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 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212         /* init args */
 213         struct file *file;
 214         struct ifreq *ifr;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static void tun_flow_init(struct tun_struct *tun);
 223 static void tun_flow_uninit(struct tun_struct *tun);
 224
 225 static int tun_napi_receive(struct napi_struct *napi, int budget)
 226 {
 227         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 228         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 229         struct sk_buff_head process_queue;
 230         struct sk_buff *skb;
 231         int received = 0;
 232
 233         __skb_queue_head_init(&process_queue);
 234
 235         spin_lock(&queue->lock);
 236         skb_queue_splice_tail_init(queue, &process_queue);
 237         spin_unlock(&queue->lock);
 238
 239         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 240                 napi_gro_receive(napi, skb);
 241                 ++received;
 242         }
 243
 244         if (!skb_queue_empty(&process_queue)) {
 245                 spin_lock(&queue->lock);
 246                 skb_queue_splice(&process_queue, queue);
 247                 spin_unlock(&queue->lock);
 248         }
 249
 250         return received;
 251 }
 252
 253 static int tun_napi_poll(struct napi_struct *napi, int budget)
 254 {
 255         unsigned int received;
 256
 257         received = tun_napi_receive(napi, budget);
 258
 259         if (received < budget)
 260                 napi_complete_done(napi, received);
 261
 262         return received;
 263 }
 264
 265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 266                           bool napi_en, bool napi_frags)
 267 {
 268         tfile->napi_enabled = napi_en;
 269         tfile->napi_frags_enabled = napi_en && napi_frags;
 270         if (napi_en) {
 271                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 272                                   NAPI_POLL_WEIGHT);
 273                 napi_enable(&tfile->napi);
 274         }
 275 }
 276
 277 static void tun_napi_disable(struct tun_file *tfile)
 278 {
 279         if (tfile->napi_enabled)
 280                 napi_disable(&tfile->napi);
 281 }
 282
 283 static void tun_napi_del(struct tun_file *tfile)
 284 {
 285         if (tfile->napi_enabled)
 286                 netif_napi_del(&tfile->napi);
 287 }
 288
 289 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 290 {
 291         return tfile->napi_frags_enabled;
 292 }
 293
 294 #ifdef CONFIG_TUN_VNET_CROSS_LE
 295 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 296 {
 297         return tun->flags & TUN_VNET_BE ? false :
 298                 virtio_legacy_is_little_endian();
 299 }
 300
 301 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 302 {
 303         int be = !!(tun->flags & TUN_VNET_BE);
 304
 305         if (put_user(be, argp))
 306                 return -EFAULT;
 307
 308         return 0;
 309 }
 310
 311 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 312 {
 313         int be;
 314
 315         if (get_user(be, argp))
 316                 return -EFAULT;
 317
 318         if (be)
 319                 tun->flags |= TUN_VNET_BE;
 320         else
 321                 tun->flags &= ~TUN_VNET_BE;
 322
 323         return 0;
 324 }
 325 #else
 326 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 327 {
 328         return virtio_legacy_is_little_endian();
 329 }
 330
 331 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 332 {
 333         return -EINVAL;
 334 }
 335
 336 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 337 {
 338         return -EINVAL;
 339 }
 340 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 341
 342 static inline bool tun_is_little_endian(struct tun_struct *tun)
 343 {
 344         return tun->flags & TUN_VNET_LE ||
 345                 tun_legacy_is_little_endian(tun);
 346 }
 347
 348 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 349 {
 350         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 351 }
 352
 353 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 354 {
 355         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 356 }
 357
 358 static inline u32 tun_hashfn(u32 rxhash)
 359 {
 360         return rxhash & TUN_MASK_FLOW_ENTRIES;
 361 }
 362
 363 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 364 {
 365         struct tun_flow_entry *e;
 366
 367         hlist_for_each_entry_rcu(e, head, hash_link) {
 368                 if (e->rxhash == rxhash)
 369                         return e;
 370         }
 371         return NULL;
 372 }
 373
 374 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 375                                               struct hlist_head *head,
 376                                               u32 rxhash, u16 queue_index)
 377 {
 378         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 379
 380         if (e) {
 381                 netif_info(tun, tx_queued, tun->dev,
 382                            "create flow: hash %u index %u\n",
 383                            rxhash, queue_index);
 384                 e->updated = jiffies;
 385                 e->rxhash = rxhash;
 386                 e->rps_rxhash = 0;
 387                 e->queue_index = queue_index;
 388                 e->tun = tun;
 389                 hlist_add_head_rcu(&e->hash_link, head);
 390                 ++tun->flow_count;
 391         }
 392         return e;
 393 }
 394
 395 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 396 {
 397         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 398                    e->rxhash, e->queue_index);
 399         hlist_del_rcu(&e->hash_link);
 400         kfree_rcu(e, rcu);
 401         --tun->flow_count;
 402 }
 403
 404 static void tun_flow_flush(struct tun_struct *tun)
 405 {
 406         int i;
 407
 408         spin_lock_bh(&tun->lock);
 409         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 410                 struct tun_flow_entry *e;
 411                 struct hlist_node *n;
 412
 413                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 414                         tun_flow_delete(tun, e);
 415         }
 416         spin_unlock_bh(&tun->lock);
 417 }
 418
 419 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 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                         if (e->queue_index == queue_index)
 430                                 tun_flow_delete(tun, e);
 431                 }
 432         }
 433         spin_unlock_bh(&tun->lock);
 434 }
 435
 436 static void tun_flow_cleanup(struct timer_list *t)
 437 {
 438         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 439         unsigned long delay = tun->ageing_time;
 440         unsigned long next_timer = jiffies + delay;
 441         unsigned long count = 0;
 442         int i;
 443
 444         spin_lock(&tun->lock);
 445         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 446                 struct tun_flow_entry *e;
 447                 struct hlist_node *n;
 448
 449                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 450                         unsigned long this_timer;
 451
 452                         this_timer = e->updated + delay;
 453                         if (time_before_eq(this_timer, jiffies)) {
 454                                 tun_flow_delete(tun, e);
 455                                 continue;
 456                         }
 457                         count++;
 458                         if (time_before(this_timer, next_timer))
 459                                 next_timer = this_timer;
 460                 }
 461         }
 462
 463         if (count)
 464                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 465         spin_unlock(&tun->lock);
 466 }
 467
 468 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 469                             struct tun_file *tfile)
 470 {
 471         struct hlist_head *head;
 472         struct tun_flow_entry *e;
 473         unsigned long delay = tun->ageing_time;
 474         u16 queue_index = tfile->queue_index;
 475
 476         head = &tun->flows[tun_hashfn(rxhash)];
 477
 478         rcu_read_lock();
 479
 480         e = tun_flow_find(head, rxhash);
 481         if (likely(e)) {
 482                 /* TODO: keep queueing to old queue until it's empty? */
 483                 if (READ_ONCE(e->queue_index) != queue_index)
 484                         WRITE_ONCE(e->queue_index, queue_index);
 485                 if (e->updated != jiffies)
 486                         e->updated = jiffies;
 487                 sock_rps_record_flow_hash(e->rps_rxhash);
 488         } else {
 489                 spin_lock_bh(&tun->lock);
 490                 if (!tun_flow_find(head, rxhash) &&
 491                     tun->flow_count < MAX_TAP_FLOWS)
 492                         tun_flow_create(tun, head, rxhash, queue_index);
 493
 494                 if (!timer_pending(&tun->flow_gc_timer))
 495                         mod_timer(&tun->flow_gc_timer,
 496                                   round_jiffies_up(jiffies + delay));
 497                 spin_unlock_bh(&tun->lock);
 498         }
 499
 500         rcu_read_unlock();
 501 }
 502
 503 /* Save the hash received in the stack receive path and update the
 504  * flow_hash table accordingly.
 505  */
 506 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 507 {
 508         if (unlikely(e->rps_rxhash != hash))
 509                 e->rps_rxhash = hash;
 510 }
 511
 512 /* We try to identify a flow through its rxhash. The reason that
 513  * we do not check rxq no. is because some cards(e.g 82599), chooses
 514  * the rxq based on the txq where the last packet of the flow comes. As
 515  * the userspace application move between processors, we may get a
 516  * different rxq no. here.
 517  */
 518 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 519 {
 520         struct tun_flow_entry *e;
 521         u32 txq = 0;
 522         u32 numqueues = 0;
 523
 524         numqueues = READ_ONCE(tun->numqueues);
 525
 526         txq = __skb_get_hash_symmetric(skb);
 527         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 528         if (e) {
 529                 tun_flow_save_rps_rxhash(e, txq);
 530                 txq = e->queue_index;
 531         } else {
 532                 /* use multiply and shift instead of expensive divide */
 533                 txq = ((u64)txq * numqueues) >> 32;
 534         }
 535
 536         return txq;
 537 }
 538
 539 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 540 {
 541         struct tun_prog *prog;
 542         u32 numqueues;
 543         u16 ret = 0;
 544
 545         numqueues = READ_ONCE(tun->numqueues);
 546         if (!numqueues)
 547                 return 0;
 548
 549         prog = rcu_dereference(tun->steering_prog);
 550         if (prog)
 551                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 552
 553         return ret % numqueues;
 554 }
 555
 556 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 557                             struct net_device *sb_dev)
 558 {
 559         struct tun_struct *tun = netdev_priv(dev);
 560         u16 ret;
 561
 562         rcu_read_lock();
 563         if (rcu_dereference(tun->steering_prog))
 564                 ret = tun_ebpf_select_queue(tun, skb);
 565         else
 566                 ret = tun_automq_select_queue(tun, skb);
 567         rcu_read_unlock();
 568
 569         return ret;
 570 }
 571
 572 static inline bool tun_not_capable(struct tun_struct *tun)
 573 {
 574         const struct cred *cred = current_cred();
 575         struct net *net = dev_net(tun->dev);
 576
 577         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 578                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 579                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 580 }
 581
 582 static void tun_set_real_num_queues(struct tun_struct *tun)
 583 {
 584         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 585         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 586 }
 587
 588 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 589 {
 590         tfile->detached = tun;
 591         list_add_tail(&tfile->next, &tun->disabled);
 592         ++tun->numdisabled;
 593 }
 594
 595 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 596 {
 597         struct tun_struct *tun = tfile->detached;
 598
 599         tfile->detached = NULL;
 600         list_del_init(&tfile->next);
 601         --tun->numdisabled;
 602         return tun;
 603 }
 604
 605 void tun_ptr_free(void *ptr)
 606 {
 607         if (!ptr)
 608                 return;
 609         if (tun_is_xdp_frame(ptr)) {
 610                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 611
 612                 xdp_return_frame(xdpf);
 613         } else {
 614                 __skb_array_destroy_skb(ptr);
 615         }
 616 }
 617 EXPORT_SYMBOL_GPL(tun_ptr_free);
 618
 619 static void tun_queue_purge(struct tun_file *tfile)
 620 {
 621         void *ptr;
 622
 623         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 624                 tun_ptr_free(ptr);
 625
 626         skb_queue_purge(&tfile->sk.sk_write_queue);
 627         skb_queue_purge(&tfile->sk.sk_error_queue);
 628 }
 629
 630 static void __tun_detach(struct tun_file *tfile, bool clean)
 631 {
 632         struct tun_file *ntfile;
 633         struct tun_struct *tun;
 634
 635         tun = rtnl_dereference(tfile->tun);
 636
 637         if (tun && clean) {
 638                 tun_napi_disable(tfile);
 639                 tun_napi_del(tfile);
 640         }
 641
 642         if (tun && !tfile->detached) {
 643                 u16 index = tfile->queue_index;
 644                 BUG_ON(index >= tun->numqueues);
 645
 646                 rcu_assign_pointer(tun->tfiles[index],
 647                                    tun->tfiles[tun->numqueues - 1]);
 648                 ntfile = rtnl_dereference(tun->tfiles[index]);
 649                 ntfile->queue_index = index;
 650                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 651                                    NULL);
 652
 653                 --tun->numqueues;
 654                 if (clean) {
 655                         RCU_INIT_POINTER(tfile->tun, NULL);
 656                         sock_put(&tfile->sk);
 657                 } else
 658                         tun_disable_queue(tun, tfile);
 659
 660                 synchronize_net();
 661                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 662                 /* Drop read queue */
 663                 tun_queue_purge(tfile);
 664                 tun_set_real_num_queues(tun);
 665         } else if (tfile->detached && clean) {
 666                 tun = tun_enable_queue(tfile);
 667                 sock_put(&tfile->sk);
 668         }
 669
 670         if (clean) {
 671                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 672                         netif_carrier_off(tun->dev);
 673
 674                         if (!(tun->flags & IFF_PERSIST) &&
 675                             tun->dev->reg_state == NETREG_REGISTERED)
 676                                 unregister_netdevice(tun->dev);
 677                 }
 678                 if (tun)
