drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <net/xdp.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/uio.h>
  76 #include <linux/skb_array.h>
  77 #include <linux/bpf.h>
  78 #include <linux/bpf_trace.h>
  79 #include <linux/mutex.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 /* Uncomment to enable debugging */
  88 /* #define TUN_DEBUG 1 */
  89
  90 #ifdef TUN_DEBUG
  91 static int debug;
  92
  93 #define tun_debug(level, tun, fmt, args...)                     \
  94 do {                                                            \
  95         if (tun->debug)                                         \
  96                 netdev_printk(level, tun->dev, fmt, ##args);    \
  97 } while (0)
  98 #define DBG1(level, fmt, args...)                               \
  99 do {                                                            \
 100         if (debug == 2)                                         \
 101                 printk(level fmt, ##args);                      \
 102 } while (0)
 103 #else
 104 #define tun_debug(level, tun, fmt, args...)                     \
 105 do {                                                            \
 106         if (0)                                                  \
 107                 netdev_printk(level, tun->dev, fmt, ##args);    \
 108 } while (0)
 109 #define DBG1(level, fmt, args...)                               \
 110 do {                                                            \
 111         if (0)                                                  \
 112                 printk(level fmt, ##args);                      \
 113 } while (0)
 114 #endif
 115
 116 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 117
 118 /* TUN device flags */
 119
 120 /* IFF_ATTACH_QUEUE is never stored in device flags,
 121  * overload it to mean fasync when stored there.
 122  */
 123 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 124 /* High bits in flags field are unused. */
 125 #define TUN_VNET_LE     0x80000000
 126 #define TUN_VNET_BE     0x40000000
 127
 128 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 129                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 130
 131 #define GOODCOPY_LEN 128
 132
 133 #define FLT_EXACT_COUNT 8
 134 struct tap_filter {
 135         unsigned int    count;    /* Number of addrs. Zero means disabled */
 136         u32             mask[2];  /* Mask of the hashed addrs */
 137         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 138 };
 139
 140 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 141  * to max number of VCPUs in guest. */
 142 #define MAX_TAP_QUEUES 256
 143 #define MAX_TAP_FLOWS  4096
 144
 145 #define TUN_FLOW_EXPIRE (3 * HZ)
 146
 147 struct tun_pcpu_stats {
 148         u64 rx_packets;
 149         u64 rx_bytes;
 150         u64 tx_packets;
 151         u64 tx_bytes;
 152         struct u64_stats_sync syncp;
 153         u32 rx_dropped;
 154         u32 tx_dropped;
 155         u32 rx_frame_errors;
 156 };
 157
 158 /* A tun_file connects an open character device to a tuntap netdevice. It
 159  * also contains all socket related structures (except sock_fprog and tap_filter)
 160  * to serve as one transmit queue for tuntap device. The sock_fprog and
 161  * tap_filter were kept in tun_struct since they were used for filtering for the
 162  * netdevice not for a specific queue (at least I didn't see the requirement for
 163  * this).
 164  *
 165  * RCU usage:
 166  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 167  * other can only be read while rcu_read_lock or rtnl_lock is held.
 168  */
 169 struct tun_file {
 170         struct sock sk;
 171         struct socket socket;
 172         struct socket_wq wq;
 173         struct tun_struct __rcu *tun;
 174         struct fasync_struct *fasync;
 175         /* only used for fasnyc */
 176         unsigned int flags;
 177         union {
 178                 u16 queue_index;
 179                 unsigned int ifindex;
 180         };
 181         struct napi_struct napi;
 182         bool napi_enabled;
 183         bool napi_frags_enabled;
 184         struct mutex napi_mutex;        /* Protects access to the above napi */
 185         struct list_head next;
 186         struct tun_struct *detached;
 187         struct ptr_ring tx_ring;
 188         struct xdp_rxq_info xdp_rxq;
 189 };
 190
 191 struct tun_page {
 192         struct page *page;
 193         int count;
 194 };
 195
 196 struct tun_flow_entry {
 197         struct hlist_node hash_link;
 198         struct rcu_head rcu;
 199         struct tun_struct *tun;
 200
 201         u32 rxhash;
 202         u32 rps_rxhash;
 203         int queue_index;
 204         unsigned long updated ____cacheline_aligned_in_smp;
 205 };
 206
 207 #define TUN_NUM_FLOW_ENTRIES 1024
 208 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 209
 210 struct tun_prog {
 211         struct rcu_head rcu;
 212         struct bpf_prog *prog;
 213 };
 214
 215 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 216  * device, socket filter, sndbuf and vnet header size were restore when the
 217  * file were attached to a persist device.
 218  */
 219 struct tun_struct {
 220         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 221         unsigned int            numqueues;
 222         unsigned int            flags;
 223         kuid_t                  owner;
 224         kgid_t                  group;
 225
 226         struct net_device       *dev;
 227         netdev_features_t       set_features;
 228 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 229                           NETIF_F_TSO6)
 230
 231         int                     align;
 232         int                     vnet_hdr_sz;
 233         int                     sndbuf;
 234         struct tap_filter       txflt;
 235         struct sock_fprog       fprog;
 236         /* protected by rtnl lock */
 237         bool                    filter_attached;
 238 #ifdef TUN_DEBUG
 239         int debug;
 240 #endif
 241         spinlock_t lock;
 242         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 243         struct timer_list flow_gc_timer;
 244         unsigned long ageing_time;
 245         unsigned int numdisabled;
 246         struct list_head disabled;
 247         void *security;
 248         u32 flow_count;
 249         u32 rx_batched;
 250         struct tun_pcpu_stats __percpu *pcpu_stats;
 251         struct bpf_prog __rcu *xdp_prog;
 252         struct tun_prog __rcu *steering_prog;
 253         struct tun_prog __rcu *filter_prog;
 254         struct ethtool_link_ksettings link_ksettings;
 255 };
 256
 257 struct veth {
 258         __be16 h_vlan_proto;
 259         __be16 h_vlan_TCI;
 260 };
 261
 262 bool tun_is_xdp_frame(void *ptr)
 263 {
 264         return (unsigned long)ptr & TUN_XDP_FLAG;
 265 }
 266 EXPORT_SYMBOL(tun_is_xdp_frame);
 267
 268 void *tun_xdp_to_ptr(void *ptr)
 269 {
 270         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 271 }
 272 EXPORT_SYMBOL(tun_xdp_to_ptr);
 273
 274 void *tun_ptr_to_xdp(void *ptr)
 275 {
 276         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 277 }
 278 EXPORT_SYMBOL(tun_ptr_to_xdp);
 279
 280 static int tun_napi_receive(struct napi_struct *napi, int budget)
 281 {
 282         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 283         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 284         struct sk_buff_head process_queue;
 285         struct sk_buff *skb;
 286         int received = 0;
 287
 288         __skb_queue_head_init(&process_queue);
 289
 290         spin_lock(&queue->lock);
 291         skb_queue_splice_tail_init(queue, &process_queue);
 292         spin_unlock(&queue->lock);
 293
 294         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 295                 napi_gro_receive(napi, skb);
 296                 ++received;
 297         }
 298
 299         if (!skb_queue_empty(&process_queue)) {
 300                 spin_lock(&queue->lock);
 301                 skb_queue_splice(&process_queue, queue);
 302                 spin_unlock(&queue->lock);
 303         }
 304
 305         return received;
 306 }
 307
 308 static int tun_napi_poll(struct napi_struct *napi, int budget)
 309 {
 310         unsigned int received;
 311
 312         received = tun_napi_receive(napi, budget);
 313
 314         if (received < budget)
 315                 napi_complete_done(napi, received);
 316
 317         return received;
 318 }
 319
 320 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 321                           bool napi_en, bool napi_frags)
 322 {
 323         tfile->napi_enabled = napi_en;
 324         tfile->napi_frags_enabled = napi_en && napi_frags;
 325         if (napi_en) {
 326                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 327                                NAPI_POLL_WEIGHT);
 328                 napi_enable(&tfile->napi);
 329         }
 330 }
 331
 332 static void tun_napi_disable(struct tun_file *tfile)
 333 {
 334         if (tfile->napi_enabled)
 335                 napi_disable(&tfile->napi);
 336 }
 337
 338 static void tun_napi_del(struct tun_file *tfile)
 339 {
 340         if (tfile->napi_enabled)
 341                 netif_napi_del(&tfile->napi);
 342 }
 343
 344 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 345 {
 346         return tfile->napi_frags_enabled;
 347 }
 348
 349 #ifdef CONFIG_TUN_VNET_CROSS_LE
 350 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 351 {
 352         return tun->flags & TUN_VNET_BE ? false :
 353                 virtio_legacy_is_little_endian();
 354 }
 355
 356 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 357 {
 358         int be = !!(tun->flags & TUN_VNET_BE);
 359
 360         if (put_user(be, argp))
 361                 return -EFAULT;
 362
 363         return 0;
 364 }
 365
 366 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 367 {
 368         int be;
 369
 370         if (get_user(be, argp))
 371                 return -EFAULT;
 372
 373         if (be)
 374                 tun->flags |= TUN_VNET_BE;
 375         else
 376                 tun->flags &= ~TUN_VNET_BE;
 377
 378         return 0;
 379 }
 380 #else
 381 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 382 {
 383         return virtio_legacy_is_little_endian();
 384 }
 385
 386 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 387 {
 388         return -EINVAL;
 389 }
 390
 391 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 392 {
 393         return -EINVAL;
 394 }
 395 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 396
 397 static inline bool tun_is_little_endian(struct tun_struct *tun)
 398 {
 399         return tun->flags & TUN_VNET_LE ||
 400                 tun_legacy_is_little_endian(tun);
 401 }
 402
 403 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 404 {
 405         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 406 }
 407
 408 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 409 {
 410         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 411 }
 412
 413 static inline u32 tun_hashfn(u32 rxhash)
 414 {
 415         return rxhash & TUN_MASK_FLOW_ENTRIES;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 419 {
 420         struct tun_flow_entry *e;
 421
 422         hlist_for_each_entry_rcu(e, head, hash_link) {
 423                 if (e->rxhash == rxhash)
 424                         return e;
 425         }
 426         return NULL;
 427 }
 428
 429 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 430                                               struct hlist_head *head,
 431                                               u32 rxhash, u16 queue_index)
 432 {
 433         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 434
 435         if (e) {
 436                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 437                           rxhash, queue_index);
 438                 e->updated = jiffies;
 439                 e->rxhash = rxhash;
 440                 e->rps_rxhash = 0;
 441                 e->queue_index = queue_index;
 442                 e->tun = tun;
 443                 hlist_add_head_rcu(&e->hash_link, head);
 444                 ++tun->flow_count;
 445         }
 446         return e;
 447 }
 448
 449 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 450 {
 451         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 452                   e->rxhash, e->queue_index);
 453         hlist_del_rcu(&e->hash_link);
 454         kfree_rcu(e, rcu);
 455         --tun->flow_count;
 456 }
 457
 458 static void tun_flow_flush(struct tun_struct *tun)
 459 {
 460         int i;
 461
 462         spin_lock_bh(&tun->lock);
 463         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 464                 struct tun_flow_entry *e;
 465                 struct hlist_node *n;
 466
 467                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 468                         tun_flow_delete(tun, e);
 469         }
 470         spin_unlock_bh(&tun->lock);
 471 }
 472
 473 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 474 {
 475         int i;
 476
 477         spin_lock_bh(&tun->lock);
 478         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 479                 struct tun_flow_entry *e;
 480                 struct hlist_node *n;
 481
 482                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 483                         if (e->queue_index == queue_index)
 484                                 tun_flow_delete(tun, e);
 485                 }
 486         }
 487         spin_unlock_bh(&tun->lock);
 488 }
 489
 490 static void tun_flow_cleanup(struct timer_list *t)
 491 {
 492         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 493         unsigned long delay = tun->ageing_time;
 494         unsigned long next_timer = jiffies + delay;
 495         unsigned long count = 0;
 496         int i;
 497
 498         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 499
 500         spin_lock(&tun->lock);
 501         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 502                 struct tun_flow_entry *e;
 503                 struct hlist_node *n;
 504
 505                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 506                         unsigned long this_timer;
 507
 508                         this_timer = e->updated + delay;
 509                         if (time_before_eq(this_timer, jiffies)) {
 510                                 tun_flow_delete(tun, e);
 511                                 continue;
 512                         }
 513                         count++;
 514                         if (time_before(this_timer, next_timer))
 515                                 next_timer = this_timer;
 516                 }
 517         }
 518
 519         if (count)
 520                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 521         spin_unlock(&tun->lock);
 522 }
 523
 524 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 525                             struct tun_file *tfile)
 526 {
 527         struct hlist_head *head;
 528         struct tun_flow_entry *e;
 529         unsigned long delay = tun->ageing_time;
 530         u16 queue_index = tfile->queue_index;
 531
 532         head = &tun->flows[tun_hashfn(rxhash)];
 533
 534         rcu_read_lock();
 535
 536         e = tun_flow_find(head, rxhash);
 537         if (likely(e)) {
 538                 /* TODO: keep queueing to old queue until it's empty? */
 539                 if (e->queue_index != queue_index)
 540                         e->queue_index = queue_index;
 541                 if (e->updated != jiffies)
 542                         e->updated = jiffies;
 543                 sock_rps_record_flow_hash(e->rps_rxhash);
 544         } else {
 545                 spin_lock_bh(&tun->lock);
 546                 if (!tun_flow_find(head, rxhash) &&
 547                     tun->flow_count < MAX_TAP_FLOWS)
 548                         tun_flow_create(tun, head, rxhash, queue_index);
 549
 550                 if (!timer_pending(&tun->flow_gc_timer))
 551                         mod_timer(&tun->flow_gc_timer,
 552                                   round_jiffies_up(jiffies + delay));
 553                 spin_unlock_bh(&tun->lock);
 554         }
 555
 556         rcu_read_unlock();
 557 }
 558
 559 /**
 560  * Save the hash received in the stack receive path and update the
 561  * flow_hash table accordingly.
