
 800         struct tun_file *tfile = file->private_data;
 801         struct net_device *dev = tun->dev;
 802         int err;
 803
 804         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 805         if (err < 0)
 806                 goto out;
 807
 808         err = -EINVAL;
 809         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 810                 goto out;
 811
 812         err = -EBUSY;
 813         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 814                 goto out;
 815
 816         err = -E2BIG;
 817         if (!tfile->detached &&
 818             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 819                 goto out;
 820
 821         err = 0;
 822
 823         /* Re-attach the filter to persist device */
 824         if (!skip_filter && (tun->filter_attached == true)) {
 825                 lock_sock(tfile->socket.sk);
 826                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 827                 release_sock(tfile->socket.sk);
 828                 if (!err)
 829                         goto out;
 830         }
 831
 832         if (!tfile->detached &&
 833             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 834                             GFP_KERNEL, tun_ptr_free)) {
 835                 err = -ENOMEM;
 836                 goto out;
 837         }
 838
 839         tfile->queue_index = tun->numqueues;
 840         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 841
 842         if (tfile->detached) {
 843                 /* Re-attach detached tfile, updating XDP queue_index */
 844                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 845
 846                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 847                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 848         } else {
 849                 /* Setup XDP RX-queue info, for new tfile getting attached */
 850                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 851                                        tun->dev, tfile->queue_index);
 852                 if (err < 0)
 853                         goto out;
 854                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 855                                                  MEM_TYPE_PAGE_SHARED, NULL);
 856                 if (err < 0) {
 857                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 858                         goto out;
 859                 }
 860                 err = 0;
 861         }
 862
 863         if (tfile->detached) {
 864                 tun_enable_queue(tfile);
 865         } else {
 866                 sock_hold(&tfile->sk);
 867                 tun_napi_init(tun, tfile, napi, napi_frags);
 868         }
 869
 870         if (rtnl_dereference(tun->xdp_prog))
 871                 sock_set_flag(&tfile->sk, SOCK_XDP);
 872
 873         /* device is allowed to go away first, so no need to hold extra
 874          * refcnt.
 875          */
 876
 877         /* Publish tfile->tun and tun->tfiles only after we've fully
 878          * initialized tfile; otherwise we risk using half-initialized
 879          * object.
 880          */
 881         rcu_assign_pointer(tfile->tun, tun);
 882         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 883         tun->numqueues++;
 884         tun_set_real_num_queues(tun);
 885 out:
 886         return err;
 887 }
 888
 889 static struct tun_struct *tun_get(struct tun_file *tfile)
 890 {
 891         struct tun_struct *tun;
 892
 893         rcu_read_lock();
 894         tun = rcu_dereference(tfile->tun);
 895         if (tun)
 896                 dev_hold(tun->dev);
 897         rcu_read_unlock();
 898
 899         return tun;
 900 }
 901
 902 static void tun_put(struct tun_struct *tun)
 903 {
 904         dev_put(tun->dev);
 905 }
 906
 907 /* TAP filtering */
 908 static void addr_hash_set(u32 *mask, const u8 *addr)
 909 {
 910         int n = ether_crc(ETH_ALEN, addr) >> 26;
 911         mask[n >> 5] |= (1 << (n & 31));
 912 }
 913
 914 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 915 {
 916         int n = ether_crc(ETH_ALEN, addr) >> 26;
 917         return mask[n >> 5] & (1 << (n & 31));
 918 }
 919
 920 static int update_filter(struct tap_filter *filter, void __user *arg)
 921 {
 922         struct { u8 u[ETH_ALEN]; } *addr;
 923         struct tun_filter uf;
 924         int err, alen, n, nexact;
 925
 926         if (copy_from_user(&uf, arg, sizeof(uf)))
 927                 return -EFAULT;
 928
 929         if (!uf.count) {
 930                 /* Disabled */
 931                 filter->count = 0;
 932                 return 0;
 933         }
 934
 935         alen = ETH_ALEN * uf.count;
 936         addr = memdup_user(arg + sizeof(uf), alen);
 937         if (IS_ERR(addr))
 938                 return PTR_ERR(addr);
 939
 940         /* The filter is updated without holding any locks. Which is
 941          * perfectly safe. We disable it first and in the worst
 942          * case we'll accept a few undesired packets. */
 943         filter->count = 0;
 944         wmb();
 945
 946         /* Use first set of addresses as an exact filter */
 947         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 948                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 949
 950         nexact = n;
 951
 952         /* Remaining multicast addresses are hashed,
 953          * unicast will leave the filter disabled. */
 954         memset(filter->mask, 0, sizeof(filter->mask));
 955         for (; n < uf.count; n++) {
 956                 if (!is_multicast_ether_addr(addr[n].u)) {
 957                         err = 0; /* no filter */
 958                         goto free_addr;
 959                 }
 960                 addr_hash_set(filter->mask, addr[n].u);
 961         }
 962
 963         /* For ALLMULTI just set the mask to all ones.
 964          * This overrides the mask populated above. */
 965         if ((uf.flags & TUN_FLT_ALLMULTI))
 966                 memset(filter->mask, ~0, sizeof(filter->mask));
 967
 968         /* Now enable the filter */
 969         wmb();
 970         filter->count = nexact;
 971
 972         /* Return the number of exact filters */
 973         err = nexact;
 974 free_addr:
 975         kfree(addr);
 976         return err;
 977 }
 978
 979 /* Returns: 0 - drop, !=0 - accept */
 980 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 981 {
 982         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 983          * at this point. */
 984         struct ethhdr *eh = (struct ethhdr *) skb->data;
 985         int i;
 986
 987         /* Exact match */
 988         for (i = 0; i < filter->count; i++)
 989                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 990                         return 1;
 991
 992         /* Inexact match (multicast only) */
 993         if (is_multicast_ether_addr(eh->h_dest))
 994                 return addr_hash_test(filter->mask, eh->h_dest);
 995
 996         return 0;
 997 }
 998
 999 /*
1000  * Checks whether the packet is accepted or not.
1001  * Returns: 0 - drop, !=0 - accept
1002  */
1003 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1004 {
1005         if (!filter->count)
1006                 return 1;
1007
1008         return run_filter(filter, skb);
1009 }
1010
1011 /* Network device part of the driver */
1012
1013 static const struct ethtool_ops tun_ethtool_ops;
1014
1015 /* Net device detach from fd. */
1016 static void tun_net_uninit(struct net_device *dev)
1017 {
1018         tun_detach_all(dev);
1019 }
1020
1021 /* Net device open. */
1022 static int tun_net_open(struct net_device *dev)
1023 {
1024         struct tun_struct *tun = netdev_priv(dev);
1025         int i;
1026
1027         netif_tx_start_all_queues(dev);
1028
1029         for (i = 0; i < tun->numqueues; i++) {
1030                 struct tun_file *tfile;
1031
1032                 tfile = rtnl_dereference(tun->tfiles[i]);
1033                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1034         }
1035
1036         return 0;
1037 }
1038
1039 /* Net device close. */
1040 static int tun_net_close(struct net_device *dev)
1041 {
1042         netif_tx_stop_all_queues(dev);
1043         return 0;
1044 }
1045
1046 /* Net device start xmit */
1047 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1048 {
1049 #ifdef CONFIG_RPS
1050         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1051                 /* Select queue was not called for the skbuff, so we extract the
1052                  * RPS hash and save it into the flow_table here.
1053                  */
1054                 struct tun_flow_entry *e;
1055                 __u32 rxhash;
1056
1057                 rxhash = __skb_get_hash_symmetric(skb);
1058                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1059                 if (e)
1060                         tun_flow_save_rps_rxhash(e, rxhash);
1061         }
1062 #endif
1063 }
1064
1065 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1066                                     struct sk_buff *skb,
1067                                     int len)
1068 {
1069         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1070
1071         if (prog)
1072                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1073
1074         return len;
1075 }
1076
1077 /* Net device start xmit */
1078 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1079 {
1080         struct tun_struct *tun = netdev_priv(dev);
1081         int txq = skb->queue_mapping;
1082         struct tun_file *tfile;
1083         int len = skb->len;
1084
1085         rcu_read_lock();
1086         tfile = rcu_dereference(tun->tfiles[txq]);
1087
1088         /* Drop packet if interface is not attached */
1089         if (txq >= tun->numqueues)
1090                 goto drop;
1091
1092         if (!rcu_dereference(tun->steering_prog))
1093                 tun_automq_xmit(tun, skb);
1094
1095         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1096
1097         BUG_ON(!tfile);
1098
1099         /* Drop if the filter does not like it.
1100          * This is a noop if the filter is disabled.
1101          * Filter can be enabled only for the TAP devices. */
1102         if (!check_filter(&tun->txflt, skb))
1103                 goto drop;
1104
1105         if (tfile->socket.sk->sk_filter &&
1106             sk_filter(tfile->socket.sk, skb))
1107                 goto drop;
1108
1109         len = run_ebpf_filter(tun, skb, len);
1110         if (len == 0 || pskb_trim(skb, len))
1111                 goto drop;
1112
1113         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1114                 goto drop;
1115
1116         skb_tx_timestamp(skb);
1117
1118         /* Orphan the skb - required as we might hang on to it
1119          * for indefinite time.
1120          */
1121         skb_orphan(skb);
1122
1123         nf_reset(skb);
1124
1125         if (ptr_ring_produce(&tfile->tx_ring, skb))
1126                 goto drop;
1127
1128         /* Notify and wake up reader process */
1129         if (tfile->flags & TUN_FASYNC)
1130                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1131         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1132
1133         rcu_read_unlock();
1134         return NETDEV_TX_OK;
1135
1136 drop:
1137         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1138         skb_tx_error(skb);
1139         kfree_skb(skb);
1140         rcu_read_unlock();
1141         return NET_XMIT_DROP;
1142 }
1143
1144 static void tun_net_mclist(struct net_device *dev)
1145 {
1146         /*
1147          * This callback is supposed to deal with mc filter in
1148          * _rx_ path and has nothing to do with the _tx_ path.
1149          * In rx path we always accept everything userspace gives us.