 679                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 680                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 681                 sock_put(&tfile->sk);
 682         }
 683 }
 684
 685 static void tun_detach(struct tun_file *tfile, bool clean)
 686 {
 687         struct tun_struct *tun;
 688         struct net_device *dev;
 689
 690         rtnl_lock();
 691         tun = rtnl_dereference(tfile->tun);
 692         dev = tun ? tun->dev : NULL;
 693         __tun_detach(tfile, clean);
 694         if (dev)
 695                 netdev_state_change(dev);
 696         rtnl_unlock();
 697 }
 698
 699 static void tun_detach_all(struct net_device *dev)
 700 {
 701         struct tun_struct *tun = netdev_priv(dev);
 702         struct tun_file *tfile, *tmp;
 703         int i, n = tun->numqueues;
 704
 705         for (i = 0; i < n; i++) {
 706                 tfile = rtnl_dereference(tun->tfiles[i]);
 707                 BUG_ON(!tfile);
 708                 tun_napi_disable(tfile);
 709                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 710                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 711                 RCU_INIT_POINTER(tfile->tun, NULL);
 712                 --tun->numqueues;
 713         }
 714         list_for_each_entry(tfile, &tun->disabled, next) {
 715                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 716                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 717                 RCU_INIT_POINTER(tfile->tun, NULL);
 718         }
 719         BUG_ON(tun->numqueues != 0);
 720
 721         synchronize_net();
 722         for (i = 0; i < n; i++) {
 723                 tfile = rtnl_dereference(tun->tfiles[i]);
 724                 tun_napi_del(tfile);
 725                 /* Drop read queue */
 726                 tun_queue_purge(tfile);
 727                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 728                 sock_put(&tfile->sk);
 729         }
 730         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 731                 tun_enable_queue(tfile);
 732                 tun_queue_purge(tfile);
 733                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 734                 sock_put(&tfile->sk);
 735         }
 736         BUG_ON(tun->numdisabled != 0);
 737
 738         if (tun->flags & IFF_PERSIST)
 739                 module_put(THIS_MODULE);
 740 }
 741
 742 static int tun_attach(struct tun_struct *tun, struct file *file,
 743                       bool skip_filter, bool napi, bool napi_frags,
 744                       bool publish_tun)
 745 {
 746         struct tun_file *tfile = file->private_data;
 747         struct net_device *dev = tun->dev;
 748         int err;
 749
 750         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 751         if (err < 0)
 752                 goto out;
 753
 754         err = -EINVAL;
 755         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 756                 goto out;
 757
 758         err = -EBUSY;
 759         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 760                 goto out;
 761
 762         err = -E2BIG;
 763         if (!tfile->detached &&
 764             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 765                 goto out;
 766
 767         err = 0;
 768
 769         /* Re-attach the filter to persist device */
 770         if (!skip_filter && (tun->filter_attached == true)) {
 771                 lock_sock(tfile->socket.sk);
 772                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 773                 release_sock(tfile->socket.sk);
 774                 if (!err)
 775                         goto out;
 776         }
 777
 778         if (!tfile->detached &&
 779             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 780                             GFP_KERNEL, tun_ptr_free)) {
 781                 err = -ENOMEM;
 782                 goto out;
 783         }
 784
 785         tfile->queue_index = tun->numqueues;
 786         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 787
 788         if (tfile->detached) {
 789                 /* Re-attach detached tfile, updating XDP queue_index */
 790                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 791
 792                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 793                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 794         } else {
 795                 /* Setup XDP RX-queue info, for new tfile getting attached */
 796                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 797                                        tun->dev, tfile->queue_index, 0);
 798                 if (err < 0)
 799                         goto out;
 800                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 801                                                  MEM_TYPE_PAGE_SHARED, NULL);
 802                 if (err < 0) {
 803                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 804                         goto out;
 805                 }
 806                 err = 0;
 807         }
 808
 809         if (tfile->detached) {
 810                 tun_enable_queue(tfile);
 811         } else {
 812                 sock_hold(&tfile->sk);
 813                 tun_napi_init(tun, tfile, napi, napi_frags);
 814         }
 815
 816         if (rtnl_dereference(tun->xdp_prog))
 817                 sock_set_flag(&tfile->sk, SOCK_XDP);
 818
 819         /* device is allowed to go away first, so no need to hold extra
 820          * refcnt.
 821          */
 822
 823         /* Publish tfile->tun and tun->tfiles only after we've fully
 824          * initialized tfile; otherwise we risk using half-initialized
 825          * object.
 826          */
 827         if (publish_tun)
 828                 rcu_assign_pointer(tfile->tun, tun);
 829         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 830         tun->numqueues++;
 831         tun_set_real_num_queues(tun);
 832 out:
 833         return err;
 834 }
 835
 836 static struct tun_struct *tun_get(struct tun_file *tfile)
 837 {
 838         struct tun_struct *tun;
 839
 840         rcu_read_lock();
 841         tun = rcu_dereference(tfile->tun);
 842         if (tun)
 843                 dev_hold(tun->dev);
 844         rcu_read_unlock();
 845
 846         return tun;
 847 }
 848
 849 static void tun_put(struct tun_struct *tun)
 850 {
 851         dev_put(tun->dev);
 852 }
 853
 854 /* TAP filtering */
 855 static void addr_hash_set(u32 *mask, const u8 *addr)
 856 {
 857         int n = ether_crc(ETH_ALEN, addr) >> 26;
 858         mask[n >> 5] |= (1 << (n & 31));
 859 }
 860
 861 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 862 {
 863         int n = ether_crc(ETH_ALEN, addr) >> 26;
 864         return mask[n >> 5] & (1 << (n & 31));
 865 }
 866
 867 static int update_filter(struct tap_filter *filter, void __user *arg)
 868 {
 869         struct { u8 u[ETH_ALEN]; } *addr;
 870         struct tun_filter uf;
 871         int err, alen, n, nexact;
 872
 873         if (copy_from_user(&uf, arg, sizeof(uf)))
 874                 return -EFAULT;
 875
 876         if (!uf.count) {
 877                 /* Disabled */
 878                 filter->count = 0;
 879                 return 0;
 880         }
 881
 882         alen = ETH_ALEN * uf.count;
 883         addr = memdup_user(arg + sizeof(uf), alen);
 884         if (IS_ERR(addr))
 885                 return PTR_ERR(addr);
 886
 887         /* The filter is updated without holding any locks. Which is
 888          * perfectly safe. We disable it first and in the worst
 889          * case we'll accept a few undesired packets. */
 890         filter->count = 0;
 891         wmb();
 892
 893         /* Use first set of addresses as an exact filter */
 894         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 895                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 896
 897         nexact = n;
 898
 899         /* Remaining multicast addresses are hashed,
 900          * unicast will leave the filter disabled. */
 901         memset(filter->mask, 0, sizeof(filter->mask));
 902         for (; n < uf.count; n++) {
 903                 if (!is_multicast_ether_addr(addr[n].u)) {
 904                         err = 0; /* no filter */
 905                         goto free_addr;
 906                 }
 907                 addr_hash_set(filter->mask, addr[n].u);
 908         }
 909
 910         /* For ALLMULTI just set the mask to all ones.
 911          * This overrides the mask populated above. */
 912         if ((uf.flags & TUN_FLT_ALLMULTI))
 913                 memset(filter->mask, ~0, sizeof(filter->mask));
 914
 915         /* Now enable the filter */
 916         wmb();
 917         filter->count = nexact;
 918
 919         /* Return the number of exact filters */
 920         err = nexact;
 921 free_addr:
 922         kfree(addr);
 923         return err;
 924 }
 925
 926 /* Returns: 0 - drop, !=0 - accept */
 927 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 928 {
 929         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 930          * at this point. */
 931         struct ethhdr *eh = (struct ethhdr *) skb->data;
 932         int i;
 933
 934         /* Exact match */
 935         for (i = 0; i < filter->count; i++)
 936                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 937                         return 1;
 938
 939         /* Inexact match (multicast only) */
 940         if (is_multicast_ether_addr(eh->h_dest))
 941                 return addr_hash_test(filter->mask, eh->h_dest);
 942
 943         return 0;
 944 }
 945
 946 /*
 947  * Checks whether the packet is accepted or not.
 948  * Returns: 0 - drop, !=0 - accept
 949  */
 950 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 951 {
 952         if (!filter->count)
 953                 return 1;
 954
 955         return run_filter(filter, skb);
 956 }
 957
 958 /* Network device part of the driver */
 959
 960 static const struct ethtool_ops tun_ethtool_ops;
 961
 962 static int tun_net_init(struct net_device *dev)
 963 {
 964         struct tun_struct *tun = netdev_priv(dev);
 965         struct ifreq *ifr = tun->ifr;
 966         int err;
 967
 968         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 969         if (!dev->tstats)
 970                 return -ENOMEM;
 971
 972         spin_lock_init(&tun->lock);
 973
 974         err = security_tun_dev_alloc_security(&tun->security);
 975         if (err < 0) {
 976                 free_percpu(dev->tstats);
 977                 return err;
 978         }
 979
 980         tun_flow_init(tun);
 981
 982         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 983                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 984                            NETIF_F_HW_VLAN_STAG_TX;
 985         dev->features = dev->hw_features | NETIF_F_LLTX;
 986         dev->vlan_features = dev->features &
 987                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 988                                NETIF_F_HW_VLAN_STAG_TX);
 989
 990         tun->flags = (tun->flags & ~TUN_FEATURES) |
 991                       (ifr->ifr_flags & TUN_FEATURES);
 992
 993         INIT_LIST_HEAD(&tun->disabled);
 994         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
 995                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
 996         if (err < 0) {
 997                 tun_flow_uninit(tun);
 998                 security_tun_dev_free_security(tun->security);
 999                 free_percpu(dev->tstats);
1000                 return err;
1001         }
1002         return 0;
1003 }
1004
1005 /* Net device detach from fd. */
1006 static void tun_net_uninit(struct net_device *dev)
1007 {
1008         tun_detach_all(dev);
1009 }
1010
1011 /* Net device open. */
1012 static int tun_net_open(struct net_device *dev)
1013 {
1014         netif_tx_start_all_queues(dev);
1015
1016         return 0;
1017 }
1018
1019 /* Net device close. */
1020 static int tun_net_close(struct net_device *dev)
1021 {
1022         netif_tx_stop_all_queues(dev);
1023         return 0;
1024 }
1025
1026 /* Net device start xmit */
1027 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1028 {
1029 #ifdef CONFIG_RPS
1030         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1031                 /* Select queue was not called for the skbuff, so we extract the
1032                  * RPS hash and save it into the flow_table here.
1033                  */
1034                 struct tun_flow_entry *e;
1035                 __u32 rxhash;
1036
1037                 rxhash = __skb_get_hash_symmetric(skb);
1038                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1039                 if (e)
1040                         tun_flow_save_rps_rxhash(e, rxhash);
1041         }
1042 #endif
1043 }
1044
1045 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1046                                     struct sk_buff *skb,
1047                                     int len)
1048 {
1049         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1050
1051         if (prog)
1052                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1053
1054         return len;
1055 }
1056
1057 /* Net device start xmit */
1058 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1059 {
1060         struct tun_struct *tun = netdev_priv(dev);
1061         enum skb_drop_reason drop_reason;
1062         int txq = skb->queue_mapping;
1063         struct netdev_queue *queue;
1064         struct tun_file *tfile;
1065         int len = skb->len;
1066
1067         rcu_read_lock();
1068         tfile = rcu_dereference(tun->tfiles[txq]);
1069
1070         /* Drop packet if interface is not attached */
1071         if (!tfile) {
1072                 drop_reason = SKB_DROP_REASON_DEV_READY;
1073                 goto drop;
1074         }
1075
1076         if (!rcu_dereference(tun->steering_prog))
1077                 tun_automq_xmit(tun, skb);
1078
1079         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1080
1081         /* Drop if the filter does not like it.
1082          * This is a noop if the filter is disabled.
1083          * Filter can be enabled only for the TAP devices. */
1084         if (!check_filter(&tun->txflt, skb)) {
1085                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1086                 goto drop;
1087         }
1088
1089         if (tfile->socket.sk->sk_filter &&
1090             sk_filter(tfile->socket.sk, skb)) {
1091                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1092                 goto drop;
1093         }
1094
1095         len = run_ebpf_filter(tun, skb, len);
1096         if (len == 0) {
1097                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1098                 goto drop;
1099         }
1100
1101         if (pskb_trim(skb, len)) {
1102                 drop_reason = SKB_DROP_REASON_NOMEM;
1103                 goto drop;
1104         }
1105
1106         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1107                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1108                 goto drop;
1109         }
1110
1111         skb_tx_timestamp(skb);
1112
1113         /* Orphan the skb - required as we might hang on to it
1114          * for indefinite time.
