drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <underley@underley.eu.org>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72
  73 #include <linux/uaccess.h>
  74 #include <linux/proc_fs.h>
  75
  76 static void tun_default_link_ksettings(struct net_device *dev,
  77                                        struct ethtool_link_ksettings *cmd);
  78
  79 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  80
  81 /* TUN device flags */
  82
  83 /* IFF_ATTACH_QUEUE is never stored in device flags,
  84  * overload it to mean fasync when stored there.
  85  */
  86 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  87 /* High bits in flags field are unused. */
  88 #define TUN_VNET_LE     0x80000000
  89 #define TUN_VNET_BE     0x40000000
  90
  91 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
  92                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
  93
  94 #define GOODCOPY_LEN 128
  95
  96 #define FLT_EXACT_COUNT 8
  97 struct tap_filter {
  98         unsigned int    count;    /* Number of addrs. Zero means disabled */
  99         u32             mask[2];  /* Mask of the hashed addrs */
 100         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 101 };
 102
 103 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 104  * to max number of VCPUs in guest. */
 105 #define MAX_TAP_QUEUES 256
 106 #define MAX_TAP_FLOWS  4096
 107
 108 #define TUN_FLOW_EXPIRE (3 * HZ)
 109
 110 struct tun_pcpu_stats {
 111         u64_stats_t rx_packets;
 112         u64_stats_t rx_bytes;
 113         u64_stats_t tx_packets;
 114         u64_stats_t tx_bytes;
 115         struct u64_stats_sync syncp;
 116         u32 rx_dropped;
 117         u32 tx_dropped;
 118         u32 rx_frame_errors;
 119 };
 120
 121 /* A tun_file connects an open character device to a tuntap netdevice. It
 122  * also contains all socket related structures (except sock_fprog and tap_filter)
 123  * to serve as one transmit queue for tuntap device. The sock_fprog and
 124  * tap_filter were kept in tun_struct since they were used for filtering for the
 125  * netdevice not for a specific queue (at least I didn't see the requirement for
 126  * this).
 127  *
 128  * RCU usage:
 129  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 130  * other can only be read while rcu_read_lock or rtnl_lock is held.
 131  */
 132 struct tun_file {
 133         struct sock sk;
 134         struct socket socket;
 135         struct tun_struct __rcu *tun;
 136         struct fasync_struct *fasync;
 137         /* only used for fasnyc */
 138         unsigned int flags;
 139         union {
 140                 u16 queue_index;
 141                 unsigned int ifindex;
 142         };
 143         struct napi_struct napi;
 144         bool napi_enabled;
 145         bool napi_frags_enabled;
 146         struct mutex napi_mutex;        /* Protects access to the above napi */
 147         struct list_head next;
 148         struct tun_struct *detached;
 149         struct ptr_ring tx_ring;
 150         struct xdp_rxq_info xdp_rxq;
 151 };
 152
 153 struct tun_page {
 154         struct page *page;
 155         int count;
 156 };
 157
 158 struct tun_flow_entry {
 159         struct hlist_node hash_link;
 160         struct rcu_head rcu;
 161         struct tun_struct *tun;
 162
 163         u32 rxhash;
 164         u32 rps_rxhash;
 165         int queue_index;
 166         unsigned long updated ____cacheline_aligned_in_smp;
 167 };
 168
 169 #define TUN_NUM_FLOW_ENTRIES 1024
 170 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 171
 172 struct tun_prog {
 173         struct rcu_head rcu;
 174         struct bpf_prog *prog;
 175 };
 176
 177 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 178  * device, socket filter, sndbuf and vnet header size were restore when the
 179  * file were attached to a persist device.
 180  */
 181 struct tun_struct {
 182         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 183         unsigned int            numqueues;
 184         unsigned int            flags;
 185         kuid_t                  owner;
 186         kgid_t                  group;
 187
 188         struct net_device       *dev;
 189         netdev_features_t       set_features;
 190 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 191                           NETIF_F_TSO6)
 192
 193         int                     align;
 194         int                     vnet_hdr_sz;
 195         int                     sndbuf;
 196         struct tap_filter       txflt;
 197         struct sock_fprog       fprog;
 198         /* protected by rtnl lock */
 199         bool                    filter_attached;
 200         u32                     msg_enable;
 201         spinlock_t lock;
 202         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 203         struct timer_list flow_gc_timer;
 204         unsigned long ageing_time;
 205         unsigned int numdisabled;
 206         struct list_head disabled;
 207         void *security;
 208         u32 flow_count;
 209         u32 rx_batched;
 210         struct tun_pcpu_stats __percpu *pcpu_stats;
 211         struct bpf_prog __rcu *xdp_prog;
 212         struct tun_prog __rcu *steering_prog;
 213         struct tun_prog __rcu *filter_prog;
 214         struct ethtool_link_ksettings link_ksettings;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 bool tun_is_xdp_frame(void *ptr)
 223 {
 224         return (unsigned long)ptr & TUN_XDP_FLAG;
 225 }
 226 EXPORT_SYMBOL(tun_is_xdp_frame);
 227
 228 void *tun_xdp_to_ptr(void *ptr)
 229 {
 230         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 231 }
 232 EXPORT_SYMBOL(tun_xdp_to_ptr);
 233
 234 void *tun_ptr_to_xdp(void *ptr)
 235 {
 236         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 237 }
 238 EXPORT_SYMBOL(tun_ptr_to_xdp);
 239
 240 static int tun_napi_receive(struct napi_struct *napi, int budget)
 241 {
 242         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 243         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 244         struct sk_buff_head process_queue;
 245         struct sk_buff *skb;
 246         int received = 0;
 247
 248         __skb_queue_head_init(&process_queue);
 249
 250         spin_lock(&queue->lock);
 251         skb_queue_splice_tail_init(queue, &process_queue);
 252         spin_unlock(&queue->lock);
 253
 254         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 255                 napi_gro_receive(napi, skb);
 256                 ++received;
 257         }
 258
 259         if (!skb_queue_empty(&process_queue)) {
 260                 spin_lock(&queue->lock);
 261                 skb_queue_splice(&process_queue, queue);
 262                 spin_unlock(&queue->lock);
 263         }
 264
 265         return received;
 266 }
 267
 268 static int tun_napi_poll(struct napi_struct *napi, int budget)
 269 {
 270         unsigned int received;
 271
 272         received = tun_napi_receive(napi, budget);
 273
 274         if (received < budget)
 275                 napi_complete_done(napi, received);
 276
 277         return received;
 278 }
 279
 280 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 281                           bool napi_en, bool napi_frags)
 282 {
 283         tfile->napi_enabled = napi_en;
 284         tfile->napi_frags_enabled = napi_en && napi_frags;
 285         if (napi_en) {
 286                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 287                                   NAPI_POLL_WEIGHT);
 288                 napi_enable(&tfile->napi);
 289         }
 290 }
 291
 292 static void tun_napi_disable(struct tun_file *tfile)
 293 {
 294         if (tfile->napi_enabled)
 295                 napi_disable(&tfile->napi);
 296 }
 297
 298 static void tun_napi_del(struct tun_file *tfile)
 299 {
 300         if (tfile->napi_enabled)
 301                 netif_napi_del(&tfile->napi);
 302 }
 303
 304 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 305 {
 306         return tfile->napi_frags_enabled;
 307 }
 308
 309 #ifdef CONFIG_TUN_VNET_CROSS_LE
 310 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 311 {
 312         return tun->flags & TUN_VNET_BE ? false :
 313                 virtio_legacy_is_little_endian();
 314 }
 315
 316 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 317 {
 318         int be = !!(tun->flags & TUN_VNET_BE);
 319
 320         if (put_user(be, argp))
 321                 return -EFAULT;
 322
 323         return 0;
 324 }
 325
 326 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 327 {
 328         int be;
 329
 330         if (get_user(be, argp))
 331                 return -EFAULT;
 332
 333         if (be)
 334                 tun->flags |= TUN_VNET_BE;
 335         else
 336                 tun->flags &= ~TUN_VNET_BE;
 337
 338         return 0;
 339 }
 340 #else
 341 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 342 {
 343         return virtio_legacy_is_little_endian();
 344 }
 345
 346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 347 {
 348         return -EINVAL;
 349 }
 350
 351 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 352 {
 353         return -EINVAL;
 354 }
 355 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 356
 357 static inline bool tun_is_little_endian(struct tun_struct *tun)
 358 {
 359         return tun->flags & TUN_VNET_LE ||
 360                 tun_legacy_is_little_endian(tun);
 361 }
 362
 363 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 364 {
 365         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 366 }
 367
 368 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 369 {
 370         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 371 }
 372
 373 static inline u32 tun_hashfn(u32 rxhash)
 374 {
 375         return rxhash & TUN_MASK_FLOW_ENTRIES;
 376 }
 377
 378 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 379 {
 380         struct tun_flow_entry *e;
 381
 382         hlist_for_each_entry_rcu(e, head, hash_link) {
 383                 if (e->rxhash == rxhash)
 384                         return e;
 385         }
 386         return NULL;
 387 }
 388
 389 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 390                                               struct hlist_head *head,
 391                                               u32 rxhash, u16 queue_index)
 392 {
 393         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 394
 395         if (e) {
 396                 netif_info(tun, tx_queued, tun->dev,
 397                            "create flow: hash %u index %u\n",
 398                            rxhash, queue_index);
 399                 e->updated = jiffies;
 400                 e->rxhash = rxhash;
 401                 e->rps_rxhash = 0;
 402                 e->queue_index = queue_index;
 403                 e->tun = tun;
 404                 hlist_add_head_rcu(&e->hash_link, head);
 405                 ++tun->flow_count;
 406         }
 407         return e;
 408 }
 409
 410 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 411 {
 412         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 413                    e->rxhash, e->queue_index);
 414         hlist_del_rcu(&e->hash_link);
 415         kfree_rcu(e, rcu);
 416         --tun->flow_count;
 417 }
 418
 419 static void tun_flow_flush(struct tun_struct *tun)
 420 {
 421         int i;
 422
 423         spin_lock_bh(&tun->lock);
 424         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 425                 struct tun_flow_entry *e;
 426                 struct hlist_node *n;
 427
 428                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 429                         tun_flow_delete(tun, e);
 430         }
 431         spin_unlock_bh(&tun->lock);
 432 }
 433
 434 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 435 {
 436         int i;
 437
 438         spin_lock_bh(&tun->lock);
 439         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 440                 struct tun_flow_entry *e;
 441                 struct hlist_node *n;
 442
 443                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 444                         if (e->queue_index == queue_index)
 445                                 tun_flow_delete(tun, e);
 446                 }
 447         }
 448         spin_unlock_bh(&tun->lock);
 449 }
 450
 451 static void tun_flow_cleanup(struct timer_list *t)
 452 {
 453         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 454         unsigned long delay = tun->ageing_time;
 455         unsigned long next_timer = jiffies + delay;
 456         unsigned long count = 0;
 457         int i;
 458
 459         spin_lock(&tun->lock);
 460         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 461                 struct tun_flow_entry *e;
 462                 struct hlist_node *n;
 463
 464                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 465                         unsigned long this_timer;
 466
 467                         this_timer = e->updated + delay;
 468                         if (time_before_eq(this_timer, jiffies)) {
 469                                 tun_flow_delete(tun, e);
 470                                 continue;
 471                         }
 472                         count++;
 473                         if (time_before(this_timer, next_timer))
 474                                 next_timer = this_timer;
 475                 }
 476         }
 477
 478         if (count)
 479                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 480         spin_unlock(&tun->lock);
 481 }
 482
 483 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 484                             struct tun_file *tfile)
 485 {
 486         struct hlist_head *head;
 487         struct tun_flow_entry *e;
 488         unsigned long delay = tun->ageing_time;
 489         u16 queue_index = tfile->queue_index;
 490
 491         head = &tun->flows[tun_hashfn(rxhash)];
 492
 493         rcu_read_lock();
 494
 495         e = tun_flow_find(head, rxhash);
 496         if (likely(e)) {
 497                 /* TODO: keep queueing to old queue until it's empty? */
 498                 if (READ_ONCE(e->queue_index) != queue_index)
 499                         WRITE_ONCE(e->queue_index, queue_index);
 500                 if (e->updated != jiffies)
 501                         e->updated = jiffies;
 502                 sock_rps_record_flow_hash(e->rps_rxhash);
 503         } else {
 504                 spin_lock_bh(&tun->lock);
 505                 if (!tun_flow_find(head, rxhash) &&
 506                     tun->flow_count < MAX_TAP_FLOWS)
 507                         tun_flow_create(tun, head, rxhash, queue_index);
 508
 509                 if (!timer_pending(&tun->flow_gc_timer))
 510                         mod_timer(&tun->flow_gc_timer,
 511                                   round_jiffies_up(jiffies + delay));
 512                 spin_unlock_bh(&tun->lock);
 513         }
 514
 515         rcu_read_unlock();
 516 }
 517
 518 /* Save the hash received in the stack receive path and update the
 519  * flow_hash table accordingly.
 520  */
 521 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 522 {
 523         if (unlikely(e->rps_rxhash != hash))
 524                 e->rps_rxhash = hash;
 525 }
 526
 527 /* We try to identify a flow through its rxhash. The reason that
 528  * we do not check rxq no. is because some cards(e.g 82599), chooses
 529  * the rxq based on the txq where the last packet of the flow comes. As
 530  * the userspace application move between processors, we may get a
 531  * different rxq no. here.