 562  */
 563 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 564 {
 565         if (unlikely(e->rps_rxhash != hash))
 566                 e->rps_rxhash = hash;
 567 }
 568
 569 /* We try to identify a flow through its rxhash. The reason that
 570  * we do not check rxq no. is because some cards(e.g 82599), chooses
 571  * the rxq based on the txq where the last packet of the flow comes. As
 572  * the userspace application move between processors, we may get a
 573  * different rxq no. here.
 574  */
 575 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 576 {
 577         struct tun_flow_entry *e;
 578         u32 txq = 0;
 579         u32 numqueues = 0;
 580
 581         numqueues = READ_ONCE(tun->numqueues);
 582
 583         txq = __skb_get_hash_symmetric(skb);
 584         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 585         if (e) {
 586                 tun_flow_save_rps_rxhash(e, txq);
 587                 txq = e->queue_index;
 588         } else {
 589                 /* use multiply and shift instead of expensive divide */
 590                 txq = ((u64)txq * numqueues) >> 32;
 591         }
 592
 593         return txq;
 594 }
 595
 596 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 597 {
 598         struct tun_prog *prog;
 599         u16 ret = 0;
 600
 601         prog = rcu_dereference(tun->steering_prog);
 602         if (prog)
 603                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 604
 605         return ret % tun->numqueues;
 606 }
 607
 608 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 609                             struct net_device *sb_dev,
 610                             select_queue_fallback_t fallback)
 611 {
 612         struct tun_struct *tun = netdev_priv(dev);
 613         u16 ret;
 614
 615         rcu_read_lock();
 616         if (rcu_dereference(tun->steering_prog))
 617                 ret = tun_ebpf_select_queue(tun, skb);
 618         else
 619                 ret = tun_automq_select_queue(tun, skb);
 620         rcu_read_unlock();
 621
 622         return ret;
 623 }
 624
 625 static inline bool tun_not_capable(struct tun_struct *tun)
 626 {
 627         const struct cred *cred = current_cred();
 628         struct net *net = dev_net(tun->dev);
 629
 630         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 631                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 632                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 633 }
 634
 635 static void tun_set_real_num_queues(struct tun_struct *tun)
 636 {
 637         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 638         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 639 }
 640
 641 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 642 {
 643         tfile->detached = tun;
 644         list_add_tail(&tfile->next, &tun->disabled);
 645         ++tun->numdisabled;
 646 }
 647
 648 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 649 {
 650         struct tun_struct *tun = tfile->detached;
 651
 652         tfile->detached = NULL;
 653         list_del_init(&tfile->next);
 654         --tun->numdisabled;
 655         return tun;
 656 }
 657
 658 void tun_ptr_free(void *ptr)
 659 {
 660         if (!ptr)
 661                 return;
 662         if (tun_is_xdp_frame(ptr)) {
 663                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 664
 665                 xdp_return_frame(xdpf);
 666         } else {
 667                 __skb_array_destroy_skb(ptr);
 668         }
 669 }
 670 EXPORT_SYMBOL_GPL(tun_ptr_free);
 671
 672 static void tun_queue_purge(struct tun_file *tfile)
 673 {
 674         void *ptr;
 675
 676         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 677                 tun_ptr_free(ptr);
 678
 679         skb_queue_purge(&tfile->sk.sk_write_queue);
 680         skb_queue_purge(&tfile->sk.sk_error_queue);
 681 }
 682
 683 static void __tun_detach(struct tun_file *tfile, bool clean)
 684 {
 685         struct tun_file *ntfile;
 686         struct tun_struct *tun;
 687
 688         tun = rtnl_dereference(tfile->tun);
 689
 690         if (tun && clean) {
 691                 tun_napi_disable(tfile);
 692                 tun_napi_del(tfile);
 693         }
 694
 695         if (tun && !tfile->detached) {
 696                 u16 index = tfile->queue_index;
 697                 BUG_ON(index >= tun->numqueues);
 698
 699                 rcu_assign_pointer(tun->tfiles[index],
 700                                    tun->tfiles[tun->numqueues - 1]);
 701                 ntfile = rtnl_dereference(tun->tfiles[index]);
 702                 ntfile->queue_index = index;
 703
 704                 --tun->numqueues;
 705                 if (clean) {
 706                         RCU_INIT_POINTER(tfile->tun, NULL);
 707                         sock_put(&tfile->sk);
 708                 } else
 709                         tun_disable_queue(tun, tfile);
 710
 711                 synchronize_net();
 712                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 713                 /* Drop read queue */
 714                 tun_queue_purge(tfile);
 715                 tun_set_real_num_queues(tun);
 716         } else if (tfile->detached && clean) {
 717                 tun = tun_enable_queue(tfile);
 718                 sock_put(&tfile->sk);
 719         }
 720
 721         if (clean) {
 722                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 723                         netif_carrier_off(tun->dev);
 724
 725                         if (!(tun->flags & IFF_PERSIST) &&
 726                             tun->dev->reg_state == NETREG_REGISTERED)
 727                                 unregister_netdevice(tun->dev);
 728                 }
 729                 if (tun)
 730                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 731                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 732                 sock_put(&tfile->sk);
 733         }
 734 }
 735
 736 static void tun_detach(struct tun_file *tfile, bool clean)
 737 {
 738         struct tun_struct *tun;
 739         struct net_device *dev;
 740
 741         rtnl_lock();
 742         tun = rtnl_dereference(tfile->tun);
 743         dev = tun ? tun->dev : NULL;
 744         __tun_detach(tfile, clean);
 745         if (dev)
 746                 netdev_state_change(dev);
 747         rtnl_unlock();
 748 }
 749
 750 static void tun_detach_all(struct net_device *dev)
 751 {
 752         struct tun_struct *tun = netdev_priv(dev);
 753         struct tun_file *tfile, *tmp;
 754         int i, n = tun->numqueues;
 755
 756         for (i = 0; i < n; i++) {
 757                 tfile = rtnl_dereference(tun->tfiles[i]);
 758                 BUG_ON(!tfile);
 759                 tun_napi_disable(tfile);
 760                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 761                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 762                 RCU_INIT_POINTER(tfile->tun, NULL);
 763                 --tun->numqueues;
 764         }
 765         list_for_each_entry(tfile, &tun->disabled, next) {
 766                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 767                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 768                 RCU_INIT_POINTER(tfile->tun, NULL);
 769         }
 770         BUG_ON(tun->numqueues != 0);
 771
 772         synchronize_net();
 773         for (i = 0; i < n; i++) {
 774                 tfile = rtnl_dereference(tun->tfiles[i]);
 775                 tun_napi_del(tfile);
 776                 /* Drop read queue */
 777                 tun_queue_purge(tfile);
 778                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 779                 sock_put(&tfile->sk);
 780         }
 781         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 782                 tun_enable_queue(tfile);
 783                 tun_queue_purge(tfile);
 784                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 785                 sock_put(&tfile->sk);
 786         }
 787         BUG_ON(tun->numdisabled != 0);
 788
 789         if (tun->flags & IFF_PERSIST)
 790                 module_put(THIS_MODULE);
 791 }
 792
 793 static int tun_attach(struct tun_struct *tun, struct file *file,
 794                       bool skip_filter, bool napi, bool napi_frags)
 795 {
 796         struct tun_file *tfile = file->private_data;
 797         struct net_device *dev = tun->dev;
 798         int err;
 799
 800         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 801         if (err < 0)
 802                 goto out;
 803
 804         err = -EINVAL;
 805         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 806                 goto out;
 807
 808         err = -EBUSY;
 809         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 810                 goto out;
 811
 812         err = -E2BIG;
 813         if (!tfile->detached &&
 814             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 815                 goto out;
 816
 817         err = 0;
 818
 819         /* Re-attach the filter to persist device */
 820         if (!skip_filter && (tun->filter_attached == true)) {
 821                 lock_sock(tfile->socket.sk);
 822                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 823                 release_sock(tfile->socket.sk);
 824                 if (!err)
 825                         goto out;
 826         }
 827
 828         if (!tfile->detached &&
 829             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 830                             GFP_KERNEL, tun_ptr_free)) {
 831                 err = -ENOMEM;
 832                 goto out;
 833         }
 834
 835         tfile->queue_index = tun->numqueues;
 836         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 837
 838         if (tfile->detached) {
 839                 /* Re-attach detached tfile, updating XDP queue_index */
 840                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 841
 842                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 843                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 844         } else {
 845                 /* Setup XDP RX-queue info, for new tfile getting attached */
 846                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 847                                        tun->dev, tfile->queue_index);
 848                 if (err < 0)
 849                         goto out;
 850                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 851                                                  MEM_TYPE_PAGE_SHARED, NULL);
 852                 if (err < 0) {
 853                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 854                         goto out;
 855                 }
 856                 err = 0;
 857         }
 858
 859         if (tfile->detached) {
 860                 tun_enable_queue(tfile);
 861         } else {
 862                 sock_hold(&tfile->sk);
 863                 tun_napi_init(tun, tfile, napi, napi_frags);
 864         }
 865
 866         if (rtnl_dereference(tun->xdp_prog))
 867                 sock_set_flag(&tfile->sk, SOCK_XDP);
 868
 869         tun_set_real_num_queues(tun);
 870
 871         /* device is allowed to go away first, so no need to hold extra
 872          * refcnt.
 873          */
 874
 875         /* Publish tfile->tun and tun->tfiles only after we've fully
 876          * initialized tfile; otherwise we risk using half-initialized
 877          * object.
 878          */
 879         rcu_assign_pointer(tfile->tun, tun);
 880         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 881         tun->numqueues++;
 882 out:
 883         return err;
 884 }
 885
 886 static struct tun_struct *tun_get(struct tun_file *tfile)
 887 {
 888         struct tun_struct *tun;
 889
 890         rcu_read_lock();
 891         tun = rcu_dereference(tfile->tun);
 892         if (tun)
 893                 dev_hold(tun->dev);
 894         rcu_read_unlock();
 895
 896         return tun;
 897 }
 898
 899 static void tun_put(struct tun_struct *tun)
 900 {
 901         dev_put(tun->dev);
 902 }
 903
 904 /* TAP filtering */
 905 static void addr_hash_set(u32 *mask, const u8 *addr)
 906 {
 907         int n = ether_crc(ETH_ALEN, addr) >> 26;
 908         mask[n >> 5] |= (1 << (n & 31));
 909 }
 910
 911 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 912 {
 913         int n = ether_crc(ETH_ALEN, addr) >> 26;
 914         return mask[n >> 5] & (1 << (n & 31));
 915 }
 916
 917 static int update_filter(struct tap_filter *filter, void __user *arg)
 918 {
 919         struct { u8 u[ETH_ALEN]; } *addr;
 920         struct tun_filter uf;
 921         int err, alen, n, nexact;
 922
 923         if (copy_from_user(&uf, arg, sizeof(uf)))
 924                 return -EFAULT;
 925
 926         if (!uf.count) {
 927                 /* Disabled */
 928                 filter->count = 0;
 929                 return 0;
 930         }
 931
 932         alen = ETH_ALEN * uf.count;
 933         addr = memdup_user(arg + sizeof(uf), alen);
 934         if (IS_ERR(addr))
 935                 return PTR_ERR(addr);
 936
 937         /* The filter is updated without holding any locks. Which is
 938          * perfectly safe. We disable it first and in the worst
 939          * case we'll accept a few undesired packets. */
 940         filter->count = 0;
 941         wmb();
 942
 943         /* Use first set of addresses as an exact filter */
 944         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 945                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 946
 947         nexact = n;
 948
 949         /* Remaining multicast addresses are hashed,
 950          * unicast will leave the filter disabled. */
 951         memset(filter->mask, 0, sizeof(filter->mask));
 952         for (; n < uf.count; n++) {
 953                 if (!is_multicast_ether_addr(addr[n].u)) {
 954                         err = 0; /* no filter */
 955                         goto free_addr;
 956                 }
 957                 addr_hash_set(filter->mask, addr[n].u);
 958         }
 959
 960         /* For ALLMULTI just set the mask to all ones.