1115          */
1116         skb_orphan(skb);
1117
1118         nf_reset_ct(skb);
1119
1120         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1121                 drop_reason = SKB_DROP_REASON_FULL_RING;
1122                 goto drop;
1123         }
1124
1125         /* NETIF_F_LLTX requires to do our own update of trans_start */
1126         queue = netdev_get_tx_queue(dev, txq);
1127         queue->trans_start = jiffies;
1128
1129         /* Notify and wake up reader process */
1130         if (tfile->flags & TUN_FASYNC)
1131                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1132         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1133
1134         rcu_read_unlock();
1135         return NETDEV_TX_OK;
1136
1137 drop:
1138         dev_core_stats_tx_dropped_inc(dev);
1139         skb_tx_error(skb);
1140         kfree_skb_reason(skb, drop_reason);
1141         rcu_read_unlock();
1142         return NET_XMIT_DROP;
1143 }
1144
1145 static void tun_net_mclist(struct net_device *dev)
1146 {
1147         /*
1148          * This callback is supposed to deal with mc filter in
1149          * _rx_ path and has nothing to do with the _tx_ path.
1150          * In rx path we always accept everything userspace gives us.
1151          */
1152 }
1153
1154 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1155         netdev_features_t features)
1156 {
1157         struct tun_struct *tun = netdev_priv(dev);
1158
1159         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1160 }
1161
1162 static void tun_set_headroom(struct net_device *dev, int new_hr)
1163 {
1164         struct tun_struct *tun = netdev_priv(dev);
1165
1166         if (new_hr < NET_SKB_PAD)
1167                 new_hr = NET_SKB_PAD;
1168
1169         tun->align = new_hr;
1170 }
1171
1172 static void
1173 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1174 {
1175         struct tun_struct *tun = netdev_priv(dev);
1176
1177         dev_get_tstats64(dev, stats);
1178
1179         stats->rx_frame_errors +=
1180                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1181 }
1182
1183 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1184                        struct netlink_ext_ack *extack)
1185 {
1186         struct tun_struct *tun = netdev_priv(dev);
1187         struct tun_file *tfile;
1188         struct bpf_prog *old_prog;
1189         int i;
1190
1191         old_prog = rtnl_dereference(tun->xdp_prog);
1192         rcu_assign_pointer(tun->xdp_prog, prog);
1193         if (old_prog)
1194                 bpf_prog_put(old_prog);
1195
1196         for (i = 0; i < tun->numqueues; i++) {
1197                 tfile = rtnl_dereference(tun->tfiles[i]);
1198                 if (prog)
1199                         sock_set_flag(&tfile->sk, SOCK_XDP);
1200                 else
1201                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1202         }
1203         list_for_each_entry(tfile, &tun->disabled, next) {
1204                 if (prog)
1205                         sock_set_flag(&tfile->sk, SOCK_XDP);
1206                 else
1207                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1208         }
1209
1210         return 0;
1211 }
1212
1213 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1214 {
1215         switch (xdp->command) {
1216         case XDP_SETUP_PROG:
1217                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1218         default:
1219                 return -EINVAL;
1220         }
1221 }
1222
1223 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1224 {
1225         if (new_carrier) {
1226                 struct tun_struct *tun = netdev_priv(dev);
1227
1228                 if (!tun->numqueues)
1229                         return -EPERM;
1230
1231                 netif_carrier_on(dev);
1232         } else {
1233                 netif_carrier_off(dev);
1234         }
1235         return 0;
1236 }
1237
1238 static const struct net_device_ops tun_netdev_ops = {
1239         .ndo_init               = tun_net_init,
1240         .ndo_uninit             = tun_net_uninit,
1241         .ndo_open               = tun_net_open,
1242         .ndo_stop               = tun_net_close,
1243         .ndo_start_xmit         = tun_net_xmit,
1244         .ndo_fix_features       = tun_net_fix_features,
1245         .ndo_select_queue       = tun_select_queue,
1246         .ndo_set_rx_headroom    = tun_set_headroom,
1247         .ndo_get_stats64        = tun_net_get_stats64,
1248         .ndo_change_carrier     = tun_net_change_carrier,
1249 };
1250
1251 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1252 {
1253         /* Notify and wake up reader process */
1254         if (tfile->flags & TUN_FASYNC)
1255                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1256         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1257 }
1258
1259 static int tun_xdp_xmit(struct net_device *dev, int n,
1260                         struct xdp_frame **frames, u32 flags)
1261 {
1262         struct tun_struct *tun = netdev_priv(dev);
1263         struct tun_file *tfile;
1264         u32 numqueues;
1265         int nxmit = 0;
1266         int i;
1267
1268         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1269                 return -EINVAL;
1270
1271         rcu_read_lock();
1272
1273 resample:
1274         numqueues = READ_ONCE(tun->numqueues);
1275         if (!numqueues) {
1276                 rcu_read_unlock();
1277                 return -ENXIO; /* Caller will free/return all frames */
1278         }
1279
1280         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1281                                             numqueues]);
1282         if (unlikely(!tfile))
1283                 goto resample;
1284
1285         spin_lock(&tfile->tx_ring.producer_lock);
1286         for (i = 0; i < n; i++) {
1287                 struct xdp_frame *xdp = frames[i];
1288                 /* Encode the XDP flag into lowest bit for consumer to differ
1289                  * XDP buffer from sk_buff.
1290                  */
1291                 void *frame = tun_xdp_to_ptr(xdp);
1292
1293                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1294                         dev_core_stats_tx_dropped_inc(dev);
1295                         break;
1296                 }
1297                 nxmit++;
1298         }
1299         spin_unlock(&tfile->tx_ring.producer_lock);
1300
1301         if (flags & XDP_XMIT_FLUSH)
1302                 __tun_xdp_flush_tfile(tfile);
1303
1304         rcu_read_unlock();
1305         return nxmit;
1306 }
1307
1308 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1309 {
1310         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1311         int nxmit;
1312
1313         if (unlikely(!frame))
1314                 return -EOVERFLOW;
1315
1316         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1317         if (!nxmit)
1318                 xdp_return_frame_rx_napi(frame);
1319         return nxmit;
1320 }
1321
1322 static const struct net_device_ops tap_netdev_ops = {
1323         .ndo_init               = tun_net_init,
1324         .ndo_uninit             = tun_net_uninit,
1325         .ndo_open               = tun_net_open,
1326         .ndo_stop               = tun_net_close,
1327         .ndo_start_xmit         = tun_net_xmit,
1328         .ndo_fix_features       = tun_net_fix_features,
1329         .ndo_set_rx_mode        = tun_net_mclist,
1330         .ndo_set_mac_address    = eth_mac_addr,
1331         .ndo_validate_addr      = eth_validate_addr,
1332         .ndo_select_queue       = tun_select_queue,
1333         .ndo_features_check     = passthru_features_check,
1334         .ndo_set_rx_headroom    = tun_set_headroom,
1335         .ndo_get_stats64        = dev_get_tstats64,
1336         .ndo_bpf                = tun_xdp,
1337         .ndo_xdp_xmit           = tun_xdp_xmit,
1338         .ndo_change_carrier     = tun_net_change_carrier,
1339 };
1340
1341 static void tun_flow_init(struct tun_struct *tun)
1342 {
1343         int i;
1344
1345         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1346                 INIT_HLIST_HEAD(&tun->flows[i]);
1347
1348         tun->ageing_time = TUN_FLOW_EXPIRE;
1349         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1350         mod_timer(&tun->flow_gc_timer,
1351                   round_jiffies_up(jiffies + tun->ageing_time));
1352 }
1353
1354 static void tun_flow_uninit(struct tun_struct *tun)
1355 {
1356         del_timer_sync(&tun->flow_gc_timer);
1357         tun_flow_flush(tun);
1358 }
1359
1360 #define MIN_MTU 68
1361 #define MAX_MTU 65535
1362
1363 /* Initialize net device. */
1364 static void tun_net_initialize(struct net_device *dev)
1365 {
1366         struct tun_struct *tun = netdev_priv(dev);
1367
1368         switch (tun->flags & TUN_TYPE_MASK) {
1369         case IFF_TUN:
1370                 dev->netdev_ops = &tun_netdev_ops;
1371                 dev->header_ops = &ip_tunnel_header_ops;
1372
1373                 /* Point-to-Point TUN Device */
1374                 dev->hard_header_len = 0;
1375                 dev->addr_len = 0;
1376                 dev->mtu = 1500;
1377
1378                 /* Zero header length */
1379                 dev->type = ARPHRD_NONE;
1380                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1381                 break;
1382
1383         case IFF_TAP:
1384                 dev->netdev_ops = &tap_netdev_ops;
1385                 /* Ethernet TAP Device */
1386                 ether_setup(dev);
1387                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1388                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1389
1390                 eth_hw_addr_random(dev);
1391
1392                 break;
1393         }
1394
1395         dev->min_mtu = MIN_MTU;
1396         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1397 }
1398
1399 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1400 {
1401         struct sock *sk = tfile->socket.sk;
1402
1403         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1404 }
1405
1406 /* Character device part */
1407
1408 /* Poll */
1409 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1410 {
1411         struct tun_file *tfile = file->private_data;
1412         struct tun_struct *tun = tun_get(tfile);
1413         struct sock *sk;
1414         __poll_t mask = 0;
1415
1416         if (!tun)
1417                 return EPOLLERR;
1418
1419         sk = tfile->socket.sk;
1420
1421         poll_wait(file, sk_sleep(sk), wait);
1422
1423         if (!ptr_ring_empty(&tfile->tx_ring))
1424                 mask |= EPOLLIN | EPOLLRDNORM;
1425
1426         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1427          * guarantee EPOLLOUT to be raised by either here or
1428          * tun_sock_write_space(). Then process could get notification
1429          * after it writes to a down device and meets -EIO.
1430          */
1431         if (tun_sock_writeable(tun, tfile) ||
1432             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1433              tun_sock_writeable(tun, tfile)))
1434                 mask |= EPOLLOUT | EPOLLWRNORM;
1435
1436         if (tun->dev->reg_state != NETREG_REGISTERED)
1437                 mask = EPOLLERR;
1438
1439         tun_put(tun);
1440         return mask;
1441 }
1442
1443 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1444                                             size_t len,
1445                                             const struct iov_iter *it)
1446 {
1447         struct sk_buff *skb;
1448         size_t linear;
1449         int err;
1450         int i;
1451
1452         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1453                 return ERR_PTR(-EMSGSIZE);
1454
1455         local_bh_disable();
1456         skb = napi_get_frags(&tfile->napi);
1457         local_bh_enable();
1458         if (!skb)
1459                 return ERR_PTR(-ENOMEM);
1460
1461         linear = iov_iter_single_seg_count(it);
1462         err = __skb_grow(skb, linear);
1463         if (err)
1464                 goto free;
1465
1466         skb->len = len;
1467         skb->data_len = len - linear;
1468         skb->truesize += skb->data_len;
1469
1470         for (i = 1; i < it->nr_segs; i++) {
1471                 size_t fragsz = it->iov[i].iov_len;
1472                 struct page *page;
1473                 void *frag;
1474
1475                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1476                         err = -EINVAL;
1477                         goto free;
1478                 }
1479                 frag = netdev_alloc_frag(fragsz);
1480                 if (!frag) {
1481                         err = -ENOMEM;
1482                         goto free;
1483                 }
1484                 page = virt_to_head_page(frag);
1485                 skb_fill_page_desc(skb, i - 1, page,
1486                                    frag - page_address(page), fragsz);
1487         }
1488
1489         return skb;
1490 free:
1491         /* frees skb and all frags allocated with napi_alloc_frag() */
1492         napi_free_frags(&tfile->napi);
1493         return ERR_PTR(err);
1494 }
1495
1496 /* prepad is the amount to reserve at front.  len is length after that.