 532  */
 533 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 534 {
 535         struct tun_flow_entry *e;
 536         u32 txq = 0;
 537         u32 numqueues = 0;
 538
 539         numqueues = READ_ONCE(tun->numqueues);
 540
 541         txq = __skb_get_hash_symmetric(skb);
 542         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 543         if (e) {
 544                 tun_flow_save_rps_rxhash(e, txq);
 545                 txq = e->queue_index;
 546         } else {
 547                 /* use multiply and shift instead of expensive divide */
 548                 txq = ((u64)txq * numqueues) >> 32;
 549         }
 550
 551         return txq;
 552 }
 553
 554 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 555 {
 556         struct tun_prog *prog;
 557         u32 numqueues;
 558         u16 ret = 0;
 559
 560         numqueues = READ_ONCE(tun->numqueues);
 561         if (!numqueues)
 562                 return 0;
 563
 564         prog = rcu_dereference(tun->steering_prog);
 565         if (prog)
 566                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 567
 568         return ret % numqueues;
 569 }
 570
 571 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 572                             struct net_device *sb_dev)
 573 {
 574         struct tun_struct *tun = netdev_priv(dev);
 575         u16 ret;
 576
 577         rcu_read_lock();
 578         if (rcu_dereference(tun->steering_prog))
 579                 ret = tun_ebpf_select_queue(tun, skb);
 580         else
 581                 ret = tun_automq_select_queue(tun, skb);
 582         rcu_read_unlock();
 583
 584         return ret;
 585 }
 586
 587 static inline bool tun_not_capable(struct tun_struct *tun)
 588 {
 589         const struct cred *cred = current_cred();
 590         struct net *net = dev_net(tun->dev);
 591
 592         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 593                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 594                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 595 }
 596
 597 static void tun_set_real_num_queues(struct tun_struct *tun)
 598 {
 599         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 600         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 601 }
 602
 603 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 604 {
 605         tfile->detached = tun;
 606         list_add_tail(&tfile->next, &tun->disabled);
 607         ++tun->numdisabled;
 608 }
 609
 610 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 611 {
 612         struct tun_struct *tun = tfile->detached;
 613
 614         tfile->detached = NULL;
 615         list_del_init(&tfile->next);
 616         --tun->numdisabled;
 617         return tun;
 618 }
 619
 620 void tun_ptr_free(void *ptr)
 621 {
 622         if (!ptr)
 623                 return;
 624         if (tun_is_xdp_frame(ptr)) {
 625                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 626
 627                 xdp_return_frame(xdpf);
 628         } else {
 629                 __skb_array_destroy_skb(ptr);
 630         }
 631 }
 632 EXPORT_SYMBOL_GPL(tun_ptr_free);
 633
 634 static void tun_queue_purge(struct tun_file *tfile)
 635 {
 636         void *ptr;
 637
 638         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 639                 tun_ptr_free(ptr);
 640
 641         skb_queue_purge(&tfile->sk.sk_write_queue);
 642         skb_queue_purge(&tfile->sk.sk_error_queue);
 643 }
 644
 645 static void __tun_detach(struct tun_file *tfile, bool clean)
 646 {
 647         struct tun_file *ntfile;
 648         struct tun_struct *tun;
 649
 650         tun = rtnl_dereference(tfile->tun);
 651
 652         if (tun && clean) {
 653                 tun_napi_disable(tfile);
 654                 tun_napi_del(tfile);
 655         }
 656
 657         if (tun && !tfile->detached) {
 658                 u16 index = tfile->queue_index;
 659                 BUG_ON(index >= tun->numqueues);
 660
 661                 rcu_assign_pointer(tun->tfiles[index],
 662                                    tun->tfiles[tun->numqueues - 1]);
 663                 ntfile = rtnl_dereference(tun->tfiles[index]);
 664                 ntfile->queue_index = index;
 665                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 666                                    NULL);
 667
 668                 --tun->numqueues;
 669                 if (clean) {
 670                         RCU_INIT_POINTER(tfile->tun, NULL);
 671                         sock_put(&tfile->sk);
 672                 } else
 673                         tun_disable_queue(tun, tfile);
 674
 675                 synchronize_net();
 676                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 677                 /* Drop read queue */
 678                 tun_queue_purge(tfile);
 679                 tun_set_real_num_queues(tun);
 680         } else if (tfile->detached && clean) {
 681                 tun = tun_enable_queue(tfile);
 682                 sock_put(&tfile->sk);
 683         }
 684
 685         if (clean) {
 686                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 687                         netif_carrier_off(tun->dev);
 688
 689                         if (!(tun->flags & IFF_PERSIST) &&
 690                             tun->dev->reg_state == NETREG_REGISTERED)
 691                                 unregister_netdevice(tun->dev);
 692                 }
 693                 if (tun)
 694                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 695                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 696                 sock_put(&tfile->sk);
 697         }
 698 }
 699
 700 static void tun_detach(struct tun_file *tfile, bool clean)
 701 {
 702         struct tun_struct *tun;
 703         struct net_device *dev;
 704
 705         rtnl_lock();
 706         tun = rtnl_dereference(tfile->tun);
 707         dev = tun ? tun->dev : NULL;
 708         __tun_detach(tfile, clean);
 709         if (dev)
 710                 netdev_state_change(dev);
 711         rtnl_unlock();
 712 }
 713
 714 static void tun_detach_all(struct net_device *dev)
 715 {
 716         struct tun_struct *tun = netdev_priv(dev);
 717         struct tun_file *tfile, *tmp;
 718         int i, n = tun->numqueues;
 719
 720         for (i = 0; i < n; i++) {
 721                 tfile = rtnl_dereference(tun->tfiles[i]);
 722                 BUG_ON(!tfile);
 723                 tun_napi_disable(tfile);
 724                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 725                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 726                 RCU_INIT_POINTER(tfile->tun, NULL);
 727                 --tun->numqueues;
 728         }
 729         list_for_each_entry(tfile, &tun->disabled, next) {
 730                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 731                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 732                 RCU_INIT_POINTER(tfile->tun, NULL);
 733         }
 734         BUG_ON(tun->numqueues != 0);
 735
 736         synchronize_net();
 737         for (i = 0; i < n; i++) {
 738                 tfile = rtnl_dereference(tun->tfiles[i]);
 739                 tun_napi_del(tfile);
 740                 /* Drop read queue */
 741                 tun_queue_purge(tfile);
 742                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 743                 sock_put(&tfile->sk);
 744         }
 745         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 746                 tun_enable_queue(tfile);
 747                 tun_queue_purge(tfile);
 748                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 749                 sock_put(&tfile->sk);
 750         }
 751         BUG_ON(tun->numdisabled != 0);
 752
 753         if (tun->flags & IFF_PERSIST)
 754                 module_put(THIS_MODULE);
 755 }
 756
 757 static int tun_attach(struct tun_struct *tun, struct file *file,
 758                       bool skip_filter, bool napi, bool napi_frags,
 759                       bool publish_tun)
 760 {
 761         struct tun_file *tfile = file->private_data;
 762         struct net_device *dev = tun->dev;
 763         int err;
 764
 765         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 766         if (err < 0)
 767                 goto out;
 768
 769         err = -EINVAL;
 770         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 771                 goto out;
 772
 773         err = -EBUSY;
 774         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 775                 goto out;
 776
 777         err = -E2BIG;
 778         if (!tfile->detached &&
 779             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 780                 goto out;
 781
 782         err = 0;
 783
 784         /* Re-attach the filter to persist device */
 785         if (!skip_filter && (tun->filter_attached == true)) {
 786                 lock_sock(tfile->socket.sk);
 787                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 788                 release_sock(tfile->socket.sk);
 789                 if (!err)
 790                         goto out;
 791         }
 792
 793         if (!tfile->detached &&
 794             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 795                             GFP_KERNEL, tun_ptr_free)) {
 796                 err = -ENOMEM;
 797                 goto out;
 798         }
 799
 800         tfile->queue_index = tun->numqueues;
 801         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 802
 803         if (tfile->detached) {
 804                 /* Re-attach detached tfile, updating XDP queue_index */
 805                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 806
 807                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 808                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 809         } else {
 810                 /* Setup XDP RX-queue info, for new tfile getting attached */
 811                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 812                                        tun->dev, tfile->queue_index);
 813                 if (err < 0)
 814                         goto out;
 815                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 816                                                  MEM_TYPE_PAGE_SHARED, NULL);
 817                 if (err < 0) {
 818                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 819                         goto out;
 820                 }
 821                 err = 0;
 822         }
 823
 824         if (tfile->detached) {
 825                 tun_enable_queue(tfile);
 826         } else {
 827                 sock_hold(&tfile->sk);
 828                 tun_napi_init(tun, tfile, napi, napi_frags);
 829         }
 830
 831         if (rtnl_dereference(tun->xdp_prog))
 832                 sock_set_flag(&tfile->sk, SOCK_XDP);
 833
 834         /* device is allowed to go away first, so no need to hold extra
 835          * refcnt.
 836          */
 837
 838         /* Publish tfile->tun and tun->tfiles only after we've fully
 839          * initialized tfile; otherwise we risk using half-initialized
 840          * object.
 841          */
 842         if (publish_tun)
 843                 rcu_assign_pointer(tfile->tun, tun);
 844         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 845         tun->numqueues++;
 846         tun_set_real_num_queues(tun);
 847 out:
 848         return err;
 849 }
 850
 851 static struct tun_struct *tun_get(struct tun_file *tfile)
 852 {
 853         struct tun_struct *tun;
 854
 855         rcu_read_lock();
 856         tun = rcu_dereference(tfile->tun);
 857         if (tun)
 858                 dev_hold(tun->dev);
 859         rcu_read_unlock();
 860
 861         return tun;
 862 }
 863
 864 static void tun_put(struct tun_struct *tun)
 865 {
 866         dev_put(tun->dev);
 867 }
 868
 869 /* TAP filtering */
 870 static void addr_hash_set(u32 *mask, const u8 *addr)
 871 {
 872         int n = ether_crc(ETH_ALEN, addr) >> 26;
 873         mask[n >> 5] |= (1 << (n & 31));
 874 }
 875
 876 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 877 {
 878         int n = ether_crc(ETH_ALEN, addr) >> 26;
 879         return mask[n >> 5] & (1 << (n & 31));
 880 }
 881
 882 static int update_filter(struct tap_filter *filter, void __user *arg)
 883 {
 884         struct { u8 u[ETH_ALEN]; } *addr;
 885         struct tun_filter uf;
 886         int err, alen, n, nexact;
 887
 888         if (copy_from_user(&uf, arg, sizeof(uf)))
 889                 return -EFAULT;
 890
 891         if (!uf.count) {
 892                 /* Disabled */
 893                 filter->count = 0;
 894                 return 0;
 895         }
 896
 897         alen = ETH_ALEN * uf.count;
 898         addr = memdup_user(arg + sizeof(uf), alen);
 899         if (IS_ERR(addr))
 900                 return PTR_ERR(addr);
 901
 902         /* The filter is updated without holding any locks. Which is
 903          * perfectly safe. We disable it first and in the worst
 904          * case we'll accept a few undesired packets. */
 905         filter->count = 0;
 906         wmb();
 907
 908         /* Use first set of addresses as an exact filter */
 909         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 910                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 911
 912         nexact = n;
 913
 914         /* Remaining multicast addresses are hashed,
 915          * unicast will leave the filter disabled. */
 916         memset(filter->mask, 0, sizeof(filter->mask));
 917         for (; n < uf.count; n++) {
 918                 if (!is_multicast_ether_addr(addr[n].u)) {
 919                         err = 0; /* no filter */
 920                         goto free_addr;
 921                 }
 922                 addr_hash_set(filter->mask, addr[n].u);
 923         }
 924
 925         /* For ALLMULTI just set the mask to all ones.
 926          * This overrides the mask populated above. */
 927         if ((uf.flags & TUN_FLT_ALLMULTI))
 928                 memset(filter->mask, ~0, sizeof(filter->mask));
 929
 930         /* Now enable the filter */
 931         wmb();
 932         filter->count = nexact;
 933
 934         /* Return the number of exact filters */
 935         err = nexact;
 936 free_addr:
 937         kfree(addr);
 938         return err;
 939 }
 940
 941 /* Returns: 0 - drop, !=0 - accept */
 942 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 943 {
 944         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 945          * at this point. */
 946         struct ethhdr *eh = (struct ethhdr *) skb->data;
 947         int i;
 948
 949         /* Exact match */
 950         for (i = 0; i < filter->count; i++)
 951                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 952                         return 1;
 953
 954         /* Inexact match (multicast only) */
 955         if (is_multicast_ether_addr(eh->h_dest))
 956                 return addr_hash_test(filter->mask, eh->h_dest);
 957
 958         return 0;
 959 }
 960
 961 /*
 962  * Checks whether the packet is accepted or not.
 963  * Returns: 0 - drop, !=0 - accept
 964  */
 965 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 966 {
 967         if (!filter->count)
 968                 return 1;
 969
 970         return run_filter(filter, skb);
 971 }
 972
 973 /* Network device part of the driver */
 974
 975 static const struct ethtool_ops tun_ethtool_ops;
 976
 977 /* Net device detach from fd. */
 978 static void tun_net_uninit(struct net_device *dev)
 979 {
 980         tun_detach_all(dev);
 981 }
 982
 983 /* Net device open. */
 984 static int tun_net_open(struct net_device *dev)
 985 {
 986         netif_tx_start_all_queues(dev);
 987
 988         return 0;
 989 }
 990
 991 /* Net device close. */
 992 static int tun_net_close(struct net_device *dev)
 993 {
 994         netif_tx_stop_all_queues(dev);
 995         return 0;
 996 }
 997
 998 /* Net device start xmit */
 999 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1000 {
1001 #ifdef CONFIG_RPS
1002         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1003                 /* Select queue was not called for the skbuff, so we extract the
1004                  * RPS hash and save it into the flow_table here.