 961          * This overrides the mask populated above. */
 962         if ((uf.flags & TUN_FLT_ALLMULTI))
 963                 memset(filter->mask, ~0, sizeof(filter->mask));
 964
 965         /* Now enable the filter */
 966         wmb();
 967         filter->count = nexact;
 968
 969         /* Return the number of exact filters */
 970         err = nexact;
 971 free_addr:
 972         kfree(addr);
 973         return err;
 974 }
 975
 976 /* Returns: 0 - drop, !=0 - accept */
 977 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 978 {
 979         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 980          * at this point. */
 981         struct ethhdr *eh = (struct ethhdr *) skb->data;
 982         int i;
 983
 984         /* Exact match */
 985         for (i = 0; i < filter->count; i++)
 986                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 987                         return 1;
 988
 989         /* Inexact match (multicast only) */
 990         if (is_multicast_ether_addr(eh->h_dest))
 991                 return addr_hash_test(filter->mask, eh->h_dest);
 992
 993         return 0;
 994 }
 995
 996 /*
 997  * Checks whether the packet is accepted or not.
 998  * Returns: 0 - drop, !=0 - accept
 999  */
1000 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1001 {
1002         if (!filter->count)
1003                 return 1;
1004
1005         return run_filter(filter, skb);
1006 }
1007
1008 /* Network device part of the driver */
1009
1010 static const struct ethtool_ops tun_ethtool_ops;
1011
1012 /* Net device detach from fd. */
1013 static void tun_net_uninit(struct net_device *dev)
1014 {
1015         tun_detach_all(dev);
1016 }
1017
1018 /* Net device open. */
1019 static int tun_net_open(struct net_device *dev)
1020 {
1021         struct tun_struct *tun = netdev_priv(dev);
1022         int i;
1023
1024         netif_tx_start_all_queues(dev);
1025
1026         for (i = 0; i < tun->numqueues; i++) {
1027                 struct tun_file *tfile;
1028
1029                 tfile = rtnl_dereference(tun->tfiles[i]);
1030                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1031         }
1032
1033         return 0;
1034 }
1035
1036 /* Net device close. */
1037 static int tun_net_close(struct net_device *dev)
1038 {
1039         netif_tx_stop_all_queues(dev);
1040         return 0;
1041 }
1042
1043 /* Net device start xmit */
1044 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1045 {
1046 #ifdef CONFIG_RPS
1047         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1048                 /* Select queue was not called for the skbuff, so we extract the
1049                  * RPS hash and save it into the flow_table here.
1050                  */
1051                 struct tun_flow_entry *e;
1052                 __u32 rxhash;
1053
1054                 rxhash = __skb_get_hash_symmetric(skb);
1055                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1056                 if (e)
1057                         tun_flow_save_rps_rxhash(e, rxhash);
1058         }
1059 #endif
1060 }
1061
1062 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1063                                     struct sk_buff *skb,
1064                                     int len)
1065 {
1066         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1067
1068         if (prog)
1069                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1070
1071         return len;
1072 }
1073
1074 /* Net device start xmit */
1075 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1076 {
1077         struct tun_struct *tun = netdev_priv(dev);
1078         int txq = skb->queue_mapping;
1079         struct tun_file *tfile;
1080         int len = skb->len;
1081
1082         rcu_read_lock();
1083         tfile = rcu_dereference(tun->tfiles[txq]);
1084
1085         /* Drop packet if interface is not attached */
1086         if (txq >= tun->numqueues)
1087                 goto drop;
1088
1089         if (!rcu_dereference(tun->steering_prog))
1090                 tun_automq_xmit(tun, skb);
1091
1092         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1093
1094         BUG_ON(!tfile);
1095
1096         /* Drop if the filter does not like it.
1097          * This is a noop if the filter is disabled.
1098          * Filter can be enabled only for the TAP devices. */
1099         if (!check_filter(&tun->txflt, skb))
1100                 goto drop;
1101
1102         if (tfile->socket.sk->sk_filter &&
1103             sk_filter(tfile->socket.sk, skb))
1104                 goto drop;
1105
1106         len = run_ebpf_filter(tun, skb, len);
1107         if (len == 0 || pskb_trim(skb, len))
1108                 goto drop;
1109
1110         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1111                 goto drop;
1112
1113         skb_tx_timestamp(skb);
1114
1115         /* Orphan the skb - required as we might hang on to it
1116          * for indefinite time.
1117          */
1118         skb_orphan(skb);
1119
1120         nf_reset(skb);
1121
1122         if (ptr_ring_produce(&tfile->tx_ring, skb))
1123                 goto drop;
1124
1125         /* Notify and wake up reader process */
1126         if (tfile->flags & TUN_FASYNC)
1127                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1128         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1129
1130         rcu_read_unlock();
1131         return NETDEV_TX_OK;
1132
1133 drop:
1134         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1135         skb_tx_error(skb);
1136         kfree_skb(skb);
1137         rcu_read_unlock();
1138         return NET_XMIT_DROP;
1139 }
1140
1141 static void tun_net_mclist(struct net_device *dev)
1142 {
1143         /*
1144          * This callback is supposed to deal with mc filter in
1145          * _rx_ path and has nothing to do with the _tx_ path.
1146          * In rx path we always accept everything userspace gives us.
1147          */
1148 }
1149
1150 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1151         netdev_features_t features)
1152 {
1153         struct tun_struct *tun = netdev_priv(dev);
1154
1155         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1156 }
1157
1158 static void tun_set_headroom(struct net_device *dev, int new_hr)
1159 {
1160         struct tun_struct *tun = netdev_priv(dev);
1161
1162         if (new_hr < NET_SKB_PAD)
1163                 new_hr = NET_SKB_PAD;
1164
1165         tun->align = new_hr;
1166 }
1167
1168 static void
1169 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1170 {
1171         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1172         struct tun_struct *tun = netdev_priv(dev);
1173         struct tun_pcpu_stats *p;
1174         int i;
1175
1176         for_each_possible_cpu(i) {
1177                 u64 rxpackets, rxbytes, txpackets, txbytes;
1178                 unsigned int start;
1179
1180                 p = per_cpu_ptr(tun->pcpu_stats, i);
1181                 do {
1182                         start = u64_stats_fetch_begin(&p->syncp);
1183                         rxpackets       = p->rx_packets;
1184                         rxbytes         = p->rx_bytes;
1185                         txpackets       = p->tx_packets;
1186                         txbytes         = p->tx_bytes;
1187                 } while (u64_stats_fetch_retry(&p->syncp, start));
1188
1189                 stats->rx_packets       += rxpackets;
1190                 stats->rx_bytes         += rxbytes;
1191                 stats->tx_packets       += txpackets;
1192                 stats->tx_bytes         += txbytes;
1193
1194                 /* u32 counters */
1195                 rx_dropped      += p->rx_dropped;
1196                 rx_frame_errors += p->rx_frame_errors;
1197                 tx_dropped      += p->tx_dropped;
1198         }
1199         stats->rx_dropped  = rx_dropped;
1200         stats->rx_frame_errors = rx_frame_errors;
1201         stats->tx_dropped = tx_dropped;
1202 }
1203
1204 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1205                        struct netlink_ext_ack *extack)
1206 {
1207         struct tun_struct *tun = netdev_priv(dev);
1208         struct tun_file *tfile;
1209         struct bpf_prog *old_prog;
1210         int i;
1211
1212         old_prog = rtnl_dereference(tun->xdp_prog);
1213         rcu_assign_pointer(tun->xdp_prog, prog);
1214         if (old_prog)
1215                 bpf_prog_put(old_prog);
1216
1217         for (i = 0; i < tun->numqueues; i++) {
1218                 tfile = rtnl_dereference(tun->tfiles[i]);
1219                 if (prog)
1220                         sock_set_flag(&tfile->sk, SOCK_XDP);
1221                 else
1222                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1223         }
1224         list_for_each_entry(tfile, &tun->disabled, next) {
1225                 if (prog)
1226                         sock_set_flag(&tfile->sk, SOCK_XDP);
1227                 else
1228                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1229         }
1230
1231         return 0;
1232 }
1233
1234 static u32 tun_xdp_query(struct net_device *dev)
1235 {
1236         struct tun_struct *tun = netdev_priv(dev);
1237         const struct bpf_prog *xdp_prog;
1238
1239         xdp_prog = rtnl_dereference(tun->xdp_prog);
1240         if (xdp_prog)
1241                 return xdp_prog->aux->id;
1242
1243         return 0;
1244 }
1245
1246 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1247 {
1248         switch (xdp->command) {
1249         case XDP_SETUP_PROG:
1250                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1251         case XDP_QUERY_PROG:
1252                 xdp->prog_id = tun_xdp_query(dev);
1253                 return 0;
1254         default:
1255                 return -EINVAL;
1256         }
1257 }
1258
1259 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1260 {
1261         if (new_carrier) {
1262                 struct tun_struct *tun = netdev_priv(dev);
1263
1264                 if (!tun->numqueues)
1265                         return -EPERM;
1266
1267                 netif_carrier_on(dev);
1268         } else {
1269                 netif_carrier_off(dev);
1270         }
1271         return 0;
1272 }
1273
1274 static const struct net_device_ops tun_netdev_ops = {
1275         .ndo_uninit             = tun_net_uninit,
1276         .ndo_open               = tun_net_open,
1277         .ndo_stop               = tun_net_close,
1278         .ndo_start_xmit         = tun_net_xmit,
1279         .ndo_fix_features       = tun_net_fix_features,
1280         .ndo_select_queue       = tun_select_queue,
1281         .ndo_set_rx_headroom    = tun_set_headroom,
1282         .ndo_get_stats64        = tun_net_get_stats64,
1283         .ndo_change_carrier     = tun_net_change_carrier,
1284 };
1285
1286 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1287 {
1288         /* Notify and wake up reader process */
1289         if (tfile->flags & TUN_FASYNC)
1290                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1291         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1292 }
1293
1294 static int tun_xdp_xmit(struct net_device *dev, int n,
1295                         struct xdp_frame **frames, u32 flags)
1296 {
1297         struct tun_struct *tun = netdev_priv(dev);
1298         struct tun_file *tfile;
1299         u32 numqueues;
1300         int drops = 0;
1301         int cnt = n;
1302         int i;
1303
1304         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1305                 return -EINVAL;
1306
1307         rcu_read_lock();
1308
1309         numqueues = READ_ONCE(tun->numqueues);
1310         if (!numqueues) {
1311                 rcu_read_unlock();
1312                 return -ENXIO; /* Caller will free/return all frames */
1313         }
1314
1315         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1316                                             numqueues]);
1317
1318         spin_lock(&tfile->tx_ring.producer_lock);
1319         for (i = 0; i < n; i++) {
1320                 struct xdp_frame *xdp = frames[i];
1321                 /* Encode the XDP flag into lowest bit for consumer to differ
1322                  * XDP buffer from sk_buff.
1323                  */
1324                 void *frame = tun_xdp_to_ptr(xdp);
1325
1326                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1327                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1328                         xdp_return_frame_rx_napi(xdp);
1329                         drops++;
1330                 }
1331         }
1332         spin_unlock(&tfile->tx_ring.producer_lock);
1333
1334         if (flags & XDP_XMIT_FLUSH)
1335                 __tun_xdp_flush_tfile(tfile);
1336
1337         rcu_read_unlock();
1338         return cnt - drops;
1339 }
1340
1341 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1342 {
1343         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1344
1345         if (unlikely(!frame))
1346                 return -EOVERFLOW;
1347
1348         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1349 }
1350
1351 static const struct net_device_ops tap_netdev_ops = {
1352         .ndo_uninit             = tun_net_uninit,
1353         .ndo_open               = tun_net_open,
1354         .ndo_stop               = tun_net_close,
1355         .ndo_start_xmit         = tun_net_xmit,
1356         .ndo_fix_features       = tun_net_fix_features,
1357         .ndo_set_rx_mode        = tun_net_mclist,
1358         .ndo_set_mac_address    = eth_mac_addr,
1359         .ndo_validate_addr      = eth_validate_addr,
1360         .ndo_select_queue       = tun_select_queue,
1361         .ndo_features_check     = passthru_features_check,
1362         .ndo_set_rx_headroom    = tun_set_headroom,
1363         .ndo_get_stats64        = tun_net_get_stats64,
1364         .ndo_bpf                = tun_xdp,
1365         .ndo_xdp_xmit           = tun_xdp_xmit,
1366         .ndo_change_carrier     = tun_net_change_carrier,
1367 };
1368
1369 static void tun_flow_init(struct tun_struct *tun)
1370 {
1371         int i;
1372
1373         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1374                 INIT_HLIST_HEAD(&tun->flows[i]);
1375
1376         tun->ageing_time = TUN_FLOW_EXPIRE;
1377         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1378         mod_timer(&tun->flow_gc_timer,
1379                   round_jiffies_up(jiffies + tun->ageing_time));
1380 }
1381
1382 static void tun_flow_uninit(struct tun_struct *tun)
1383 {
1384         del_timer_sync(&tun->flow_gc_timer);
1385         tun_flow_flush(tun);
1386 }
1387
1388 #define MIN_MTU 68
1389 #define MAX_MTU 65535
1390
1391 /* Initialize net device. */
1392 static void tun_net_init(struct net_device *dev)
1393 {
1394         struct tun_struct *tun = netdev_priv(dev);
1395
1396         switch (tun->flags & TUN_TYPE_MASK) {
1397         case IFF_TUN:
1398                 dev->netdev_ops = &tun_netdev_ops;
1399
1400                 /* Point-to-Point TUN Device */
1401                 dev->hard_header_len = 0;
1402                 dev->addr_len = 0;
1403                 dev->mtu = 1500;
1404
1405                 /* Zero header length */
1406                 dev->type = ARPHRD_NONE;
1407                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1408                 break;
1409
1410         case IFF_TAP:
1411                 dev->netdev_ops = &tap_netdev_ops;
1412                 /* Ethernet TAP Device */
1413                 ether_setup(dev);
1414                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1415                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1416
1417                 eth_hw_addr_random(dev);
1418
1419                 break;
1420         }
1421
1422         dev->min_mtu = MIN_MTU;
1423         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1424 }
1425
1426 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1427 {
1428         struct sock *sk = tfile->socket.sk;
1429
1430         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1431 }
1432
1433 /* Character device part */
1434
1435 /* Poll */
1436 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1437 {
1438         struct tun_file *tfile = file->private_data;
1439         struct tun_struct *tun = tun_get(tfile);
1440         struct sock *sk;
1441         __poll_t mask = 0;
1442
1443         if (!tun)
1444                 return EPOLLERR;
1445
1446         sk = tfile->socket.sk;
1447
1448         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1449
1450         poll_wait(file, sk_sleep(sk), wait);
1451
1452         if (!ptr_ring_empty(&tfile->tx_ring))
1453                 mask |= EPOLLIN | EPOLLRDNORM;
1454
1455         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1456          * guarantee EPOLLOUT to be raised by either here or
1457          * tun_sock_write_space(). Then process could get notification
1458          * after it writes to a down device and meets -EIO.