1150          */
1151 }
1152
1153 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1154         netdev_features_t features)
1155 {
1156         struct tun_struct *tun = netdev_priv(dev);
1157
1158         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1159 }
1160
1161 static void tun_set_headroom(struct net_device *dev, int new_hr)
1162 {
1163         struct tun_struct *tun = netdev_priv(dev);
1164
1165         if (new_hr < NET_SKB_PAD)
1166                 new_hr = NET_SKB_PAD;
1167
1168         tun->align = new_hr;
1169 }
1170
1171 static void
1172 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1173 {
1174         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1175         struct tun_struct *tun = netdev_priv(dev);
1176         struct tun_pcpu_stats *p;
1177         int i;
1178
1179         for_each_possible_cpu(i) {
1180                 u64 rxpackets, rxbytes, txpackets, txbytes;
1181                 unsigned int start;
1182
1183                 p = per_cpu_ptr(tun->pcpu_stats, i);
1184                 do {
1185                         start = u64_stats_fetch_begin(&p->syncp);
1186                         rxpackets       = p->rx_packets;
1187                         rxbytes         = p->rx_bytes;
1188                         txpackets       = p->tx_packets;
1189                         txbytes         = p->tx_bytes;
1190                 } while (u64_stats_fetch_retry(&p->syncp, start));
1191
1192                 stats->rx_packets       += rxpackets;
1193                 stats->rx_bytes         += rxbytes;
1194                 stats->tx_packets       += txpackets;
1195                 stats->tx_bytes         += txbytes;
1196
1197                 /* u32 counters */
1198                 rx_dropped      += p->rx_dropped;
1199                 rx_frame_errors += p->rx_frame_errors;
1200                 tx_dropped      += p->tx_dropped;
1201         }
1202         stats->rx_dropped  = rx_dropped;
1203         stats->rx_frame_errors = rx_frame_errors;
1204         stats->tx_dropped = tx_dropped;
1205 }
1206
1207 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1208                        struct netlink_ext_ack *extack)
1209 {
1210         struct tun_struct *tun = netdev_priv(dev);
1211         struct tun_file *tfile;
1212         struct bpf_prog *old_prog;
1213         int i;
1214
1215         old_prog = rtnl_dereference(tun->xdp_prog);
1216         rcu_assign_pointer(tun->xdp_prog, prog);
1217         if (old_prog)
1218                 bpf_prog_put(old_prog);
1219
1220         for (i = 0; i < tun->numqueues; i++) {
1221                 tfile = rtnl_dereference(tun->tfiles[i]);
1222                 if (prog)
1223                         sock_set_flag(&tfile->sk, SOCK_XDP);
1224                 else
1225                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1226         }
1227         list_for_each_entry(tfile, &tun->disabled, next) {
1228                 if (prog)
1229                         sock_set_flag(&tfile->sk, SOCK_XDP);
1230                 else
1231                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1232         }
1233
1234         return 0;
1235 }
1236
1237 static u32 tun_xdp_query(struct net_device *dev)
1238 {
1239         struct tun_struct *tun = netdev_priv(dev);
1240         const struct bpf_prog *xdp_prog;
1241
1242         xdp_prog = rtnl_dereference(tun->xdp_prog);
1243         if (xdp_prog)
1244                 return xdp_prog->aux->id;
1245
1246         return 0;
1247 }
1248
1249 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1250 {
1251         switch (xdp->command) {
1252         case XDP_SETUP_PROG:
1253                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1254         case XDP_QUERY_PROG:
1255                 xdp->prog_id = tun_xdp_query(dev);
1256                 return 0;
1257         default:
1258                 return -EINVAL;
1259         }
1260 }
1261
1262 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1263 {
1264         if (new_carrier) {
1265                 struct tun_struct *tun = netdev_priv(dev);
1266
1267                 if (!tun->numqueues)
1268                         return -EPERM;
1269
1270                 netif_carrier_on(dev);
1271         } else {
1272                 netif_carrier_off(dev);
1273         }
1274         return 0;
1275 }
1276
1277 static const struct net_device_ops tun_netdev_ops = {
1278         .ndo_uninit             = tun_net_uninit,
1279         .ndo_open               = tun_net_open,
1280         .ndo_stop               = tun_net_close,
1281         .ndo_start_xmit         = tun_net_xmit,
1282         .ndo_fix_features       = tun_net_fix_features,
1283         .ndo_select_queue       = tun_select_queue,
1284         .ndo_set_rx_headroom    = tun_set_headroom,
1285         .ndo_get_stats64        = tun_net_get_stats64,
1286         .ndo_change_carrier     = tun_net_change_carrier,
1287 };
1288
1289 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1290 {
1291         /* Notify and wake up reader process */
1292         if (tfile->flags & TUN_FASYNC)
1293                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1294         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1295 }
1296
1297 static int tun_xdp_xmit(struct net_device *dev, int n,
1298                         struct xdp_frame **frames, u32 flags)
1299 {
1300         struct tun_struct *tun = netdev_priv(dev);
1301         struct tun_file *tfile;
1302         u32 numqueues;
1303         int drops = 0;
1304         int cnt = n;
1305         int i;
1306
1307         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1308                 return -EINVAL;
1309
1310         rcu_read_lock();
1311
1312         numqueues = READ_ONCE(tun->numqueues);
1313         if (!numqueues) {
1314                 rcu_read_unlock();
1315                 return -ENXIO; /* Caller will free/return all frames */
1316         }
1317
1318         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1319                                             numqueues]);
1320
1321         spin_lock(&tfile->tx_ring.producer_lock);
1322         for (i = 0; i < n; i++) {
1323                 struct xdp_frame *xdp = frames[i];
1324                 /* Encode the XDP flag into lowest bit for consumer to differ
1325                  * XDP buffer from sk_buff.
1326                  */
1327                 void *frame = tun_xdp_to_ptr(xdp);
1328
1329                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1330                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1331                         xdp_return_frame_rx_napi(xdp);
1332                         drops++;
1333                 }
1334         }
1335         spin_unlock(&tfile->tx_ring.producer_lock);
1336
1337         if (flags & XDP_XMIT_FLUSH)
1338                 __tun_xdp_flush_tfile(tfile);
1339
1340         rcu_read_unlock();
1341         return cnt - drops;
1342 }
1343
1344 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1345 {
1346         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1347
1348         if (unlikely(!frame))
1349                 return -EOVERFLOW;
1350
1351         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1352 }
1353
1354 static const struct net_device_ops tap_netdev_ops = {
1355         .ndo_uninit             = tun_net_uninit,
1356         .ndo_open               = tun_net_open,
1357         .ndo_stop               = tun_net_close,
1358         .ndo_start_xmit         = tun_net_xmit,
1359         .ndo_fix_features       = tun_net_fix_features,
1360         .ndo_set_rx_mode        = tun_net_mclist,
1361         .ndo_set_mac_address    = eth_mac_addr,
1362         .ndo_validate_addr      = eth_validate_addr,
1363         .ndo_select_queue       = tun_select_queue,
1364         .ndo_features_check     = passthru_features_check,
1365         .ndo_set_rx_headroom    = tun_set_headroom,
1366         .ndo_get_stats64        = tun_net_get_stats64,
1367         .ndo_bpf                = tun_xdp,
1368         .ndo_xdp_xmit           = tun_xdp_xmit,
1369         .ndo_change_carrier     = tun_net_change_carrier,
1370 };
1371
1372 static void tun_flow_init(struct tun_struct *tun)
1373 {
1374         int i;
1375
1376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1377                 INIT_HLIST_HEAD(&tun->flows[i]);
1378
1379         tun->ageing_time = TUN_FLOW_EXPIRE;
1380         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1381         mod_timer(&tun->flow_gc_timer,
1382                   round_jiffies_up(jiffies + tun->ageing_time));
1383 }
1384
1385 static void tun_flow_uninit(struct tun_struct *tun)
1386 {
1387         del_timer_sync(&tun->flow_gc_timer);
1388         tun_flow_flush(tun);
1389 }
1390
1391 #define MIN_MTU 68
1392 #define MAX_MTU 65535
1393
1394 /* Initialize net device. */
1395 static void tun_net_init(struct net_device *dev)
1396 {
1397         struct tun_struct *tun = netdev_priv(dev);
1398
1399         switch (tun->flags & TUN_TYPE_MASK) {
1400         case IFF_TUN:
1401                 dev->netdev_ops = &tun_netdev_ops;
1402
1403                 /* Point-to-Point TUN Device */
1404                 dev->hard_header_len = 0;
1405                 dev->addr_len = 0;
1406                 dev->mtu = 1500;
1407
1408                 /* Zero header length */
1409                 dev->type = ARPHRD_NONE;
1410                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1411                 break;
1412
1413         case IFF_TAP:
1414                 dev->netdev_ops = &tap_netdev_ops;
1415                 /* Ethernet TAP Device */
1416                 ether_setup(dev);
1417                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1418                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1419
1420                 eth_hw_addr_random(dev);
1421
1422                 break;
1423         }
1424
1425         dev->min_mtu = MIN_MTU;
1426         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1427 }
1428
1429 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1430 {
1431         struct sock *sk = tfile->socket.sk;
1432
1433         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1434 }
1435
1436 /* Character device part */
1437
1438 /* Poll */
1439 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1440 {
1441         struct tun_file *tfile = file->private_data;
1442         struct tun_struct *tun = tun_get(tfile);
1443         struct sock *sk;
1444         __poll_t mask = 0;
1445
1446         if (!tun)
1447                 return EPOLLERR;
1448
1449         sk = tfile->socket.sk;
1450
1451         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1452
1453         poll_wait(file, sk_sleep(sk), wait);
1454
1455         if (!ptr_ring_empty(&tfile->tx_ring))
1456                 mask |= EPOLLIN | EPOLLRDNORM;
1457
1458         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1459          * guarantee EPOLLOUT to be raised by either here or
1460          * tun_sock_write_space(). Then process could get notification
1461          * after it writes to a down device and meets -EIO.
1462          */
1463         if (tun_sock_writeable(tun, tfile) ||
1464             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1465              tun_sock_writeable(tun, tfile)))
1466                 mask |= EPOLLOUT | EPOLLWRNORM;
1467
1468         if (tun->dev->reg_state != NETREG_REGISTERED)
1469                 mask = EPOLLERR;
1470
1471         tun_put(tun);
1472         return mask;
1473 }
1474
1475 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1476                                             size_t len,
1477                                             const struct iov_iter *it)
1478 {
1479         struct sk_buff *skb;
1480         size_t linear;
1481         int err;
1482         int i;
1483
1484         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1485                 return ERR_PTR(-ENOMEM);
1486
1487         local_bh_disable();
1488         skb = napi_get_frags(&tfile->napi);
1489         local_bh_enable();
1490         if (!skb)
1491                 return ERR_PTR(-ENOMEM);
1492
1493         linear = iov_iter_single_seg_count(it);
1494         err = __skb_grow(skb, linear);
1495         if (err)
1496                 goto free;
1497
1498         skb->len = len;
1499         skb->data_len = len - linear;
1500         skb->truesize += skb->data_len;
1501
1502         for (i = 1; i < it->nr_segs; i++) {
1503                 size_t fragsz = it->iov[i].iov_len;
1504                 struct page *page;
1505                 void *frag;
1506
1507                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1508                         err = -EINVAL;
1509                         goto free;
1510                 }
1511                 frag = netdev_alloc_frag(fragsz);
1512                 if (!frag) {
1513                         err = -ENOMEM;
1514                         goto free;
1515                 }
1516                 page = virt_to_head_page(frag);
1517                 skb_fill_page_desc(skb, i - 1, page,
1518                                    frag - page_address(page), fragsz);
1519         }
1520
1521         return skb;
1522 free:
1523         /* frees skb and all frags allocated with napi_alloc_frag() */
1524         napi_free_frags(&tfile->napi);
1525         return ERR_PTR(err);
1526 }
1527
1528 /* prepad is the amount to reserve at front.  len is length after that.