1005                  */
1006                 struct tun_flow_entry *e;
1007                 __u32 rxhash;
1008
1009                 rxhash = __skb_get_hash_symmetric(skb);
1010                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1011                 if (e)
1012                         tun_flow_save_rps_rxhash(e, rxhash);
1013         }
1014 #endif
1015 }
1016
1017 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1018                                     struct sk_buff *skb,
1019                                     int len)
1020 {
1021         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1022
1023         if (prog)
1024                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1025
1026         return len;
1027 }
1028
1029 /* Net device start xmit */
1030 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1031 {
1032         struct tun_struct *tun = netdev_priv(dev);
1033         int txq = skb->queue_mapping;
1034         struct tun_file *tfile;
1035         int len = skb->len;
1036
1037         rcu_read_lock();
1038         tfile = rcu_dereference(tun->tfiles[txq]);
1039
1040         /* Drop packet if interface is not attached */
1041         if (!tfile)
1042                 goto drop;
1043
1044         if (!rcu_dereference(tun->steering_prog))
1045                 tun_automq_xmit(tun, skb);
1046
1047         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1048
1049         /* Drop if the filter does not like it.
1050          * This is a noop if the filter is disabled.
1051          * Filter can be enabled only for the TAP devices. */
1052         if (!check_filter(&tun->txflt, skb))
1053                 goto drop;
1054
1055         if (tfile->socket.sk->sk_filter &&
1056             sk_filter(tfile->socket.sk, skb))
1057                 goto drop;
1058
1059         len = run_ebpf_filter(tun, skb, len);
1060         if (len == 0 || pskb_trim(skb, len))
1061                 goto drop;
1062
1063         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1064                 goto drop;
1065
1066         skb_tx_timestamp(skb);
1067
1068         /* Orphan the skb - required as we might hang on to it
1069          * for indefinite time.
1070          */
1071         skb_orphan(skb);
1072
1073         nf_reset_ct(skb);
1074
1075         if (ptr_ring_produce(&tfile->tx_ring, skb))
1076                 goto drop;
1077
1078         /* Notify and wake up reader process */
1079         if (tfile->flags & TUN_FASYNC)
1080                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1081         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1082
1083         rcu_read_unlock();
1084         return NETDEV_TX_OK;
1085
1086 drop:
1087         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1088         skb_tx_error(skb);
1089         kfree_skb(skb);
1090         rcu_read_unlock();
1091         return NET_XMIT_DROP;
1092 }
1093
1094 static void tun_net_mclist(struct net_device *dev)
1095 {
1096         /*
1097          * This callback is supposed to deal with mc filter in
1098          * _rx_ path and has nothing to do with the _tx_ path.
1099          * In rx path we always accept everything userspace gives us.
1100          */
1101 }
1102
1103 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1104         netdev_features_t features)
1105 {
1106         struct tun_struct *tun = netdev_priv(dev);
1107
1108         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1109 }
1110
1111 static void tun_set_headroom(struct net_device *dev, int new_hr)
1112 {
1113         struct tun_struct *tun = netdev_priv(dev);
1114
1115         if (new_hr < NET_SKB_PAD)
1116                 new_hr = NET_SKB_PAD;
1117
1118         tun->align = new_hr;
1119 }
1120
1121 static void
1122 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1123 {
1124         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1125         struct tun_struct *tun = netdev_priv(dev);
1126         struct tun_pcpu_stats *p;
1127         int i;
1128
1129         for_each_possible_cpu(i) {
1130                 u64 rxpackets, rxbytes, txpackets, txbytes;
1131                 unsigned int start;
1132
1133                 p = per_cpu_ptr(tun->pcpu_stats, i);
1134                 do {
1135                         start = u64_stats_fetch_begin(&p->syncp);
1136                         rxpackets       = u64_stats_read(&p->rx_packets);
1137                         rxbytes         = u64_stats_read(&p->rx_bytes);
1138                         txpackets       = u64_stats_read(&p->tx_packets);
1139                         txbytes         = u64_stats_read(&p->tx_bytes);
1140                 } while (u64_stats_fetch_retry(&p->syncp, start));
1141
1142                 stats->rx_packets       += rxpackets;
1143                 stats->rx_bytes         += rxbytes;
1144                 stats->tx_packets       += txpackets;
1145                 stats->tx_bytes         += txbytes;
1146
1147                 /* u32 counters */
1148                 rx_dropped      += p->rx_dropped;
1149                 rx_frame_errors += p->rx_frame_errors;
1150                 tx_dropped      += p->tx_dropped;
1151         }
1152         stats->rx_dropped  = rx_dropped;
1153         stats->rx_frame_errors = rx_frame_errors;
1154         stats->tx_dropped = tx_dropped;
1155 }
1156
1157 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1158                        struct netlink_ext_ack *extack)
1159 {
1160         struct tun_struct *tun = netdev_priv(dev);
1161         struct tun_file *tfile;
1162         struct bpf_prog *old_prog;
1163         int i;
1164
1165         old_prog = rtnl_dereference(tun->xdp_prog);
1166         rcu_assign_pointer(tun->xdp_prog, prog);
1167         if (old_prog)
1168                 bpf_prog_put(old_prog);
1169
1170         for (i = 0; i < tun->numqueues; i++) {
1171                 tfile = rtnl_dereference(tun->tfiles[i]);
1172                 if (prog)
1173                         sock_set_flag(&tfile->sk, SOCK_XDP);
1174                 else
1175                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1176         }
1177         list_for_each_entry(tfile, &tun->disabled, next) {
1178                 if (prog)
1179                         sock_set_flag(&tfile->sk, SOCK_XDP);
1180                 else
1181                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1182         }
1183
1184         return 0;
1185 }
1186
1187 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1188 {
1189         switch (xdp->command) {
1190         case XDP_SETUP_PROG:
1191                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1192         default:
1193                 return -EINVAL;
1194         }
1195 }
1196
1197 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1198 {
1199         if (new_carrier) {
1200                 struct tun_struct *tun = netdev_priv(dev);
1201
1202                 if (!tun->numqueues)
1203                         return -EPERM;
1204
1205                 netif_carrier_on(dev);
1206         } else {
1207                 netif_carrier_off(dev);
1208         }
1209         return 0;
1210 }
1211
1212 static const struct net_device_ops tun_netdev_ops = {
1213         .ndo_uninit             = tun_net_uninit,
1214         .ndo_open               = tun_net_open,
1215         .ndo_stop               = tun_net_close,
1216         .ndo_start_xmit         = tun_net_xmit,
1217         .ndo_fix_features       = tun_net_fix_features,
1218         .ndo_select_queue       = tun_select_queue,
1219         .ndo_set_rx_headroom    = tun_set_headroom,
1220         .ndo_get_stats64        = tun_net_get_stats64,
1221         .ndo_change_carrier     = tun_net_change_carrier,
1222 };
1223
1224 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1225 {
1226         /* Notify and wake up reader process */
1227         if (tfile->flags & TUN_FASYNC)
1228                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1229         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1230 }
1231
1232 static int tun_xdp_xmit(struct net_device *dev, int n,
1233                         struct xdp_frame **frames, u32 flags)
1234 {
1235         struct tun_struct *tun = netdev_priv(dev);
1236         struct tun_file *tfile;
1237         u32 numqueues;
1238         int drops = 0;
1239         int cnt = n;
1240         int i;
1241
1242         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1243                 return -EINVAL;
1244
1245         rcu_read_lock();
1246
1247 resample:
1248         numqueues = READ_ONCE(tun->numqueues);
1249         if (!numqueues) {
1250                 rcu_read_unlock();
1251                 return -ENXIO; /* Caller will free/return all frames */
1252         }
1253
1254         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1255                                             numqueues]);
1256         if (unlikely(!tfile))
1257                 goto resample;
1258
1259         spin_lock(&tfile->tx_ring.producer_lock);
1260         for (i = 0; i < n; i++) {
1261                 struct xdp_frame *xdp = frames[i];
1262                 /* Encode the XDP flag into lowest bit for consumer to differ
1263                  * XDP buffer from sk_buff.
1264                  */
1265                 void *frame = tun_xdp_to_ptr(xdp);
1266
1267                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1268                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1269                         xdp_return_frame_rx_napi(xdp);
1270                         drops++;
1271                 }
1272         }
1273         spin_unlock(&tfile->tx_ring.producer_lock);
1274
1275         if (flags & XDP_XMIT_FLUSH)
1276                 __tun_xdp_flush_tfile(tfile);
1277
1278         rcu_read_unlock();
1279         return cnt - drops;
1280 }
1281
1282 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1283 {
1284         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1285
1286         if (unlikely(!frame))
1287                 return -EOVERFLOW;
1288
1289         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1290 }
1291
1292 static const struct net_device_ops tap_netdev_ops = {
1293         .ndo_uninit             = tun_net_uninit,
1294         .ndo_open               = tun_net_open,
1295         .ndo_stop               = tun_net_close,
1296         .ndo_start_xmit         = tun_net_xmit,
1297         .ndo_fix_features       = tun_net_fix_features,
1298         .ndo_set_rx_mode        = tun_net_mclist,
1299         .ndo_set_mac_address    = eth_mac_addr,
1300         .ndo_validate_addr      = eth_validate_addr,
1301         .ndo_select_queue       = tun_select_queue,
1302         .ndo_features_check     = passthru_features_check,
1303         .ndo_set_rx_headroom    = tun_set_headroom,
1304         .ndo_get_stats64        = tun_net_get_stats64,
1305         .ndo_bpf                = tun_xdp,
1306         .ndo_xdp_xmit           = tun_xdp_xmit,
1307         .ndo_change_carrier     = tun_net_change_carrier,
1308 };
1309
1310 static void tun_flow_init(struct tun_struct *tun)
1311 {
1312         int i;
1313
1314         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1315                 INIT_HLIST_HEAD(&tun->flows[i]);
1316
1317         tun->ageing_time = TUN_FLOW_EXPIRE;
1318         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1319         mod_timer(&tun->flow_gc_timer,
1320                   round_jiffies_up(jiffies + tun->ageing_time));
1321 }
1322
1323 static void tun_flow_uninit(struct tun_struct *tun)
1324 {
1325         del_timer_sync(&tun->flow_gc_timer);
1326         tun_flow_flush(tun);
1327 }
1328
1329 #define MIN_MTU 68
1330 #define MAX_MTU 65535
1331
1332 /* Initialize net device. */
1333 static void tun_net_init(struct net_device *dev)
1334 {
1335         struct tun_struct *tun = netdev_priv(dev);
1336
1337         switch (tun->flags & TUN_TYPE_MASK) {
1338         case IFF_TUN:
1339                 dev->netdev_ops = &tun_netdev_ops;
1340                 dev->header_ops = &ip_tunnel_header_ops;
1341
1342                 /* Point-to-Point TUN Device */
1343                 dev->hard_header_len = 0;
1344                 dev->addr_len = 0;
1345                 dev->mtu = 1500;
1346
1347                 /* Zero header length */
1348                 dev->type = ARPHRD_NONE;
1349                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1350                 break;
1351
1352         case IFF_TAP:
1353                 dev->netdev_ops = &tap_netdev_ops;
1354                 /* Ethernet TAP Device */
1355                 ether_setup(dev);
1356                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1357                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1358
1359                 eth_hw_addr_random(dev);
1360
1361                 break;
1362         }
1363
1364         dev->min_mtu = MIN_MTU;
1365         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1366 }
1367
1368 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1369 {
1370         struct sock *sk = tfile->socket.sk;
1371
1372         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1373 }
1374
1375 /* Character device part */
1376
1377 /* Poll */
1378 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1379 {
1380         struct tun_file *tfile = file->private_data;
1381         struct tun_struct *tun = tun_get(tfile);
1382         struct sock *sk;
1383         __poll_t mask = 0;
1384
1385         if (!tun)
1386                 return EPOLLERR;
1387
1388         sk = tfile->socket.sk;
1389
1390         poll_wait(file, sk_sleep(sk), wait);
1391
1392         if (!ptr_ring_empty(&tfile->tx_ring))
1393                 mask |= EPOLLIN | EPOLLRDNORM;
1394
1395         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1396          * guarantee EPOLLOUT to be raised by either here or
1397          * tun_sock_write_space(). Then process could get notification
1398          * after it writes to a down device and meets -EIO.