1497  * linear is a hint as to how much to copy (usually headers). */
1498 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1499                                      size_t prepad, size_t len,
1500                                      size_t linear, int noblock)
1501 {
1502         struct sock *sk = tfile->socket.sk;
1503         struct sk_buff *skb;
1504         int err;
1505
1506         /* Under a page?  Don't bother with paged skb. */
1507         if (prepad + len < PAGE_SIZE || !linear)
1508                 linear = len;
1509
1510         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1511                                    &err, 0);
1512         if (!skb)
1513                 return ERR_PTR(err);
1514
1515         skb_reserve(skb, prepad);
1516         skb_put(skb, linear);
1517         skb->data_len = len - linear;
1518         skb->len += len - linear;
1519
1520         return skb;
1521 }
1522
1523 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1524                            struct sk_buff *skb, int more)
1525 {
1526         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1527         struct sk_buff_head process_queue;
1528         u32 rx_batched = tun->rx_batched;
1529         bool rcv = false;
1530
1531         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1532                 local_bh_disable();
1533                 skb_record_rx_queue(skb, tfile->queue_index);
1534                 netif_receive_skb(skb);
1535                 local_bh_enable();
1536                 return;
1537         }
1538
1539         spin_lock(&queue->lock);
1540         if (!more || skb_queue_len(queue) == rx_batched) {
1541                 __skb_queue_head_init(&process_queue);
1542                 skb_queue_splice_tail_init(queue, &process_queue);
1543                 rcv = true;
1544         } else {
1545                 __skb_queue_tail(queue, skb);
1546         }
1547         spin_unlock(&queue->lock);
1548
1549         if (rcv) {
1550                 struct sk_buff *nskb;
1551
1552                 local_bh_disable();
1553                 while ((nskb = __skb_dequeue(&process_queue))) {
1554                         skb_record_rx_queue(nskb, tfile->queue_index);
1555                         netif_receive_skb(nskb);
1556                 }
1557                 skb_record_rx_queue(skb, tfile->queue_index);
1558                 netif_receive_skb(skb);
1559                 local_bh_enable();
1560         }
1561 }
1562
1563 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1564                               int len, int noblock, bool zerocopy)
1565 {
1566         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1567                 return false;
1568
1569         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1570                 return false;
1571
1572         if (!noblock)
1573                 return false;
1574
1575         if (zerocopy)
1576                 return false;
1577
1578         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1579             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1580                 return false;
1581
1582         return true;
1583 }
1584
1585 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1586                                        struct page_frag *alloc_frag, char *buf,
1587                                        int buflen, int len, int pad)
1588 {
1589         struct sk_buff *skb = build_skb(buf, buflen);
1590
1591         if (!skb)
1592                 return ERR_PTR(-ENOMEM);
1593
1594         skb_reserve(skb, pad);
1595         skb_put(skb, len);
1596         skb_set_owner_w(skb, tfile->socket.sk);
1597
1598         get_page(alloc_frag->page);
1599         alloc_frag->offset += buflen;
1600
1601         return skb;
1602 }
1603
1604 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1605                        struct xdp_buff *xdp, u32 act)
1606 {
1607         int err;
1608
1609         switch (act) {
1610         case XDP_REDIRECT:
1611                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1612                 if (err)
1613                         return err;
1614                 break;
1615         case XDP_TX:
1616                 err = tun_xdp_tx(tun->dev, xdp);
1617                 if (err < 0)
1618                         return err;
1619                 break;
1620         case XDP_PASS:
1621                 break;
1622         default:
1623                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1624                 fallthrough;
1625         case XDP_ABORTED:
1626                 trace_xdp_exception(tun->dev, xdp_prog, act);
1627                 fallthrough;
1628         case XDP_DROP:
1629                 dev_core_stats_rx_dropped_inc(tun->dev);
1630                 break;
1631         }
1632
1633         return act;
1634 }
1635
1636 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1637                                      struct tun_file *tfile,
1638                                      struct iov_iter *from,
1639                                      struct virtio_net_hdr *hdr,
1640                                      int len, int *skb_xdp)
1641 {
1642         struct page_frag *alloc_frag = &current->task_frag;
1643         struct bpf_prog *xdp_prog;
1644         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1645         char *buf;
1646         size_t copied;
1647         int pad = TUN_RX_PAD;
1648         int err = 0;
1649
1650         rcu_read_lock();
1651         xdp_prog = rcu_dereference(tun->xdp_prog);
1652         if (xdp_prog)
1653                 pad += XDP_PACKET_HEADROOM;
1654         buflen += SKB_DATA_ALIGN(len + pad);
1655         rcu_read_unlock();
1656
1657         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1658         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1659                 return ERR_PTR(-ENOMEM);
1660
1661         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1662         copied = copy_page_from_iter(alloc_frag->page,
1663                                      alloc_frag->offset + pad,
1664                                      len, from);
1665         if (copied != len)
1666                 return ERR_PTR(-EFAULT);
1667
1668         /* There's a small window that XDP may be set after the check
1669          * of xdp_prog above, this should be rare and for simplicity
1670          * we do XDP on skb in case the headroom is not enough.
1671          */
1672         if (hdr->gso_type || !xdp_prog) {
1673                 *skb_xdp = 1;
1674                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1675                                        pad);
1676         }
1677
1678         *skb_xdp = 0;
1679
1680         local_bh_disable();
1681         rcu_read_lock();
1682         xdp_prog = rcu_dereference(tun->xdp_prog);
1683         if (xdp_prog) {
1684                 struct xdp_buff xdp;
1685                 u32 act;
1686
1687                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1688                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1689
1690                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1691                 if (act == XDP_REDIRECT || act == XDP_TX) {
1692                         get_page(alloc_frag->page);
1693                         alloc_frag->offset += buflen;
1694                 }
1695                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1696                 if (err < 0) {
1697                         if (act == XDP_REDIRECT || act == XDP_TX)
1698                                 put_page(alloc_frag->page);
1699                         goto out;
1700                 }
1701
1702                 if (err == XDP_REDIRECT)
1703                         xdp_do_flush();
1704                 if (err != XDP_PASS)
1705                         goto out;
1706
1707                 pad = xdp.data - xdp.data_hard_start;
1708                 len = xdp.data_end - xdp.data;
1709         }
1710         rcu_read_unlock();
1711         local_bh_enable();
1712
1713         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1714
1715 out:
1716         rcu_read_unlock();
1717         local_bh_enable();
1718         return NULL;
1719 }
1720
1721 /* Get packet from user space buffer */
1722 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1723                             void *msg_control, struct iov_iter *from,
1724                             int noblock, bool more)
1725 {
1726         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1727         struct sk_buff *skb;
1728         size_t total_len = iov_iter_count(from);
1729         size_t len = total_len, align = tun->align, linear;
1730         struct virtio_net_hdr gso = { 0 };
1731         int good_linear;
1732         int copylen;
1733         bool zerocopy = false;
1734         int err;
1735         u32 rxhash = 0;
1736         int skb_xdp = 1;
1737         bool frags = tun_napi_frags_enabled(tfile);
1738         enum skb_drop_reason drop_reason;
1739
1740         if (!(tun->flags & IFF_NO_PI)) {
1741                 if (len < sizeof(pi))
1742                         return -EINVAL;
1743                 len -= sizeof(pi);
1744
1745                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1746                         return -EFAULT;
1747         }
1748
1749         if (tun->flags & IFF_VNET_HDR) {
1750                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1751
1752                 if (len < vnet_hdr_sz)
1753                         return -EINVAL;
1754                 len -= vnet_hdr_sz;
1755
1756                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1757                         return -EFAULT;
1758
1759                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1760                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1761                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1762
1763                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1764                         return -EINVAL;
1765                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1766         }
1767
1768         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1769                 align += NET_IP_ALIGN;
1770                 if (unlikely(len < ETH_HLEN ||
1771                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1772                         return -EINVAL;
1773         }
1774
1775         good_linear = SKB_MAX_HEAD(align);
1776
1777         if (msg_control) {
1778                 struct iov_iter i = *from;
1779
1780                 /* There are 256 bytes to be copied in skb, so there is
1781                  * enough room for skb expand head in case it is used.
1782                  * The rest of the buffer is mapped from userspace.
1783                  */
1784                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1785                 if (copylen > good_linear)
1786                         copylen = good_linear;
1787                 linear = copylen;
1788                 iov_iter_advance(&i, copylen);
1789                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1790                         zerocopy = true;
1791         }
1792
1793         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1794                 /* For the packet that is not easy to be processed
1795                  * (e.g gso or jumbo packet), we will do it at after
1796                  * skb was created with generic XDP routine.
1797                  */
1798                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1799                 if (IS_ERR(skb)) {
1800                         dev_core_stats_rx_dropped_inc(tun->dev);
1801                         return PTR_ERR(skb);
1802                 }
1803                 if (!skb)
1804                         return total_len;
1805         } else {
1806                 if (!zerocopy) {
1807                         copylen = len;
1808                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1809                                 linear = good_linear;
1810                         else
1811                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1812                 }
1813
1814                 if (frags) {
1815                         mutex_lock(&tfile->napi_mutex);
1816                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1817                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1818                          * If zerocopy is enabled, then this layout will be
1819                          * overwritten by zerocopy_sg_from_iter().
1820                          */
1821                         zerocopy = false;
1822                 } else {
1823                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1824                                             noblock);
1825                 }
1826
1827                 if (IS_ERR(skb)) {
1828                         if (PTR_ERR(skb) != -EAGAIN)
1829                                 dev_core_stats_rx_dropped_inc(tun->dev);
1830                         if (frags)
1831                                 mutex_unlock(&tfile->napi_mutex);
1832                         return PTR_ERR(skb);
1833                 }
1834
1835                 if (zerocopy)
1836                         err = zerocopy_sg_from_iter(skb, from);
1837                 else
1838                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1839
1840                 if (err) {
1841                         err = -EFAULT;
1842                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1843 drop:
1844                         dev_core_stats_rx_dropped_inc(tun->dev);
1845                         kfree_skb_reason(skb, drop_reason);
1846                         if (frags) {
1847                                 tfile->napi.skb = NULL;
1848                                 mutex_unlock(&tfile->napi_mutex);
1849                         }
1850
1851                         return err;
1852                 }
1853         }
1854
1855         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1856                 atomic_long_inc(&tun->rx_frame_errors);
1857                 kfree_skb(skb);
1858                 if (frags) {
1859                         tfile->napi.skb = NULL;
1860                         mutex_unlock(&tfile->napi_mutex);
1861                 }
1862
1863                 return -EINVAL;
1864         }
1865
1866         switch (tun->flags & TUN_TYPE_MASK) {
1867         case IFF_TUN:
1868                 if (tun->flags & IFF_NO_PI) {
1869                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1870
1871                         switch (ip_version) {
1872                         case 4:
1873                                 pi.proto = htons(ETH_P_IP);
1874                                 break;
1875                         case 6:
1876                                 pi.proto = htons(ETH_P_IPV6);
1877                                 break;
1878                         default:
1879                                 dev_core_stats_rx_dropped_inc(tun->dev);
1880                                 kfree_skb(skb);
1881                                 return -EINVAL;
1882                         }
1883                 }
1884
1885                 skb_reset_mac_header(skb);
1886                 skb->protocol = pi.proto;
1887                 skb->dev = tun->dev;
1888                 break;
1889         case IFF_TAP:
1890                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1891                         err = -ENOMEM;
1892                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1893                         goto drop;
1894                 }
1895                 skb->protocol = eth_type_trans(skb, tun->dev);
1896                 break;
1897         }
1898
1899         /* copy skb_ubuf_info for callback when skb has no error */
1900         if (zerocopy) {
1901                 skb_zcopy_init(skb, msg_control);
1902         } else if (msg_control) {
1903                 struct ubuf_info *uarg = msg_control;
1904                 uarg->callback(NULL, uarg, false);
1905         }
1906
1907         skb_reset_network_header(skb);
1908         skb_probe_transport_header(skb);
1909         skb_record_rx_queue(skb, tfile->queue_index);
1910
1911         if (skb_xdp) {
1912                 struct bpf_prog *xdp_prog;
1913                 int ret;
1914
1915                 local_bh_disable();
1916                 rcu_read_lock();
1917                 xdp_prog = rcu_dereference(tun->xdp_prog);
1918                 if (xdp_prog) {
1919                         ret = do_xdp_generic(xdp_prog, skb);
1920                         if (ret != XDP_PASS) {
1921                                 rcu_read_unlock();
1922                                 local_bh_enable();
1923                                 if (frags) {
1924                                         tfile->napi.skb = NULL;
1925                                         mutex_unlock(&tfile->napi_mutex);
1926                                 }
1927                                 return total_len;
1928                         }
1929                 }
1930                 rcu_read_unlock();
1931                 local_bh_enable();
1932         }
1933
1934         /* Compute the costly rx hash only if needed for flow updates.
1935          * We may get a very small possibility of OOO during switching, not
1936          * worth to optimize.