1529  * linear is a hint as to how much to copy (usually headers). */
1530 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1531                                      size_t prepad, size_t len,
1532                                      size_t linear, int noblock)
1533 {
1534         struct sock *sk = tfile->socket.sk;
1535         struct sk_buff *skb;
1536         int err;
1537
1538         /* Under a page?  Don't bother with paged skb. */
1539         if (prepad + len < PAGE_SIZE || !linear)
1540                 linear = len;
1541
1542         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1543                                    &err, 0);
1544         if (!skb)
1545                 return ERR_PTR(err);
1546
1547         skb_reserve(skb, prepad);
1548         skb_put(skb, linear);
1549         skb->data_len = len - linear;
1550         skb->len += len - linear;
1551
1552         return skb;
1553 }
1554
1555 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1556                            struct sk_buff *skb, int more)
1557 {
1558         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1559         struct sk_buff_head process_queue;
1560         u32 rx_batched = tun->rx_batched;
1561         bool rcv = false;
1562
1563         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1564                 local_bh_disable();
1565                 skb_record_rx_queue(skb, tfile->queue_index);
1566                 netif_receive_skb(skb);
1567                 local_bh_enable();
1568                 return;
1569         }
1570
1571         spin_lock(&queue->lock);
1572         if (!more || skb_queue_len(queue) == rx_batched) {
1573                 __skb_queue_head_init(&process_queue);
1574                 skb_queue_splice_tail_init(queue, &process_queue);
1575                 rcv = true;
1576         } else {
1577                 __skb_queue_tail(queue, skb);
1578         }
1579         spin_unlock(&queue->lock);
1580
1581         if (rcv) {
1582                 struct sk_buff *nskb;
1583
1584                 local_bh_disable();
1585                 while ((nskb = __skb_dequeue(&process_queue))) {
1586                         skb_record_rx_queue(nskb, tfile->queue_index);
1587                         netif_receive_skb(nskb);
1588                 }
1589                 skb_record_rx_queue(skb, tfile->queue_index);
1590                 netif_receive_skb(skb);
1591                 local_bh_enable();
1592         }
1593 }
1594
1595 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1596                               int len, int noblock, bool zerocopy)
1597 {
1598         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1599                 return false;
1600
1601         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1602                 return false;
1603
1604         if (!noblock)
1605                 return false;
1606
1607         if (zerocopy)
1608                 return false;
1609
1610         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1611             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1612                 return false;
1613
1614         return true;
1615 }
1616
1617 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1618                                        int buflen, int len, int pad)
1619 {
1620         struct sk_buff *skb = build_skb(buf, buflen);
1621
1622         if (!skb)
1623                 return ERR_PTR(-ENOMEM);
1624
1625         skb_reserve(skb, pad);
1626         skb_put(skb, len);
1627
1628         get_page(alloc_frag->page);
1629         alloc_frag->offset += buflen;
1630
1631         return skb;
1632 }
1633
1634 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1635                        struct xdp_buff *xdp, u32 act)
1636 {
1637         int err;
1638
1639         switch (act) {
1640         case XDP_REDIRECT:
1641                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1642                 if (err)
1643                         return err;
1644                 break;
1645         case XDP_TX:
1646                 err = tun_xdp_tx(tun->dev, xdp);
1647                 if (err < 0)
1648                         return err;
1649                 break;
1650         case XDP_PASS:
1651                 break;
1652         default:
1653                 bpf_warn_invalid_xdp_action(act);
1654                 /* fall through */
1655         case XDP_ABORTED:
1656                 trace_xdp_exception(tun->dev, xdp_prog, act);
1657                 /* fall through */
1658         case XDP_DROP:
1659                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1660                 break;
1661         }
1662
1663         return act;
1664 }
1665
1666 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1667                                      struct tun_file *tfile,
1668                                      struct iov_iter *from,
1669                                      struct virtio_net_hdr *hdr,
1670                                      int len, int *skb_xdp)
1671 {
1672         struct page_frag *alloc_frag = &current->task_frag;
1673         struct bpf_prog *xdp_prog;
1674         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1675         char *buf;
1676         size_t copied;
1677         int pad = TUN_RX_PAD;
1678         int err = 0;
1679
1680         rcu_read_lock();
1681         xdp_prog = rcu_dereference(tun->xdp_prog);
1682         if (xdp_prog)
1683                 pad += XDP_PACKET_HEADROOM;
1684         buflen += SKB_DATA_ALIGN(len + pad);
1685         rcu_read_unlock();
1686
1687         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1688         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1689                 return ERR_PTR(-ENOMEM);
1690
1691         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1692         copied = copy_page_from_iter(alloc_frag->page,
1693                                      alloc_frag->offset + pad,
1694                                      len, from);
1695         if (copied != len)
1696                 return ERR_PTR(-EFAULT);
1697
1698         /* There's a small window that XDP may be set after the check
1699          * of xdp_prog above, this should be rare and for simplicity
1700          * we do XDP on skb in case the headroom is not enough.
1701          */
1702         if (hdr->gso_type || !xdp_prog) {
1703                 *skb_xdp = 1;
1704                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1705         }
1706
1707         *skb_xdp = 0;
1708
1709         local_bh_disable();
1710         rcu_read_lock();
1711         xdp_prog = rcu_dereference(tun->xdp_prog);
1712         if (xdp_prog) {
1713                 struct xdp_buff xdp;
1714                 u32 act;
1715
1716                 xdp.data_hard_start = buf;
1717                 xdp.data = buf + pad;
1718                 xdp_set_data_meta_invalid(&xdp);
1719                 xdp.data_end = xdp.data + len;
1720                 xdp.rxq = &tfile->xdp_rxq;
1721
1722                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1723                 if (act == XDP_REDIRECT || act == XDP_TX) {
1724                         get_page(alloc_frag->page);
1725                         alloc_frag->offset += buflen;
1726                 }
1727                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1728                 if (err < 0)
1729                         goto err_xdp;
1730                 if (err == XDP_REDIRECT)
1731                         xdp_do_flush_map();
1732                 if (err != XDP_PASS)
1733                         goto out;
1734
1735                 pad = xdp.data - xdp.data_hard_start;
1736                 len = xdp.data_end - xdp.data;
1737         }
1738         rcu_read_unlock();
1739         local_bh_enable();
1740
1741         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1742
1743 err_xdp:
1744         put_page(alloc_frag->page);
1745 out:
1746         rcu_read_unlock();
1747         local_bh_enable();
1748         return NULL;
1749 }
1750
1751 /* Get packet from user space buffer */
1752 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1753                             void *msg_control, struct iov_iter *from,
1754                             int noblock, bool more)
1755 {
1756         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1757         struct sk_buff *skb;
1758         size_t total_len = iov_iter_count(from);
1759         size_t len = total_len, align = tun->align, linear;
1760         struct virtio_net_hdr gso = { 0 };
1761         struct tun_pcpu_stats *stats;
1762         int good_linear;
1763         int copylen;
1764         bool zerocopy = false;
1765         int err;
1766         u32 rxhash = 0;
1767         int skb_xdp = 1;
1768         bool frags = tun_napi_frags_enabled(tfile);
1769
1770         if (!(tun->flags & IFF_NO_PI)) {
1771                 if (len < sizeof(pi))
1772                         return -EINVAL;
1773                 len -= sizeof(pi);
1774
1775                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1776                         return -EFAULT;
1777         }
1778
1779         if (tun->flags & IFF_VNET_HDR) {
1780                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1781
1782                 if (len < vnet_hdr_sz)
1783                         return -EINVAL;
1784                 len -= vnet_hdr_sz;
1785
1786                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1787                         return -EFAULT;
1788
1789                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1790                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1791                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1792
1793                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1794                         return -EINVAL;
1795                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1796         }
1797
1798         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1799                 align += NET_IP_ALIGN;
1800                 if (unlikely(len < ETH_HLEN ||
1801                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1802                         return -EINVAL;
1803         }
1804
1805         good_linear = SKB_MAX_HEAD(align);
1806
1807         if (msg_control) {
1808                 struct iov_iter i = *from;
1809
1810                 /* There are 256 bytes to be copied in skb, so there is
1811                  * enough room for skb expand head in case it is used.
1812                  * The rest of the buffer is mapped from userspace.
1813                  */
1814                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1815                 if (copylen > good_linear)
1816                         copylen = good_linear;
1817                 linear = copylen;
1818                 iov_iter_advance(&i, copylen);
1819                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1820                         zerocopy = true;
1821         }
1822
1823         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1824                 /* For the packet that is not easy to be processed
1825                  * (e.g gso or jumbo packet), we will do it at after
1826                  * skb was created with generic XDP routine.
1827                  */
1828                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1829                 if (IS_ERR(skb)) {
1830                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1831                         return PTR_ERR(skb);
1832                 }
1833                 if (!skb)
1834                         return total_len;
1835         } else {
1836                 if (!zerocopy) {
1837                         copylen = len;
1838                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1839                                 linear = good_linear;
1840                         else
1841                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1842                 }
1843
1844                 if (frags) {
1845                         mutex_lock(&tfile->napi_mutex);
1846                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1847                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1848                          * If zerocopy is enabled, then this layout will be
1849                          * overwritten by zerocopy_sg_from_iter().