1399          */
1400         if (tun_sock_writeable(tun, tfile) ||
1401             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1402              tun_sock_writeable(tun, tfile)))
1403                 mask |= EPOLLOUT | EPOLLWRNORM;
1404
1405         if (tun->dev->reg_state != NETREG_REGISTERED)
1406                 mask = EPOLLERR;
1407
1408         tun_put(tun);
1409         return mask;
1410 }
1411
1412 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1413                                             size_t len,
1414                                             const struct iov_iter *it)
1415 {
1416         struct sk_buff *skb;
1417         size_t linear;
1418         int err;
1419         int i;
1420
1421         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1422                 return ERR_PTR(-ENOMEM);
1423
1424         local_bh_disable();
1425         skb = napi_get_frags(&tfile->napi);
1426         local_bh_enable();
1427         if (!skb)
1428                 return ERR_PTR(-ENOMEM);
1429
1430         linear = iov_iter_single_seg_count(it);
1431         err = __skb_grow(skb, linear);
1432         if (err)
1433                 goto free;
1434
1435         skb->len = len;
1436         skb->data_len = len - linear;
1437         skb->truesize += skb->data_len;
1438
1439         for (i = 1; i < it->nr_segs; i++) {
1440                 size_t fragsz = it->iov[i].iov_len;
1441                 struct page *page;
1442                 void *frag;
1443
1444                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1445                         err = -EINVAL;
1446                         goto free;
1447                 }
1448                 frag = netdev_alloc_frag(fragsz);
1449                 if (!frag) {
1450                         err = -ENOMEM;
1451                         goto free;
1452                 }
1453                 page = virt_to_head_page(frag);
1454                 skb_fill_page_desc(skb, i - 1, page,
1455                                    frag - page_address(page), fragsz);
1456         }
1457
1458         return skb;
1459 free:
1460         /* frees skb and all frags allocated with napi_alloc_frag() */
1461         napi_free_frags(&tfile->napi);
1462         return ERR_PTR(err);
1463 }
1464
1465 /* prepad is the amount to reserve at front.  len is length after that.
1466  * linear is a hint as to how much to copy (usually headers). */
1467 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1468                                      size_t prepad, size_t len,
1469                                      size_t linear, int noblock)
1470 {
1471         struct sock *sk = tfile->socket.sk;
1472         struct sk_buff *skb;
1473         int err;
1474
1475         /* Under a page?  Don't bother with paged skb. */
1476         if (prepad + len < PAGE_SIZE || !linear)
1477                 linear = len;
1478
1479         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1480                                    &err, 0);
1481         if (!skb)
1482                 return ERR_PTR(err);
1483
1484         skb_reserve(skb, prepad);
1485         skb_put(skb, linear);
1486         skb->data_len = len - linear;
1487         skb->len += len - linear;
1488
1489         return skb;
1490 }
1491
1492 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1493                            struct sk_buff *skb, int more)
1494 {
1495         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1496         struct sk_buff_head process_queue;
1497         u32 rx_batched = tun->rx_batched;
1498         bool rcv = false;
1499
1500         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1501                 local_bh_disable();
1502                 skb_record_rx_queue(skb, tfile->queue_index);
1503                 netif_receive_skb(skb);
1504                 local_bh_enable();
1505                 return;
1506         }
1507
1508         spin_lock(&queue->lock);
1509         if (!more || skb_queue_len(queue) == rx_batched) {
1510                 __skb_queue_head_init(&process_queue);
1511                 skb_queue_splice_tail_init(queue, &process_queue);
1512                 rcv = true;
1513         } else {
1514                 __skb_queue_tail(queue, skb);
1515         }
1516         spin_unlock(&queue->lock);
1517
1518         if (rcv) {
1519                 struct sk_buff *nskb;
1520
1521                 local_bh_disable();
1522                 while ((nskb = __skb_dequeue(&process_queue))) {
1523                         skb_record_rx_queue(nskb, tfile->queue_index);
1524                         netif_receive_skb(nskb);
1525                 }
1526                 skb_record_rx_queue(skb, tfile->queue_index);
1527                 netif_receive_skb(skb);
1528                 local_bh_enable();
1529         }
1530 }
1531
1532 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1533                               int len, int noblock, bool zerocopy)
1534 {
1535         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1536                 return false;
1537
1538         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1539                 return false;
1540
1541         if (!noblock)
1542                 return false;
1543
1544         if (zerocopy)
1545                 return false;
1546
1547         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1548             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1549                 return false;
1550
1551         return true;
1552 }
1553
1554 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1555                                        struct page_frag *alloc_frag, char *buf,
1556                                        int buflen, int len, int pad)
1557 {
1558         struct sk_buff *skb = build_skb(buf, buflen);
1559
1560         if (!skb)
1561                 return ERR_PTR(-ENOMEM);
1562
1563         skb_reserve(skb, pad);
1564         skb_put(skb, len);
1565         skb_set_owner_w(skb, tfile->socket.sk);
1566
1567         get_page(alloc_frag->page);
1568         alloc_frag->offset += buflen;
1569
1570         return skb;
1571 }
1572
1573 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1574                        struct xdp_buff *xdp, u32 act)
1575 {
1576         int err;
1577
1578         switch (act) {
1579         case XDP_REDIRECT:
1580                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1581                 if (err)
1582                         return err;
1583                 break;
1584         case XDP_TX:
1585                 err = tun_xdp_tx(tun->dev, xdp);
1586                 if (err < 0)
1587                         return err;
1588                 break;
1589         case XDP_PASS:
1590                 break;
1591         default:
1592                 bpf_warn_invalid_xdp_action(act);
1593                 fallthrough;
1594         case XDP_ABORTED:
1595                 trace_xdp_exception(tun->dev, xdp_prog, act);
1596                 fallthrough;
1597         case XDP_DROP:
1598                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1599                 break;
1600         }
1601
1602         return act;
1603 }
1604
1605 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1606                                      struct tun_file *tfile,
1607                                      struct iov_iter *from,
1608                                      struct virtio_net_hdr *hdr,
1609                                      int len, int *skb_xdp)
1610 {
1611         struct page_frag *alloc_frag = &current->task_frag;
1612         struct bpf_prog *xdp_prog;
1613         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1614         char *buf;
1615         size_t copied;
1616         int pad = TUN_RX_PAD;
1617         int err = 0;
1618
1619         rcu_read_lock();
1620         xdp_prog = rcu_dereference(tun->xdp_prog);
1621         if (xdp_prog)
1622                 pad += XDP_PACKET_HEADROOM;
1623         buflen += SKB_DATA_ALIGN(len + pad);
1624         rcu_read_unlock();
1625
1626         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628                 return ERR_PTR(-ENOMEM);
1629
1630         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631         copied = copy_page_from_iter(alloc_frag->page,
1632                                      alloc_frag->offset + pad,
1633                                      len, from);
1634         if (copied != len)
1635                 return ERR_PTR(-EFAULT);
1636
1637         /* There's a small window that XDP may be set after the check
1638          * of xdp_prog above, this should be rare and for simplicity
1639          * we do XDP on skb in case the headroom is not enough.
1640          */
1641         if (hdr->gso_type || !xdp_prog) {
1642                 *skb_xdp = 1;
1643                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1644                                        pad);
1645         }
1646
1647         *skb_xdp = 0;
1648
1649         local_bh_disable();
1650         rcu_read_lock();
1651         xdp_prog = rcu_dereference(tun->xdp_prog);
1652         if (xdp_prog) {
1653                 struct xdp_buff xdp;
1654                 u32 act;
1655
1656                 xdp.data_hard_start = buf;
1657                 xdp.data = buf + pad;
1658                 xdp_set_data_meta_invalid(&xdp);
1659                 xdp.data_end = xdp.data + len;
1660                 xdp.rxq = &tfile->xdp_rxq;
1661                 xdp.frame_sz = buflen;
1662
1663                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1664                 if (act == XDP_REDIRECT || act == XDP_TX) {
1665                         get_page(alloc_frag->page);
1666                         alloc_frag->offset += buflen;
1667                 }
1668                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1669                 if (err < 0) {
1670                         if (act == XDP_REDIRECT || act == XDP_TX)
1671                                 put_page(alloc_frag->page);
1672                         goto out;
1673                 }
1674
1675                 if (err == XDP_REDIRECT)
1676                         xdp_do_flush();
1677                 if (err != XDP_PASS)
1678                         goto out;
1679
1680                 pad = xdp.data - xdp.data_hard_start;
1681                 len = xdp.data_end - xdp.data;
1682         }
1683         rcu_read_unlock();
1684         local_bh_enable();
1685
1686         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1687
1688 out:
1689         rcu_read_unlock();
1690         local_bh_enable();
1691         return NULL;
1692 }
1693
1694 /* Get packet from user space buffer */
1695 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1696                             void *msg_control, struct iov_iter *from,
1697                             int noblock, bool more)
1698 {
1699         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1700         struct sk_buff *skb;
1701         size_t total_len = iov_iter_count(from);
1702         size_t len = total_len, align = tun->align, linear;
1703         struct virtio_net_hdr gso = { 0 };
1704         struct tun_pcpu_stats *stats;
1705         int good_linear;
1706         int copylen;
1707         bool zerocopy = false;
1708         int err;
1709         u32 rxhash = 0;
1710         int skb_xdp = 1;
1711         bool frags = tun_napi_frags_enabled(tfile);
1712
1713         if (!(tun->flags & IFF_NO_PI)) {
1714                 if (len < sizeof(pi))
1715                         return -EINVAL;
1716                 len -= sizeof(pi);
1717
1718                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1719                         return -EFAULT;
1720         }
1721
1722         if (tun->flags & IFF_VNET_HDR) {
1723                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1724
1725                 if (len < vnet_hdr_sz)
1726                         return -EINVAL;
1727                 len -= vnet_hdr_sz;
1728
1729                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1730                         return -EFAULT;
1731
1732                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1733                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1734                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1735
1736                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1737                         return -EINVAL;
1738                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1739         }
1740
1741         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1742                 align += NET_IP_ALIGN;
1743                 if (unlikely(len < ETH_HLEN ||
1744                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1745                         return -EINVAL;
1746         }
1747
1748         good_linear = SKB_MAX_HEAD(align);
1749
1750         if (msg_control) {
1751                 struct iov_iter i = *from;
1752
1753                 /* There are 256 bytes to be copied in skb, so there is
1754                  * enough room for skb expand head in case it is used.
1755                  * The rest of the buffer is mapped from userspace.
1756                  */
1757                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1758                 if (copylen > good_linear)
1759                         copylen = good_linear;
1760                 linear = copylen;
1761                 iov_iter_advance(&i, copylen);
1762                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1763                         zerocopy = true;
1764         }
1765
1766         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1767                 /* For the packet that is not easy to be processed
1768                  * (e.g gso or jumbo packet), we will do it at after
1769                  * skb was created with generic XDP routine.
1770                  */
1771                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1772                 if (IS_ERR(skb)) {
1773                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1774                         return PTR_ERR(skb);
1775                 }
1776                 if (!skb)
1777                         return total_len;
1778         } else {
1779                 if (!zerocopy) {
1780                         copylen = len;
1781                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1782                                 linear = good_linear;
1783                         else
1784                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1785                 }
1786
1787                 if (frags) {
1788                         mutex_lock(&tfile->napi_mutex);
1789                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1790                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1791                          * If zerocopy is enabled, then this layout will be
1792                          * overwritten by zerocopy_sg_from_iter().
1793                          */
1794                         zerocopy = false;
1795                 } else {
1796                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1797                                             noblock);
1798                 }
1799
1800                 if (IS_ERR(skb)) {
1801                         if (PTR_ERR(skb) != -EAGAIN)
1802                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1803                         if (frags)
1804                                 mutex_unlock(&tfile->napi_mutex);
1805                         return PTR_ERR(skb);
1806                 }
1807
1808                 if (zerocopy)
1809                         err = zerocopy_sg_from_iter(skb, from);
1810                 else
1811                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1812
1813                 if (err) {
1814                         err = -EFAULT;
1815 drop:
1816                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1817                         kfree_skb(skb);
1818                         if (frags) {
1819                                 tfile->napi.skb = NULL;
1820                                 mutex_unlock(&tfile->napi_mutex);
1821                         }
1822
1823                         return err;
1824                 }
1825         }
1826
1827         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1828                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1829                 kfree_skb(skb);
1830                 if (frags) {
1831                         tfile->napi.skb = NULL;
1832                         mutex_unlock(&tfile->napi_mutex);
1833                 }
1834
1835                 return -EINVAL;
1836         }
1837
1838         switch (tun->flags & TUN_TYPE_MASK) {
1839         case IFF_TUN:
1840                 if (tun->flags & IFF_NO_PI) {
1841                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1842
1843                         switch (ip_version) {
1844                         case 4:
1845                                 pi.proto = htons(ETH_P_IP);
1846                                 break;
1847                         case 6:
1848                                 pi.proto = htons(ETH_P_IPV6);
1849                                 break;
1850                         default:
1851                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1852                                 kfree_skb(skb);
1853                                 return -EINVAL;
1854                         }
1855                 }
1856
1857                 skb_reset_mac_header(skb);
1858                 skb->protocol = pi.proto;
1859                 skb->dev = tun->dev;
1860                 break;
1861         case IFF_TAP:
1862                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1863                         err = -ENOMEM;
1864                         goto drop;
1865                 }
1866                 skb->protocol = eth_type_trans(skb, tun->dev);
1867                 break;
1868         }
1869
1870         /* copy skb_ubuf_info for callback when skb has no error */
1871         if (zerocopy) {
1872                 skb_shinfo(skb)->destructor_arg = msg_control;
1873                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1874                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1875         } else if (msg_control) {
1876                 struct ubuf_info *uarg = msg_control;
1877                 uarg->callback(uarg, false);
1878         }
1879
1880         skb_reset_network_header(skb);
1881         skb_probe_transport_header(skb);
1882         skb_record_rx_queue(skb, tfile->queue_index);
1883
1884         if (skb_xdp) {
1885                 struct bpf_prog *xdp_prog;
1886                 int ret;
1887
1888                 local_bh_disable();
1889                 rcu_read_lock();
1890                 xdp_prog = rcu_dereference(tun->xdp_prog);
1891                 if (xdp_prog) {
1892                         ret = do_xdp_generic(xdp_prog, skb);
1893                         if (ret != XDP_PASS) {
1894                                 rcu_read_unlock();
1895                                 local_bh_enable();
1896                                 if (frags) {
1897                                         tfile->napi.skb = NULL;
1898                                         mutex_unlock(&tfile->napi_mutex);
1899                                 }
1900                                 return total_len;
1901                         }
1902                 }
1903                 rcu_read_unlock();
1904                 local_bh_enable();
1905         }
1906
1907         /* Compute the costly rx hash only if needed for flow updates.