1459          */
1460         if (tun_sock_writeable(tun, tfile) ||
1461             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1462              tun_sock_writeable(tun, tfile)))
1463                 mask |= EPOLLOUT | EPOLLWRNORM;
1464
1465         if (tun->dev->reg_state != NETREG_REGISTERED)
1466                 mask = EPOLLERR;
1467
1468         tun_put(tun);
1469         return mask;
1470 }
1471
1472 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1473                                             size_t len,
1474                                             const struct iov_iter *it)
1475 {
1476         struct sk_buff *skb;
1477         size_t linear;
1478         int err;
1479         int i;
1480
1481         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1482                 return ERR_PTR(-ENOMEM);
1483
1484         local_bh_disable();
1485         skb = napi_get_frags(&tfile->napi);
1486         local_bh_enable();
1487         if (!skb)
1488                 return ERR_PTR(-ENOMEM);
1489
1490         linear = iov_iter_single_seg_count(it);
1491         err = __skb_grow(skb, linear);
1492         if (err)
1493                 goto free;
1494
1495         skb->len = len;
1496         skb->data_len = len - linear;
1497         skb->truesize += skb->data_len;
1498
1499         for (i = 1; i < it->nr_segs; i++) {
1500                 size_t fragsz = it->iov[i].iov_len;
1501                 struct page *page;
1502                 void *frag;
1503
1504                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1505                         err = -EINVAL;
1506                         goto free;
1507                 }
1508                 frag = netdev_alloc_frag(fragsz);
1509                 if (!frag) {
1510                         err = -ENOMEM;
1511                         goto free;
1512                 }
1513                 page = virt_to_head_page(frag);
1514                 skb_fill_page_desc(skb, i - 1, page,
1515                                    frag - page_address(page), fragsz);
1516         }
1517
1518         return skb;
1519 free:
1520         /* frees skb and all frags allocated with napi_alloc_frag() */
1521         napi_free_frags(&tfile->napi);
1522         return ERR_PTR(err);
1523 }
1524
1525 /* prepad is the amount to reserve at front.  len is length after that.
1526  * linear is a hint as to how much to copy (usually headers). */
1527 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1528                                      size_t prepad, size_t len,
1529                                      size_t linear, int noblock)
1530 {
1531         struct sock *sk = tfile->socket.sk;
1532         struct sk_buff *skb;
1533         int err;
1534
1535         /* Under a page?  Don't bother with paged skb. */
1536         if (prepad + len < PAGE_SIZE || !linear)
1537                 linear = len;
1538
1539         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1540                                    &err, 0);
1541         if (!skb)
1542                 return ERR_PTR(err);
1543
1544         skb_reserve(skb, prepad);
1545         skb_put(skb, linear);
1546         skb->data_len = len - linear;
1547         skb->len += len - linear;
1548
1549         return skb;
1550 }
1551
1552 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1553                            struct sk_buff *skb, int more)
1554 {
1555         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1556         struct sk_buff_head process_queue;
1557         u32 rx_batched = tun->rx_batched;
1558         bool rcv = false;
1559
1560         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1561                 local_bh_disable();
1562                 skb_record_rx_queue(skb, tfile->queue_index);
1563                 netif_receive_skb(skb);
1564                 local_bh_enable();
1565                 return;
1566         }
1567
1568         spin_lock(&queue->lock);
1569         if (!more || skb_queue_len(queue) == rx_batched) {
1570                 __skb_queue_head_init(&process_queue);
1571                 skb_queue_splice_tail_init(queue, &process_queue);
1572                 rcv = true;
1573         } else {
1574                 __skb_queue_tail(queue, skb);
1575         }
1576         spin_unlock(&queue->lock);
1577
1578         if (rcv) {
1579                 struct sk_buff *nskb;
1580
1581                 local_bh_disable();
1582                 while ((nskb = __skb_dequeue(&process_queue))) {
1583                         skb_record_rx_queue(nskb, tfile->queue_index);
1584                         netif_receive_skb(nskb);
1585                 }
1586                 skb_record_rx_queue(skb, tfile->queue_index);
1587                 netif_receive_skb(skb);
1588                 local_bh_enable();
1589         }
1590 }
1591
1592 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1593                               int len, int noblock, bool zerocopy)
1594 {
1595         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1596                 return false;
1597
1598         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1599                 return false;
1600
1601         if (!noblock)
1602                 return false;
1603
1604         if (zerocopy)
1605                 return false;
1606
1607         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1608             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1609                 return false;
1610
1611         return true;
1612 }
1613
1614 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1615                                        int buflen, int len, int pad)
1616 {
1617         struct sk_buff *skb = build_skb(buf, buflen);
1618
1619         if (!skb)
1620                 return ERR_PTR(-ENOMEM);
1621
1622         skb_reserve(skb, pad);
1623         skb_put(skb, len);
1624
1625         get_page(alloc_frag->page);
1626         alloc_frag->offset += buflen;
1627
1628         return skb;
1629 }
1630
1631 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1632                        struct xdp_buff *xdp, u32 act)
1633 {
1634         int err;
1635
1636         switch (act) {
1637         case XDP_REDIRECT:
1638                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1639                 if (err)
1640                         return err;
1641                 break;
1642         case XDP_TX:
1643                 err = tun_xdp_tx(tun->dev, xdp);
1644                 if (err < 0)
1645                         return err;
1646                 break;
1647         case XDP_PASS:
1648                 break;
1649         default:
1650                 bpf_warn_invalid_xdp_action(act);
1651                 /* fall through */
1652         case XDP_ABORTED:
1653                 trace_xdp_exception(tun->dev, xdp_prog, act);
1654                 /* fall through */
1655         case XDP_DROP:
1656                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1657                 break;
1658         }
1659
1660         return act;
1661 }
1662
1663 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1664                                      struct tun_file *tfile,
1665                                      struct iov_iter *from,
1666                                      struct virtio_net_hdr *hdr,
1667                                      int len, int *skb_xdp)
1668 {
1669         struct page_frag *alloc_frag = &current->task_frag;
1670         struct bpf_prog *xdp_prog;
1671         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1672         char *buf;
1673         size_t copied;
1674         int pad = TUN_RX_PAD;
1675         int err = 0;
1676
1677         rcu_read_lock();
1678         xdp_prog = rcu_dereference(tun->xdp_prog);
1679         if (xdp_prog)
1680                 pad += XDP_PACKET_HEADROOM;
1681         buflen += SKB_DATA_ALIGN(len + pad);
1682         rcu_read_unlock();
1683
1684         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1685         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1686                 return ERR_PTR(-ENOMEM);
1687
1688         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1689         copied = copy_page_from_iter(alloc_frag->page,
1690                                      alloc_frag->offset + pad,
1691                                      len, from);
1692         if (copied != len)
1693                 return ERR_PTR(-EFAULT);
1694
1695         /* There's a small window that XDP may be set after the check
1696          * of xdp_prog above, this should be rare and for simplicity
1697          * we do XDP on skb in case the headroom is not enough.
1698          */
1699         if (hdr->gso_type || !xdp_prog) {
1700                 *skb_xdp = 1;
1701                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1702         }
1703
1704         *skb_xdp = 0;
1705
1706         local_bh_disable();
1707         rcu_read_lock();
1708         xdp_prog = rcu_dereference(tun->xdp_prog);
1709         if (xdp_prog) {
1710                 struct xdp_buff xdp;
1711                 u32 act;
1712
1713                 xdp.data_hard_start = buf;
1714                 xdp.data = buf + pad;
1715                 xdp_set_data_meta_invalid(&xdp);
1716                 xdp.data_end = xdp.data + len;
1717                 xdp.rxq = &tfile->xdp_rxq;
1718
1719                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1720                 if (act == XDP_REDIRECT || act == XDP_TX) {
1721                         get_page(alloc_frag->page);
1722                         alloc_frag->offset += buflen;
1723                 }
1724                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1725                 if (err < 0)
1726                         goto err_xdp;
1727                 if (err == XDP_REDIRECT)
1728                         xdp_do_flush_map();
1729                 if (err != XDP_PASS)
1730                         goto out;
1731
1732                 pad = xdp.data - xdp.data_hard_start;
1733                 len = xdp.data_end - xdp.data;
1734         }
1735         rcu_read_unlock();
1736         local_bh_enable();
1737
1738         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1739
1740 err_xdp:
1741         put_page(alloc_frag->page);
1742 out:
1743         rcu_read_unlock();
1744         local_bh_enable();
1745         return NULL;
1746 }
1747
1748 /* Get packet from user space buffer */
1749 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1750                             void *msg_control, struct iov_iter *from,
1751                             int noblock, bool more)
1752 {
1753         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1754         struct sk_buff *skb;
1755         size_t total_len = iov_iter_count(from);
1756         size_t len = total_len, align = tun->align, linear;
1757         struct virtio_net_hdr gso = { 0 };
1758         struct tun_pcpu_stats *stats;
1759         int good_linear;
1760         int copylen;
1761         bool zerocopy = false;
1762         int err;
1763         u32 rxhash = 0;
1764         int skb_xdp = 1;
1765         bool frags = tun_napi_frags_enabled(tfile);
1766
1767         if (!(tun->dev->flags & IFF_UP))
1768                 return -EIO;
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                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1872                         kfree_skb(skb);
1873                         if (frags) {
1874                                 tfile->napi.skb = NULL;
1875                                 mutex_unlock(&tfile->napi_mutex);
1876                         }
1877
1878                         return -EFAULT;
1879                 }
1880         }
1881
1882         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1883                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1884                 kfree_skb(skb);
1885                 if (frags) {
1886                         tfile->napi.skb = NULL;
1887                         mutex_unlock(&tfile->napi_mutex);
1888                 }
1889
1890                 return -EINVAL;
1891         }
1892
1893         switch (tun->flags & TUN_TYPE_MASK) {
1894         case IFF_TUN:
1895                 if (tun->flags & IFF_NO_PI) {
1896                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1897
1898                         switch (ip_version) {
1899                         case 4:
1900                                 pi.proto = htons(ETH_P_IP);
1901                                 break;
1902                         case 6:
1903                                 pi.proto = htons(ETH_P_IPV6);
1904                                 break;
1905                         default:
1906                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1907                                 kfree_skb(skb);
1908                                 return -EINVAL;
1909                         }
1910                 }
1911
1912                 skb_reset_mac_header(skb);
1913                 skb->protocol = pi.proto;
1914                 skb->dev = tun->dev;
1915                 break;
1916         case IFF_TAP:
1917                 if (!frags)
1918                         skb->protocol = eth_type_trans(skb, tun->dev);
1919                 break;
1920         }
1921
1922         /* copy skb_ubuf_info for callback when skb has no error */
1923         if (zerocopy) {
1924                 skb_shinfo(skb)->destructor_arg = msg_control;
1925                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1926                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1927         } else if (msg_control) {
1928                 struct ubuf_info *uarg = msg_control;
1929                 uarg->callback(uarg, false);
1930         }
1931
1932         skb_reset_network_header(skb);
1933         skb_probe_transport_header(skb, 0);
1934
1935         if (skb_xdp) {
1936                 struct bpf_prog *xdp_prog;
1937                 int ret;
1938
1939                 local_bh_disable();
1940                 rcu_read_lock();
1941                 xdp_prog = rcu_dereference(tun->xdp_prog);
1942                 if (xdp_prog) {
1943                         ret = do_xdp_generic(xdp_prog, skb);
1944                         if (ret != XDP_PASS) {
1945                                 rcu_read_unlock();
1946                                 local_bh_enable();
1947                                 return total_len;
1948                         }
1949                 }
1950                 rcu_read_unlock();
1951                 local_bh_enable();
1952         }
1953
1954         /* Compute the costly rx hash only if needed for flow updates.