1850                          */
1851                         zerocopy = false;
1852                 } else {
1853                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1854                                             noblock);
1855                 }
1856
1857                 if (IS_ERR(skb)) {
1858                         if (PTR_ERR(skb) != -EAGAIN)
1859                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1860                         if (frags)
1861                                 mutex_unlock(&tfile->napi_mutex);
1862                         return PTR_ERR(skb);
1863                 }
1864
1865                 if (zerocopy)
1866                         err = zerocopy_sg_from_iter(skb, from);
1867                 else
1868                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1869
1870                 if (err) {
1871                         err = -EFAULT;
1872 drop:
1873                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1874                         kfree_skb(skb);
1875                         if (frags) {
1876                                 tfile->napi.skb = NULL;
1877                                 mutex_unlock(&tfile->napi_mutex);
1878                         }
1879
1880                         return err;
1881                 }
1882         }
1883
1884         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1885                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1886                 kfree_skb(skb);
1887                 if (frags) {
1888                         tfile->napi.skb = NULL;
1889                         mutex_unlock(&tfile->napi_mutex);
1890                 }
1891
1892                 return -EINVAL;
1893         }
1894
1895         switch (tun->flags & TUN_TYPE_MASK) {
1896         case IFF_TUN:
1897                 if (tun->flags & IFF_NO_PI) {
1898                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1899
1900                         switch (ip_version) {
1901                         case 4:
1902                                 pi.proto = htons(ETH_P_IP);
1903                                 break;
1904                         case 6:
1905                                 pi.proto = htons(ETH_P_IPV6);
1906                                 break;
1907                         default:
1908                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1909                                 kfree_skb(skb);
1910                                 return -EINVAL;
1911                         }
1912                 }
1913
1914                 skb_reset_mac_header(skb);
1915                 skb->protocol = pi.proto;
1916                 skb->dev = tun->dev;
1917                 break;
1918         case IFF_TAP:
1919                 if (!frags)
1920                         skb->protocol = eth_type_trans(skb, tun->dev);
1921                 break;
1922         }
1923
1924         /* copy skb_ubuf_info for callback when skb has no error */
1925         if (zerocopy) {
1926                 skb_shinfo(skb)->destructor_arg = msg_control;
1927                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1928                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1929         } else if (msg_control) {
1930                 struct ubuf_info *uarg = msg_control;
1931                 uarg->callback(uarg, false);
1932         }
1933
1934         skb_reset_network_header(skb);
1935         skb_probe_transport_header(skb);
1936
1937         if (skb_xdp) {
1938                 struct bpf_prog *xdp_prog;
1939                 int ret;
1940
1941                 local_bh_disable();
1942                 rcu_read_lock();
1943                 xdp_prog = rcu_dereference(tun->xdp_prog);
1944                 if (xdp_prog) {
1945                         ret = do_xdp_generic(xdp_prog, skb);
1946                         if (ret != XDP_PASS) {
1947                                 rcu_read_unlock();
1948                                 local_bh_enable();
1949                                 return total_len;
1950                         }
1951                 }
1952                 rcu_read_unlock();
1953                 local_bh_enable();
1954         }
1955
1956         /* Compute the costly rx hash only if needed for flow updates.
1957          * We may get a very small possibility of OOO during switching, not
1958          * worth to optimize.
1959          */
1960         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1961             !tfile->detached)
1962                 rxhash = __skb_get_hash_symmetric(skb);
1963
1964         rcu_read_lock();
1965         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1966                 err = -EIO;
1967                 rcu_read_unlock();
1968                 goto drop;
1969         }
1970
1971         if (frags) {
1972                 /* Exercise flow dissector code path. */
1973                 u32 headlen = eth_get_headlen(tun->dev, skb->data,
1974                                               skb_headlen(skb));
1975
1976                 if (unlikely(headlen > skb_headlen(skb))) {
1977                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1978                         napi_free_frags(&tfile->napi);
1979                         rcu_read_unlock();
1980                         mutex_unlock(&tfile->napi_mutex);
1981                         WARN_ON(1);
1982                         return -ENOMEM;
1983                 }
1984
1985                 local_bh_disable();
1986                 napi_gro_frags(&tfile->napi);
1987                 local_bh_enable();
1988                 mutex_unlock(&tfile->napi_mutex);
1989         } else if (tfile->napi_enabled) {
1990                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1991                 int queue_len;
1992
1993                 spin_lock_bh(&queue->lock);
1994                 __skb_queue_tail(queue, skb);
1995                 queue_len = skb_queue_len(queue);
1996                 spin_unlock(&queue->lock);
1997
1998                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1999                         napi_schedule(&tfile->napi);
2000
2001                 local_bh_enable();
2002         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2003                 tun_rx_batched(tun, tfile, skb, more);
2004         } else {
2005                 netif_rx_ni(skb);
2006         }
2007         rcu_read_unlock();
2008
2009         stats = get_cpu_ptr(tun->pcpu_stats);
2010         u64_stats_update_begin(&stats->syncp);
2011         stats->rx_packets++;
2012         stats->rx_bytes += len;
2013         u64_stats_update_end(&stats->syncp);
2014         put_cpu_ptr(stats);
2015
2016         if (rxhash)
2017                 tun_flow_update(tun, rxhash, tfile);
2018
2019         return total_len;
2020 }
2021
2022 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2023 {
2024         struct file *file = iocb->ki_filp;
2025         struct tun_file *tfile = file->private_data;
2026         struct tun_struct *tun = tun_get(tfile);
2027         ssize_t result;
2028
2029         if (!tun)
2030                 return -EBADFD;
2031
2032         result = tun_get_user(tun, tfile, NULL, from,
2033                               file->f_flags & O_NONBLOCK, false);
2034
2035         tun_put(tun);
2036         return result;
2037 }
2038
2039 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2040                                 struct tun_file *tfile,
2041                                 struct xdp_frame *xdp_frame,
2042                                 struct iov_iter *iter)
2043 {
2044         int vnet_hdr_sz = 0;
2045         size_t size = xdp_frame->len;
2046         struct tun_pcpu_stats *stats;
2047         size_t ret;
2048
2049         if (tun->flags & IFF_VNET_HDR) {
2050                 struct virtio_net_hdr gso = { 0 };
2051
2052                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2053                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2054                         return -EINVAL;
2055                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2056                              sizeof(gso)))
2057                         return -EFAULT;
2058                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2059         }
2060
2061         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2062
2063         stats = get_cpu_ptr(tun->pcpu_stats);
2064         u64_stats_update_begin(&stats->syncp);
2065         stats->tx_packets++;
2066         stats->tx_bytes += ret;
2067         u64_stats_update_end(&stats->syncp);
2068         put_cpu_ptr(tun->pcpu_stats);
2069
2070         return ret;
2071 }
2072
2073 /* Put packet to the user space buffer */
2074 static ssize_t tun_put_user(struct tun_struct *tun,
2075                             struct tun_file *tfile,
2076                             struct sk_buff *skb,
2077                             struct iov_iter *iter)
2078 {
2079         struct tun_pi pi = { 0, skb->protocol };
2080         struct tun_pcpu_stats *stats;
2081         ssize_t total;
2082         int vlan_offset = 0;
2083         int vlan_hlen = 0;
2084         int vnet_hdr_sz = 0;
2085
2086         if (skb_vlan_tag_present(skb))
2087                 vlan_hlen = VLAN_HLEN;
2088
2089         if (tun->flags & IFF_VNET_HDR)
2090                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2091
2092         total = skb->len + vlan_hlen + vnet_hdr_sz;
2093
2094         if (!(tun->flags & IFF_NO_PI)) {
2095                 if (iov_iter_count(iter) < sizeof(pi))
2096                         return -EINVAL;
2097
2098                 total += sizeof(pi);
2099                 if (iov_iter_count(iter) < total) {
2100                         /* Packet will be striped */
2101                         pi.flags |= TUN_PKT_STRIP;
2102                 }
2103
2104                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2105                         return -EFAULT;
2106         }
2107
2108         if (vnet_hdr_sz) {
2109                 struct virtio_net_hdr gso;
2110
2111                 if (iov_iter_count(iter) < vnet_hdr_sz)
2112                         return -EINVAL;
2113
2114                 if (virtio_net_hdr_from_skb(skb, &gso,
2115                                             tun_is_little_endian(tun), true,
2116                                             vlan_hlen)) {
2117                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2118                         pr_err("unexpected GSO type: "
2119                                "0x%x, gso_size %d, hdr_len %d\n",
2120                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2121                                tun16_to_cpu(tun, gso.hdr_len));
2122                         print_hex_dump(KERN_ERR, "tun: ",
2123                                        DUMP_PREFIX_NONE,
2124                                        16, 1, skb->head,
2125                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2126                         WARN_ON_ONCE(1);
2127                         return -EINVAL;
2128                 }
2129
2130                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2131                         return -EFAULT;
2132
2133                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2134         }
2135
2136         if (vlan_hlen) {
2137                 int ret;
2138                 struct veth veth;
2139
2140                 veth.h_vlan_proto = skb->vlan_proto;
2141                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2142
2143                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2144
2145                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2146                 if (ret || !iov_iter_count(iter))
2147                         goto done;
2148
2149                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2150                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2151                         goto done;
2152         }
2153
2154         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2155
2156 done:
2157         /* caller is in process context, */
2158         stats = get_cpu_ptr(tun->pcpu_stats);
2159         u64_stats_update_begin(&stats->syncp);
2160         stats->tx_packets++;
2161         stats->tx_bytes += skb->len + vlan_hlen;
2162         u64_stats_update_end(&stats->syncp);
2163         put_cpu_ptr(tun->pcpu_stats);
2164
2165         return total;
2166 }
2167
2168 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2169 {
2170         DECLARE_WAITQUEUE(wait, current);
2171         void *ptr = NULL;
2172         int error = 0;
2173
2174         ptr = ptr_ring_consume(&tfile->tx_ring);
2175         if (ptr)
2176                 goto out;
2177         if (noblock) {
2178                 error = -EAGAIN;
2179                 goto out;
2180         }
2181
2182         add_wait_queue(&tfile->wq.wait, &wait);
2183
2184         while (1) {
2185                 set_current_state(TASK_INTERRUPTIBLE);
2186                 ptr = ptr_ring_consume(&tfile->tx_ring);
2187                 if (ptr)
2188                         break;
2189                 if (signal_pending(current)) {
2190                         error = -ERESTARTSYS;
2191                         break;
2192                 }
2193                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2194                         error = -EFAULT;
2195                         break;
2196                 }
2197
2198                 schedule();
2199         }
2200
2201         __set_current_state(TASK_RUNNING);
2202         remove_wait_queue(&tfile->wq.wait, &wait);
2203
2204 out:
2205         *err = error;
2206         return ptr;
2207 }
2208
2209 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2210                            struct iov_iter *to,
2211                            int noblock, void *ptr)
2212 {
2213         ssize_t ret;
2214         int err;
2215
2216         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2217
2218         if (!iov_iter_count(to)) {
2219                 tun_ptr_free(ptr);
2220                 return 0;
2221         }
2222
2223         if (!ptr) {
2224                 /* Read frames from ring */
2225                 ptr = tun_ring_recv(tfile, noblock, &err);
2226                 if (!ptr)
2227                         return err;
2228         }
2229
2230         if (tun_is_xdp_frame(ptr)) {
2231                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2232
2233                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2234                 xdp_return_frame(xdpf);
2235         } else {
2236                 struct sk_buff *skb = ptr;
2237
2238                 ret = tun_put_user(tun, tfile, skb, to);
2239                 if (unlikely(ret < 0))
2240                         kfree_skb(skb);
2241                 else
2242                         consume_skb(skb);
2243         }
2244
2245         return ret;
2246 }
2247
2248 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2249 {
2250         struct file *file = iocb->ki_filp;
2251         struct tun_file *tfile = file->private_data;
2252         struct tun_struct *tun = tun_get(tfile);
2253         ssize_t len = iov_iter_count(to), ret;
2254
2255         if (!tun)
2256                 return -EBADFD;
2257         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2258         ret = min_t(ssize_t, ret, len);
2259         if (ret > 0)
2260                 iocb->ki_pos = ret;
2261         tun_put(tun);
2262         return ret;
2263 }
2264
2265 static void tun_prog_free(struct rcu_head *rcu)
2266 {
2267         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2268
2269         bpf_prog_destroy(prog->prog);
2270         kfree(prog);
2271 }
2272
2273 static int __tun_set_ebpf(struct tun_struct *tun,
2274                           struct tun_prog __rcu **prog_p,
2275                           struct bpf_prog *prog)
2276 {
2277         struct tun_prog *old, *new = NULL;
2278
2279         if (prog) {
2280                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2281                 if (!new)
2282                         return -ENOMEM;
2283                 new->prog = prog;
2284         }
2285
2286         spin_lock_bh(&tun->lock);
2287         old = rcu_dereference_protected(*prog_p,
2288                                         lockdep_is_held(&tun->lock));
2289         rcu_assign_pointer(*prog_p, new);
2290         spin_unlock_bh(&tun->lock);
2291
2292         if (old)
2293                 call_rcu(&old->rcu, tun_prog_free);
2294
2295         return 0;
2296 }
2297
2298 static void tun_free_netdev(struct net_device *dev)
2299 {
2300         struct tun_struct *tun = netdev_priv(dev);
2301
2302         BUG_ON(!(list_empty(&tun->disabled)));
2303         free_percpu(tun->pcpu_stats);
2304         tun_flow_uninit(tun);
2305         security_tun_dev_free_security(tun->security);
2306         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2307         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2308 }
2309
2310 static void tun_setup(struct net_device *dev)
2311 {
2312         struct tun_struct *tun = netdev_priv(dev);
2313
2314         tun->owner = INVALID_UID;
2315         tun->group = INVALID_GID;
2316         tun_default_link_ksettings(dev, &tun->link_ksettings);
2317
2318         dev->ethtool_ops = &tun_ethtool_ops;
2319         dev->needs_free_netdev = true;
2320         dev->priv_destructor = tun_free_netdev;
2321         /* We prefer our own queue length */
2322         dev->tx_queue_len = TUN_READQ_SIZE;
2323 }
2324
2325 /* Trivial set of netlink ops to allow deleting tun or tap
2326  * device with netlink.
