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