1908          * We may get a very small possibility of OOO during switching, not
1909          * worth to optimize.
1910          */
1911         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1912             !tfile->detached)
1913                 rxhash = __skb_get_hash_symmetric(skb);
1914
1915         rcu_read_lock();
1916         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1917                 err = -EIO;
1918                 rcu_read_unlock();
1919                 goto drop;
1920         }
1921
1922         if (frags) {
1923                 u32 headlen;
1924
1925                 /* Exercise flow dissector code path. */
1926                 skb_push(skb, ETH_HLEN);
1927                 headlen = eth_get_headlen(tun->dev, skb->data,
1928                                           skb_headlen(skb));
1929
1930                 if (unlikely(headlen > skb_headlen(skb))) {
1931                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1932                         napi_free_frags(&tfile->napi);
1933                         rcu_read_unlock();
1934                         mutex_unlock(&tfile->napi_mutex);
1935                         WARN_ON(1);
1936                         return -ENOMEM;
1937                 }
1938
1939                 local_bh_disable();
1940                 napi_gro_frags(&tfile->napi);
1941                 local_bh_enable();
1942                 mutex_unlock(&tfile->napi_mutex);
1943         } else if (tfile->napi_enabled) {
1944                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1945                 int queue_len;
1946
1947                 spin_lock_bh(&queue->lock);
1948                 __skb_queue_tail(queue, skb);
1949                 queue_len = skb_queue_len(queue);
1950                 spin_unlock(&queue->lock);
1951
1952                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1953                         napi_schedule(&tfile->napi);
1954
1955                 local_bh_enable();
1956         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1957                 tun_rx_batched(tun, tfile, skb, more);
1958         } else {
1959                 netif_rx_ni(skb);
1960         }
1961         rcu_read_unlock();
1962
1963         stats = get_cpu_ptr(tun->pcpu_stats);
1964         u64_stats_update_begin(&stats->syncp);
1965         u64_stats_inc(&stats->rx_packets);
1966         u64_stats_add(&stats->rx_bytes, len);
1967         u64_stats_update_end(&stats->syncp);
1968         put_cpu_ptr(stats);
1969
1970         if (rxhash)
1971                 tun_flow_update(tun, rxhash, tfile);
1972
1973         return total_len;
1974 }
1975
1976 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1977 {
1978         struct file *file = iocb->ki_filp;
1979         struct tun_file *tfile = file->private_data;
1980         struct tun_struct *tun = tun_get(tfile);
1981         ssize_t result;
1982
1983         if (!tun)
1984                 return -EBADFD;
1985
1986         result = tun_get_user(tun, tfile, NULL, from,
1987                               file->f_flags & O_NONBLOCK, false);
1988
1989         tun_put(tun);
1990         return result;
1991 }
1992
1993 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1994                                 struct tun_file *tfile,
1995                                 struct xdp_frame *xdp_frame,
1996                                 struct iov_iter *iter)
1997 {
1998         int vnet_hdr_sz = 0;
1999         size_t size = xdp_frame->len;
2000         struct tun_pcpu_stats *stats;
2001         size_t ret;
2002
2003         if (tun->flags & IFF_VNET_HDR) {
2004                 struct virtio_net_hdr gso = { 0 };
2005
2006                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2007                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2008                         return -EINVAL;
2009                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2010                              sizeof(gso)))
2011                         return -EFAULT;
2012                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2013         }
2014
2015         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2016
2017         stats = get_cpu_ptr(tun->pcpu_stats);
2018         u64_stats_update_begin(&stats->syncp);
2019         u64_stats_inc(&stats->tx_packets);
2020         u64_stats_add(&stats->tx_bytes, ret);
2021         u64_stats_update_end(&stats->syncp);
2022         put_cpu_ptr(tun->pcpu_stats);
2023
2024         return ret;
2025 }
2026
2027 /* Put packet to the user space buffer */
2028 static ssize_t tun_put_user(struct tun_struct *tun,
2029                             struct tun_file *tfile,
2030                             struct sk_buff *skb,
2031                             struct iov_iter *iter)
2032 {
2033         struct tun_pi pi = { 0, skb->protocol };
2034         struct tun_pcpu_stats *stats;
2035         ssize_t total;
2036         int vlan_offset = 0;
2037         int vlan_hlen = 0;
2038         int vnet_hdr_sz = 0;
2039
2040         if (skb_vlan_tag_present(skb))
2041                 vlan_hlen = VLAN_HLEN;
2042
2043         if (tun->flags & IFF_VNET_HDR)
2044                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2045
2046         total = skb->len + vlan_hlen + vnet_hdr_sz;
2047
2048         if (!(tun->flags & IFF_NO_PI)) {
2049                 if (iov_iter_count(iter) < sizeof(pi))
2050                         return -EINVAL;
2051
2052                 total += sizeof(pi);
2053                 if (iov_iter_count(iter) < total) {
2054                         /* Packet will be striped */
2055                         pi.flags |= TUN_PKT_STRIP;
2056                 }
2057
2058                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2059                         return -EFAULT;
2060         }
2061
2062         if (vnet_hdr_sz) {
2063                 struct virtio_net_hdr gso;
2064
2065                 if (iov_iter_count(iter) < vnet_hdr_sz)
2066                         return -EINVAL;
2067
2068                 if (virtio_net_hdr_from_skb(skb, &gso,
2069                                             tun_is_little_endian(tun), true,
2070                                             vlan_hlen)) {
2071                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2072                         pr_err("unexpected GSO type: "
2073                                "0x%x, gso_size %d, hdr_len %d\n",
2074                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2075                                tun16_to_cpu(tun, gso.hdr_len));
2076                         print_hex_dump(KERN_ERR, "tun: ",
2077                                        DUMP_PREFIX_NONE,
2078                                        16, 1, skb->head,
2079                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2080                         WARN_ON_ONCE(1);
2081                         return -EINVAL;
2082                 }
2083
2084                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2085                         return -EFAULT;
2086
2087                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2088         }
2089
2090         if (vlan_hlen) {
2091                 int ret;
2092                 struct veth veth;
2093
2094                 veth.h_vlan_proto = skb->vlan_proto;
2095                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2096
2097                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2098
2099                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2100                 if (ret || !iov_iter_count(iter))
2101                         goto done;
2102
2103                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2104                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2105                         goto done;
2106         }
2107
2108         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2109
2110 done:
2111         /* caller is in process context, */
2112         stats = get_cpu_ptr(tun->pcpu_stats);
2113         u64_stats_update_begin(&stats->syncp);
2114         u64_stats_inc(&stats->tx_packets);
2115         u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2116         u64_stats_update_end(&stats->syncp);
2117         put_cpu_ptr(tun->pcpu_stats);
2118
2119         return total;
2120 }
2121
2122 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2123 {
2124         DECLARE_WAITQUEUE(wait, current);
2125         void *ptr = NULL;
2126         int error = 0;
2127
2128         ptr = ptr_ring_consume(&tfile->tx_ring);
2129         if (ptr)
2130                 goto out;
2131         if (noblock) {
2132                 error = -EAGAIN;
2133                 goto out;
2134         }
2135
2136         add_wait_queue(&tfile->socket.wq.wait, &wait);
2137
2138         while (1) {
2139                 set_current_state(TASK_INTERRUPTIBLE);
2140                 ptr = ptr_ring_consume(&tfile->tx_ring);
2141                 if (ptr)
2142                         break;
2143                 if (signal_pending(current)) {
2144                         error = -ERESTARTSYS;
2145                         break;
2146                 }
2147                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2148                         error = -EFAULT;
2149                         break;
2150                 }
2151
2152                 schedule();
2153         }
2154
2155         __set_current_state(TASK_RUNNING);
2156         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2157
2158 out:
2159         *err = error;
2160         return ptr;
2161 }
2162
2163 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2164                            struct iov_iter *to,
2165                            int noblock, void *ptr)
2166 {
2167         ssize_t ret;
2168         int err;
2169
2170         if (!iov_iter_count(to)) {
2171                 tun_ptr_free(ptr);
2172                 return 0;
2173         }
2174
2175         if (!ptr) {
2176                 /* Read frames from ring */
2177                 ptr = tun_ring_recv(tfile, noblock, &err);
2178                 if (!ptr)
2179                         return err;
2180         }
2181
2182         if (tun_is_xdp_frame(ptr)) {
2183                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2184
2185                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2186                 xdp_return_frame(xdpf);
2187         } else {
2188                 struct sk_buff *skb = ptr;
2189
2190                 ret = tun_put_user(tun, tfile, skb, to);
2191                 if (unlikely(ret < 0))
2192                         kfree_skb(skb);
2193                 else
2194                         consume_skb(skb);
2195         }
2196
2197         return ret;
2198 }
2199
2200 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2201 {
2202         struct file *file = iocb->ki_filp;
2203         struct tun_file *tfile = file->private_data;
2204         struct tun_struct *tun = tun_get(tfile);
2205         ssize_t len = iov_iter_count(to), ret;
2206
2207         if (!tun)
2208                 return -EBADFD;
2209         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2210         ret = min_t(ssize_t, ret, len);
2211         if (ret > 0)
2212                 iocb->ki_pos = ret;
2213         tun_put(tun);
2214         return ret;
2215 }
2216
2217 static void tun_prog_free(struct rcu_head *rcu)
2218 {
2219         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2220
2221         bpf_prog_destroy(prog->prog);
2222         kfree(prog);
2223 }
2224
2225 static int __tun_set_ebpf(struct tun_struct *tun,
2226                           struct tun_prog __rcu **prog_p,
2227                           struct bpf_prog *prog)
2228 {
2229         struct tun_prog *old, *new = NULL;
2230
2231         if (prog) {
2232                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2233                 if (!new)
2234                         return -ENOMEM;
2235                 new->prog = prog;
2236         }
2237
2238         spin_lock_bh(&tun->lock);
2239         old = rcu_dereference_protected(*prog_p,
2240                                         lockdep_is_held(&tun->lock));
2241         rcu_assign_pointer(*prog_p, new);
2242         spin_unlock_bh(&tun->lock);
2243
2244         if (old)
2245                 call_rcu(&old->rcu, tun_prog_free);
2246
2247         return 0;
2248 }
2249
2250 static void tun_free_netdev(struct net_device *dev)
2251 {
2252         struct tun_struct *tun = netdev_priv(dev);
2253
2254         BUG_ON(!(list_empty(&tun->disabled)));
2255
2256         free_percpu(tun->pcpu_stats);
2257         /* We clear pcpu_stats so that tun_set_iff() can tell if
2258          * tun_free_netdev() has been called from register_netdevice().
2259          */
2260         tun->pcpu_stats = NULL;
2261
2262         tun_flow_uninit(tun);
2263         security_tun_dev_free_security(tun->security);
2264         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2265         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2266 }
2267
2268 static void tun_setup(struct net_device *dev)
2269 {
2270         struct tun_struct *tun = netdev_priv(dev);
2271
2272         tun->owner = INVALID_UID;
2273         tun->group = INVALID_GID;
2274         tun_default_link_ksettings(dev, &tun->link_ksettings);
2275
2276         dev->ethtool_ops = &tun_ethtool_ops;
2277         dev->needs_free_netdev = true;
2278         dev->priv_destructor = tun_free_netdev;
2279         /* We prefer our own queue length */
2280         dev->tx_queue_len = TUN_READQ_SIZE;
2281 }
2282
2283 /* Trivial set of netlink ops to allow deleting tun or tap
2284  * device with netlink.