2327  */
2328 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2329                         struct netlink_ext_ack *extack)
2330 {
2331         NL_SET_ERR_MSG(extack,
2332                        "tun/tap creation via rtnetlink is not supported.");
2333         return -EOPNOTSUPP;
2334 }
2335
2336 static size_t tun_get_size(const struct net_device *dev)
2337 {
2338         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2339         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2340
2341         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2342                nla_total_size(sizeof(gid_t)) + /* GROUP */
2343                nla_total_size(sizeof(u8)) + /* TYPE */
2344                nla_total_size(sizeof(u8)) + /* PI */
2345                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2346                nla_total_size(sizeof(u8)) + /* PERSIST */
2347                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2348                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2349                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2350                0;
2351 }
2352
2353 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2354 {
2355         struct tun_struct *tun = netdev_priv(dev);
2356
2357         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2358                 goto nla_put_failure;
2359         if (uid_valid(tun->owner) &&
2360             nla_put_u32(skb, IFLA_TUN_OWNER,
2361                         from_kuid_munged(current_user_ns(), tun->owner)))
2362                 goto nla_put_failure;
2363         if (gid_valid(tun->group) &&
2364             nla_put_u32(skb, IFLA_TUN_GROUP,
2365                         from_kgid_munged(current_user_ns(), tun->group)))
2366                 goto nla_put_failure;
2367         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2368                 goto nla_put_failure;
2369         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2370                 goto nla_put_failure;
2371         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2372                 goto nla_put_failure;
2373         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2374                        !!(tun->flags & IFF_MULTI_QUEUE)))
2375                 goto nla_put_failure;
2376         if (tun->flags & IFF_MULTI_QUEUE) {
2377                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2378                         goto nla_put_failure;
2379                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2380                                 tun->numdisabled))
2381                         goto nla_put_failure;
2382         }
2383
2384         return 0;
2385
2386 nla_put_failure:
2387         return -EMSGSIZE;
2388 }
2389
2390 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2391         .kind           = DRV_NAME,
2392         .priv_size      = sizeof(struct tun_struct),
2393         .setup          = tun_setup,
2394         .validate       = tun_validate,
2395         .get_size       = tun_get_size,
2396         .fill_info      = tun_fill_info,
2397 };
2398
2399 static void tun_sock_write_space(struct sock *sk)
2400 {
2401         struct tun_file *tfile;
2402         wait_queue_head_t *wqueue;
2403
2404         if (!sock_writeable(sk))
2405                 return;
2406
2407         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2408                 return;
2409
2410         wqueue = sk_sleep(sk);
2411         if (wqueue && waitqueue_active(wqueue))
2412                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2413                                                 EPOLLWRNORM | EPOLLWRBAND);
2414
2415         tfile = container_of(sk, struct tun_file, sk);
2416         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2417 }
2418
2419 static void tun_put_page(struct tun_page *tpage)
2420 {
2421         if (tpage->page)
2422                 __page_frag_cache_drain(tpage->page, tpage->count);
2423 }
2424
2425 static int tun_xdp_one(struct tun_struct *tun,
2426                        struct tun_file *tfile,
2427                        struct xdp_buff *xdp, int *flush,
2428                        struct tun_page *tpage)
2429 {
2430         unsigned int datasize = xdp->data_end - xdp->data;
2431         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2432         struct virtio_net_hdr *gso = &hdr->gso;
2433         struct tun_pcpu_stats *stats;
2434         struct bpf_prog *xdp_prog;
2435         struct sk_buff *skb = NULL;
2436         u32 rxhash = 0, act;
2437         int buflen = hdr->buflen;
2438         int err = 0;
2439         bool skb_xdp = false;
2440         struct page *page;
2441
2442         xdp_prog = rcu_dereference(tun->xdp_prog);
2443         if (xdp_prog) {
2444                 if (gso->gso_type) {
2445                         skb_xdp = true;
2446                         goto build;
2447                 }
2448                 xdp_set_data_meta_invalid(xdp);
2449                 xdp->rxq = &tfile->xdp_rxq;
2450
2451                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2452                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2453                 if (err < 0) {
2454                         put_page(virt_to_head_page(xdp->data));
2455                         return err;
2456                 }
2457
2458                 switch (err) {
2459                 case XDP_REDIRECT:
2460                         *flush = true;
2461                         /* fall through */
2462                 case XDP_TX:
2463                         return 0;
2464                 case XDP_PASS:
2465                         break;
2466                 default:
2467                         page = virt_to_head_page(xdp->data);
2468                         if (tpage->page == page) {
2469                                 ++tpage->count;
2470                         } else {
2471                                 tun_put_page(tpage);
2472                                 tpage->page = page;
2473                                 tpage->count = 1;
2474                         }
2475                         return 0;
2476                 }
2477         }
2478
2479 build:
2480         skb = build_skb(xdp->data_hard_start, buflen);
2481         if (!skb) {
2482                 err = -ENOMEM;
2483                 goto out;
2484         }
2485
2486         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2487         skb_put(skb, xdp->data_end - xdp->data);
2488
2489         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2490                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2491                 kfree_skb(skb);
2492                 err = -EINVAL;
2493                 goto out;
2494         }
2495
2496         skb->protocol = eth_type_trans(skb, tun->dev);
2497         skb_reset_network_header(skb);
2498         skb_probe_transport_header(skb);
2499
2500         if (skb_xdp) {
2501                 err = do_xdp_generic(xdp_prog, skb);
2502                 if (err != XDP_PASS)
2503                         goto out;
2504         }
2505
2506         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2507             !tfile->detached)
2508                 rxhash = __skb_get_hash_symmetric(skb);
2509
2510         skb_record_rx_queue(skb, tfile->queue_index);
2511         netif_receive_skb(skb);
2512
2513         /* No need for get_cpu_ptr() here since this function is
2514          * always called with bh disabled
2515          */
2516         stats = this_cpu_ptr(tun->pcpu_stats);
2517         u64_stats_update_begin(&stats->syncp);
2518         stats->rx_packets++;
2519         stats->rx_bytes += datasize;
2520         u64_stats_update_end(&stats->syncp);
2521
2522         if (rxhash)
2523                 tun_flow_update(tun, rxhash, tfile);
2524
2525 out:
2526         return err;
2527 }
2528
2529 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2530 {
2531         int ret, i;
2532         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2533         struct tun_struct *tun = tun_get(tfile);
2534         struct tun_msg_ctl *ctl = m->msg_control;
2535         struct xdp_buff *xdp;
2536
2537         if (!tun)
2538                 return -EBADFD;
2539
2540         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2541                 struct tun_page tpage;
2542                 int n = ctl->num;
2543                 int flush = 0;
2544
2545                 memset(&tpage, 0, sizeof(tpage));
2546
2547                 local_bh_disable();
2548                 rcu_read_lock();
2549
2550                 for (i = 0; i < n; i++) {
2551                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2552                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2553                 }
2554
2555                 if (flush)
2556                         xdp_do_flush_map();
2557
2558                 rcu_read_unlock();
2559                 local_bh_enable();
2560
2561                 tun_put_page(&tpage);
2562
2563                 ret = total_len;
2564                 goto out;
2565         }
2566
2567         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2568                            m->msg_flags & MSG_DONTWAIT,
2569                            m->msg_flags & MSG_MORE);
2570 out:
2571         tun_put(tun);
2572         return ret;
2573 }
2574
2575 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2576                        int flags)
2577 {
2578         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2579         struct tun_struct *tun = tun_get(tfile);
2580         void *ptr = m->msg_control;
2581         int ret;
2582
2583         if (!tun) {
2584                 ret = -EBADFD;
2585                 goto out_free;
2586         }
2587
2588         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2589                 ret = -EINVAL;
2590                 goto out_put_tun;
2591         }
2592         if (flags & MSG_ERRQUEUE) {
2593                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2594                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2595                 goto out;
2596         }
2597         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2598         if (ret > (ssize_t)total_len) {
2599                 m->msg_flags |= MSG_TRUNC;
2600                 ret = flags & MSG_TRUNC ? ret : total_len;
2601         }
2602 out:
2603         tun_put(tun);
2604         return ret;
2605
2606 out_put_tun:
2607         tun_put(tun);
2608 out_free:
2609         tun_ptr_free(ptr);
2610         return ret;
2611 }
2612
2613 static int tun_ptr_peek_len(void *ptr)
2614 {
2615         if (likely(ptr)) {
2616                 if (tun_is_xdp_frame(ptr)) {
2617                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2618
2619                         return xdpf->len;
2620                 }
2621                 return __skb_array_len_with_tag(ptr);
2622         } else {
2623                 return 0;
2624         }
2625 }
2626
2627 static int tun_peek_len(struct socket *sock)
2628 {
2629         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2630         struct tun_struct *tun;
2631         int ret = 0;
2632
2633         tun = tun_get(tfile);
2634         if (!tun)
2635                 return 0;
2636
2637         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2638         tun_put(tun);
2639
2640         return ret;
2641 }
2642
2643 /* Ops structure to mimic raw sockets with tun */
2644 static const struct proto_ops tun_socket_ops = {
2645         .peek_len = tun_peek_len,
2646         .sendmsg = tun_sendmsg,
2647         .recvmsg = tun_recvmsg,
2648 };
2649
2650 static struct proto tun_proto = {
2651         .name           = "tun",
2652         .owner          = THIS_MODULE,
2653         .obj_size       = sizeof(struct tun_file),
2654 };
2655
2656 static int tun_flags(struct tun_struct *tun)
2657 {
2658         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2659 }
2660
2661 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2662                               char *buf)
2663 {
2664         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2665         return sprintf(buf, "0x%x\n", tun_flags(tun));
2666 }
2667
2668 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2669                               char *buf)
2670 {
2671         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2672         return uid_valid(tun->owner)?