1937          */
1938         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1939             !tfile->detached)
1940                 rxhash = __skb_get_hash_symmetric(skb);
1941
1942         rcu_read_lock();
1943         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1944                 err = -EIO;
1945                 rcu_read_unlock();
1946                 drop_reason = SKB_DROP_REASON_DEV_READY;
1947                 goto drop;
1948         }
1949
1950         if (frags) {
1951                 u32 headlen;
1952
1953                 /* Exercise flow dissector code path. */
1954                 skb_push(skb, ETH_HLEN);
1955                 headlen = eth_get_headlen(tun->dev, skb->data,
1956                                           skb_headlen(skb));
1957
1958                 if (unlikely(headlen > skb_headlen(skb))) {
1959                         dev_core_stats_rx_dropped_inc(tun->dev);
1960                         napi_free_frags(&tfile->napi);
1961                         rcu_read_unlock();
1962                         mutex_unlock(&tfile->napi_mutex);
1963                         WARN_ON(1);
1964                         return -ENOMEM;
1965                 }
1966
1967                 local_bh_disable();
1968                 napi_gro_frags(&tfile->napi);
1969                 local_bh_enable();
1970                 mutex_unlock(&tfile->napi_mutex);
1971         } else if (tfile->napi_enabled) {
1972                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1973                 int queue_len;
1974
1975                 spin_lock_bh(&queue->lock);
1976                 __skb_queue_tail(queue, skb);
1977                 queue_len = skb_queue_len(queue);
1978                 spin_unlock(&queue->lock);
1979
1980                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1981                         napi_schedule(&tfile->napi);
1982
1983                 local_bh_enable();
1984         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1985                 tun_rx_batched(tun, tfile, skb, more);
1986         } else {
1987                 netif_rx(skb);
1988         }
1989         rcu_read_unlock();
1990
1991         preempt_disable();
1992         dev_sw_netstats_rx_add(tun->dev, len);
1993         preempt_enable();
1994
1995         if (rxhash)
1996                 tun_flow_update(tun, rxhash, tfile);
1997
1998         return total_len;
1999 }
2000
2001 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2002 {
2003         struct file *file = iocb->ki_filp;
2004         struct tun_file *tfile = file->private_data;
2005         struct tun_struct *tun = tun_get(tfile);
2006         ssize_t result;
2007         int noblock = 0;
2008
2009         if (!tun)
2010                 return -EBADFD;
2011
2012         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2013                 noblock = 1;
2014
2015         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2016
2017         tun_put(tun);
2018         return result;
2019 }
2020
2021 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2022                                 struct tun_file *tfile,
2023                                 struct xdp_frame *xdp_frame,
2024                                 struct iov_iter *iter)
2025 {
2026         int vnet_hdr_sz = 0;
2027         size_t size = xdp_frame->len;
2028         size_t ret;
2029
2030         if (tun->flags & IFF_VNET_HDR) {
2031                 struct virtio_net_hdr gso = { 0 };
2032
2033                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2034                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2035                         return -EINVAL;
2036                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2037                              sizeof(gso)))
2038                         return -EFAULT;
2039                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2040         }
2041
2042         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2043
2044         preempt_disable();
2045         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2046         preempt_enable();
2047
2048         return ret;
2049 }
2050
2051 /* Put packet to the user space buffer */
2052 static ssize_t tun_put_user(struct tun_struct *tun,
2053                             struct tun_file *tfile,
2054                             struct sk_buff *skb,
2055                             struct iov_iter *iter)
2056 {
2057         struct tun_pi pi = { 0, skb->protocol };
2058         ssize_t total;
2059         int vlan_offset = 0;
2060         int vlan_hlen = 0;
2061         int vnet_hdr_sz = 0;
2062
2063         if (skb_vlan_tag_present(skb))
2064                 vlan_hlen = VLAN_HLEN;
2065
2066         if (tun->flags & IFF_VNET_HDR)
2067                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2068
2069         total = skb->len + vlan_hlen + vnet_hdr_sz;
2070
2071         if (!(tun->flags & IFF_NO_PI)) {
2072                 if (iov_iter_count(iter) < sizeof(pi))
2073                         return -EINVAL;
2074
2075                 total += sizeof(pi);
2076                 if (iov_iter_count(iter) < total) {
2077                         /* Packet will be striped */
2078                         pi.flags |= TUN_PKT_STRIP;
2079                 }
2080
2081                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2082                         return -EFAULT;
2083         }
2084
2085         if (vnet_hdr_sz) {
2086                 struct virtio_net_hdr gso;
2087
2088                 if (iov_iter_count(iter) < vnet_hdr_sz)
2089                         return -EINVAL;
2090
2091                 if (virtio_net_hdr_from_skb(skb, &gso,
2092                                             tun_is_little_endian(tun), true,
2093                                             vlan_hlen)) {
2094                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2095                         pr_err("unexpected GSO type: "
2096                                "0x%x, gso_size %d, hdr_len %d\n",
2097                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2098                                tun16_to_cpu(tun, gso.hdr_len));
2099                         print_hex_dump(KERN_ERR, "tun: ",
2100                                        DUMP_PREFIX_NONE,
2101                                        16, 1, skb->head,
2102                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2103                         WARN_ON_ONCE(1);
2104                         return -EINVAL;
2105                 }
2106
2107                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2108                         return -EFAULT;
2109
2110                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2111         }
2112
2113         if (vlan_hlen) {
2114                 int ret;
2115                 struct veth veth;
2116
2117                 veth.h_vlan_proto = skb->vlan_proto;
2118                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2119
2120                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2121
2122                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2123                 if (ret || !iov_iter_count(iter))
2124                         goto done;
2125
2126                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2127                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2128                         goto done;
2129         }
2130
2131         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2132
2133 done:
2134         /* caller is in process context, */
2135         preempt_disable();
2136         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2137         preempt_enable();
2138
2139         return total;
2140 }
2141
2142 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2143 {
2144         DECLARE_WAITQUEUE(wait, current);
2145         void *ptr = NULL;
2146         int error = 0;
2147
2148         ptr = ptr_ring_consume(&tfile->tx_ring);
2149         if (ptr)
2150                 goto out;
2151         if (noblock) {
2152                 error = -EAGAIN;
2153                 goto out;
2154         }
2155
2156         add_wait_queue(&tfile->socket.wq.wait, &wait);
2157
2158         while (1) {
2159                 set_current_state(TASK_INTERRUPTIBLE);
2160                 ptr = ptr_ring_consume(&tfile->tx_ring);
2161                 if (ptr)
2162                         break;
2163                 if (signal_pending(current)) {
2164                         error = -ERESTARTSYS;
2165                         break;
2166                 }
2167                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2168                         error = -EFAULT;
2169                         break;
2170                 }
2171
2172                 schedule();
2173         }
2174
2175         __set_current_state(TASK_RUNNING);
2176         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2177
2178 out:
2179         *err = error;
2180         return ptr;
2181 }
2182
2183 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2184                            struct iov_iter *to,
2185                            int noblock, void *ptr)
2186 {
2187         ssize_t ret;
2188         int err;
2189
2190         if (!iov_iter_count(to)) {
2191                 tun_ptr_free(ptr);
2192                 return 0;
2193         }
2194
2195         if (!ptr) {
2196                 /* Read frames from ring */
2197                 ptr = tun_ring_recv(tfile, noblock, &err);
2198                 if (!ptr)
2199                         return err;
2200         }
2201
2202         if (tun_is_xdp_frame(ptr)) {
2203                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2204
2205                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2206                 xdp_return_frame(xdpf);
2207         } else {
2208                 struct sk_buff *skb = ptr;
2209
2210                 ret = tun_put_user(tun, tfile, skb, to);
2211                 if (unlikely(ret < 0))
2212                         kfree_skb(skb);
2213                 else
2214                         consume_skb(skb);
2215         }
2216
2217         return ret;
2218 }
2219
2220 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2221 {
2222         struct file *file = iocb->ki_filp;
2223         struct tun_file *tfile = file->private_data;
2224         struct tun_struct *tun = tun_get(tfile);
2225         ssize_t len = iov_iter_count(to), ret;
2226         int noblock = 0;
2227
2228         if (!tun)
2229                 return -EBADFD;
2230
2231         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2232                 noblock = 1;
2233
2234         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2235         ret = min_t(ssize_t, ret, len);
2236         if (ret > 0)
2237                 iocb->ki_pos = ret;
2238         tun_put(tun);
2239         return ret;
2240 }
2241
2242 static void tun_prog_free(struct rcu_head *rcu)
2243 {
2244         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2245
2246         bpf_prog_destroy(prog->prog);
2247         kfree(prog);
2248 }
2249
2250 static int __tun_set_ebpf(struct tun_struct *tun,
2251                           struct tun_prog __rcu **prog_p,
2252                           struct bpf_prog *prog)
2253 {
2254         struct tun_prog *old, *new = NULL;
2255
2256         if (prog) {
2257                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2258                 if (!new)
2259                         return -ENOMEM;
2260                 new->prog = prog;
2261         }
2262
2263         spin_lock_bh(&tun->lock);
2264         old = rcu_dereference_protected(*prog_p,
2265                                         lockdep_is_held(&tun->lock));
2266         rcu_assign_pointer(*prog_p, new);
2267         spin_unlock_bh(&tun->lock);
2268
2269         if (old)
2270                 call_rcu(&old->rcu, tun_prog_free);
2271
2272         return 0;
2273 }
2274
2275 static void tun_free_netdev(struct net_device *dev)
2276 {
2277         struct tun_struct *tun = netdev_priv(dev);
2278
2279         BUG_ON(!(list_empty(&tun->disabled)));
2280
2281         free_percpu(dev->tstats);
2282         tun_flow_uninit(tun);
2283         security_tun_dev_free_security(tun->security);
2284         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2285         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2286 }
2287
2288 static void tun_setup(struct net_device *dev)
2289 {
2290         struct tun_struct *tun = netdev_priv(dev);
2291
2292         tun->owner = INVALID_UID;
2293         tun->group = INVALID_GID;
2294         tun_default_link_ksettings(dev, &tun->link_ksettings);
2295
2296         dev->ethtool_ops = &tun_ethtool_ops;
2297         dev->needs_free_netdev = true;
2298         dev->priv_destructor = tun_free_netdev;
2299         /* We prefer our own queue length */
2300         dev->tx_queue_len = TUN_READQ_SIZE;
2301 }
2302
2303 /* Trivial set of netlink ops to allow deleting tun or tap
2304  * device with netlink.
2305  */
2306 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2307                         struct netlink_ext_ack *extack)
2308 {
2309         NL_SET_ERR_MSG(extack,
2310                        "tun/tap creation via rtnetlink is not supported.");
2311         return -EOPNOTSUPP;
2312 }
2313
2314 static size_t tun_get_size(const struct net_device *dev)
2315 {
2316         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2317         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2318
2319         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2320                nla_total_size(sizeof(gid_t)) + /* GROUP */
2321                nla_total_size(sizeof(u8)) + /* TYPE */
2322                nla_total_size(sizeof(u8)) + /* PI */
2323                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2324                nla_total_size(sizeof(u8)) + /* PERSIST */
2325                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2326                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2327                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2328                0;
2329 }
2330
2331 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2332 {
2333         struct tun_struct *tun = netdev_priv(dev);
2334
2335         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2336                 goto nla_put_failure;
2337         if (uid_valid(tun->owner) &&
2338             nla_put_u32(skb, IFLA_TUN_OWNER,
2339                         from_kuid_munged(current_user_ns(), tun->owner)))
2340                 goto nla_put_failure;
2341         if (gid_valid(tun->group) &&
2342             nla_put_u32(skb, IFLA_TUN_GROUP,
2343                         from_kgid_munged(current_user_ns(), tun->group)))
2344                 goto nla_put_failure;
2345         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2346                 goto nla_put_failure;
2347         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2348                 goto nla_put_failure;
2349         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2350                 goto nla_put_failure;
2351         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2352                        !!(tun->flags & IFF_MULTI_QUEUE)))
2353                 goto nla_put_failure;
2354         if (tun->flags & IFF_MULTI_QUEUE) {
2355                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2356                         goto nla_put_failure;
2357                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2358                                 tun->numdisabled))
2359                         goto nla_put_failure;
2360         }
2361
2362         return 0;
2363
2364 nla_put_failure:
2365         return -EMSGSIZE;
2366 }
2367
2368 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2369         .kind           = DRV_NAME,
2370         .priv_size      = sizeof(struct tun_struct),
2371         .setup          = tun_setup,
2372         .validate       = tun_validate,
2373         .get_size       = tun_get_size,
2374         .fill_info      = tun_fill_info,
2375 };
2376