2285  */
2286 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2287                         struct netlink_ext_ack *extack)
2288 {
2289         NL_SET_ERR_MSG(extack,
2290                        "tun/tap creation via rtnetlink is not supported.");
2291         return -EOPNOTSUPP;
2292 }
2293
2294 static size_t tun_get_size(const struct net_device *dev)
2295 {
2296         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2297         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2298
2299         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2300                nla_total_size(sizeof(gid_t)) + /* GROUP */
2301                nla_total_size(sizeof(u8)) + /* TYPE */
2302                nla_total_size(sizeof(u8)) + /* PI */
2303                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2304                nla_total_size(sizeof(u8)) + /* PERSIST */
2305                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2306                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2307                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2308                0;
2309 }
2310
2311 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2312 {
2313         struct tun_struct *tun = netdev_priv(dev);
2314
2315         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2316                 goto nla_put_failure;
2317         if (uid_valid(tun->owner) &&
2318             nla_put_u32(skb, IFLA_TUN_OWNER,
2319                         from_kuid_munged(current_user_ns(), tun->owner)))
2320                 goto nla_put_failure;
2321         if (gid_valid(tun->group) &&
2322             nla_put_u32(skb, IFLA_TUN_GROUP,
2323                         from_kgid_munged(current_user_ns(), tun->group)))
2324                 goto nla_put_failure;
2325         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2326                 goto nla_put_failure;
2327         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2328                 goto nla_put_failure;
2329         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2330                 goto nla_put_failure;
2331         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2332                        !!(tun->flags & IFF_MULTI_QUEUE)))
2333                 goto nla_put_failure;
2334         if (tun->flags & IFF_MULTI_QUEUE) {
2335                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2336                         goto nla_put_failure;
2337                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2338                                 tun->numdisabled))
2339                         goto nla_put_failure;
2340         }
2341
2342         return 0;
2343
2344 nla_put_failure:
2345         return -EMSGSIZE;
2346 }
2347
2348 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2349         .kind           = DRV_NAME,
2350         .priv_size      = sizeof(struct tun_struct),
2351         .setup          = tun_setup,
2352         .validate       = tun_validate,
2353         .get_size       = tun_get_size,
2354         .fill_info      = tun_fill_info,
2355 };
2356
2357 static void tun_sock_write_space(struct sock *sk)
2358 {
2359         struct tun_file *tfile;
2360         wait_queue_head_t *wqueue;
2361
2362         if (!sock_writeable(sk))
2363                 return;
2364
2365         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2366                 return;
2367
2368         wqueue = sk_sleep(sk);
2369         if (wqueue && waitqueue_active(wqueue))
2370                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2371                                                 EPOLLWRNORM | EPOLLWRBAND);
2372
2373         tfile = container_of(sk, struct tun_file, sk);
2374         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2375 }
2376
2377 static void tun_put_page(struct tun_page *tpage)
2378 {
2379         if (tpage->page)
2380                 __page_frag_cache_drain(tpage->page, tpage->count);
2381 }
2382
2383 static int tun_xdp_one(struct tun_struct *tun,
2384                        struct tun_file *tfile,
2385                        struct xdp_buff *xdp, int *flush,
2386                        struct tun_page *tpage)
2387 {
2388         unsigned int datasize = xdp->data_end - xdp->data;
2389         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2390         struct virtio_net_hdr *gso = &hdr->gso;
2391         struct tun_pcpu_stats *stats;
2392         struct bpf_prog *xdp_prog;
2393         struct sk_buff *skb = NULL;
2394         u32 rxhash = 0, act;
2395         int buflen = hdr->buflen;
2396         int err = 0;
2397         bool skb_xdp = false;
2398         struct page *page;
2399
2400         xdp_prog = rcu_dereference(tun->xdp_prog);
2401         if (xdp_prog) {
2402                 if (gso->gso_type) {
2403                         skb_xdp = true;
2404                         goto build;
2405                 }
2406                 xdp_set_data_meta_invalid(xdp);
2407                 xdp->rxq = &tfile->xdp_rxq;
2408                 xdp->frame_sz = buflen;
2409
2410                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2411                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2412                 if (err < 0) {
2413                         put_page(virt_to_head_page(xdp->data));
2414                         return err;
2415                 }
2416
2417                 switch (err) {
2418                 case XDP_REDIRECT:
2419                         *flush = true;
2420                         fallthrough;
2421                 case XDP_TX:
2422                         return 0;
2423                 case XDP_PASS:
2424                         break;
2425                 default:
2426                         page = virt_to_head_page(xdp->data);
2427                         if (tpage->page == page) {
2428                                 ++tpage->count;
2429                         } else {
2430                                 tun_put_page(tpage);
2431                                 tpage->page = page;
2432                                 tpage->count = 1;
2433                         }
2434                         return 0;
2435                 }
2436         }
2437
2438 build:
2439         skb = build_skb(xdp->data_hard_start, buflen);
2440         if (!skb) {
2441                 err = -ENOMEM;
2442                 goto out;
2443         }
2444
2445         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2446         skb_put(skb, xdp->data_end - xdp->data);
2447
2448         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2449                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2450                 kfree_skb(skb);
2451                 err = -EINVAL;
2452                 goto out;
2453         }
2454
2455         skb->protocol = eth_type_trans(skb, tun->dev);
2456         skb_reset_network_header(skb);
2457         skb_probe_transport_header(skb);
2458         skb_record_rx_queue(skb, tfile->queue_index);
2459
2460         if (skb_xdp) {
2461                 err = do_xdp_generic(xdp_prog, skb);
2462                 if (err != XDP_PASS)
2463                         goto out;
2464         }
2465
2466         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2467             !tfile->detached)
2468                 rxhash = __skb_get_hash_symmetric(skb);
2469
2470         netif_receive_skb(skb);
2471
2472         /* No need for get_cpu_ptr() here since this function is
2473          * always called with bh disabled
2474          */
2475         stats = this_cpu_ptr(tun->pcpu_stats);
2476         u64_stats_update_begin(&stats->syncp);
2477         u64_stats_inc(&stats->rx_packets);
2478         u64_stats_add(&stats->rx_bytes, datasize);
2479         u64_stats_update_end(&stats->syncp);
2480
2481         if (rxhash)
2482                 tun_flow_update(tun, rxhash, tfile);
2483
2484 out:
2485         return err;
2486 }
2487
2488 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2489 {
2490         int ret, i;
2491         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2492         struct tun_struct *tun = tun_get(tfile);
2493         struct tun_msg_ctl *ctl = m->msg_control;
2494         struct xdp_buff *xdp;
2495
2496         if (!tun)
2497                 return -EBADFD;
2498
2499         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2500                 struct tun_page tpage;
2501                 int n = ctl->num;
2502                 int flush = 0;
2503
2504                 memset(&tpage, 0, sizeof(tpage));
2505
2506                 local_bh_disable();
2507                 rcu_read_lock();
2508
2509                 for (i = 0; i < n; i++) {
2510                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2511                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2512                 }
2513
2514                 if (flush)
2515                         xdp_do_flush();
2516
2517                 rcu_read_unlock();
2518                 local_bh_enable();
2519
2520                 tun_put_page(&tpage);
2521
2522                 ret = total_len;
2523                 goto out;
2524         }
2525
2526         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2527                            m->msg_flags & MSG_DONTWAIT,
2528                            m->msg_flags & MSG_MORE);
2529 out:
2530         tun_put(tun);
2531         return ret;
2532 }
2533
2534 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2535                        int flags)
2536 {
2537         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2538         struct tun_struct *tun = tun_get(tfile);
2539         void *ptr = m->msg_control;
2540         int ret;
2541
2542         if (!tun) {
2543                 ret = -EBADFD;
2544                 goto out_free;
2545         }
2546
2547         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2548                 ret = -EINVAL;
2549                 goto out_put_tun;
2550         }
2551         if (flags & MSG_ERRQUEUE) {
2552                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2553                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2554                 goto out;
2555         }
2556         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2557         if (ret > (ssize_t)total_len) {
2558                 m->msg_flags |= MSG_TRUNC;
2559                 ret = flags & MSG_TRUNC ? ret : total_len;
2560         }
2561 out:
2562         tun_put(tun);
2563         return ret;
2564
2565 out_put_tun:
2566         tun_put(tun);
2567 out_free:
2568         tun_ptr_free(ptr);
2569         return ret;
2570 }
2571
2572 static int tun_ptr_peek_len(void *ptr)
2573 {
2574         if (likely(ptr)) {
2575                 if (tun_is_xdp_frame(ptr)) {
2576                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2577
2578                         return xdpf->len;
2579                 }
2580                 return __skb_array_len_with_tag(ptr);
2581         } else {
2582                 return 0;
2583         }
2584 }
2585
2586 static int tun_peek_len(struct socket *sock)
2587 {
2588         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2589         struct tun_struct *tun;
2590         int ret = 0;
2591
2592         tun = tun_get(tfile);
2593         if (!tun)
2594                 return 0;
2595
2596         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2597         tun_put(tun);
2598
2599         return ret;
2600 }
2601
2602 /* Ops structure to mimic raw sockets with tun */
2603 static const struct proto_ops tun_socket_ops = {
2604         .peek_len = tun_peek_len,
2605         .sendmsg = tun_sendmsg,
2606         .recvmsg = tun_recvmsg,
2607 };
2608
2609 static struct proto tun_proto = {
2610         .name           = "tun",
2611         .owner          = THIS_MODULE,
2612         .obj_size       = sizeof(struct tun_file),
2613 };
2614
2615 static int tun_flags(struct tun_struct *tun)
2616 {
2617         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2618 }
2619
2620 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2621                               char *buf)
2622 {
2623         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2624         return sprintf(buf, "0x%x\n", tun_flags(tun));
2625 }
2626
2627 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2628                               char *buf)
2629 {
2630         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2631         return uid_valid(tun->owner)?
2632                 sprintf(buf, "%u\n",
2633                         from_kuid_munged(current_user_ns(), tun->owner)):
2634                 sprintf(buf, "-1\n");
2635 }
2636
2637 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2638                               char *buf)
2639 {
2640         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2641         return gid_valid(tun->group) ?
2642                 sprintf(buf, "%u\n",
2643                         from_kgid_munged(current_user_ns(), tun->group)):
2644                 sprintf(buf, "-1\n");
2645 }
2646
2647 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2648 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2649 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2650
2651 static struct attribute *tun_dev_attrs[] = {
2652         &dev_attr_tun_flags.attr,
2653         &dev_attr_owner.attr,
2654         &dev_attr_group.attr,
2655         NULL
2656 };
2657
2658 static const struct attribute_group tun_attr_group = {
2659         .attrs = tun_dev_attrs
2660 };
2661
2662 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2663 {
2664         struct tun_struct *tun;
2665         struct tun_file *tfile = file->private_data;
2666         struct net_device *dev;
2667         int err;
2668
2669         if (tfile->detached)
2670                 return -EINVAL;
2671
2672         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2673                 if (!capable(CAP_NET_ADMIN))
2674                         return -EPERM;
2675
2676                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2677                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2678                         return -EINVAL;
2679         }
2680
2681         dev = __dev_get_by_name(net, ifr->ifr_name);
2682         if (dev) {
2683                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2684                         return -EBUSY;
2685                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2686                         tun = netdev_priv(dev);
2687                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2688                         tun = netdev_priv(dev);
2689                 else
2690                         return -EINVAL;
2691
2692                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2693                     !!(tun->flags & IFF_MULTI_QUEUE))
2694                         return -EINVAL;
2695
2696                 if (tun_not_capable(tun))
2697                         return -EPERM;
2698                 err = security_tun_dev_open(tun->security);
2699                 if (err < 0)
2700                         return err;
2701
2702                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2703                                  ifr->ifr_flags & IFF_NAPI,
2704                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2705                 if (err < 0)
2706                         return err;
2707
2708                 if (tun->flags & IFF_MULTI_QUEUE &&
2709                     (tun->numqueues + tun->numdisabled > 1)) {
2710                         /* One or more queue has already been attached, no need
2711                          * to initialize the device again.
2712                          */
2713                         netdev_state_change(dev);
2714                         return 0;
2715                 }
2716
2717                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2718                               (ifr->ifr_flags & TUN_FEATURES);
2719
2720                 netdev_state_change(dev);
2721         } else {
2722                 char *name;
2723                 unsigned long flags = 0;
2724                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2725                              MAX_TAP_QUEUES : 1;
2726
2727                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2728                         return -EPERM;
2729                 err = security_tun_dev_create();
2730                 if (err < 0)
2731                         return err;
2732
2733                 /* Set dev type */
2734                 if (ifr->ifr_flags & IFF_TUN) {
2735                         /* TUN device */
2736                         flags |= IFF_TUN;
2737                         name = "tun%d";
2738                 } else if (ifr->ifr_flags & IFF_TAP) {
2739                         /* TAP device */
2740                         flags |= IFF_TAP;
2741                         name = "tap%d";
2742                 } else
2743                         return -EINVAL;
2744
2745                 if (*ifr->ifr_name)
2746                         name = ifr->ifr_name;
2747
2748                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2749                                        NET_NAME_UNKNOWN, tun_setup, queues,
2750                                        queues);
2751
2752                 if (!dev)
2753                         return -ENOMEM;
2754
2755                 dev_net_set(dev, net);
2756                 dev->rtnl_link_ops = &tun_link_ops;
2757                 dev->ifindex = tfile->ifindex;
2758                 dev->sysfs_groups[0] = &tun_attr_group;
2759
2760                 tun = netdev_priv(dev);
2761                 tun->dev = dev;
2762                 tun->flags = flags;
2763                 tun->txflt.count = 0;
2764                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2765
2766                 tun->align = NET_SKB_PAD;
2767                 tun->filter_attached = false;
2768                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2769                 tun->rx_batched = 0;
2770                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2771
2772                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2773                 if (!tun->pcpu_stats) {
2774                         err = -ENOMEM;
2775                         goto err_free_dev;
2776                 }
2777
2778                 spin_lock_init(&tun->lock);
2779
2780                 err = security_tun_dev_alloc_security(&tun->security);
2781                 if (err < 0)
2782                         goto err_free_stat;
2783
2784                 tun_net_init(dev);
2785                 tun_flow_init(tun);
2786
2787                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2788                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2789                                    NETIF_F_HW_VLAN_STAG_TX;
2790                 dev->features = dev->hw_features | NETIF_F_LLTX;
2791                 dev->vlan_features = dev->features &
2792                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2793                                        NETIF_F_HW_VLAN_STAG_TX);
2794
2795                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2796                               (ifr->ifr_flags & TUN_FEATURES);
2797
2798                 INIT_LIST_HEAD(&tun->disabled);
2799                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2800                                  ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2801                 if (err < 0)
2802                         goto err_free_flow;
2803
2804                 err = register_netdevice(tun->dev);
2805                 if (err < 0)
2806                         goto err_detach;
2807                 /* free_netdev() won't check refcnt, to aovid race
2808                  * with dev_put() we need publish tun after registration.