2673                 sprintf(buf, "%u\n",
2674                         from_kuid_munged(current_user_ns(), tun->owner)):
2675                 sprintf(buf, "-1\n");
2676 }
2677
2678 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2679                               char *buf)
2680 {
2681         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2682         return gid_valid(tun->group) ?
2683                 sprintf(buf, "%u\n",
2684                         from_kgid_munged(current_user_ns(), tun->group)):
2685                 sprintf(buf, "-1\n");
2686 }
2687
2688 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2689 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2690 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2691
2692 static struct attribute *tun_dev_attrs[] = {
2693         &dev_attr_tun_flags.attr,
2694         &dev_attr_owner.attr,
2695         &dev_attr_group.attr,
2696         NULL
2697 };
2698
2699 static const struct attribute_group tun_attr_group = {
2700         .attrs = tun_dev_attrs
2701 };
2702
2703 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2704 {
2705         struct tun_struct *tun;
2706         struct tun_file *tfile = file->private_data;
2707         struct net_device *dev;
2708         int err;
2709
2710         if (tfile->detached)
2711                 return -EINVAL;
2712
2713         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2714                 if (!capable(CAP_NET_ADMIN))
2715                         return -EPERM;
2716
2717                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2718                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2719                         return -EINVAL;
2720         }
2721
2722         dev = __dev_get_by_name(net, ifr->ifr_name);
2723         if (dev) {
2724                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2725                         return -EBUSY;
2726                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2727                         tun = netdev_priv(dev);
2728                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2729                         tun = netdev_priv(dev);
2730                 else
2731                         return -EINVAL;
2732
2733                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2734                     !!(tun->flags & IFF_MULTI_QUEUE))
2735                         return -EINVAL;
2736
2737                 if (tun_not_capable(tun))
2738                         return -EPERM;
2739                 err = security_tun_dev_open(tun->security);
2740                 if (err < 0)
2741                         return err;
2742
2743                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2744                                  ifr->ifr_flags & IFF_NAPI,
2745                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2746                 if (err < 0)
2747                         return err;
2748
2749                 if (tun->flags & IFF_MULTI_QUEUE &&
2750                     (tun->numqueues + tun->numdisabled > 1)) {
2751                         /* One or more queue has already been attached, no need
2752                          * to initialize the device again.
2753                          */
2754                         netdev_state_change(dev);
2755                         return 0;
2756                 }
2757
2758                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2759                               (ifr->ifr_flags & TUN_FEATURES);
2760
2761                 netdev_state_change(dev);
2762         } else {
2763                 char *name;
2764                 unsigned long flags = 0;
2765                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2766                              MAX_TAP_QUEUES : 1;
2767
2768                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2769                         return -EPERM;
2770                 err = security_tun_dev_create();
2771                 if (err < 0)
2772                         return err;
2773
2774                 /* Set dev type */
2775                 if (ifr->ifr_flags & IFF_TUN) {
2776                         /* TUN device */
2777                         flags |= IFF_TUN;
2778                         name = "tun%d";
2779                 } else if (ifr->ifr_flags & IFF_TAP) {
2780                         /* TAP device */
2781                         flags |= IFF_TAP;
2782                         name = "tap%d";
2783                 } else
2784                         return -EINVAL;
2785
2786                 if (*ifr->ifr_name)
2787                         name = ifr->ifr_name;
2788
2789                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2790                                        NET_NAME_UNKNOWN, tun_setup, queues,
2791                                        queues);
2792
2793                 if (!dev)
2794                         return -ENOMEM;
2795                 err = dev_get_valid_name(net, dev, name);
2796                 if (err < 0)
2797                         goto err_free_dev;
2798
2799                 dev_net_set(dev, net);
2800                 dev->rtnl_link_ops = &tun_link_ops;
2801                 dev->ifindex = tfile->ifindex;
2802                 dev->sysfs_groups[0] = &tun_attr_group;
2803
2804                 tun = netdev_priv(dev);
2805                 tun->dev = dev;
2806                 tun->flags = flags;
2807                 tun->txflt.count = 0;
2808                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2809
2810                 tun->align = NET_SKB_PAD;
2811                 tun->filter_attached = false;
2812                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2813                 tun->rx_batched = 0;
2814                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2815
2816                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2817                 if (!tun->pcpu_stats) {
2818                         err = -ENOMEM;
2819                         goto err_free_dev;
2820                 }
2821
2822                 spin_lock_init(&tun->lock);
2823
2824                 err = security_tun_dev_alloc_security(&tun->security);
2825                 if (err < 0)
2826                         goto err_free_stat;
2827
2828                 tun_net_init(dev);
2829                 tun_flow_init(tun);
2830
2831                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2832                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2833                                    NETIF_F_HW_VLAN_STAG_TX;
2834                 dev->features = dev->hw_features | NETIF_F_LLTX;
2835                 dev->vlan_features = dev->features &
2836                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2837                                        NETIF_F_HW_VLAN_STAG_TX);
2838
2839                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2840                               (ifr->ifr_flags & TUN_FEATURES);
2841
2842                 INIT_LIST_HEAD(&tun->disabled);
2843                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2844                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2845                 if (err < 0)
2846                         goto err_free_flow;
2847
2848                 err = register_netdevice(tun->dev);
2849                 if (err < 0)
2850                         goto err_detach;
2851         }
2852
2853         netif_carrier_on(tun->dev);
2854
2855         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2856
2857         /* Make sure persistent devices do not get stuck in
2858          * xoff state.
2859          */
2860         if (netif_running(tun->dev))
2861                 netif_tx_wake_all_queues(tun->dev);
2862
2863         strcpy(ifr->ifr_name, tun->dev->name);
2864         return 0;
2865
2866 err_detach:
2867         tun_detach_all(dev);
2868         /* register_netdevice() already called tun_free_netdev() */
2869         goto err_free_dev;
2870
2871 err_free_flow:
2872         tun_flow_uninit(tun);
2873         security_tun_dev_free_security(tun->security);
2874 err_free_stat:
2875         free_percpu(tun->pcpu_stats);
2876 err_free_dev:
2877         free_netdev(dev);
2878         return err;
2879 }
2880
2881 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2882 {
2883         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2884
2885         strcpy(ifr->ifr_name, tun->dev->name);
2886
2887         ifr->ifr_flags = tun_flags(tun);
2888
2889 }
2890
2891 /* This is like a cut-down ethtool ops, except done via tun fd so no
2892  * privs required. */
2893 static int set_offload(struct tun_struct *tun, unsigned long arg)
2894 {
2895         netdev_features_t features = 0;
2896
2897         if (arg & TUN_F_CSUM) {
2898                 features |= NETIF_F_HW_CSUM;
2899                 arg &= ~TUN_F_CSUM;
2900
2901                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2902                         if (arg & TUN_F_TSO_ECN) {
2903                                 features |= NETIF_F_TSO_ECN;
2904                                 arg &= ~TUN_F_TSO_ECN;
2905                         }
2906                         if (arg & TUN_F_TSO4)
2907                                 features |= NETIF_F_TSO;
2908                         if (arg & TUN_F_TSO6)
2909                                 features |= NETIF_F_TSO6;
2910                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2911                 }
2912
2913                 arg &= ~TUN_F_UFO;
2914         }
2915
2916         /* This gives the user a way to test for new features in future by
2917          * trying to set them. */
2918         if (arg)
2919                 return -EINVAL;
2920
2921         tun->set_features = features;
2922         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2923         tun->dev->wanted_features |= features;
2924         netdev_update_features(tun->dev);
2925
2926         return 0;
2927 }
2928
2929 static void tun_detach_filter(struct tun_struct *tun, int n)
2930 {
2931         int i;
2932         struct tun_file *tfile;
2933
2934         for (i = 0; i < n; i++) {
2935                 tfile = rtnl_dereference(tun->tfiles[i]);
2936                 lock_sock(tfile->socket.sk);
2937                 sk_detach_filter(tfile->socket.sk);
2938                 release_sock(tfile->socket.sk);
2939         }
2940
2941         tun->filter_attached = false;
2942 }
2943
2944 static int tun_attach_filter(struct tun_struct *tun)
2945 {
2946         int i, ret = 0;
2947         struct tun_file *tfile;
2948
2949         for (i = 0; i < tun->numqueues; i++) {
2950                 tfile = rtnl_dereference(tun->tfiles[i]);
2951                 lock_sock(tfile->socket.sk);
2952                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2953                 release_sock(tfile->socket.sk);
2954                 if (ret) {
2955                         tun_detach_filter(tun, i);
2956                         return ret;
2957                 }
2958         }
2959
2960         tun->filter_attached = true;
2961         return ret;
2962 }
2963
2964 static void tun_set_sndbuf(struct tun_struct *tun)
2965 {
2966         struct tun_file *tfile;
2967         int i;
2968
2969         for (i = 0; i < tun->numqueues; i++) {
2970                 tfile = rtnl_dereference(tun->tfiles[i]);
2971                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2972         }
2973 }
2974
2975 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2976 {
2977         struct tun_file *tfile = file->private_data;
2978         struct tun_struct *tun;
2979         int ret = 0;
2980
2981         rtnl_lock();
2982
2983         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2984                 tun = tfile->detached;
2985                 if (!tun) {
2986                         ret = -EINVAL;
2987                         goto unlock;
2988                 }
2989                 ret = security_tun_dev_attach_queue(tun->security);
2990                 if (ret < 0)
2991                         goto unlock;
2992                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2993                                  tun->flags & IFF_NAPI_FRAGS);
2994         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2995                 tun = rtnl_dereference(tfile->tun);
2996                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2997                         ret = -EINVAL;
2998                 else
2999                         __tun_detach(tfile, false);
3000         } else
3001                 ret = -EINVAL;
3002
3003         if (ret >= 0)
3004                 netdev_state_change(tun->dev);
3005
3006 unlock:
3007         rtnl_unlock();
3008         return ret;
3009 }
3010
3011 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3012                         void __user *data)
3013 {
3014         struct bpf_prog *prog;
3015         int fd;
3016
3017         if (copy_from_user(&fd, data, sizeof(fd)))
3018                 return -EFAULT;
3019
3020         if (fd == -1) {
3021                 prog = NULL;
3022         } else {
3023                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3024                 if (IS_ERR(prog))
3025                         return PTR_ERR(prog);
3026         }
3027
3028         return __tun_set_ebpf(tun, prog_p, prog);
3029 }
3030
3031 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3032                             unsigned long arg, int ifreq_len)
3033 {
3034         struct tun_file *tfile = file->private_data;
3035         struct net *net = sock_net(&tfile->sk);
3036         struct tun_struct *tun;
3037         void __user* argp = (void __user*)arg;
3038         unsigned int ifindex, carrier;
3039         struct ifreq ifr;
3040         kuid_t owner;
3041         kgid_t group;
3042         int sndbuf;
3043         int vnet_hdr_sz;
3044         int le;
3045         int ret;
3046         bool do_notify = false;
3047
3048         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3049             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3050                 if (copy_from_user(&ifr, argp, ifreq_len))
3051                         return -EFAULT;
3052         } else {
3053                 memset(&ifr, 0, sizeof(ifr));
3054         }
3055         if (cmd == TUNGETFEATURES) {
3056                 /* Currently this just means: "what IFF flags are valid?".