2377 static void tun_sock_write_space(struct sock *sk)
2378 {
2379         struct tun_file *tfile;
2380         wait_queue_head_t *wqueue;
2381
2382         if (!sock_writeable(sk))
2383                 return;
2384
2385         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2386                 return;
2387
2388         wqueue = sk_sleep(sk);
2389         if (wqueue && waitqueue_active(wqueue))
2390                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2391                                                 EPOLLWRNORM | EPOLLWRBAND);
2392
2393         tfile = container_of(sk, struct tun_file, sk);
2394         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2395 }
2396
2397 static void tun_put_page(struct tun_page *tpage)
2398 {
2399         if (tpage->page)
2400                 __page_frag_cache_drain(tpage->page, tpage->count);
2401 }
2402
2403 static int tun_xdp_one(struct tun_struct *tun,
2404                        struct tun_file *tfile,
2405                        struct xdp_buff *xdp, int *flush,
2406                        struct tun_page *tpage)
2407 {
2408         unsigned int datasize = xdp->data_end - xdp->data;
2409         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2410         struct virtio_net_hdr *gso = &hdr->gso;
2411         struct bpf_prog *xdp_prog;
2412         struct sk_buff *skb = NULL;
2413         struct sk_buff_head *queue;
2414         u32 rxhash = 0, act;
2415         int buflen = hdr->buflen;
2416         int ret = 0;
2417         bool skb_xdp = false;
2418         struct page *page;
2419
2420         xdp_prog = rcu_dereference(tun->xdp_prog);
2421         if (xdp_prog) {
2422                 if (gso->gso_type) {
2423                         skb_xdp = true;
2424                         goto build;
2425                 }
2426
2427                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2428                 xdp_set_data_meta_invalid(xdp);
2429
2430                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2431                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2432                 if (ret < 0) {
2433                         put_page(virt_to_head_page(xdp->data));
2434                         return ret;
2435                 }
2436
2437                 switch (ret) {
2438                 case XDP_REDIRECT:
2439                         *flush = true;
2440                         fallthrough;
2441                 case XDP_TX:
2442                         return 0;
2443                 case XDP_PASS:
2444                         break;
2445                 default:
2446                         page = virt_to_head_page(xdp->data);
2447                         if (tpage->page == page) {
2448                                 ++tpage->count;
2449                         } else {
2450                                 tun_put_page(tpage);
2451                                 tpage->page = page;
2452                                 tpage->count = 1;
2453                         }
2454                         return 0;
2455                 }
2456         }
2457
2458 build:
2459         skb = build_skb(xdp->data_hard_start, buflen);
2460         if (!skb) {
2461                 ret = -ENOMEM;
2462                 goto out;
2463         }
2464
2465         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2466         skb_put(skb, xdp->data_end - xdp->data);
2467
2468         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2469                 atomic_long_inc(&tun->rx_frame_errors);
2470                 kfree_skb(skb);
2471                 ret = -EINVAL;
2472                 goto out;
2473         }
2474
2475         skb->protocol = eth_type_trans(skb, tun->dev);
2476         skb_reset_network_header(skb);
2477         skb_probe_transport_header(skb);
2478         skb_record_rx_queue(skb, tfile->queue_index);
2479
2480         if (skb_xdp) {
2481                 ret = do_xdp_generic(xdp_prog, skb);
2482                 if (ret != XDP_PASS) {
2483                         ret = 0;
2484                         goto out;
2485                 }
2486         }
2487
2488         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2489             !tfile->detached)
2490                 rxhash = __skb_get_hash_symmetric(skb);
2491
2492         if (tfile->napi_enabled) {
2493                 queue = &tfile->sk.sk_write_queue;
2494                 spin_lock(&queue->lock);
2495                 __skb_queue_tail(queue, skb);
2496                 spin_unlock(&queue->lock);
2497                 ret = 1;
2498         } else {
2499                 netif_receive_skb(skb);
2500                 ret = 0;
2501         }
2502
2503         /* No need to disable preemption here since this function is
2504          * always called with bh disabled
2505          */
2506         dev_sw_netstats_rx_add(tun->dev, datasize);
2507
2508         if (rxhash)
2509                 tun_flow_update(tun, rxhash, tfile);
2510
2511 out:
2512         return ret;
2513 }
2514
2515 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2516 {
2517         int ret, i;
2518         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2519         struct tun_struct *tun = tun_get(tfile);
2520         struct tun_msg_ctl *ctl = m->msg_control;
2521         struct xdp_buff *xdp;
2522
2523         if (!tun)
2524                 return -EBADFD;
2525
2526         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2527             ctl && ctl->type == TUN_MSG_PTR) {
2528                 struct tun_page tpage;
2529                 int n = ctl->num;
2530                 int flush = 0, queued = 0;
2531
2532                 memset(&tpage, 0, sizeof(tpage));
2533
2534                 local_bh_disable();
2535                 rcu_read_lock();
2536
2537                 for (i = 0; i < n; i++) {
2538                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2539                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2540                         if (ret > 0)
2541                                 queued += ret;
2542                 }
2543
2544                 if (flush)
2545                         xdp_do_flush();
2546
2547                 if (tfile->napi_enabled && queued > 0)
2548                         napi_schedule(&tfile->napi);
2549
2550                 rcu_read_unlock();
2551                 local_bh_enable();
2552
2553                 tun_put_page(&tpage);
2554
2555                 ret = total_len;
2556                 goto out;
2557         }
2558
2559         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2560                            m->msg_flags & MSG_DONTWAIT,
2561                            m->msg_flags & MSG_MORE);
2562 out:
2563         tun_put(tun);
2564         return ret;
2565 }
2566
2567 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2568                        int flags)
2569 {
2570         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2571         struct tun_struct *tun = tun_get(tfile);
2572         void *ptr = m->msg_control;
2573         int ret;
2574
2575         if (!tun) {
2576                 ret = -EBADFD;
2577                 goto out_free;
2578         }
2579
2580         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2581                 ret = -EINVAL;
2582                 goto out_put_tun;
2583         }
2584         if (flags & MSG_ERRQUEUE) {
2585                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2586                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2587                 goto out;
2588         }
2589         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2590         if (ret > (ssize_t)total_len) {
2591                 m->msg_flags |= MSG_TRUNC;
2592                 ret = flags & MSG_TRUNC ? ret : total_len;
2593         }
2594 out:
2595         tun_put(tun);
2596         return ret;
2597
2598 out_put_tun:
2599         tun_put(tun);
2600 out_free:
2601         tun_ptr_free(ptr);
2602         return ret;
2603 }
2604
2605 static int tun_ptr_peek_len(void *ptr)
2606 {
2607         if (likely(ptr)) {
2608                 if (tun_is_xdp_frame(ptr)) {
2609                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2610
2611                         return xdpf->len;
2612                 }
2613                 return __skb_array_len_with_tag(ptr);
2614         } else {
2615                 return 0;
2616         }
2617 }
2618
2619 static int tun_peek_len(struct socket *sock)
2620 {
2621         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2622         struct tun_struct *tun;
2623         int ret = 0;
2624
2625         tun = tun_get(tfile);
2626         if (!tun)
2627                 return 0;
2628
2629         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2630         tun_put(tun);
2631
2632         return ret;
2633 }
2634
2635 /* Ops structure to mimic raw sockets with tun */
2636 static const struct proto_ops tun_socket_ops = {
2637         .peek_len = tun_peek_len,
2638         .sendmsg = tun_sendmsg,
2639         .recvmsg = tun_recvmsg,
2640 };
2641
2642 static struct proto tun_proto = {
2643         .name           = "tun",
2644         .owner          = THIS_MODULE,
2645         .obj_size       = sizeof(struct tun_file),
2646 };
2647
2648 static int tun_flags(struct tun_struct *tun)
2649 {
2650         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2651 }
2652
2653 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2654                               char *buf)
2655 {
2656         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2657         return sprintf(buf, "0x%x\n", tun_flags(tun));
2658 }
2659
2660 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2661                           char *buf)
2662 {
2663         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2664         return uid_valid(tun->owner)?
2665                 sprintf(buf, "%u\n",
2666                         from_kuid_munged(current_user_ns(), tun->owner)):
2667                 sprintf(buf, "-1\n");
2668 }
2669
2670 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2671                           char *buf)
2672 {
2673         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2674         return gid_valid(tun->group) ?
2675                 sprintf(buf, "%u\n",
2676                         from_kgid_munged(current_user_ns(), tun->group)):
2677                 sprintf(buf, "-1\n");
2678 }
2679
2680 static DEVICE_ATTR_RO(tun_flags);
2681 static DEVICE_ATTR_RO(owner);
2682 static DEVICE_ATTR_RO(group);
2683
2684 static struct attribute *tun_dev_attrs[] = {
2685         &dev_attr_tun_flags.attr,
2686         &dev_attr_owner.attr,
2687         &dev_attr_group.attr,
2688         NULL
2689 };
2690
2691 static const struct attribute_group tun_attr_group = {
2692         .attrs = tun_dev_attrs
2693 };
2694
2695 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2696 {
2697         struct tun_struct *tun;
2698         struct tun_file *tfile = file->private_data;
2699         struct net_device *dev;
2700         int err;
2701
2702         if (tfile->detached)
2703                 return -EINVAL;
2704
2705         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2706                 if (!capable(CAP_NET_ADMIN))
2707                         return -EPERM;
2708
2709                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2710                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2711                         return -EINVAL;
2712         }
2713
2714         dev = __dev_get_by_name(net, ifr->ifr_name);
2715         if (dev) {
2716                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2717                         return -EBUSY;
2718                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2719                         tun = netdev_priv(dev);
2720                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2721                         tun = netdev_priv(dev);
2722                 else
2723                         return -EINVAL;
2724
2725                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2726                     !!(tun->flags & IFF_MULTI_QUEUE))
2727                         return -EINVAL;
2728
2729                 if (tun_not_capable(tun))
2730                         return -EPERM;
2731                 err = security_tun_dev_open(tun->security);
2732                 if (err < 0)
2733                         return err;
2734
2735                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2736                                  ifr->ifr_flags & IFF_NAPI,
2737                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2738                 if (err < 0)
2739                         return err;
2740
2741                 if (tun->flags & IFF_MULTI_QUEUE &&
2742                     (tun->numqueues + tun->numdisabled > 1)) {
2743                         /* One or more queue has already been attached, no need
2744                          * to initialize the device again.
2745                          */
2746                         netdev_state_change(dev);
2747                         return 0;
2748                 }
2749
2750                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2751                               (ifr->ifr_flags & TUN_FEATURES);
2752
2753                 netdev_state_change(dev);
2754         } else {
2755                 char *name;
2756                 unsigned long flags = 0;
2757                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2758                              MAX_TAP_QUEUES : 1;
2759
2760                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2761                         return -EPERM;
2762                 err = security_tun_dev_create();
2763                 if (err < 0)
2764                         return err;
2765
2766                 /* Set dev type */
2767                 if (ifr->ifr_flags & IFF_TUN) {
2768                         /* TUN device */
2769                         flags |= IFF_TUN;
2770                         name = "tun%d";
2771                 } else if (ifr->ifr_flags & IFF_TAP) {
2772                         /* TAP device */
2773                         flags |= IFF_TAP;
2774                         name = "tap%d";
2775                 } else
2776                         return -EINVAL;
2777
2778                 if (*ifr->ifr_name)
2779                         name = ifr->ifr_name;
2780
2781                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2782                                        NET_NAME_UNKNOWN, tun_setup, queues,
2783                                        queues);
2784
2785                 if (!dev)
2786                         return -ENOMEM;
2787
2788                 dev_net_set(dev, net);
2789                 dev->rtnl_link_ops = &tun_link_ops;
2790                 dev->ifindex = tfile->ifindex;
2791                 dev->sysfs_groups[0] = &tun_attr_group;
2792
2793                 tun = netdev_priv(dev);
2794                 tun->dev = dev;
2795                 tun->flags = flags;
2796                 tun->txflt.count = 0;
2797                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2798
2799                 tun->align = NET_SKB_PAD;
2800                 tun->filter_attached = false;
2801                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2802                 tun->rx_batched = 0;
2803                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2804
2805                 tun->ifr = ifr;
2806                 tun->file = file;
2807
2808                 tun_net_initialize(dev);
2809
2810                 err = register_netdevice(tun->dev);
2811                 if (err < 0) {
2812                         free_netdev(dev);
2813                         return err;
2814                 }
2815                 /* free_netdev() won't check refcnt, to avoid race
2816                  * with dev_put() we need publish tun after registration.
2817                  */
2818                 rcu_assign_pointer(tfile->tun, tun);
2819         }
2820
2821         netif_carrier_on(tun->dev);
2822
2823         /* Make sure persistent devices do not get stuck in
2824          * xoff state.