1955          * We may get a very small possibility of OOO during switching, not
1956          * worth to optimize.
1957          */
1958         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1959             !tfile->detached)
1960                 rxhash = __skb_get_hash_symmetric(skb);
1961
1962         if (frags) {
1963                 /* Exercise flow dissector code path. */
1964                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1965
1966                 if (unlikely(headlen > skb_headlen(skb))) {
1967                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1968                         napi_free_frags(&tfile->napi);
1969                         mutex_unlock(&tfile->napi_mutex);
1970                         WARN_ON(1);
1971                         return -ENOMEM;
1972                 }
1973
1974                 local_bh_disable();
1975                 napi_gro_frags(&tfile->napi);
1976                 local_bh_enable();
1977                 mutex_unlock(&tfile->napi_mutex);
1978         } else if (tfile->napi_enabled) {
1979                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1980                 int queue_len;
1981
1982                 spin_lock_bh(&queue->lock);
1983                 __skb_queue_tail(queue, skb);
1984                 queue_len = skb_queue_len(queue);
1985                 spin_unlock(&queue->lock);
1986
1987                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1988                         napi_schedule(&tfile->napi);
1989
1990                 local_bh_enable();
1991         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1992                 tun_rx_batched(tun, tfile, skb, more);
1993         } else {
1994                 netif_rx_ni(skb);
1995         }
1996
1997         stats = get_cpu_ptr(tun->pcpu_stats);
1998         u64_stats_update_begin(&stats->syncp);
1999         stats->rx_packets++;
2000         stats->rx_bytes += len;
2001         u64_stats_update_end(&stats->syncp);
2002         put_cpu_ptr(stats);
2003
2004         if (rxhash)
2005                 tun_flow_update(tun, rxhash, tfile);
2006
2007         return total_len;
2008 }
2009
2010 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2011 {
2012         struct file *file = iocb->ki_filp;
2013         struct tun_file *tfile = file->private_data;
2014         struct tun_struct *tun = tun_get(tfile);
2015         ssize_t result;
2016
2017         if (!tun)
2018                 return -EBADFD;
2019
2020         result = tun_get_user(tun, tfile, NULL, from,
2021                               file->f_flags & O_NONBLOCK, false);
2022
2023         tun_put(tun);
2024         return result;
2025 }
2026
2027 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2028                                 struct tun_file *tfile,
2029                                 struct xdp_frame *xdp_frame,
2030                                 struct iov_iter *iter)
2031 {
2032         int vnet_hdr_sz = 0;
2033         size_t size = xdp_frame->len;
2034         struct tun_pcpu_stats *stats;
2035         size_t ret;
2036
2037         if (tun->flags & IFF_VNET_HDR) {
2038                 struct virtio_net_hdr gso = { 0 };
2039
2040                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2041                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2042                         return -EINVAL;
2043                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2044                              sizeof(gso)))
2045                         return -EFAULT;
2046                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2047         }
2048
2049         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2050
2051         stats = get_cpu_ptr(tun->pcpu_stats);
2052         u64_stats_update_begin(&stats->syncp);
2053         stats->tx_packets++;
2054         stats->tx_bytes += ret;
2055         u64_stats_update_end(&stats->syncp);
2056         put_cpu_ptr(tun->pcpu_stats);
2057
2058         return ret;
2059 }
2060
2061 /* Put packet to the user space buffer */
2062 static ssize_t tun_put_user(struct tun_struct *tun,
2063                             struct tun_file *tfile,
2064                             struct sk_buff *skb,
2065                             struct iov_iter *iter)
2066 {
2067         struct tun_pi pi = { 0, skb->protocol };
2068         struct tun_pcpu_stats *stats;
2069         ssize_t total;
2070         int vlan_offset = 0;
2071         int vlan_hlen = 0;
2072         int vnet_hdr_sz = 0;
2073
2074         if (skb_vlan_tag_present(skb))
2075                 vlan_hlen = VLAN_HLEN;
2076
2077         if (tun->flags & IFF_VNET_HDR)
2078                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2079
2080         total = skb->len + vlan_hlen + vnet_hdr_sz;
2081
2082         if (!(tun->flags & IFF_NO_PI)) {
2083                 if (iov_iter_count(iter) < sizeof(pi))
2084                         return -EINVAL;
2085
2086                 total += sizeof(pi);
2087                 if (iov_iter_count(iter) < total) {
2088                         /* Packet will be striped */
2089                         pi.flags |= TUN_PKT_STRIP;
2090                 }
2091
2092                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2093                         return -EFAULT;
2094         }
2095
2096         if (vnet_hdr_sz) {
2097                 struct virtio_net_hdr gso;
2098
2099                 if (iov_iter_count(iter) < vnet_hdr_sz)
2100                         return -EINVAL;
2101
2102                 if (virtio_net_hdr_from_skb(skb, &gso,
2103                                             tun_is_little_endian(tun), true,
2104                                             vlan_hlen)) {
2105                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2106                         pr_err("unexpected GSO type: "
2107                                "0x%x, gso_size %d, hdr_len %d\n",
2108                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2109                                tun16_to_cpu(tun, gso.hdr_len));
2110                         print_hex_dump(KERN_ERR, "tun: ",
2111                                        DUMP_PREFIX_NONE,
2112                                        16, 1, skb->head,
2113                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2114                         WARN_ON_ONCE(1);
2115                         return -EINVAL;
2116                 }
2117
2118                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2119                         return -EFAULT;
2120
2121                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2122         }
2123
2124         if (vlan_hlen) {
2125                 int ret;
2126                 struct veth veth;
2127
2128                 veth.h_vlan_proto = skb->vlan_proto;
2129                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2130
2131                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2132
2133                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2134                 if (ret || !iov_iter_count(iter))
2135                         goto done;
2136
2137                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2138                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2139                         goto done;
2140         }
2141
2142         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2143
2144 done:
2145         /* caller is in process context, */
2146         stats = get_cpu_ptr(tun->pcpu_stats);
2147         u64_stats_update_begin(&stats->syncp);
2148         stats->tx_packets++;
2149         stats->tx_bytes += skb->len + vlan_hlen;
2150         u64_stats_update_end(&stats->syncp);
2151         put_cpu_ptr(tun->pcpu_stats);
2152
2153         return total;
2154 }
2155
2156 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2157 {
2158         DECLARE_WAITQUEUE(wait, current);
2159         void *ptr = NULL;
2160         int error = 0;
2161
2162         ptr = ptr_ring_consume(&tfile->tx_ring);
2163         if (ptr)
2164                 goto out;
2165         if (noblock) {
2166                 error = -EAGAIN;
2167                 goto out;
2168         }
2169
2170         add_wait_queue(&tfile->wq.wait, &wait);
2171         current->state = TASK_INTERRUPTIBLE;
2172
2173         while (1) {
2174                 ptr = ptr_ring_consume(&tfile->tx_ring);
2175                 if (ptr)
2176                         break;
2177                 if (signal_pending(current)) {
2178                         error = -ERESTARTSYS;
2179                         break;
2180                 }
2181                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2182                         error = -EFAULT;
2183                         break;
2184                 }
2185
2186                 schedule();
2187         }
2188
2189         current->state = TASK_RUNNING;
2190         remove_wait_queue(&tfile->wq.wait, &wait);
2191
2192 out:
2193         *err = error;
2194         return ptr;
2195 }
2196
2197 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2198                            struct iov_iter *to,
2199                            int noblock, void *ptr)
2200 {
2201         ssize_t ret;
2202         int err;
2203
2204         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2205
2206         if (!iov_iter_count(to)) {
2207                 tun_ptr_free(ptr);
2208                 return 0;
2209         }
2210
2211         if (!ptr) {
2212                 /* Read frames from ring */
2213                 ptr = tun_ring_recv(tfile, noblock, &err);
2214                 if (!ptr)
2215                         return err;
2216         }
2217
2218         if (tun_is_xdp_frame(ptr)) {
2219                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2220
2221                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2222                 xdp_return_frame(xdpf);
2223         } else {
2224                 struct sk_buff *skb = ptr;
2225
2226                 ret = tun_put_user(tun, tfile, skb, to);
2227                 if (unlikely(ret < 0))
2228                         kfree_skb(skb);
2229                 else
2230                         consume_skb(skb);
2231         }
2232
2233         return ret;
2234 }
2235
2236 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2237 {
2238         struct file *file = iocb->ki_filp;
2239         struct tun_file *tfile = file->private_data;
2240         struct tun_struct *tun = tun_get(tfile);
2241         ssize_t len = iov_iter_count(to), ret;
2242
2243         if (!tun)
2244                 return -EBADFD;
2245         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2246         ret = min_t(ssize_t, ret, len);
2247         if (ret > 0)
2248                 iocb->ki_pos = ret;
2249         tun_put(tun);
2250         return ret;
2251 }
2252
2253 static void tun_prog_free(struct rcu_head *rcu)
2254 {
2255         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2256
2257         bpf_prog_destroy(prog->prog);
2258         kfree(prog);
2259 }
2260
2261 static int __tun_set_ebpf(struct tun_struct *tun,
2262                           struct tun_prog __rcu **prog_p,
2263                           struct bpf_prog *prog)
2264 {
2265         struct tun_prog *old, *new = NULL;
2266
2267         if (prog) {
2268                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2269                 if (!new)
2270                         return -ENOMEM;
2271                 new->prog = prog;
2272         }
2273
2274         spin_lock_bh(&tun->lock);
2275         old = rcu_dereference_protected(*prog_p,
2276                                         lockdep_is_held(&tun->lock));
2277         rcu_assign_pointer(*prog_p, new);
2278         spin_unlock_bh(&tun->lock);
2279
2280         if (old)
2281                 call_rcu(&old->rcu, tun_prog_free);
2282
2283         return 0;
2284 }
2285
2286 static void tun_free_netdev(struct net_device *dev)
2287 {
2288         struct tun_struct *tun = netdev_priv(dev);
2289
2290         BUG_ON(!(list_empty(&tun->disabled)));
2291         free_percpu(tun->pcpu_stats);
2292         tun_flow_uninit(tun);
2293         security_tun_dev_free_security(tun->security);
2294         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2295         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2296 }
2297
2298 static void tun_setup(struct net_device *dev)
2299 {
2300         struct tun_struct *tun = netdev_priv(dev);
2301
2302         tun->owner = INVALID_UID;
2303         tun->group = INVALID_GID;
2304         tun_default_link_ksettings(dev, &tun->link_ksettings);
2305
2306         dev->ethtool_ops = &tun_ethtool_ops;
2307         dev->needs_free_netdev = true;
2308         dev->priv_destructor = tun_free_netdev;
2309         /* We prefer our own queue length */
2310         dev->tx_queue_len = TUN_READQ_SIZE;
2311 }
2312
2313 /* Trivial set of netlink ops to allow deleting tun or tap
2314  * device with netlink.