2809                  */
2810                 rcu_assign_pointer(tfile->tun, tun);
2811         }
2812
2813         netif_carrier_on(tun->dev);
2814
2815         /* Make sure persistent devices do not get stuck in
2816          * xoff state.
2817          */
2818         if (netif_running(tun->dev))
2819                 netif_tx_wake_all_queues(tun->dev);
2820
2821         strcpy(ifr->ifr_name, tun->dev->name);
2822         return 0;
2823
2824 err_detach:
2825         tun_detach_all(dev);
2826         /* We are here because register_netdevice() has failed.
2827          * If register_netdevice() already called tun_free_netdev()
2828          * while dealing with the error, tun->pcpu_stats has been cleared.
2829          */
2830         if (!tun->pcpu_stats)
2831                 goto err_free_dev;
2832
2833 err_free_flow:
2834         tun_flow_uninit(tun);
2835         security_tun_dev_free_security(tun->security);
2836 err_free_stat:
2837         free_percpu(tun->pcpu_stats);
2838 err_free_dev:
2839         free_netdev(dev);
2840         return err;
2841 }
2842
2843 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2844 {
2845         strcpy(ifr->ifr_name, tun->dev->name);
2846
2847         ifr->ifr_flags = tun_flags(tun);
2848
2849 }
2850
2851 /* This is like a cut-down ethtool ops, except done via tun fd so no
2852  * privs required. */
2853 static int set_offload(struct tun_struct *tun, unsigned long arg)
2854 {
2855         netdev_features_t features = 0;
2856
2857         if (arg & TUN_F_CSUM) {
2858                 features |= NETIF_F_HW_CSUM;
2859                 arg &= ~TUN_F_CSUM;
2860
2861                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2862                         if (arg & TUN_F_TSO_ECN) {
2863                                 features |= NETIF_F_TSO_ECN;
2864                                 arg &= ~TUN_F_TSO_ECN;
2865                         }
2866                         if (arg & TUN_F_TSO4)
2867                                 features |= NETIF_F_TSO;
2868                         if (arg & TUN_F_TSO6)
2869                                 features |= NETIF_F_TSO6;
2870                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2871                 }
2872
2873                 arg &= ~TUN_F_UFO;
2874         }
2875
2876         /* This gives the user a way to test for new features in future by
2877          * trying to set them. */
2878         if (arg)
2879                 return -EINVAL;
2880
2881         tun->set_features = features;
2882         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2883         tun->dev->wanted_features |= features;
2884         netdev_update_features(tun->dev);
2885
2886         return 0;
2887 }
2888
2889 static void tun_detach_filter(struct tun_struct *tun, int n)
2890 {
2891         int i;
2892         struct tun_file *tfile;
2893
2894         for (i = 0; i < n; i++) {
2895                 tfile = rtnl_dereference(tun->tfiles[i]);
2896                 lock_sock(tfile->socket.sk);
2897                 sk_detach_filter(tfile->socket.sk);
2898                 release_sock(tfile->socket.sk);
2899         }
2900
2901         tun->filter_attached = false;
2902 }
2903
2904 static int tun_attach_filter(struct tun_struct *tun)
2905 {
2906         int i, ret = 0;
2907         struct tun_file *tfile;
2908
2909         for (i = 0; i < tun->numqueues; i++) {
2910                 tfile = rtnl_dereference(tun->tfiles[i]);
2911                 lock_sock(tfile->socket.sk);
2912                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2913                 release_sock(tfile->socket.sk);
2914                 if (ret) {
2915                         tun_detach_filter(tun, i);
2916                         return ret;
2917                 }
2918         }
2919
2920         tun->filter_attached = true;
2921         return ret;
2922 }
2923
2924 static void tun_set_sndbuf(struct tun_struct *tun)
2925 {
2926         struct tun_file *tfile;
2927         int i;
2928
2929         for (i = 0; i < tun->numqueues; i++) {
2930                 tfile = rtnl_dereference(tun->tfiles[i]);
2931                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2932         }
2933 }
2934
2935 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2936 {
2937         struct tun_file *tfile = file->private_data;
2938         struct tun_struct *tun;
2939         int ret = 0;
2940
2941         rtnl_lock();
2942
2943         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2944                 tun = tfile->detached;
2945                 if (!tun) {
2946                         ret = -EINVAL;
2947                         goto unlock;
2948                 }
2949                 ret = security_tun_dev_attach_queue(tun->security);
2950                 if (ret < 0)
2951                         goto unlock;
2952                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2953                                  tun->flags & IFF_NAPI_FRAGS, true);
2954         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2955                 tun = rtnl_dereference(tfile->tun);
2956                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2957                         ret = -EINVAL;
2958                 else
2959                         __tun_detach(tfile, false);
2960         } else
2961                 ret = -EINVAL;
2962
2963         if (ret >= 0)
2964                 netdev_state_change(tun->dev);
2965
2966 unlock:
2967         rtnl_unlock();
2968         return ret;
2969 }
2970
2971 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2972                         void __user *data)
2973 {
2974         struct bpf_prog *prog;
2975         int fd;
2976
2977         if (copy_from_user(&fd, data, sizeof(fd)))
2978                 return -EFAULT;
2979
2980         if (fd == -1) {
2981                 prog = NULL;
2982         } else {
2983                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2984                 if (IS_ERR(prog))
2985                         return PTR_ERR(prog);
2986         }
2987
2988         return __tun_set_ebpf(tun, prog_p, prog);
2989 }
2990
2991 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2992                             unsigned long arg, int ifreq_len)
2993 {
2994         struct tun_file *tfile = file->private_data;
2995         struct net *net = sock_net(&tfile->sk);
2996         struct tun_struct *tun;
2997         void __user* argp = (void __user*)arg;
2998         unsigned int ifindex, carrier;
2999         struct ifreq ifr;
3000         kuid_t owner;
3001         kgid_t group;
3002         int sndbuf;
3003         int vnet_hdr_sz;
3004         int le;
3005         int ret;
3006         bool do_notify = false;
3007
3008         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3009             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3010                 if (copy_from_user(&ifr, argp, ifreq_len))
3011                         return -EFAULT;
3012         } else {
3013                 memset(&ifr, 0, sizeof(ifr));
3014         }
3015         if (cmd == TUNGETFEATURES) {
3016                 /* Currently this just means: "what IFF flags are valid?".
3017                  * This is needed because we never checked for invalid flags on
3018                  * TUNSETIFF.
3019                  */
3020                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3021                                 (unsigned int __user*)argp);
3022         } else if (cmd == TUNSETQUEUE) {
3023                 return tun_set_queue(file, &ifr);
3024         } else if (cmd == SIOCGSKNS) {
3025                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3026                         return -EPERM;
3027                 return open_related_ns(&net->ns, get_net_ns);
3028         }
3029
3030         ret = 0;
3031         rtnl_lock();
3032
3033         tun = tun_get(tfile);
3034         if (cmd == TUNSETIFF) {
3035                 ret = -EEXIST;
3036                 if (tun)
3037                         goto unlock;
3038
3039                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3040
3041                 ret = tun_set_iff(net, file, &ifr);
3042
3043                 if (ret)
3044                         goto unlock;
3045
3046                 if (copy_to_user(argp, &ifr, ifreq_len))
3047                         ret = -EFAULT;
3048                 goto unlock;
3049         }
3050         if (cmd == TUNSETIFINDEX) {
3051                 ret = -EPERM;
3052                 if (tun)
3053                         goto unlock;
3054
3055                 ret = -EFAULT;
3056                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3057                         goto unlock;
3058
3059                 ret = 0;
3060                 tfile->ifindex = ifindex;
3061                 goto unlock;
3062         }
3063
3064         ret = -EBADFD;
3065         if (!tun)
3066                 goto unlock;
3067
3068         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3069
3070         net = dev_net(tun->dev);
3071         ret = 0;
3072         switch (cmd) {
3073         case TUNGETIFF:
3074                 tun_get_iff(tun, &ifr);
3075
3076                 if (tfile->detached)
3077                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3078                 if (!tfile->socket.sk->sk_filter)
3079                         ifr.ifr_flags |= IFF_NOFILTER;
3080
3081                 if (copy_to_user(argp, &ifr, ifreq_len))
3082                         ret = -EFAULT;
3083                 break;
3084
3085         case TUNSETNOCSUM:
3086                 /* Disable/Enable checksum */
3087
3088                 /* [unimplemented] */
3089                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3090                            arg ? "disabled" : "enabled");
3091                 break;
3092
3093         case TUNSETPERSIST:
3094                 /* Disable/Enable persist mode. Keep an extra reference to the
3095                  * module to prevent the module being unprobed.
3096                  */
3097                 if (arg && !(tun->flags & IFF_PERSIST)) {
3098                         tun->flags |= IFF_PERSIST;
3099                         __module_get(THIS_MODULE);
3100                         do_notify = true;
3101                 }
3102                 if (!arg && (tun->flags & IFF_PERSIST)) {
3103                         tun->flags &= ~IFF_PERSIST;
3104                         module_put(THIS_MODULE);
3105                         do_notify = true;
3106                 }
3107
3108                 netif_info(tun, drv, tun->dev, "persist %s\n",
3109                            arg ? "enabled" : "disabled");
3110                 break;
3111
3112         case TUNSETOWNER:
3113                 /* Set owner of the device */
3114                 owner = make_kuid(current_user_ns(), arg);
3115                 if (!uid_valid(owner)) {
3116                         ret = -EINVAL;
3117                         break;
3118                 }
3119                 tun->owner = owner;
3120                 do_notify = true;
3121                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3122                            from_kuid(&init_user_ns, tun->owner));
3123                 break;
3124
3125         case TUNSETGROUP:
3126                 /* Set group of the device */
3127                 group = make_kgid(current_user_ns(), arg);
3128                 if (!gid_valid(group)) {
3129                         ret = -EINVAL;
3130                         break;
3131                 }
3132                 tun->group = group;
3133                 do_notify = true;
3134                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3135                            from_kgid(&init_user_ns, tun->group));
3136                 break;
3137
3138         case TUNSETLINK:
3139                 /* Only allow setting the type when the interface is down */
3140                 if (tun->dev->flags & IFF_UP) {
3141                         netif_info(tun, drv, tun->dev,
3142                                    "Linktype set failed because interface is up\n");
3143                         ret = -EBUSY;
3144                 } else {
3145                         tun->dev->type = (int) arg;
3146                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3147                                    tun->dev->type);
3148                         ret = 0;
3149                 }
3150                 break;
3151
3152         case TUNSETDEBUG:
3153                 tun->msg_enable = (u32)arg;
3154                 break;
3155
3156         case TUNSETOFFLOAD:
3157                 ret = set_offload(tun, arg);
3158                 break;
3159
3160         case TUNSETTXFILTER:
3161                 /* Can be set only for TAPs */
3162                 ret = -EINVAL;
3163                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3164                         break;
3165                 ret = update_filter(&tun->txflt, (void __user *)arg);
3166                 break;
3167
3168         case SIOCGIFHWADDR:
3169                 /* Get hw address */
3170                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3171                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3172                 if (copy_to_user(argp, &ifr, ifreq_len))
3173                         ret = -EFAULT;
3174                 break;
3175
3176         case SIOCSIFHWADDR:
3177                 /* Set hw address */
3178                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3179                 break;
3180
3181         case TUNGETSNDBUF:
3182                 sndbuf = tfile->socket.sk->sk_sndbuf;
3183                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3184                         ret = -EFAULT;
3185                 break;
3186
3187         case TUNSETSNDBUF:
3188                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3189                         ret = -EFAULT;
3190                         break;
3191                 }
3192                 if (sndbuf <= 0) {
3193                         ret = -EINVAL;
3194                         break;
3195                 }
3196
3197                 tun->sndbuf = sndbuf;
3198                 tun_set_sndbuf(tun);
3199                 break;
3200
3201         case TUNGETVNETHDRSZ:
3202                 vnet_hdr_sz = tun->vnet_hdr_sz;
3203                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3204                         ret = -EFAULT;
3205                 break;
3206
3207         case TUNSETVNETHDRSZ:
3208                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3209                         ret = -EFAULT;
3210                         break;
3211                 }
3212                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3213                         ret = -EINVAL;
3214                         break;
3215                 }
3216
3217                 tun->vnet_hdr_sz = vnet_hdr_sz;
3218                 break;
3219
3220         case TUNGETVNETLE:
3221                 le = !!(tun->flags & TUN_VNET_LE);
3222                 if (put_user(le, (int __user *)argp))
3223                         ret = -EFAULT;
3224                 break;
3225
3226         case TUNSETVNETLE:
3227                 if (get_user(le, (int __user *)argp)) {
3228                         ret = -EFAULT;
3229                         break;
3230                 }
3231                 if (le)
3232                         tun->flags |= TUN_VNET_LE;
3233                 else
3234                         tun->flags &= ~TUN_VNET_LE;
3235                 break;
3236
3237         case TUNGETVNETBE:
3238                 ret = tun_get_vnet_be(tun, argp);
3239                 break;
3240
3241         case TUNSETVNETBE:
3242                 ret = tun_set_vnet_be(tun, argp);
3243                 break;
3244
3245         case TUNATTACHFILTER:
3246                 /* Can be set only for TAPs */
3247                 ret = -EINVAL;
3248                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3249                         break;
3250                 ret = -EFAULT;
3251                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3252                         break;
3253
3254                 ret = tun_attach_filter(tun);
3255                 break;
3256
3257         case TUNDETACHFILTER:
3258                 /* Can be set only for TAPs */
3259                 ret = -EINVAL;
3260                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3261                         break;
3262                 ret = 0;
3263                 tun_detach_filter(tun, tun->numqueues);
3264                 break;
3265
3266         case TUNGETFILTER:
3267                 ret = -EINVAL;
3268                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3269                         break;
3270                 ret = -EFAULT;
3271                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3272                         break;
3273                 ret = 0;
3274                 break;
3275
3276         case TUNSETSTEERINGEBPF:
3277                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3278                 break;
3279
3280         case TUNSETFILTEREBPF:
3281                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3282                 break;
3283
3284         case TUNSETCARRIER:
3285                 ret = -EFAULT;
3286                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3287                         goto unlock;
3288
3289                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3290                 break;
3291
3292         case TUNGETDEVNETNS:
3293                 ret = -EPERM;
3294                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3295                         goto unlock;
3296                 ret = open_related_ns(&net->ns, get_net_ns);
3297                 break;
3298
3299         default:
3300                 ret = -EINVAL;
3301                 break;
3302         }
3303
3304         if (do_notify)
3305                 netdev_state_change(tun->dev);
3306
3307 unlock:
3308         rtnl_unlock();
3309         if (tun)
3310                 tun_put(tun);
3311         return ret;
3312 }
3313
3314 static long tun_chr_ioctl(struct file *file,
3315                           unsigned int cmd, unsigned long arg)
3316 {
3317         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3318 }
3319
3320 #ifdef CONFIG_COMPAT
3321 static long tun_chr_compat_ioctl(struct file *file,
3322                          unsigned int cmd, unsigned long arg)
3323 {
3324         switch (cmd) {
3325         case TUNSETIFF:
3326         case TUNGETIFF:
3327         case TUNSETTXFILTER:
3328         case TUNGETSNDBUF:
3329         case TUNSETSNDBUF:
3330         case SIOCGIFHWADDR:
3331         case SIOCSIFHWADDR:
3332                 arg = (unsigned long)compat_ptr(arg);
3333                 break;
3334         default:
3335                 arg = (compat_ulong_t)arg;
3336                 break;
3337         }
3338
3339         /*
3340          * compat_ifreq is shorter than ifreq, so we must not access beyond
3341          * the end of that structure. All fields that are used in this
3342          * driver are compatible though, we don't need to convert the
3343          * contents.