2825          */
2826         if (netif_running(tun->dev))
2827                 netif_tx_wake_all_queues(tun->dev);
2828
2829         strcpy(ifr->ifr_name, tun->dev->name);
2830         return 0;
2831 }
2832
2833 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2834 {
2835         strcpy(ifr->ifr_name, tun->dev->name);
2836
2837         ifr->ifr_flags = tun_flags(tun);
2838
2839 }
2840
2841 /* This is like a cut-down ethtool ops, except done via tun fd so no
2842  * privs required. */
2843 static int set_offload(struct tun_struct *tun, unsigned long arg)
2844 {
2845         netdev_features_t features = 0;
2846
2847         if (arg & TUN_F_CSUM) {
2848                 features |= NETIF_F_HW_CSUM;
2849                 arg &= ~TUN_F_CSUM;
2850
2851                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2852                         if (arg & TUN_F_TSO_ECN) {
2853                                 features |= NETIF_F_TSO_ECN;
2854                                 arg &= ~TUN_F_TSO_ECN;
2855                         }
2856                         if (arg & TUN_F_TSO4)
2857                                 features |= NETIF_F_TSO;
2858                         if (arg & TUN_F_TSO6)
2859                                 features |= NETIF_F_TSO6;
2860                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2861                 }
2862
2863                 arg &= ~TUN_F_UFO;
2864         }
2865
2866         /* This gives the user a way to test for new features in future by
2867          * trying to set them. */
2868         if (arg)
2869                 return -EINVAL;
2870
2871         tun->set_features = features;
2872         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2873         tun->dev->wanted_features |= features;
2874         netdev_update_features(tun->dev);
2875
2876         return 0;
2877 }
2878
2879 static void tun_detach_filter(struct tun_struct *tun, int n)
2880 {
2881         int i;
2882         struct tun_file *tfile;
2883
2884         for (i = 0; i < n; i++) {
2885                 tfile = rtnl_dereference(tun->tfiles[i]);
2886                 lock_sock(tfile->socket.sk);
2887                 sk_detach_filter(tfile->socket.sk);
2888                 release_sock(tfile->socket.sk);
2889         }
2890
2891         tun->filter_attached = false;
2892 }
2893
2894 static int tun_attach_filter(struct tun_struct *tun)
2895 {
2896         int i, ret = 0;
2897         struct tun_file *tfile;
2898
2899         for (i = 0; i < tun->numqueues; i++) {
2900                 tfile = rtnl_dereference(tun->tfiles[i]);
2901                 lock_sock(tfile->socket.sk);
2902                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2903                 release_sock(tfile->socket.sk);
2904                 if (ret) {
2905                         tun_detach_filter(tun, i);
2906                         return ret;
2907                 }
2908         }
2909
2910         tun->filter_attached = true;
2911         return ret;
2912 }
2913
2914 static void tun_set_sndbuf(struct tun_struct *tun)
2915 {
2916         struct tun_file *tfile;
2917         int i;
2918
2919         for (i = 0; i < tun->numqueues; i++) {
2920                 tfile = rtnl_dereference(tun->tfiles[i]);
2921                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2922         }
2923 }
2924
2925 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2926 {
2927         struct tun_file *tfile = file->private_data;
2928         struct tun_struct *tun;
2929         int ret = 0;
2930
2931         rtnl_lock();
2932
2933         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2934                 tun = tfile->detached;
2935                 if (!tun) {
2936                         ret = -EINVAL;
2937                         goto unlock;
2938                 }
2939                 ret = security_tun_dev_attach_queue(tun->security);
2940                 if (ret < 0)
2941                         goto unlock;
2942                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2943                                  tun->flags & IFF_NAPI_FRAGS, true);
2944         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2945                 tun = rtnl_dereference(tfile->tun);
2946                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2947                         ret = -EINVAL;
2948                 else
2949                         __tun_detach(tfile, false);
2950         } else
2951                 ret = -EINVAL;
2952
2953         if (ret >= 0)
2954                 netdev_state_change(tun->dev);
2955
2956 unlock:
2957         rtnl_unlock();
2958         return ret;
2959 }
2960
2961 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2962                         void __user *data)
2963 {
2964         struct bpf_prog *prog;
2965         int fd;
2966
2967         if (copy_from_user(&fd, data, sizeof(fd)))
2968                 return -EFAULT;
2969
2970         if (fd == -1) {
2971                 prog = NULL;
2972         } else {
2973                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2974                 if (IS_ERR(prog))
2975                         return PTR_ERR(prog);
2976         }
2977
2978         return __tun_set_ebpf(tun, prog_p, prog);
2979 }
2980
2981 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2982 static unsigned char tun_get_addr_len(unsigned short type)
2983 {
2984         switch (type) {
2985         case ARPHRD_IP6GRE:
2986         case ARPHRD_TUNNEL6:
2987                 return sizeof(struct in6_addr);
2988         case ARPHRD_IPGRE:
2989         case ARPHRD_TUNNEL:
2990         case ARPHRD_SIT:
2991                 return 4;
2992         case ARPHRD_ETHER:
2993                 return ETH_ALEN;
2994         case ARPHRD_IEEE802154:
2995         case ARPHRD_IEEE802154_MONITOR:
2996                 return IEEE802154_EXTENDED_ADDR_LEN;
2997         case ARPHRD_PHONET_PIPE:
2998         case ARPHRD_PPP:
2999         case ARPHRD_NONE:
3000                 return 0;
3001         case ARPHRD_6LOWPAN:
3002                 return EUI64_ADDR_LEN;
3003         case ARPHRD_FDDI:
3004                 return FDDI_K_ALEN;
3005         case ARPHRD_HIPPI:
3006                 return HIPPI_ALEN;
3007         case ARPHRD_IEEE802:
3008                 return FC_ALEN;
3009         case ARPHRD_ROSE:
3010                 return ROSE_ADDR_LEN;
3011         case ARPHRD_NETROM:
3012                 return AX25_ADDR_LEN;
3013         case ARPHRD_LOCALTLK:
3014                 return LTALK_ALEN;
3015         default:
3016                 return 0;
3017         }
3018 }
3019
3020 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3021                             unsigned long arg, int ifreq_len)
3022 {
3023         struct tun_file *tfile = file->private_data;
3024         struct net *net = sock_net(&tfile->sk);
3025         struct tun_struct *tun;
3026         void __user* argp = (void __user*)arg;
3027         unsigned int ifindex, carrier;
3028         struct ifreq ifr;
3029         kuid_t owner;
3030         kgid_t group;
3031         int sndbuf;
3032         int vnet_hdr_sz;
3033         int le;
3034         int ret;
3035         bool do_notify = false;
3036
3037         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3038             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3039                 if (copy_from_user(&ifr, argp, ifreq_len))
3040                         return -EFAULT;
3041         } else {
3042                 memset(&ifr, 0, sizeof(ifr));
3043         }
3044         if (cmd == TUNGETFEATURES) {
3045                 /* Currently this just means: "what IFF flags are valid?".
3046                  * This is needed because we never checked for invalid flags on
3047                  * TUNSETIFF.
3048                  */
3049                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3050                                 (unsigned int __user*)argp);
3051         } else if (cmd == TUNSETQUEUE) {
3052                 return tun_set_queue(file, &ifr);
3053         } else if (cmd == SIOCGSKNS) {
3054                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3055                         return -EPERM;
3056                 return open_related_ns(&net->ns, get_net_ns);
3057         }
3058
3059         rtnl_lock();
3060
3061         tun = tun_get(tfile);
3062         if (cmd == TUNSETIFF) {
3063                 ret = -EEXIST;
3064                 if (tun)
3065                         goto unlock;
3066
3067                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3068
3069                 ret = tun_set_iff(net, file, &ifr);
3070
3071                 if (ret)
3072                         goto unlock;
3073
3074                 if (copy_to_user(argp, &ifr, ifreq_len))
3075                         ret = -EFAULT;
3076                 goto unlock;
3077         }
3078         if (cmd == TUNSETIFINDEX) {
3079                 ret = -EPERM;
3080                 if (tun)
3081                         goto unlock;
3082
3083                 ret = -EFAULT;
3084                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3085                         goto unlock;
3086
3087                 ret = 0;
3088                 tfile->ifindex = ifindex;
3089                 goto unlock;
3090         }
3091
3092         ret = -EBADFD;
3093         if (!tun)
3094                 goto unlock;
3095
3096         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3097
3098         net = dev_net(tun->dev);
3099         ret = 0;
3100         switch (cmd) {
3101         case TUNGETIFF:
3102                 tun_get_iff(tun, &ifr);
3103
3104                 if (tfile->detached)
3105                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3106                 if (!tfile->socket.sk->sk_filter)
3107                         ifr.ifr_flags |= IFF_NOFILTER;
3108
3109                 if (copy_to_user(argp, &ifr, ifreq_len))
3110                         ret = -EFAULT;
3111                 break;
3112
3113         case TUNSETNOCSUM:
3114                 /* Disable/Enable checksum */
3115
3116                 /* [unimplemented] */
3117                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3118                            arg ? "disabled" : "enabled");
3119                 break;
3120
3121         case TUNSETPERSIST:
3122                 /* Disable/Enable persist mode. Keep an extra reference to the
3123                  * module to prevent the module being unprobed.
3124                  */
3125                 if (arg && !(tun->flags & IFF_PERSIST)) {
3126                         tun->flags |= IFF_PERSIST;
3127                         __module_get(THIS_MODULE);
3128                         do_notify = true;
3129                 }
3130                 if (!arg && (tun->flags & IFF_PERSIST)) {
3131                         tun->flags &= ~IFF_PERSIST;
3132                         module_put(THIS_MODULE);
3133                         do_notify = true;
3134                 }
3135
3136                 netif_info(tun, drv, tun->dev, "persist %s\n",
3137                            arg ? "enabled" : "disabled");
3138                 break;
3139
3140         case TUNSETOWNER:
3141                 /* Set owner of the device */
3142                 owner = make_kuid(current_user_ns(), arg);
3143                 if (!uid_valid(owner)) {
3144                         ret = -EINVAL;
3145                         break;
3146                 }
3147                 tun->owner = owner;
3148                 do_notify = true;
3149                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3150                            from_kuid(&init_user_ns, tun->owner));
3151                 break;
3152
3153         case TUNSETGROUP:
3154                 /* Set group of the device */
3155                 group = make_kgid(current_user_ns(), arg);
3156                 if (!gid_valid(group)) {
3157                         ret = -EINVAL;
3158                         break;
3159                 }
3160                 tun->group = group;
3161                 do_notify = true;
3162                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3163                            from_kgid(&init_user_ns, tun->group));
3164                 break;
3165
3166         case TUNSETLINK:
3167                 /* Only allow setting the type when the interface is down */
3168                 if (tun->dev->flags & IFF_UP) {
3169                         netif_info(tun, drv, tun->dev,
3170                                    "Linktype set failed because interface is up\n");
3171                         ret = -EBUSY;
3172                 } else {
3173                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3174                                                        tun->dev);
3175                         ret = notifier_to_errno(ret);
3176                         if (ret) {
3177                                 netif_info(tun, drv, tun->dev,
3178                                            "Refused to change device type\n");
3179                                 break;
3180                         }
3181                         tun->dev->type = (int) arg;
3182                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3183                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3184                                    tun->dev->type);
3185                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3186                                                  tun->dev);
3187                 }
3188                 break;
3189
3190         case TUNSETDEBUG:
3191                 tun->msg_enable = (u32)arg;
3192                 break;
3193
3194         case TUNSETOFFLOAD:
3195                 ret = set_offload(tun, arg);
3196                 break;
3197
3198         case TUNSETTXFILTER:
3199                 /* Can be set only for TAPs */
3200                 ret = -EINVAL;
3201                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3202                         break;
3203                 ret = update_filter(&tun->txflt, (void __user *)arg);
3204                 break;
3205
3206         case SIOCGIFHWADDR:
3207                 /* Get hw address */
3208                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3209                 if (copy_to_user(argp, &ifr, ifreq_len))
3210                         ret = -EFAULT;
3211                 break;
3212
3213         case SIOCSIFHWADDR:
3214                 /* Set hw address */
3215                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3216                 break;
3217
3218         case TUNGETSNDBUF:
3219                 sndbuf = tfile->socket.sk->sk_sndbuf;
3220                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3221                         ret = -EFAULT;
3222                 break;
3223
3224         case TUNSETSNDBUF:
3225                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3226                         ret = -EFAULT;
3227                         break;
3228                 }
3229                 if (sndbuf <= 0) {
3230                         ret = -EINVAL;
3231                         break;
3232                 }
3233
3234                 tun->sndbuf = sndbuf;
3235                 tun_set_sndbuf(tun);
3236                 break;
3237
3238         case TUNGETVNETHDRSZ:
3239                 vnet_hdr_sz = tun->vnet_hdr_sz;
3240                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3241                         ret = -EFAULT;
3242                 break;
3243
3244         case TUNSETVNETHDRSZ:
3245                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3246                         ret = -EFAULT;
3247                         break;
3248                 }
3249                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3250                         ret = -EINVAL;
3251                         break;
3252                 }
3253
3254                 tun->vnet_hdr_sz = vnet_hdr_sz;
3255                 break;
3256
3257         case TUNGETVNETLE:
3258                 le = !!(tun->flags & TUN_VNET_LE);
3259                 if (put_user(le, (int __user *)argp))
3260                         ret = -EFAULT;
3261                 break;
3262
3263         case TUNSETVNETLE:
3264                 if (get_user(le, (int __user *)argp)) {
3265                         ret = -EFAULT;
3266                         break;
3267                 }
3268                 if (le)
3269                         tun->flags |= TUN_VNET_LE;
3270                 else
3271                         tun->flags &= ~TUN_VNET_LE;
3272                 break;
3273
3274         case TUNGETVNETBE:
3275                 ret = tun_get_vnet_be(tun, argp);
3276                 break;
3277
3278         case TUNSETVNETBE:
3279                 ret = tun_set_vnet_be(tun, argp);
3280                 break;
3281
3282         case TUNATTACHFILTER:
3283                 /* Can be set only for TAPs */
3284                 ret = -EINVAL;
3285                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3286                         break;
3287                 ret = -EFAULT;
3288                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3289                         break;
3290
3291                 ret = tun_attach_filter(tun);
3292                 break;
3293
3294         case TUNDETACHFILTER:
3295                 /* Can be set only for TAPs */
3296                 ret = -EINVAL;
3297                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3298                         break;
3299                 ret = 0;
3300                 tun_detach_filter(tun, tun->numqueues);
3301                 break;
3302
3303         case TUNGETFILTER:
3304                 ret = -EINVAL;
3305                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3306                         break;
3307                 ret = -EFAULT;
3308                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3309                         break;
3310                 ret = 0;
3311                 break;
3312
3313         case TUNSETSTEERINGEBPF:
3314                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3315                 break;
3316
3317         case TUNSETFILTEREBPF:
3318                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3319                 break;
3320
3321         case TUNSETCARRIER:
3322                 ret = -EFAULT;
3323                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3324                         goto unlock;
3325
3326                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3327                 break;
3328
3329         case TUNGETDEVNETNS:
3330                 ret = -EPERM;
3331                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3332                         goto unlock;
3333                 ret = open_related_ns(&net->ns, get_net_ns);
3334                 break;
3335
3336         default:
3337                 ret = -EINVAL;
3338                 break;
3339         }
3340
3341         if (do_notify)
3342                 netdev_state_change(tun->dev);
3343
3344 unlock:
3345         rtnl_unlock();
3346         if (tun)
3347                 tun_put(tun);
3348         return ret;
3349 }
3350
3351 static long tun_chr_ioctl(struct file *file,
3352                           unsigned int cmd, unsigned long arg)
3353 {
3354         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3355 }
3356
3357 #ifdef CONFIG_COMPAT
3358 static long tun_chr_compat_ioctl(struct file *file,
3359                          unsigned int cmd, unsigned long arg)
3360 {
3361         switch (cmd) {
3362         case TUNSETIFF:
3363         case TUNGETIFF:
3364         case TUNSETTXFILTER:
3365         case TUNGETSNDBUF:
3366         case TUNSETSNDBUF:
3367         case SIOCGIFHWADDR:
3368         case SIOCSIFHWADDR:
3369                 arg = (unsigned long)compat_ptr(arg);
3370                 break;
3371         default:
3372                 arg = (compat_ulong_t)arg;
3373                 break;
3374         }
3375
3376         /*
3377          * compat_ifreq is shorter than ifreq, so we must not access beyond
3378          * the end of that structure. All fields that are used in this
3379          * driver are compatible though, we don't need to convert the
3380          * contents.