2315  */
2316 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2317                         struct netlink_ext_ack *extack)
2318 {
2319         NL_SET_ERR_MSG(extack,
2320                        "tun/tap creation via rtnetlink is not supported.");
2321         return -EOPNOTSUPP;
2322 }
2323
2324 static size_t tun_get_size(const struct net_device *dev)
2325 {
2326         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2327         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2328
2329         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2330                nla_total_size(sizeof(gid_t)) + /* GROUP */
2331                nla_total_size(sizeof(u8)) + /* TYPE */
2332                nla_total_size(sizeof(u8)) + /* PI */
2333                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2334                nla_total_size(sizeof(u8)) + /* PERSIST */
2335                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2336                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2337                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2338                0;
2339 }
2340
2341 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2342 {
2343         struct tun_struct *tun = netdev_priv(dev);
2344
2345         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2346                 goto nla_put_failure;
2347         if (uid_valid(tun->owner) &&
2348             nla_put_u32(skb, IFLA_TUN_OWNER,
2349                         from_kuid_munged(current_user_ns(), tun->owner)))
2350                 goto nla_put_failure;
2351         if (gid_valid(tun->group) &&
2352             nla_put_u32(skb, IFLA_TUN_GROUP,
2353                         from_kgid_munged(current_user_ns(), tun->group)))
2354                 goto nla_put_failure;
2355         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2356                 goto nla_put_failure;
2357         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2358                 goto nla_put_failure;
2359         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2360                 goto nla_put_failure;
2361         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2362                        !!(tun->flags & IFF_MULTI_QUEUE)))
2363                 goto nla_put_failure;
2364         if (tun->flags & IFF_MULTI_QUEUE) {
2365                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2366                         goto nla_put_failure;
2367                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2368                                 tun->numdisabled))
2369                         goto nla_put_failure;
2370         }
2371
2372         return 0;
2373
2374 nla_put_failure:
2375         return -EMSGSIZE;
2376 }
2377
2378 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2379         .kind           = DRV_NAME,
2380         .priv_size      = sizeof(struct tun_struct),
2381         .setup          = tun_setup,
2382         .validate       = tun_validate,
2383         .get_size       = tun_get_size,
2384         .fill_info      = tun_fill_info,
2385 };
2386
2387 static void tun_sock_write_space(struct sock *sk)
2388 {
2389         struct tun_file *tfile;
2390         wait_queue_head_t *wqueue;
2391
2392         if (!sock_writeable(sk))
2393                 return;
2394
2395         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2396                 return;
2397
2398         wqueue = sk_sleep(sk);
2399         if (wqueue && waitqueue_active(wqueue))
2400                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2401                                                 EPOLLWRNORM | EPOLLWRBAND);
2402
2403         tfile = container_of(sk, struct tun_file, sk);
2404         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2405 }
2406
2407 static void tun_put_page(struct tun_page *tpage)
2408 {
2409         if (tpage->page)
2410                 __page_frag_cache_drain(tpage->page, tpage->count);
2411 }
2412
2413 static int tun_xdp_one(struct tun_struct *tun,
2414                        struct tun_file *tfile,
2415                        struct xdp_buff *xdp, int *flush,
2416                        struct tun_page *tpage)
2417 {
2418         unsigned int datasize = xdp->data_end - xdp->data;
2419         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2420         struct virtio_net_hdr *gso = &hdr->gso;
2421         struct tun_pcpu_stats *stats;
2422         struct bpf_prog *xdp_prog;
2423         struct sk_buff *skb = NULL;
2424         u32 rxhash = 0, act;
2425         int buflen = hdr->buflen;
2426         int err = 0;
2427         bool skb_xdp = false;
2428         struct page *page;
2429
2430         xdp_prog = rcu_dereference(tun->xdp_prog);
2431         if (xdp_prog) {
2432                 if (gso->gso_type) {
2433                         skb_xdp = true;
2434                         goto build;
2435                 }
2436                 xdp_set_data_meta_invalid(xdp);
2437                 xdp->rxq = &tfile->xdp_rxq;
2438
2439                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2440                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2441                 if (err < 0) {
2442                         put_page(virt_to_head_page(xdp->data));
2443                         return err;
2444                 }
2445
2446                 switch (err) {
2447                 case XDP_REDIRECT:
2448                         *flush = true;
2449                         /* fall through */
2450                 case XDP_TX:
2451                         return 0;
2452                 case XDP_PASS:
2453                         break;
2454                 default:
2455                         page = virt_to_head_page(xdp->data);
2456                         if (tpage->page == page) {
2457                                 ++tpage->count;
2458                         } else {
2459                                 tun_put_page(tpage);
2460                                 tpage->page = page;
2461                                 tpage->count = 1;
2462                         }
2463                         return 0;
2464                 }
2465         }
2466
2467 build:
2468         skb = build_skb(xdp->data_hard_start, buflen);
2469         if (!skb) {
2470                 err = -ENOMEM;
2471                 goto out;
2472         }
2473
2474         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2475         skb_put(skb, xdp->data_end - xdp->data);
2476
2477         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2478                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2479                 kfree_skb(skb);
2480                 err = -EINVAL;
2481                 goto out;
2482         }
2483
2484         skb->protocol = eth_type_trans(skb, tun->dev);
2485         skb_reset_network_header(skb);
2486         skb_probe_transport_header(skb, 0);
2487
2488         if (skb_xdp) {
2489                 err = do_xdp_generic(xdp_prog, skb);
2490                 if (err != XDP_PASS)
2491                         goto out;
2492         }
2493
2494         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2495             !tfile->detached)
2496                 rxhash = __skb_get_hash_symmetric(skb);
2497
2498         skb_record_rx_queue(skb, tfile->queue_index);
2499         netif_receive_skb(skb);
2500
2501         /* No need for get_cpu_ptr() here since this function is
2502          * always called with bh disabled
2503          */
2504         stats = this_cpu_ptr(tun->pcpu_stats);
2505         u64_stats_update_begin(&stats->syncp);
2506         stats->rx_packets++;
2507         stats->rx_bytes += datasize;
2508         u64_stats_update_end(&stats->syncp);
2509
2510         if (rxhash)
2511                 tun_flow_update(tun, rxhash, tfile);
2512
2513 out:
2514         return err;
2515 }
2516
2517 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2518 {
2519         int ret, i;
2520         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2521         struct tun_struct *tun = tun_get(tfile);
2522         struct tun_msg_ctl *ctl = m->msg_control;
2523         struct xdp_buff *xdp;
2524
2525         if (!tun)
2526                 return -EBADFD;
2527
2528         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2529                 struct tun_page tpage;
2530                 int n = ctl->num;
2531                 int flush = 0;
2532
2533                 memset(&tpage, 0, sizeof(tpage));
2534
2535                 local_bh_disable();
2536                 rcu_read_lock();
2537
2538                 for (i = 0; i < n; i++) {
2539                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2540                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2541                 }
2542
2543                 if (flush)
2544                         xdp_do_flush_map();
2545
2546                 rcu_read_unlock();
2547                 local_bh_enable();
2548
2549                 tun_put_page(&tpage);
2550
2551                 ret = total_len;
2552                 goto out;
2553         }
2554
2555         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2556                            m->msg_flags & MSG_DONTWAIT,
2557                            m->msg_flags & MSG_MORE);
2558 out:
2559         tun_put(tun);
2560         return ret;
2561 }
2562
2563 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2564                        int flags)
2565 {
2566         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2567         struct tun_struct *tun = tun_get(tfile);
2568         void *ptr = m->msg_control;
2569         int ret;
2570
2571         if (!tun) {
2572                 ret = -EBADFD;
2573                 goto out_free;
2574         }
2575
2576         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2577                 ret = -EINVAL;
2578                 goto out_put_tun;
2579         }
2580         if (flags & MSG_ERRQUEUE) {
2581                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2582                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2583                 goto out;
2584         }
2585         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2586         if (ret > (ssize_t)total_len) {
2587                 m->msg_flags |= MSG_TRUNC;
2588                 ret = flags & MSG_TRUNC ? ret : total_len;
2589         }
2590 out:
2591         tun_put(tun);
2592         return ret;
2593
2594 out_put_tun:
2595         tun_put(tun);
2596 out_free:
2597         tun_ptr_free(ptr);
2598         return ret;
2599 }
2600
2601 static int tun_ptr_peek_len(void *ptr)
2602 {
2603         if (likely(ptr)) {
2604                 if (tun_is_xdp_frame(ptr)) {
2605                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2606
2607                         return xdpf->len;
2608                 }
2609                 return __skb_array_len_with_tag(ptr);
2610         } else {
2611                 return 0;
2612         }
2613 }
2614
2615 static int tun_peek_len(struct socket *sock)
2616 {
2617         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2618         struct tun_struct *tun;
2619         int ret = 0;
2620
2621         tun = tun_get(tfile);
2622         if (!tun)
2623                 return 0;
2624
2625         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2626         tun_put(tun);
2627
2628         return ret;
2629 }
2630
2631 /* Ops structure to mimic raw sockets with tun */
2632 static const struct proto_ops tun_socket_ops = {
2633         .peek_len = tun_peek_len,
2634         .sendmsg = tun_sendmsg,
2635         .recvmsg = tun_recvmsg,
2636 };
2637
2638 static struct proto tun_proto = {
2639         .name           = "tun",
2640         .owner          = THIS_MODULE,
2641         .obj_size       = sizeof(struct tun_file),
2642 };
2643
2644 static int tun_flags(struct tun_struct *tun)
2645 {
2646         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2647 }
2648
2649 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2650                               char *buf)
2651 {
2652         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2653         return sprintf(buf, "0x%x\n", tun_flags(tun));
2654 }
2655
2656 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2657                               char *buf)
2658 {
2659         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2660         return uid_valid(tun->owner)?
2661                 sprintf(buf, "%u\n",
2662                         from_kuid_munged(current_user_ns(), tun->owner)):
2663                 sprintf(buf, "-1\n");
2664 }
2665
2666 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2667                               char *buf)
2668 {
2669         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2670         return gid_valid(tun->group) ?
2671                 sprintf(buf, "%u\n",
2672                         from_kgid_munged(current_user_ns(), tun->group)):
2673                 sprintf(buf, "-1\n");
2674 }
2675
2676 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2677 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2678 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2679
2680 static struct attribute *tun_dev_attrs[] = {
2681         &dev_attr_tun_flags.attr,
2682         &dev_attr_owner.attr,
2683         &dev_attr_group.attr,
2684         NULL
2685 };
2686
2687 static const struct attribute_group tun_attr_group = {
2688         .attrs = tun_dev_attrs
2689 };
2690
2691 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2692 {
2693         struct tun_struct *tun;
2694         struct tun_file *tfile = file->private_data;
2695         struct net_device *dev;
2696         int err;
2697
2698         if (tfile->detached)
2699                 return -EINVAL;
2700
2701         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2702                 if (!capable(CAP_NET_ADMIN))
2703                         return -EPERM;
2704
2705                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2706                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2707                         return -EINVAL;
2708         }
2709
2710         dev = __dev_get_by_name(net, ifr->ifr_name);
2711         if (dev) {
2712                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2713                         return -EBUSY;
2714                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2715                         tun = netdev_priv(dev);
2716                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2717                         tun = netdev_priv(dev);
2718                 else
2719                         return -EINVAL;
2720
2721                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2722                     !!(tun->flags & IFF_MULTI_QUEUE))
2723                         return -EINVAL;
2724
2725                 if (tun_not_capable(tun))
2726                         return -EPERM;
2727                 err = security_tun_dev_open(tun->security);
2728                 if (err < 0)
2729                         return err;
2730
2731                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2732                                  ifr->ifr_flags & IFF_NAPI,
2733                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2734                 if (err < 0)
2735                         return err;
2736
2737                 if (tun->flags & IFF_MULTI_QUEUE &&
2738                     (tun->numqueues + tun->numdisabled > 1)) {
2739                         /* One or more queue has already been attached, no need
2740                          * to initialize the device again.
2741                          */
2742                         netdev_state_change(dev);
2743                         return 0;
2744                 }
2745
2746                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2747                               (ifr->ifr_flags & TUN_FEATURES);
2748
2749                 netdev_state_change(dev);
2750         } else {
2751                 char *name;
2752                 unsigned long flags = 0;
2753                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2754                              MAX_TAP_QUEUES : 1;
2755
2756                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2757                         return -EPERM;
2758                 err = security_tun_dev_create();
2759                 if (err < 0)
2760                         return err;
2761
2762                 /* Set dev type */
2763                 if (ifr->ifr_flags & IFF_TUN) {
2764                         /* TUN device */
2765                         flags |= IFF_TUN;
2766                         name = "tun%d";
2767                 } else if (ifr->ifr_flags & IFF_TAP) {
2768                         /* TAP device */
2769                         flags |= IFF_TAP;
2770                         name = "tap%d";
2771                 } else
2772                         return -EINVAL;
2773
2774                 if (*ifr->ifr_name)
2775                         name = ifr->ifr_name;
2776
2777                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2778                                        NET_NAME_UNKNOWN, tun_setup, queues,
2779                                        queues);
2780
2781                 if (!dev)
2782                         return -ENOMEM;
2783                 err = dev_get_valid_name(net, dev, name);
2784                 if (err < 0)
2785                         goto err_free_dev;
2786
2787                 dev_net_set(dev, net);
2788                 dev->rtnl_link_ops = &tun_link_ops;
2789                 dev->ifindex = tfile->ifindex;
2790                 dev->sysfs_groups[0] = &tun_attr_group;
2791
2792                 tun = netdev_priv(dev);
2793                 tun->dev = dev;
2794                 tun->flags = flags;
2795                 tun->txflt.count = 0;
2796                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2797
2798                 tun->align = NET_SKB_PAD;
2799                 tun->filter_attached = false;
2800                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2801                 tun->rx_batched = 0;
2802                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2803
2804                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2805                 if (!tun->pcpu_stats) {
2806                         err = -ENOMEM;
2807                         goto err_free_dev;
2808                 }
2809
2810                 spin_lock_init(&tun->lock);
2811
2812                 err = security_tun_dev_alloc_security(&tun->security);
2813                 if (err < 0)
2814                         goto err_free_stat;
2815
2816                 tun_net_init(dev);
2817                 tun_flow_init(tun);
2818
2819                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2820                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2821                                    NETIF_F_HW_VLAN_STAG_TX;
2822                 dev->features = dev->hw_features | NETIF_F_LLTX;
2823                 dev->vlan_features = dev->features &
2824                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2825                                        NETIF_F_HW_VLAN_STAG_TX);
2826
2827                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2828                               (ifr->ifr_flags & TUN_FEATURES);
2829
2830                 INIT_LIST_HEAD(&tun->disabled);
2831                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2832                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2833                 if (err < 0)
2834                         goto err_free_flow;
2835
2836                 err = register_netdevice(tun->dev);
2837                 if (err < 0)
2838                         goto err_detach;
2839         }
2840
2841         netif_carrier_on(tun->dev);
2842
2843         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2844
2845         /* Make sure persistent devices do not get stuck in
2846          * xoff state.