3057                  * This is needed because we never checked for invalid flags on
3058                  * TUNSETIFF.
3059                  */
3060                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3061                                 (unsigned int __user*)argp);
3062         } else if (cmd == TUNSETQUEUE) {
3063                 return tun_set_queue(file, &ifr);
3064         } else if (cmd == SIOCGSKNS) {
3065                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3066                         return -EPERM;
3067                 return open_related_ns(&net->ns, get_net_ns);
3068         }
3069
3070         ret = 0;
3071         rtnl_lock();
3072
3073         tun = tun_get(tfile);
3074         if (cmd == TUNSETIFF) {
3075                 ret = -EEXIST;
3076                 if (tun)
3077                         goto unlock;
3078
3079                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3080
3081                 ret = tun_set_iff(net, file, &ifr);
3082
3083                 if (ret)
3084                         goto unlock;
3085
3086                 if (copy_to_user(argp, &ifr, ifreq_len))
3087                         ret = -EFAULT;
3088                 goto unlock;
3089         }
3090         if (cmd == TUNSETIFINDEX) {
3091                 ret = -EPERM;
3092                 if (tun)
3093                         goto unlock;
3094
3095                 ret = -EFAULT;
3096                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3097                         goto unlock;
3098
3099                 ret = 0;
3100                 tfile->ifindex = ifindex;
3101                 goto unlock;
3102         }
3103
3104         ret = -EBADFD;
3105         if (!tun)
3106                 goto unlock;
3107
3108         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3109
3110         net = dev_net(tun->dev);
3111         ret = 0;
3112         switch (cmd) {
3113         case TUNGETIFF:
3114                 tun_get_iff(tun, &ifr);
3115
3116                 if (tfile->detached)
3117                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3118                 if (!tfile->socket.sk->sk_filter)
3119                         ifr.ifr_flags |= IFF_NOFILTER;
3120
3121                 if (copy_to_user(argp, &ifr, ifreq_len))
3122                         ret = -EFAULT;
3123                 break;
3124
3125         case TUNSETNOCSUM:
3126                 /* Disable/Enable checksum */
3127
3128                 /* [unimplemented] */
3129                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3130                           arg ? "disabled" : "enabled");
3131                 break;
3132
3133         case TUNSETPERSIST:
3134                 /* Disable/Enable persist mode. Keep an extra reference to the
3135                  * module to prevent the module being unprobed.
3136                  */
3137                 if (arg && !(tun->flags & IFF_PERSIST)) {
3138                         tun->flags |= IFF_PERSIST;
3139                         __module_get(THIS_MODULE);
3140                         do_notify = true;
3141                 }
3142                 if (!arg && (tun->flags & IFF_PERSIST)) {
3143                         tun->flags &= ~IFF_PERSIST;
3144                         module_put(THIS_MODULE);
3145                         do_notify = true;
3146                 }
3147
3148                 tun_debug(KERN_INFO, tun, "persist %s\n",
3149                           arg ? "enabled" : "disabled");
3150                 break;
3151
3152         case TUNSETOWNER:
3153                 /* Set owner of the device */
3154                 owner = make_kuid(current_user_ns(), arg);
3155                 if (!uid_valid(owner)) {
3156                         ret = -EINVAL;
3157                         break;
3158                 }
3159                 tun->owner = owner;
3160                 do_notify = true;
3161                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3162                           from_kuid(&init_user_ns, tun->owner));
3163                 break;
3164
3165         case TUNSETGROUP:
3166                 /* Set group of the device */
3167                 group = make_kgid(current_user_ns(), arg);
3168                 if (!gid_valid(group)) {
3169                         ret = -EINVAL;
3170                         break;
3171                 }
3172                 tun->group = group;
3173                 do_notify = true;
3174                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3175                           from_kgid(&init_user_ns, tun->group));
3176                 break;
3177
3178         case TUNSETLINK:
3179                 /* Only allow setting the type when the interface is down */
3180                 if (tun->dev->flags & IFF_UP) {
3181                         tun_debug(KERN_INFO, tun,
3182                                   "Linktype set failed because interface is up\n");
3183                         ret = -EBUSY;
3184                 } else {
3185                         tun->dev->type = (int) arg;
3186                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3187                                   tun->dev->type);
3188                         ret = 0;
3189                 }
3190                 break;
3191
3192 #ifdef TUN_DEBUG
3193         case TUNSETDEBUG:
3194                 tun->debug = arg;
3195                 break;
3196 #endif
3197         case TUNSETOFFLOAD:
3198                 ret = set_offload(tun, arg);
3199                 break;
3200
3201         case TUNSETTXFILTER:
3202                 /* Can be set only for TAPs */
3203                 ret = -EINVAL;
3204                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3205                         break;
3206                 ret = update_filter(&tun->txflt, (void __user *)arg);
3207                 break;
3208
3209         case SIOCGIFHWADDR:
3210                 /* Get hw address */
3211                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3212                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3213                 if (copy_to_user(argp, &ifr, ifreq_len))
3214                         ret = -EFAULT;
3215                 break;
3216
3217         case SIOCSIFHWADDR:
3218                 /* Set hw address */
3219                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3220                           ifr.ifr_hwaddr.sa_data);
3221
3222                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3223                 break;
3224
3225         case TUNGETSNDBUF:
3226                 sndbuf = tfile->socket.sk->sk_sndbuf;
3227                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3228                         ret = -EFAULT;
3229                 break;
3230
3231         case TUNSETSNDBUF:
3232                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3233                         ret = -EFAULT;
3234                         break;
3235                 }
3236                 if (sndbuf <= 0) {
3237                         ret = -EINVAL;
3238                         break;
3239                 }
3240
3241                 tun->sndbuf = sndbuf;
3242                 tun_set_sndbuf(tun);
3243                 break;
3244
3245         case TUNGETVNETHDRSZ:
3246                 vnet_hdr_sz = tun->vnet_hdr_sz;
3247                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3248                         ret = -EFAULT;
3249                 break;
3250
3251         case TUNSETVNETHDRSZ:
3252                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3253                         ret = -EFAULT;
3254                         break;
3255                 }
3256                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3257                         ret = -EINVAL;
3258                         break;
3259                 }
3260
3261                 tun->vnet_hdr_sz = vnet_hdr_sz;
3262                 break;
3263
3264         case TUNGETVNETLE:
3265                 le = !!(tun->flags & TUN_VNET_LE);
3266                 if (put_user(le, (int __user *)argp))
3267                         ret = -EFAULT;
3268                 break;
3269
3270         case TUNSETVNETLE:
3271                 if (get_user(le, (int __user *)argp)) {
3272                         ret = -EFAULT;
3273                         break;
3274                 }
3275                 if (le)
3276                         tun->flags |= TUN_VNET_LE;
3277                 else
3278                         tun->flags &= ~TUN_VNET_LE;
3279                 break;
3280
3281         case TUNGETVNETBE:
3282                 ret = tun_get_vnet_be(tun, argp);
3283                 break;
3284
3285         case TUNSETVNETBE:
3286                 ret = tun_set_vnet_be(tun, argp);
3287                 break;
3288
3289         case TUNATTACHFILTER:
3290                 /* Can be set only for TAPs */
3291                 ret = -EINVAL;
3292                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3293                         break;
3294                 ret = -EFAULT;
3295                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3296                         break;
3297
3298                 ret = tun_attach_filter(tun);
3299                 break;
3300
3301         case TUNDETACHFILTER:
3302                 /* Can be set only for TAPs */
3303                 ret = -EINVAL;
3304                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3305                         break;
3306                 ret = 0;
3307                 tun_detach_filter(tun, tun->numqueues);
3308                 break;
3309
3310         case TUNGETFILTER:
3311                 ret = -EINVAL;
3312                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3313                         break;
3314                 ret = -EFAULT;
3315                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3316                         break;
3317                 ret = 0;
3318                 break;
3319
3320         case TUNSETSTEERINGEBPF:
3321                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3322                 break;
3323
3324         case TUNSETFILTEREBPF:
3325                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3326                 break;
3327
3328         case TUNSETCARRIER:
3329                 ret = -EFAULT;
3330                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3331                         goto unlock;
3332
3333                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3334                 break;
3335
3336         case TUNGETDEVNETNS:
3337                 ret = -EPERM;
3338                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3339                         goto unlock;
3340                 ret = open_related_ns(&net->ns, get_net_ns);
3341                 break;
3342
3343         default:
3344                 ret = -EINVAL;
3345                 break;
3346         }
3347
3348         if (do_notify)
3349                 netdev_state_change(tun->dev);
3350
3351 unlock:
3352         rtnl_unlock();
3353         if (tun)
3354                 tun_put(tun);
3355         return ret;
3356 }
3357
3358 static long tun_chr_ioctl(struct file *file,
3359                           unsigned int cmd, unsigned long arg)
3360 {
3361         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3362 }
3363
3364 #ifdef CONFIG_COMPAT
3365 static long tun_chr_compat_ioctl(struct file *file,
3366                          unsigned int cmd, unsigned long arg)
3367 {
3368         switch (cmd) {
3369         case TUNSETIFF:
3370         case TUNGETIFF:
3371         case TUNSETTXFILTER:
3372         case TUNGETSNDBUF:
3373         case TUNSETSNDBUF:
3374         case SIOCGIFHWADDR:
3375         case SIOCSIFHWADDR:
3376                 arg = (unsigned long)compat_ptr(arg);
3377                 break;
3378         default:
3379                 arg = (compat_ulong_t)arg;
3380                 break;
3381         }
3382
3383         /*
3384          * compat_ifreq is shorter than ifreq, so we must not access beyond
3385          * the end of that structure. All fields that are used in this
3386          * driver are compatible though, we don't need to convert the
3387          * contents.