3344          */
3345         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3346 }
3347 #endif /* CONFIG_COMPAT */
3348
3349 static int tun_chr_fasync(int fd, struct file *file, int on)
3350 {
3351         struct tun_file *tfile = file->private_data;
3352         int ret;
3353
3354         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3355                 goto out;
3356
3357         if (on) {
3358                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3359                 tfile->flags |= TUN_FASYNC;
3360         } else
3361                 tfile->flags &= ~TUN_FASYNC;
3362         ret = 0;
3363 out:
3364         return ret;
3365 }
3366
3367 static int tun_chr_open(struct inode *inode, struct file * file)
3368 {
3369         struct net *net = current->nsproxy->net_ns;
3370         struct tun_file *tfile;
3371
3372         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3373                                             &tun_proto, 0);
3374         if (!tfile)
3375                 return -ENOMEM;
3376         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3377                 sk_free(&tfile->sk);
3378                 return -ENOMEM;
3379         }
3380
3381         mutex_init(&tfile->napi_mutex);
3382         RCU_INIT_POINTER(tfile->tun, NULL);
3383         tfile->flags = 0;
3384         tfile->ifindex = 0;
3385
3386         init_waitqueue_head(&tfile->socket.wq.wait);
3387
3388         tfile->socket.file = file;
3389         tfile->socket.ops = &tun_socket_ops;
3390
3391         sock_init_data(&tfile->socket, &tfile->sk);
3392
3393         tfile->sk.sk_write_space = tun_sock_write_space;
3394         tfile->sk.sk_sndbuf = INT_MAX;
3395
3396         file->private_data = tfile;
3397         INIT_LIST_HEAD(&tfile->next);
3398
3399         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3400
3401         return 0;
3402 }
3403
3404 static int tun_chr_close(struct inode *inode, struct file *file)
3405 {
3406         struct tun_file *tfile = file->private_data;
3407
3408         tun_detach(tfile, true);
3409
3410         return 0;
3411 }
3412
3413 #ifdef CONFIG_PROC_FS
3414 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3415 {
3416         struct tun_file *tfile = file->private_data;
3417         struct tun_struct *tun;
3418         struct ifreq ifr;
3419
3420         memset(&ifr, 0, sizeof(ifr));
3421
3422         rtnl_lock();
3423         tun = tun_get(tfile);
3424         if (tun)
3425                 tun_get_iff(tun, &ifr);
3426         rtnl_unlock();
3427
3428         if (tun)
3429                 tun_put(tun);
3430
3431         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3432 }
3433 #endif
3434
3435 static const struct file_operations tun_fops = {
3436         .owner  = THIS_MODULE,
3437         .llseek = no_llseek,
3438         .read_iter  = tun_chr_read_iter,
3439         .write_iter = tun_chr_write_iter,
3440         .poll   = tun_chr_poll,
3441         .unlocked_ioctl = tun_chr_ioctl,
3442 #ifdef CONFIG_COMPAT
3443         .compat_ioctl = tun_chr_compat_ioctl,
3444 #endif
3445         .open   = tun_chr_open,
3446         .release = tun_chr_close,
3447         .fasync = tun_chr_fasync,
3448 #ifdef CONFIG_PROC_FS
3449         .show_fdinfo = tun_chr_show_fdinfo,
3450 #endif
3451 };
3452
3453 static struct miscdevice tun_miscdev = {
3454         .minor = TUN_MINOR,
3455         .name = "tun",
3456         .nodename = "net/tun",
3457         .fops = &tun_fops,
3458 };
3459
3460 /* ethtool interface */
3461
3462 static void tun_default_link_ksettings(struct net_device *dev,
3463                                        struct ethtool_link_ksettings *cmd)
3464 {
3465         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3466         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3467         cmd->base.speed         = SPEED_10;
3468         cmd->base.duplex        = DUPLEX_FULL;
3469         cmd->base.port          = PORT_TP;
3470         cmd->base.phy_address   = 0;
3471         cmd->base.autoneg       = AUTONEG_DISABLE;
3472 }
3473
3474 static int tun_get_link_ksettings(struct net_device *dev,
3475                                   struct ethtool_link_ksettings *cmd)
3476 {
3477         struct tun_struct *tun = netdev_priv(dev);
3478
3479         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3480         return 0;
3481 }
3482
3483 static int tun_set_link_ksettings(struct net_device *dev,
3484                                   const struct ethtool_link_ksettings *cmd)
3485 {
3486         struct tun_struct *tun = netdev_priv(dev);
3487
3488         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3489         return 0;
3490 }
3491
3492 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3493 {
3494         struct tun_struct *tun = netdev_priv(dev);
3495
3496         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3497         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3498
3499         switch (tun->flags & TUN_TYPE_MASK) {
3500         case IFF_TUN:
3501                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3502                 break;
3503         case IFF_TAP:
3504                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3505                 break;
3506         }
3507 }
3508
3509 static u32 tun_get_msglevel(struct net_device *dev)
3510 {
3511         struct tun_struct *tun = netdev_priv(dev);
3512
3513         return tun->msg_enable;
3514 }
3515
3516 static void tun_set_msglevel(struct net_device *dev, u32 value)
3517 {
3518         struct tun_struct *tun = netdev_priv(dev);
3519
3520         tun->msg_enable = value;
3521 }
3522
3523 static int tun_get_coalesce(struct net_device *dev,
3524                             struct ethtool_coalesce *ec)
3525 {
3526         struct tun_struct *tun = netdev_priv(dev);
3527
3528         ec->rx_max_coalesced_frames = tun->rx_batched;
3529
3530         return 0;
3531 }
3532
3533 static int tun_set_coalesce(struct net_device *dev,
3534                             struct ethtool_coalesce *ec)
3535 {
3536         struct tun_struct *tun = netdev_priv(dev);
3537
3538         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3539                 tun->rx_batched = NAPI_POLL_WEIGHT;
3540         else
3541                 tun->rx_batched = ec->rx_max_coalesced_frames;
3542
3543         return 0;
3544 }
3545
3546 static const struct ethtool_ops tun_ethtool_ops = {
3547         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3548         .get_drvinfo    = tun_get_drvinfo,
3549         .get_msglevel   = tun_get_msglevel,
3550         .set_msglevel   = tun_set_msglevel,
3551         .get_link       = ethtool_op_get_link,
3552         .get_ts_info    = ethtool_op_get_ts_info,
3553         .get_coalesce   = tun_get_coalesce,
3554         .set_coalesce   = tun_set_coalesce,
3555         .get_link_ksettings = tun_get_link_ksettings,
3556         .set_link_ksettings = tun_set_link_ksettings,
3557 };
3558
3559 static int tun_queue_resize(struct tun_struct *tun)
3560 {
3561         struct net_device *dev = tun->dev;
3562         struct tun_file *tfile;
3563         struct ptr_ring **rings;
3564         int n = tun->numqueues + tun->numdisabled;
3565         int ret, i;
3566
3567         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3568         if (!rings)
3569                 return -ENOMEM;
3570
3571         for (i = 0; i < tun->numqueues; i++) {
3572                 tfile = rtnl_dereference(tun->tfiles[i]);
3573                 rings[i] = &tfile->tx_ring;
3574         }
3575         list_for_each_entry(tfile, &tun->disabled, next)
3576                 rings[i++] = &tfile->tx_ring;
3577
3578         ret = ptr_ring_resize_multiple(rings, n,
3579                                        dev->tx_queue_len, GFP_KERNEL,
3580                                        tun_ptr_free);
3581
3582         kfree(rings);
3583         return ret;
3584 }
3585
3586 static int tun_device_event(struct notifier_block *unused,
3587                             unsigned long event, void *ptr)
3588 {
3589         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3590         struct tun_struct *tun = netdev_priv(dev);
3591         int i;
3592
3593         if (dev->rtnl_link_ops != &tun_link_ops)
3594                 return NOTIFY_DONE;
3595
3596         switch (event) {
3597         case NETDEV_CHANGE_TX_QUEUE_LEN:
3598                 if (tun_queue_resize(tun))
3599                         return NOTIFY_BAD;
3600                 break;
3601         case NETDEV_UP:
3602                 for (i = 0; i < tun->numqueues; i++) {
3603                         struct tun_file *tfile;
3604
3605                         tfile = rtnl_dereference(tun->tfiles[i]);
3606                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3607                 }
3608                 break;
3609         default:
3610                 break;
3611         }
3612
3613         return NOTIFY_DONE;
3614 }
3615
3616 static struct notifier_block tun_notifier_block __read_mostly = {
3617         .notifier_call  = tun_device_event,
3618 };
3619
3620 static int __init tun_init(void)
3621 {
3622         int ret = 0;
3623
3624         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3625
3626         ret = rtnl_link_register(&tun_link_ops);
3627         if (ret) {
3628                 pr_err("Can't register link_ops\n");
3629                 goto err_linkops;
3630         }
3631
3632         ret = misc_register(&tun_miscdev);
3633         if (ret) {
3634                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3635                 goto err_misc;
3636         }
3637
3638         ret = register_netdevice_notifier(&tun_notifier_block);
3639         if (ret) {
3640                 pr_err("Can't register netdevice notifier\n");
3641                 goto err_notifier;
3642         }
3643
3644         return  0;
3645
3646 err_notifier:
3647         misc_deregister(&tun_miscdev);
3648 err_misc:
3649         rtnl_link_unregister(&tun_link_ops);
3650 err_linkops:
3651         return ret;
3652 }
3653
3654 static void tun_cleanup(void)
3655 {
3656         misc_deregister(&tun_miscdev);
3657         rtnl_link_unregister(&tun_link_ops);
3658         unregister_netdevice_notifier(&tun_notifier_block);
3659 }
3660
3661 /* Get an underlying socket object from tun file.  Returns error unless file is
3662  * attached to a device.  The returned object works like a packet socket, it
3663  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3664  * holding a reference to the file for as long as the socket is in use. */
3665 struct socket *tun_get_socket(struct file *file)
3666 {
3667         struct tun_file *tfile;
3668         if (file->f_op != &tun_fops)
3669                 return ERR_PTR(-EINVAL);
3670         tfile = file->private_data;
3671         if (!tfile)
3672                 return ERR_PTR(-EBADFD);
3673         return &tfile->socket;
3674 }
3675 EXPORT_SYMBOL_GPL(tun_get_socket);
3676
3677 struct ptr_ring *tun_get_tx_ring(struct file *file)
3678 {
3679         struct tun_file *tfile;
3680
3681         if (file->f_op != &tun_fops)
3682                 return ERR_PTR(-EINVAL);
3683         tfile = file->private_data;
3684         if (!tfile)
3685                 return ERR_PTR(-EBADFD);
3686         return &tfile->tx_ring;
3687 }
3688 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3689
3690 module_init(tun_init);
3691 module_exit(tun_cleanup);
3692 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3693 MODULE_AUTHOR(DRV_COPYRIGHT);
3694 MODULE_LICENSE("GPL");
3695 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3696 MODULE_ALIAS("devname:net/tun");