2847          */
2848         if (netif_running(tun->dev))
2849                 netif_tx_wake_all_queues(tun->dev);
2850
2851         strcpy(ifr->ifr_name, tun->dev->name);
2852         return 0;
2853
2854 err_detach:
2855         tun_detach_all(dev);
2856         /* register_netdevice() already called tun_free_netdev() */
2857         goto err_free_dev;
2858
2859 err_free_flow:
2860         tun_flow_uninit(tun);
2861         security_tun_dev_free_security(tun->security);
2862 err_free_stat:
2863         free_percpu(tun->pcpu_stats);
2864 err_free_dev:
2865         free_netdev(dev);
2866         return err;
2867 }
2868
2869 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2870                        struct ifreq *ifr)
2871 {
2872         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2873
2874         strcpy(ifr->ifr_name, tun->dev->name);
2875
2876         ifr->ifr_flags = tun_flags(tun);
2877
2878 }
2879
2880 /* This is like a cut-down ethtool ops, except done via tun fd so no
2881  * privs required. */
2882 static int set_offload(struct tun_struct *tun, unsigned long arg)
2883 {
2884         netdev_features_t features = 0;
2885
2886         if (arg & TUN_F_CSUM) {
2887                 features |= NETIF_F_HW_CSUM;
2888                 arg &= ~TUN_F_CSUM;
2889
2890                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2891                         if (arg & TUN_F_TSO_ECN) {
2892                                 features |= NETIF_F_TSO_ECN;
2893                                 arg &= ~TUN_F_TSO_ECN;
2894                         }
2895                         if (arg & TUN_F_TSO4)
2896                                 features |= NETIF_F_TSO;
2897                         if (arg & TUN_F_TSO6)
2898                                 features |= NETIF_F_TSO6;
2899                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2900                 }
2901
2902                 arg &= ~TUN_F_UFO;
2903         }
2904
2905         /* This gives the user a way to test for new features in future by
2906          * trying to set them. */
2907         if (arg)
2908                 return -EINVAL;
2909
2910         tun->set_features = features;
2911         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2912         tun->dev->wanted_features |= features;
2913         netdev_update_features(tun->dev);
2914
2915         return 0;
2916 }
2917
2918 static void tun_detach_filter(struct tun_struct *tun, int n)
2919 {
2920         int i;
2921         struct tun_file *tfile;
2922
2923         for (i = 0; i < n; i++) {
2924                 tfile = rtnl_dereference(tun->tfiles[i]);
2925                 lock_sock(tfile->socket.sk);
2926                 sk_detach_filter(tfile->socket.sk);
2927                 release_sock(tfile->socket.sk);
2928         }
2929
2930         tun->filter_attached = false;
2931 }
2932
2933 static int tun_attach_filter(struct tun_struct *tun)
2934 {
2935         int i, ret = 0;
2936         struct tun_file *tfile;
2937
2938         for (i = 0; i < tun->numqueues; i++) {
2939                 tfile = rtnl_dereference(tun->tfiles[i]);
2940                 lock_sock(tfile->socket.sk);
2941                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2942                 release_sock(tfile->socket.sk);
2943                 if (ret) {
2944                         tun_detach_filter(tun, i);
2945                         return ret;
2946                 }
2947         }
2948
2949         tun->filter_attached = true;
2950         return ret;
2951 }
2952
2953 static void tun_set_sndbuf(struct tun_struct *tun)
2954 {
2955         struct tun_file *tfile;
2956         int i;
2957
2958         for (i = 0; i < tun->numqueues; i++) {
2959                 tfile = rtnl_dereference(tun->tfiles[i]);
2960                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2961         }
2962 }
2963
2964 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2965 {
2966         struct tun_file *tfile = file->private_data;
2967         struct tun_struct *tun;
2968         int ret = 0;
2969
2970         rtnl_lock();
2971
2972         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2973                 tun = tfile->detached;
2974                 if (!tun) {
2975                         ret = -EINVAL;
2976                         goto unlock;
2977                 }
2978                 ret = security_tun_dev_attach_queue(tun->security);
2979                 if (ret < 0)
2980                         goto unlock;
2981                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2982                                  tun->flags & IFF_NAPI_FRAGS);
2983         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2984                 tun = rtnl_dereference(tfile->tun);
2985                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2986                         ret = -EINVAL;
2987                 else
2988                         __tun_detach(tfile, false);
2989         } else
2990                 ret = -EINVAL;
2991
2992         if (ret >= 0)
2993                 netdev_state_change(tun->dev);
2994
2995 unlock:
2996         rtnl_unlock();
2997         return ret;
2998 }
2999
3000 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3001                         void __user *data)
3002 {
3003         struct bpf_prog *prog;
3004         int fd;
3005
3006         if (copy_from_user(&fd, data, sizeof(fd)))
3007                 return -EFAULT;
3008
3009         if (fd == -1) {
3010                 prog = NULL;
3011         } else {
3012                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3013                 if (IS_ERR(prog))
3014                         return PTR_ERR(prog);
3015         }
3016
3017         return __tun_set_ebpf(tun, prog_p, prog);
3018 }
3019
3020 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3021                             unsigned long arg, int ifreq_len)
3022 {
3023         struct tun_file *tfile = file->private_data;
3024         struct net *net = sock_net(&tfile->sk);
3025         struct tun_struct *tun;
3026         void __user* argp = (void __user*)arg;
3027         unsigned int ifindex, carrier;
3028         struct ifreq ifr;
3029         kuid_t owner;
3030         kgid_t group;
3031         int sndbuf;
3032         int vnet_hdr_sz;
3033         int le;
3034         int ret;
3035         bool do_notify = false;
3036
3037         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3038             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3039                 if (copy_from_user(&ifr, argp, ifreq_len))
3040                         return -EFAULT;
3041         } else {
3042                 memset(&ifr, 0, sizeof(ifr));
3043         }
3044         if (cmd == TUNGETFEATURES) {
3045                 /* Currently this just means: "what IFF flags are valid?".
3046                  * This is needed because we never checked for invalid flags on
3047                  * TUNSETIFF.
3048                  */
3049                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3050                                 (unsigned int __user*)argp);
3051         } else if (cmd == TUNSETQUEUE) {
3052                 return tun_set_queue(file, &ifr);
3053         } else if (cmd == SIOCGSKNS) {
3054                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3055                         return -EPERM;
3056                 return open_related_ns(&net->ns, get_net_ns);
3057         }
3058
3059         ret = 0;
3060         rtnl_lock();
3061
3062         tun = tun_get(tfile);
3063         if (cmd == TUNSETIFF) {
3064                 ret = -EEXIST;
3065                 if (tun)
3066                         goto unlock;
3067
3068                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3069
3070                 ret = tun_set_iff(net, file, &ifr);
3071
3072                 if (ret)
3073                         goto unlock;
3074
3075                 if (copy_to_user(argp, &ifr, ifreq_len))
3076                         ret = -EFAULT;
3077                 goto unlock;
3078         }
3079         if (cmd == TUNSETIFINDEX) {
3080                 ret = -EPERM;
3081                 if (tun)
3082                         goto unlock;
3083
3084                 ret = -EFAULT;
3085                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3086                         goto unlock;
3087
3088                 ret = 0;
3089                 tfile->ifindex = ifindex;
3090                 goto unlock;
3091         }
3092
3093         ret = -EBADFD;
3094         if (!tun)
3095                 goto unlock;
3096
3097         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3098
3099         ret = 0;
3100         switch (cmd) {
3101         case TUNGETIFF:
3102                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3103
3104                 if (tfile->detached)
3105                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3106                 if (!tfile->socket.sk->sk_filter)
3107                         ifr.ifr_flags |= IFF_NOFILTER;
3108
3109                 if (copy_to_user(argp, &ifr, ifreq_len))
3110                         ret = -EFAULT;
3111                 break;
3112
3113         case TUNSETNOCSUM:
3114                 /* Disable/Enable checksum */
3115
3116                 /* [unimplemented] */
3117                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3118                           arg ? "disabled" : "enabled");
3119                 break;
3120
3121         case TUNSETPERSIST:
3122                 /* Disable/Enable persist mode. Keep an extra reference to the
3123                  * module to prevent the module being unprobed.
3124                  */
3125                 if (arg && !(tun->flags & IFF_PERSIST)) {
3126                         tun->flags |= IFF_PERSIST;
3127                         __module_get(THIS_MODULE);
3128                         do_notify = true;
3129                 }
3130                 if (!arg && (tun->flags & IFF_PERSIST)) {
3131                         tun->flags &= ~IFF_PERSIST;
3132                         module_put(THIS_MODULE);
3133                         do_notify = true;
3134                 }
3135
3136                 tun_debug(KERN_INFO, tun, "persist %s\n",
3137                           arg ? "enabled" : "disabled");
3138                 break;
3139
3140         case TUNSETOWNER:
3141                 /* Set owner of the device */
3142                 owner = make_kuid(current_user_ns(), arg);
3143                 if (!uid_valid(owner)) {
3144                         ret = -EINVAL;
3145                         break;
3146                 }
3147                 tun->owner = owner;
3148                 do_notify = true;
3149                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3150                           from_kuid(&init_user_ns, tun->owner));
3151                 break;
3152
3153         case TUNSETGROUP:
3154                 /* Set group of the device */
3155                 group = make_kgid(current_user_ns(), arg);
3156                 if (!gid_valid(group)) {
3157                         ret = -EINVAL;
3158                         break;
3159                 }
3160                 tun->group = group;
3161                 do_notify = true;
3162                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3163                           from_kgid(&init_user_ns, tun->group));
3164                 break;
3165
3166         case TUNSETLINK:
3167                 /* Only allow setting the type when the interface is down */
3168                 if (tun->dev->flags & IFF_UP) {
3169                         tun_debug(KERN_INFO, tun,
3170                                   "Linktype set failed because interface is up\n");
3171                         ret = -EBUSY;
3172                 } else {
3173                         tun->dev->type = (int) arg;
3174                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3175                                   tun->dev->type);
3176                         ret = 0;
3177                 }
3178                 break;
3179
3180 #ifdef TUN_DEBUG
3181         case TUNSETDEBUG:
3182                 tun->debug = arg;
3183                 break;
3184 #endif
3185         case TUNSETOFFLOAD:
3186                 ret = set_offload(tun, arg);
3187                 break;
3188
3189         case TUNSETTXFILTER:
3190                 /* Can be set only for TAPs */
3191                 ret = -EINVAL;
3192                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3193                         break;
3194                 ret = update_filter(&tun->txflt, (void __user *)arg);
3195                 break;
3196
3197         case SIOCGIFHWADDR:
3198                 /* Get hw address */
3199                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3200                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3201                 if (copy_to_user(argp, &ifr, ifreq_len))
3202                         ret = -EFAULT;
3203                 break;
3204
3205         case SIOCSIFHWADDR:
3206                 /* Set hw address */
3207                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3208                           ifr.ifr_hwaddr.sa_data);
3209
3210                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3211                 break;
3212
3213         case TUNGETSNDBUF:
3214                 sndbuf = tfile->socket.sk->sk_sndbuf;
3215                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3216                         ret = -EFAULT;
3217                 break;
3218
3219         case TUNSETSNDBUF:
3220                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3221                         ret = -EFAULT;
3222                         break;
3223                 }
3224                 if (sndbuf <= 0) {
3225                         ret = -EINVAL;
3226                         break;
3227                 }
3228
3229                 tun->sndbuf = sndbuf;
3230                 tun_set_sndbuf(tun);
3231                 break;
3232
3233         case TUNGETVNETHDRSZ:
3234                 vnet_hdr_sz = tun->vnet_hdr_sz;
3235                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3236                         ret = -EFAULT;
3237                 break;
3238
3239         case TUNSETVNETHDRSZ:
3240                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3241                         ret = -EFAULT;
3242                         break;
3243                 }
3244                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3245                         ret = -EINVAL;
3246                         break;
3247                 }
3248
3249                 tun->vnet_hdr_sz = vnet_hdr_sz;
3250                 break;
3251
3252         case TUNGETVNETLE:
3253                 le = !!(tun->flags & TUN_VNET_LE);
3254                 if (put_user(le, (int __user *)argp))
3255                         ret = -EFAULT;
3256                 break;
3257
3258         case TUNSETVNETLE:
3259                 if (get_user(le, (int __user *)argp)) {
3260                         ret = -EFAULT;
3261                         break;
3262                 }
3263                 if (le)
3264                         tun->flags |= TUN_VNET_LE;
3265                 else
3266                         tun->flags &= ~TUN_VNET_LE;
3267                 break;
3268
3269         case TUNGETVNETBE:
3270                 ret = tun_get_vnet_be(tun, argp);
3271                 break;
3272
3273         case TUNSETVNETBE:
3274                 ret = tun_set_vnet_be(tun, argp);
3275                 break;
3276
3277         case TUNATTACHFILTER:
3278                 /* Can be set only for TAPs */
3279                 ret = -EINVAL;
3280                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3281                         break;
3282                 ret = -EFAULT;
3283                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3284                         break;
3285
3286                 ret = tun_attach_filter(tun);
3287                 break;
3288
3289         case TUNDETACHFILTER:
3290                 /* Can be set only for TAPs */
3291                 ret = -EINVAL;
3292                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3293                         break;
3294                 ret = 0;
3295                 tun_detach_filter(tun, tun->numqueues);
3296                 break;
3297
3298         case TUNGETFILTER:
3299                 ret = -EINVAL;
3300                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3301                         break;
3302                 ret = -EFAULT;
3303                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3304                         break;
3305                 ret = 0;
3306                 break;
3307
3308         case TUNSETSTEERINGEBPF:
3309                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3310                 break;
3311
3312         case TUNSETFILTEREBPF:
3313                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3314                 break;
3315
3316         case TUNSETCARRIER:
3317                 ret = -EFAULT;
3318                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3319                         goto unlock;
3320
3321                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3322                 break;
3323
3324         default:
3325                 ret = -EINVAL;
3326                 break;
3327         }
3328
3329         if (do_notify)
3330                 netdev_state_change(tun->dev);
3331
3332 unlock:
3333         rtnl_unlock();
3334         if (tun)
3335                 tun_put(tun);
3336         return ret;
3337 }
3338
3339 static long tun_chr_ioctl(struct file *file,
3340                           unsigned int cmd, unsigned long arg)
3341 {
3342         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3343 }
3344
3345 #ifdef CONFIG_COMPAT
3346 static long tun_chr_compat_ioctl(struct file *file,
3347                          unsigned int cmd, unsigned long arg)
3348 {
3349         switch (cmd) {
3350         case TUNSETIFF:
3351         case TUNGETIFF:
3352         case TUNSETTXFILTER:
3353         case TUNGETSNDBUF:
3354         case TUNSETSNDBUF:
3355         case SIOCGIFHWADDR:
3356         case SIOCSIFHWADDR:
3357                 arg = (unsigned long)compat_ptr(arg);
3358                 break;
3359         default:
3360                 arg = (compat_ulong_t)arg;
3361                 break;
3362         }
3363
3364         /*
3365          * compat_ifreq is shorter than ifreq, so we must not access beyond
3366          * the end of that structure. All fields that are used in this
3367          * driver are compatible though, we don't need to convert the
3368          * contents.