3381          */
3382         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3383 }
3384 #endif /* CONFIG_COMPAT */
3385
3386 static int tun_chr_fasync(int fd, struct file *file, int on)
3387 {
3388         struct tun_file *tfile = file->private_data;
3389         int ret;
3390
3391         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3392                 goto out;
3393
3394         if (on) {
3395                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3396                 tfile->flags |= TUN_FASYNC;
3397         } else
3398                 tfile->flags &= ~TUN_FASYNC;
3399         ret = 0;
3400 out:
3401         return ret;
3402 }
3403
3404 static int tun_chr_open(struct inode *inode, struct file * file)
3405 {
3406         struct net *net = current->nsproxy->net_ns;
3407         struct tun_file *tfile;
3408
3409         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3410                                             &tun_proto, 0);
3411         if (!tfile)
3412                 return -ENOMEM;
3413         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3414                 sk_free(&tfile->sk);
3415                 return -ENOMEM;
3416         }
3417
3418         mutex_init(&tfile->napi_mutex);
3419         RCU_INIT_POINTER(tfile->tun, NULL);
3420         tfile->flags = 0;
3421         tfile->ifindex = 0;
3422
3423         init_waitqueue_head(&tfile->socket.wq.wait);
3424
3425         tfile->socket.file = file;
3426         tfile->socket.ops = &tun_socket_ops;
3427
3428         sock_init_data(&tfile->socket, &tfile->sk);
3429
3430         tfile->sk.sk_write_space = tun_sock_write_space;
3431         tfile->sk.sk_sndbuf = INT_MAX;
3432
3433         file->private_data = tfile;
3434         INIT_LIST_HEAD(&tfile->next);
3435
3436         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3437
3438         return 0;
3439 }
3440
3441 static int tun_chr_close(struct inode *inode, struct file *file)
3442 {
3443         struct tun_file *tfile = file->private_data;
3444
3445         tun_detach(tfile, true);
3446
3447         return 0;
3448 }
3449
3450 #ifdef CONFIG_PROC_FS
3451 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3452 {
3453         struct tun_file *tfile = file->private_data;
3454         struct tun_struct *tun;
3455         struct ifreq ifr;
3456
3457         memset(&ifr, 0, sizeof(ifr));
3458
3459         rtnl_lock();
3460         tun = tun_get(tfile);
3461         if (tun)
3462                 tun_get_iff(tun, &ifr);
3463         rtnl_unlock();
3464
3465         if (tun)
3466                 tun_put(tun);
3467
3468         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3469 }
3470 #endif
3471
3472 static const struct file_operations tun_fops = {
3473         .owner  = THIS_MODULE,
3474         .llseek = no_llseek,
3475         .read_iter  = tun_chr_read_iter,
3476         .write_iter = tun_chr_write_iter,
3477         .poll   = tun_chr_poll,
3478         .unlocked_ioctl = tun_chr_ioctl,
3479 #ifdef CONFIG_COMPAT
3480         .compat_ioctl = tun_chr_compat_ioctl,
3481 #endif
3482         .open   = tun_chr_open,
3483         .release = tun_chr_close,
3484         .fasync = tun_chr_fasync,
3485 #ifdef CONFIG_PROC_FS
3486         .show_fdinfo = tun_chr_show_fdinfo,
3487 #endif
3488 };
3489
3490 static struct miscdevice tun_miscdev = {
3491         .minor = TUN_MINOR,
3492         .name = "tun",
3493         .nodename = "net/tun",
3494         .fops = &tun_fops,
3495 };
3496
3497 /* ethtool interface */
3498
3499 static void tun_default_link_ksettings(struct net_device *dev,
3500                                        struct ethtool_link_ksettings *cmd)
3501 {
3502         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3503         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3504         cmd->base.speed         = SPEED_10;
3505         cmd->base.duplex        = DUPLEX_FULL;
3506         cmd->base.port          = PORT_TP;
3507         cmd->base.phy_address   = 0;
3508         cmd->base.autoneg       = AUTONEG_DISABLE;
3509 }
3510
3511 static int tun_get_link_ksettings(struct net_device *dev,
3512                                   struct ethtool_link_ksettings *cmd)
3513 {
3514         struct tun_struct *tun = netdev_priv(dev);
3515
3516         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3517         return 0;
3518 }
3519
3520 static int tun_set_link_ksettings(struct net_device *dev,
3521                                   const struct ethtool_link_ksettings *cmd)
3522 {
3523         struct tun_struct *tun = netdev_priv(dev);
3524
3525         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3526         return 0;
3527 }
3528
3529 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3530 {
3531         struct tun_struct *tun = netdev_priv(dev);
3532
3533         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3534         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3535
3536         switch (tun->flags & TUN_TYPE_MASK) {
3537         case IFF_TUN:
3538                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3539                 break;
3540         case IFF_TAP:
3541                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3542                 break;
3543         }
3544 }
3545
3546 static u32 tun_get_msglevel(struct net_device *dev)
3547 {
3548         struct tun_struct *tun = netdev_priv(dev);
3549
3550         return tun->msg_enable;
3551 }
3552
3553 static void tun_set_msglevel(struct net_device *dev, u32 value)
3554 {
3555         struct tun_struct *tun = netdev_priv(dev);
3556
3557         tun->msg_enable = value;
3558 }
3559
3560 static int tun_get_coalesce(struct net_device *dev,
3561                             struct ethtool_coalesce *ec,
3562                             struct kernel_ethtool_coalesce *kernel_coal,
3563                             struct netlink_ext_ack *extack)
3564 {
3565         struct tun_struct *tun = netdev_priv(dev);
3566
3567         ec->rx_max_coalesced_frames = tun->rx_batched;
3568
3569         return 0;
3570 }
3571
3572 static int tun_set_coalesce(struct net_device *dev,
3573                             struct ethtool_coalesce *ec,
3574                             struct kernel_ethtool_coalesce *kernel_coal,
3575                             struct netlink_ext_ack *extack)
3576 {
3577         struct tun_struct *tun = netdev_priv(dev);
3578
3579         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3580                 tun->rx_batched = NAPI_POLL_WEIGHT;
3581         else
3582                 tun->rx_batched = ec->rx_max_coalesced_frames;
3583
3584         return 0;
3585 }
3586
3587 static const struct ethtool_ops tun_ethtool_ops = {
3588         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3589         .get_drvinfo    = tun_get_drvinfo,
3590         .get_msglevel   = tun_get_msglevel,
3591         .set_msglevel   = tun_set_msglevel,
3592         .get_link       = ethtool_op_get_link,
3593         .get_ts_info    = ethtool_op_get_ts_info,
3594         .get_coalesce   = tun_get_coalesce,
3595         .set_coalesce   = tun_set_coalesce,
3596         .get_link_ksettings = tun_get_link_ksettings,
3597         .set_link_ksettings = tun_set_link_ksettings,
3598 };
3599
3600 static int tun_queue_resize(struct tun_struct *tun)
3601 {
3602         struct net_device *dev = tun->dev;
3603         struct tun_file *tfile;
3604         struct ptr_ring **rings;
3605         int n = tun->numqueues + tun->numdisabled;
3606         int ret, i;
3607
3608         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3609         if (!rings)
3610                 return -ENOMEM;
3611
3612         for (i = 0; i < tun->numqueues; i++) {
3613                 tfile = rtnl_dereference(tun->tfiles[i]);
3614                 rings[i] = &tfile->tx_ring;
3615         }
3616         list_for_each_entry(tfile, &tun->disabled, next)
3617                 rings[i++] = &tfile->tx_ring;
3618
3619         ret = ptr_ring_resize_multiple(rings, n,
3620                                        dev->tx_queue_len, GFP_KERNEL,
3621                                        tun_ptr_free);
3622
3623         kfree(rings);
3624         return ret;
3625 }
3626
3627 static int tun_device_event(struct notifier_block *unused,
3628                             unsigned long event, void *ptr)
3629 {
3630         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3631         struct tun_struct *tun = netdev_priv(dev);
3632         int i;
3633
3634         if (dev->rtnl_link_ops != &tun_link_ops)
3635                 return NOTIFY_DONE;
3636
3637         switch (event) {
3638         case NETDEV_CHANGE_TX_QUEUE_LEN:
3639                 if (tun_queue_resize(tun))
3640                         return NOTIFY_BAD;
3641                 break;
3642         case NETDEV_UP:
3643                 for (i = 0; i < tun->numqueues; i++) {
3644                         struct tun_file *tfile;
3645
3646                         tfile = rtnl_dereference(tun->tfiles[i]);
3647                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3648                 }
3649                 break;
3650         default:
3651                 break;
3652         }
3653
3654         return NOTIFY_DONE;
3655 }
3656
3657 static struct notifier_block tun_notifier_block __read_mostly = {
3658         .notifier_call  = tun_device_event,
3659 };
3660
3661 static int __init tun_init(void)
3662 {
3663         int ret = 0;
3664
3665         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3666
3667         ret = rtnl_link_register(&tun_link_ops);
3668         if (ret) {
3669                 pr_err("Can't register link_ops\n");
3670                 goto err_linkops;
3671         }
3672
3673         ret = misc_register(&tun_miscdev);
3674         if (ret) {
3675                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3676                 goto err_misc;
3677         }
3678
3679         ret = register_netdevice_notifier(&tun_notifier_block);
3680         if (ret) {
3681                 pr_err("Can't register netdevice notifier\n");
3682                 goto err_notifier;
3683         }
3684
3685         return  0;
3686
3687 err_notifier:
3688         misc_deregister(&tun_miscdev);
3689 err_misc:
3690         rtnl_link_unregister(&tun_link_ops);
3691 err_linkops:
3692         return ret;
3693 }
3694
3695 static void tun_cleanup(void)
3696 {
3697         misc_deregister(&tun_miscdev);
3698         rtnl_link_unregister(&tun_link_ops);
3699         unregister_netdevice_notifier(&tun_notifier_block);
3700 }
3701
3702 /* Get an underlying socket object from tun file.  Returns error unless file is
3703  * attached to a device.  The returned object works like a packet socket, it
3704  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3705  * holding a reference to the file for as long as the socket is in use. */
3706 struct socket *tun_get_socket(struct file *file)
3707 {
3708         struct tun_file *tfile;
3709         if (file->f_op != &tun_fops)
3710                 return ERR_PTR(-EINVAL);
3711         tfile = file->private_data;
3712         if (!tfile)
3713                 return ERR_PTR(-EBADFD);
3714         return &tfile->socket;
3715 }
3716 EXPORT_SYMBOL_GPL(tun_get_socket);
3717
3718 struct ptr_ring *tun_get_tx_ring(struct file *file)
3719 {
3720         struct tun_file *tfile;
3721
3722         if (file->f_op != &tun_fops)
3723                 return ERR_PTR(-EINVAL);
3724         tfile = file->private_data;
3725         if (!tfile)
3726                 return ERR_PTR(-EBADFD);
3727         return &tfile->tx_ring;
3728 }
3729 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3730
3731 module_init(tun_init);
3732 module_exit(tun_cleanup);
3733 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3734 MODULE_AUTHOR(DRV_COPYRIGHT);
3735 MODULE_LICENSE("GPL");
3736 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3737 MODULE_ALIAS("devname:net/tun");