3388          */
3389         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3390 }
3391 #endif /* CONFIG_COMPAT */
3392
3393 static int tun_chr_fasync(int fd, struct file *file, int on)
3394 {
3395         struct tun_file *tfile = file->private_data;
3396         int ret;
3397
3398         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3399                 goto out;
3400
3401         if (on) {
3402                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3403                 tfile->flags |= TUN_FASYNC;
3404         } else
3405                 tfile->flags &= ~TUN_FASYNC;
3406         ret = 0;
3407 out:
3408         return ret;
3409 }
3410
3411 static int tun_chr_open(struct inode *inode, struct file * file)
3412 {
3413         struct net *net = current->nsproxy->net_ns;
3414         struct tun_file *tfile;
3415
3416         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3417
3418         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3419                                             &tun_proto, 0);
3420         if (!tfile)
3421                 return -ENOMEM;
3422         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3423                 sk_free(&tfile->sk);
3424                 return -ENOMEM;
3425         }
3426
3427         mutex_init(&tfile->napi_mutex);
3428         RCU_INIT_POINTER(tfile->tun, NULL);
3429         tfile->flags = 0;
3430         tfile->ifindex = 0;
3431
3432         init_waitqueue_head(&tfile->wq.wait);
3433         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3434
3435         tfile->socket.file = file;
3436         tfile->socket.ops = &tun_socket_ops;
3437
3438         sock_init_data(&tfile->socket, &tfile->sk);
3439
3440         tfile->sk.sk_write_space = tun_sock_write_space;
3441         tfile->sk.sk_sndbuf = INT_MAX;
3442
3443         file->private_data = tfile;
3444         INIT_LIST_HEAD(&tfile->next);
3445
3446         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3447
3448         return 0;
3449 }
3450
3451 static int tun_chr_close(struct inode *inode, struct file *file)
3452 {
3453         struct tun_file *tfile = file->private_data;
3454
3455         tun_detach(tfile, true);
3456
3457         return 0;
3458 }
3459
3460 #ifdef CONFIG_PROC_FS
3461 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3462 {
3463         struct tun_file *tfile = file->private_data;
3464         struct tun_struct *tun;
3465         struct ifreq ifr;
3466
3467         memset(&ifr, 0, sizeof(ifr));
3468
3469         rtnl_lock();
3470         tun = tun_get(tfile);
3471         if (tun)
3472                 tun_get_iff(tun, &ifr);
3473         rtnl_unlock();
3474
3475         if (tun)
3476                 tun_put(tun);
3477
3478         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3479 }
3480 #endif
3481
3482 static const struct file_operations tun_fops = {
3483         .owner  = THIS_MODULE,
3484         .llseek = no_llseek,
3485         .read_iter  = tun_chr_read_iter,
3486         .write_iter = tun_chr_write_iter,
3487         .poll   = tun_chr_poll,
3488         .unlocked_ioctl = tun_chr_ioctl,
3489 #ifdef CONFIG_COMPAT
3490         .compat_ioctl = tun_chr_compat_ioctl,
3491 #endif
3492         .open   = tun_chr_open,
3493         .release = tun_chr_close,
3494         .fasync = tun_chr_fasync,
3495 #ifdef CONFIG_PROC_FS
3496         .show_fdinfo = tun_chr_show_fdinfo,
3497 #endif
3498 };
3499
3500 static struct miscdevice tun_miscdev = {
3501         .minor = TUN_MINOR,
3502         .name = "tun",
3503         .nodename = "net/tun",
3504         .fops = &tun_fops,
3505 };
3506
3507 /* ethtool interface */
3508
3509 static void tun_default_link_ksettings(struct net_device *dev,
3510                                        struct ethtool_link_ksettings *cmd)
3511 {
3512         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3513         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3514         cmd->base.speed         = SPEED_10;
3515         cmd->base.duplex        = DUPLEX_FULL;
3516         cmd->base.port          = PORT_TP;
3517         cmd->base.phy_address   = 0;
3518         cmd->base.autoneg       = AUTONEG_DISABLE;
3519 }
3520
3521 static int tun_get_link_ksettings(struct net_device *dev,
3522                                   struct ethtool_link_ksettings *cmd)
3523 {
3524         struct tun_struct *tun = netdev_priv(dev);
3525
3526         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3527         return 0;
3528 }
3529
3530 static int tun_set_link_ksettings(struct net_device *dev,
3531                                   const struct ethtool_link_ksettings *cmd)
3532 {
3533         struct tun_struct *tun = netdev_priv(dev);
3534
3535         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3536         return 0;
3537 }
3538
3539 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3540 {
3541         struct tun_struct *tun = netdev_priv(dev);
3542
3543         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3544         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3545
3546         switch (tun->flags & TUN_TYPE_MASK) {
3547         case IFF_TUN:
3548                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3549                 break;
3550         case IFF_TAP:
3551                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3552                 break;
3553         }
3554 }
3555
3556 static u32 tun_get_msglevel(struct net_device *dev)
3557 {
3558 #ifdef TUN_DEBUG
3559         struct tun_struct *tun = netdev_priv(dev);
3560         return tun->debug;
3561 #else
3562         return -EOPNOTSUPP;
3563 #endif
3564 }
3565
3566 static void tun_set_msglevel(struct net_device *dev, u32 value)
3567 {
3568 #ifdef TUN_DEBUG
3569         struct tun_struct *tun = netdev_priv(dev);
3570         tun->debug = value;
3571 #endif
3572 }
3573
3574 static int tun_get_coalesce(struct net_device *dev,
3575                             struct ethtool_coalesce *ec)
3576 {
3577         struct tun_struct *tun = netdev_priv(dev);
3578
3579         ec->rx_max_coalesced_frames = tun->rx_batched;
3580
3581         return 0;
3582 }
3583
3584 static int tun_set_coalesce(struct net_device *dev,
3585                             struct ethtool_coalesce *ec)
3586 {
3587         struct tun_struct *tun = netdev_priv(dev);
3588
3589         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3590                 tun->rx_batched = NAPI_POLL_WEIGHT;
3591         else
3592                 tun->rx_batched = ec->rx_max_coalesced_frames;
3593
3594         return 0;
3595 }
3596
3597 static const struct ethtool_ops tun_ethtool_ops = {
3598         .get_drvinfo    = tun_get_drvinfo,
3599         .get_msglevel   = tun_get_msglevel,
3600         .set_msglevel   = tun_set_msglevel,
3601         .get_link       = ethtool_op_get_link,
3602         .get_ts_info    = ethtool_op_get_ts_info,
3603         .get_coalesce   = tun_get_coalesce,
3604         .set_coalesce   = tun_set_coalesce,
3605         .get_link_ksettings = tun_get_link_ksettings,
3606         .set_link_ksettings = tun_set_link_ksettings,
3607 };
3608
3609 static int tun_queue_resize(struct tun_struct *tun)
3610 {
3611         struct net_device *dev = tun->dev;
3612         struct tun_file *tfile;
3613         struct ptr_ring **rings;
3614         int n = tun->numqueues + tun->numdisabled;
3615         int ret, i;
3616
3617         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3618         if (!rings)
3619                 return -ENOMEM;
3620
3621         for (i = 0; i < tun->numqueues; i++) {
3622                 tfile = rtnl_dereference(tun->tfiles[i]);
3623                 rings[i] = &tfile->tx_ring;
3624         }
3625         list_for_each_entry(tfile, &tun->disabled, next)
3626                 rings[i++] = &tfile->tx_ring;
3627
3628         ret = ptr_ring_resize_multiple(rings, n,
3629                                        dev->tx_queue_len, GFP_KERNEL,
3630                                        tun_ptr_free);
3631
3632         kfree(rings);
3633         return ret;
3634 }
3635
3636 static int tun_device_event(struct notifier_block *unused,
3637                             unsigned long event, void *ptr)
3638 {
3639         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3640         struct tun_struct *tun = netdev_priv(dev);
3641
3642         if (dev->rtnl_link_ops != &tun_link_ops)
3643                 return NOTIFY_DONE;
3644
3645         switch (event) {
3646         case NETDEV_CHANGE_TX_QUEUE_LEN:
3647                 if (tun_queue_resize(tun))
3648                         return NOTIFY_BAD;
3649                 break;
3650         default:
3651                 break;
3652         }
3653
3654         return NOTIFY_DONE;
3655 }
3656
3657 static struct notifier_block tun_notifier_block __read_mostly = {
3658         .notifier_call  = tun_device_event,
3659 };
3660
3661 static int __init tun_init(void)
3662 {
3663         int ret = 0;
3664
3665         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3666
3667         ret = rtnl_link_register(&tun_link_ops);
3668         if (ret) {
3669                 pr_err("Can't register link_ops\n");
3670                 goto err_linkops;
3671         }
3672
3673         ret = misc_register(&tun_miscdev);
3674         if (ret) {
3675                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3676                 goto err_misc;
3677         }
3678
3679         ret = register_netdevice_notifier(&tun_notifier_block);
3680         if (ret) {
3681                 pr_err("Can't register netdevice notifier\n");
3682                 goto err_notifier;
3683         }
3684
3685         return  0;
3686
3687 err_notifier:
3688         misc_deregister(&tun_miscdev);
3689 err_misc:
3690         rtnl_link_unregister(&tun_link_ops);
3691 err_linkops:
3692         return ret;
3693 }
3694
3695 static void tun_cleanup(void)
3696 {
3697         misc_deregister(&tun_miscdev);
3698         rtnl_link_unregister(&tun_link_ops);
3699         unregister_netdevice_notifier(&tun_notifier_block);
3700 }
3701
3702 /* Get an underlying socket object from tun file.  Returns error unless file is
3703  * attached to a device.  The returned object works like a packet socket, it
3704  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3705  * holding a reference to the file for as long as the socket is in use. */
3706 struct socket *tun_get_socket(struct file *file)
3707 {
3708         struct tun_file *tfile;
3709         if (file->f_op != &tun_fops)
3710                 return ERR_PTR(-EINVAL);
3711         tfile = file->private_data;
3712         if (!tfile)
3713                 return ERR_PTR(-EBADFD);
3714         return &tfile->socket;
3715 }
3716 EXPORT_SYMBOL_GPL(tun_get_socket);
3717
3718 struct ptr_ring *tun_get_tx_ring(struct file *file)
3719 {
3720         struct tun_file *tfile;
3721
3722         if (file->f_op != &tun_fops)
3723                 return ERR_PTR(-EINVAL);
3724         tfile = file->private_data;
3725         if (!tfile)
3726                 return ERR_PTR(-EBADFD);
3727         return &tfile->tx_ring;
3728 }
3729 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3730
3731 module_init(tun_init);
3732 module_exit(tun_cleanup);
3733 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3734 MODULE_AUTHOR(DRV_COPYRIGHT);
3735 MODULE_LICENSE("GPL");
3736 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3737 MODULE_ALIAS("devname:net/tun");