3369          */
3370         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3371 }
3372 #endif /* CONFIG_COMPAT */
3373
3374 static int tun_chr_fasync(int fd, struct file *file, int on)
3375 {
3376         struct tun_file *tfile = file->private_data;
3377         int ret;
3378
3379         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3380                 goto out;
3381
3382         if (on) {
3383                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3384                 tfile->flags |= TUN_FASYNC;
3385         } else
3386                 tfile->flags &= ~TUN_FASYNC;
3387         ret = 0;
3388 out:
3389         return ret;
3390 }
3391
3392 static int tun_chr_open(struct inode *inode, struct file * file)
3393 {
3394         struct net *net = current->nsproxy->net_ns;
3395         struct tun_file *tfile;
3396
3397         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3398
3399         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3400                                             &tun_proto, 0);
3401         if (!tfile)
3402                 return -ENOMEM;
3403         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3404                 sk_free(&tfile->sk);
3405                 return -ENOMEM;
3406         }
3407
3408         mutex_init(&tfile->napi_mutex);
3409         RCU_INIT_POINTER(tfile->tun, NULL);
3410         tfile->flags = 0;
3411         tfile->ifindex = 0;
3412
3413         init_waitqueue_head(&tfile->wq.wait);
3414         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3415
3416         tfile->socket.file = file;
3417         tfile->socket.ops = &tun_socket_ops;
3418
3419         sock_init_data(&tfile->socket, &tfile->sk);
3420
3421         tfile->sk.sk_write_space = tun_sock_write_space;
3422         tfile->sk.sk_sndbuf = INT_MAX;
3423
3424         file->private_data = tfile;
3425         INIT_LIST_HEAD(&tfile->next);
3426
3427         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3428
3429         return 0;
3430 }
3431
3432 static int tun_chr_close(struct inode *inode, struct file *file)
3433 {
3434         struct tun_file *tfile = file->private_data;
3435
3436         tun_detach(tfile, true);
3437
3438         return 0;
3439 }
3440
3441 #ifdef CONFIG_PROC_FS
3442 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3443 {
3444         struct tun_file *tfile = file->private_data;
3445         struct tun_struct *tun;
3446         struct ifreq ifr;
3447
3448         memset(&ifr, 0, sizeof(ifr));
3449
3450         rtnl_lock();
3451         tun = tun_get(tfile);
3452         if (tun)
3453                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3454         rtnl_unlock();
3455
3456         if (tun)
3457                 tun_put(tun);
3458
3459         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3460 }
3461 #endif
3462
3463 static const struct file_operations tun_fops = {
3464         .owner  = THIS_MODULE,
3465         .llseek = no_llseek,
3466         .read_iter  = tun_chr_read_iter,
3467         .write_iter = tun_chr_write_iter,
3468         .poll   = tun_chr_poll,
3469         .unlocked_ioctl = tun_chr_ioctl,
3470 #ifdef CONFIG_COMPAT
3471         .compat_ioctl = tun_chr_compat_ioctl,
3472 #endif
3473         .open   = tun_chr_open,
3474         .release = tun_chr_close,
3475         .fasync = tun_chr_fasync,
3476 #ifdef CONFIG_PROC_FS
3477         .show_fdinfo = tun_chr_show_fdinfo,
3478 #endif
3479 };
3480
3481 static struct miscdevice tun_miscdev = {
3482         .minor = TUN_MINOR,
3483         .name = "tun",
3484         .nodename = "net/tun",
3485         .fops = &tun_fops,
3486 };
3487
3488 /* ethtool interface */
3489
3490 static void tun_default_link_ksettings(struct net_device *dev,
3491                                        struct ethtool_link_ksettings *cmd)
3492 {
3493         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3494         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3495         cmd->base.speed         = SPEED_10;
3496         cmd->base.duplex        = DUPLEX_FULL;
3497         cmd->base.port          = PORT_TP;
3498         cmd->base.phy_address   = 0;
3499         cmd->base.autoneg       = AUTONEG_DISABLE;
3500 }
3501
3502 static int tun_get_link_ksettings(struct net_device *dev,
3503                                   struct ethtool_link_ksettings *cmd)
3504 {
3505         struct tun_struct *tun = netdev_priv(dev);
3506
3507         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3508         return 0;
3509 }
3510
3511 static int tun_set_link_ksettings(struct net_device *dev,
3512                                   const struct ethtool_link_ksettings *cmd)
3513 {
3514         struct tun_struct *tun = netdev_priv(dev);
3515
3516         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3517         return 0;
3518 }
3519
3520 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3521 {
3522         struct tun_struct *tun = netdev_priv(dev);
3523
3524         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3525         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3526
3527         switch (tun->flags & TUN_TYPE_MASK) {
3528         case IFF_TUN:
3529                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3530                 break;
3531         case IFF_TAP:
3532                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3533                 break;
3534         }
3535 }
3536
3537 static u32 tun_get_msglevel(struct net_device *dev)
3538 {
3539 #ifdef TUN_DEBUG
3540         struct tun_struct *tun = netdev_priv(dev);
3541         return tun->debug;
3542 #else
3543         return -EOPNOTSUPP;
3544 #endif
3545 }
3546
3547 static void tun_set_msglevel(struct net_device *dev, u32 value)
3548 {
3549 #ifdef TUN_DEBUG
3550         struct tun_struct *tun = netdev_priv(dev);
3551         tun->debug = value;
3552 #endif
3553 }
3554
3555 static int tun_get_coalesce(struct net_device *dev,
3556                             struct ethtool_coalesce *ec)
3557 {
3558         struct tun_struct *tun = netdev_priv(dev);
3559
3560         ec->rx_max_coalesced_frames = tun->rx_batched;
3561
3562         return 0;
3563 }
3564
3565 static int tun_set_coalesce(struct net_device *dev,
3566                             struct ethtool_coalesce *ec)
3567 {
3568         struct tun_struct *tun = netdev_priv(dev);
3569
3570         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3571                 tun->rx_batched = NAPI_POLL_WEIGHT;
3572         else
3573                 tun->rx_batched = ec->rx_max_coalesced_frames;
3574
3575         return 0;
3576 }
3577
3578 static const struct ethtool_ops tun_ethtool_ops = {
3579         .get_drvinfo    = tun_get_drvinfo,
3580         .get_msglevel   = tun_get_msglevel,
3581         .set_msglevel   = tun_set_msglevel,
3582         .get_link       = ethtool_op_get_link,
3583         .get_ts_info    = ethtool_op_get_ts_info,
3584         .get_coalesce   = tun_get_coalesce,
3585         .set_coalesce   = tun_set_coalesce,
3586         .get_link_ksettings = tun_get_link_ksettings,
3587         .set_link_ksettings = tun_set_link_ksettings,
3588 };
3589
3590 static int tun_queue_resize(struct tun_struct *tun)
3591 {
3592         struct net_device *dev = tun->dev;
3593         struct tun_file *tfile;
3594         struct ptr_ring **rings;
3595         int n = tun->numqueues + tun->numdisabled;
3596         int ret, i;
3597
3598         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3599         if (!rings)
3600                 return -ENOMEM;
3601
3602         for (i = 0; i < tun->numqueues; i++) {
3603                 tfile = rtnl_dereference(tun->tfiles[i]);
3604                 rings[i] = &tfile->tx_ring;
3605         }
3606         list_for_each_entry(tfile, &tun->disabled, next)
3607                 rings[i++] = &tfile->tx_ring;
3608
3609         ret = ptr_ring_resize_multiple(rings, n,
3610                                        dev->tx_queue_len, GFP_KERNEL,
3611                                        tun_ptr_free);
3612
3613         kfree(rings);
3614         return ret;
3615 }
3616
3617 static int tun_device_event(struct notifier_block *unused,
3618                             unsigned long event, void *ptr)
3619 {
3620         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3621         struct tun_struct *tun = netdev_priv(dev);
3622
3623         if (dev->rtnl_link_ops != &tun_link_ops)
3624                 return NOTIFY_DONE;
3625
3626         switch (event) {
3627         case NETDEV_CHANGE_TX_QUEUE_LEN:
3628                 if (tun_queue_resize(tun))
3629                         return NOTIFY_BAD;
3630                 break;
3631         default:
3632                 break;
3633         }
3634
3635         return NOTIFY_DONE;
3636 }
3637
3638 static struct notifier_block tun_notifier_block __read_mostly = {
3639         .notifier_call  = tun_device_event,
3640 };
3641
3642 static int __init tun_init(void)
3643 {
3644         int ret = 0;
3645
3646         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3647
3648         ret = rtnl_link_register(&tun_link_ops);
3649         if (ret) {
3650                 pr_err("Can't register link_ops\n");
3651                 goto err_linkops;
3652         }
3653
3654         ret = misc_register(&tun_miscdev);
3655         if (ret) {
3656                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3657                 goto err_misc;
3658         }
3659
3660         ret = register_netdevice_notifier(&tun_notifier_block);
3661         if (ret) {
3662                 pr_err("Can't register netdevice notifier\n");
3663                 goto err_notifier;
3664         }
3665
3666         return  0;
3667
3668 err_notifier:
3669         misc_deregister(&tun_miscdev);
3670 err_misc:
3671         rtnl_link_unregister(&tun_link_ops);
3672 err_linkops:
3673         return ret;
3674 }
3675
3676 static void tun_cleanup(void)
3677 {
3678         misc_deregister(&tun_miscdev);
3679         rtnl_link_unregister(&tun_link_ops);
3680         unregister_netdevice_notifier(&tun_notifier_block);
3681 }
3682
3683 /* Get an underlying socket object from tun file.  Returns error unless file is
3684  * attached to a device.  The returned object works like a packet socket, it
3685  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3686  * holding a reference to the file for as long as the socket is in use. */
3687 struct socket *tun_get_socket(struct file *file)
3688 {
3689         struct tun_file *tfile;
3690         if (file->f_op != &tun_fops)
3691                 return ERR_PTR(-EINVAL);
3692         tfile = file->private_data;
3693         if (!tfile)
3694                 return ERR_PTR(-EBADFD);
3695         return &tfile->socket;
3696 }
3697 EXPORT_SYMBOL_GPL(tun_get_socket);
3698
3699 struct ptr_ring *tun_get_tx_ring(struct file *file)
3700 {
3701         struct tun_file *tfile;
3702
3703         if (file->f_op != &tun_fops)
3704                 return ERR_PTR(-EINVAL);
3705         tfile = file->private_data;
3706         if (!tfile)
3707                 return ERR_PTR(-EBADFD);
3708         return &tfile->tx_ring;
3709 }
3710 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3711
3712 module_init(tun_init);
3713 module_exit(tun_cleanup);
3714 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3715 MODULE_AUTHOR(DRV_COPYRIGHT);
3716 MODULE_LICENSE("GPL");
3717 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3718 MODULE_ALIAS("devname:net/tun");