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 struct tun_pcpu_stats {
 119         u64_stats_t rx_packets;
 120         u64_stats_t rx_bytes;
 121         u64_stats_t tx_packets;
 122         u64_stats_t tx_bytes;
 123         struct u64_stats_sync syncp;
 124         u32 rx_dropped;
 125         u32 tx_dropped;
 126         u32 rx_frame_errors;
 127 };
 128
 129 /* A tun_file connects an open character device to a tuntap netdevice. It
 130  * also contains all socket related structures (except sock_fprog and tap_filter)
 131  * to serve as one transmit queue for tuntap device. The sock_fprog and
 132  * tap_filter were kept in tun_struct since they were used for filtering for the
 133  * netdevice not for a specific queue (at least I didn't see the requirement for
 134  * this).
 135  *
 136  * RCU usage:
 137  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 138  * other can only be read while rcu_read_lock or rtnl_lock is held.
 139  */
 140 struct tun_file {
 141         struct sock sk;
 142         struct socket socket;
 143         struct tun_struct __rcu *tun;
 144         struct fasync_struct *fasync;
 145         /* only used for fasnyc */
 146         unsigned int flags;
 147         union {
 148                 u16 queue_index;
 149                 unsigned int ifindex;
 150         };
 151         struct napi_struct napi;
 152         bool napi_enabled;
 153         bool napi_frags_enabled;
 154         struct mutex napi_mutex;        /* Protects access to the above napi */
 155         struct list_head next;
 156         struct tun_struct *detached;
 157         struct ptr_ring tx_ring;
 158         struct xdp_rxq_info xdp_rxq;
 159 };
 160
 161 struct tun_page {
 162         struct page *page;
 163         int count;
 164 };
 165
 166 struct tun_flow_entry {
 167         struct hlist_node hash_link;
 168         struct rcu_head rcu;
 169         struct tun_struct *tun;
 170
 171         u32 rxhash;
 172         u32 rps_rxhash;
 173         int queue_index;
 174         unsigned long updated ____cacheline_aligned_in_smp;
 175 };
 176
 177 #define TUN_NUM_FLOW_ENTRIES 1024
 178 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 179
 180 struct tun_prog {
 181         struct rcu_head rcu;
 182         struct bpf_prog *prog;
 183 };
 184
 185 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 186  * device, socket filter, sndbuf and vnet header size were restore when the
 187  * file were attached to a persist device.
 188  */
 189 struct tun_struct {
 190         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 191         unsigned int            numqueues;
 192         unsigned int            flags;
 193         kuid_t                  owner;
 194         kgid_t                  group;
 195
 196         struct net_device       *dev;
 197         netdev_features_t       set_features;
 198 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 199                           NETIF_F_TSO6)
 200
 201         int                     align;
 202         int                     vnet_hdr_sz;
 203         int                     sndbuf;
 204         struct tap_filter       txflt;
 205         struct sock_fprog       fprog;
 206         /* protected by rtnl lock */
 207         bool                    filter_attached;
 208         u32                     msg_enable;
 209         spinlock_t lock;
 210         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 211         struct timer_list flow_gc_timer;
 212         unsigned long ageing_time;
 213         unsigned int numdisabled;
 214         struct list_head disabled;
 215         void *security;
 216         u32 flow_count;
 217         u32 rx_batched;
 218         struct tun_pcpu_stats __percpu *pcpu_stats;
 219         struct bpf_prog __rcu *xdp_prog;
 220         struct tun_prog __rcu *steering_prog;
 221         struct tun_prog __rcu *filter_prog;
 222         struct ethtool_link_ksettings link_ksettings;
 223         /* init args */
 224         struct file *file;
 225         struct ifreq *ifr;
 226 };
 227
 228 struct veth {
 229         __be16 h_vlan_proto;
 230         __be16 h_vlan_TCI;
 231 };
 232
 233 static void tun_flow_init(struct tun_struct *tun);
 234 static void tun_flow_uninit(struct tun_struct *tun);
 235
 236 static int tun_napi_receive(struct napi_struct *napi, int budget)
 237 {
 238         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 239         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 240         struct sk_buff_head process_queue;
 241         struct sk_buff *skb;
 242         int received = 0;
 243
 244         __skb_queue_head_init(&process_queue);
 245
 246         spin_lock(&queue->lock);
 247         skb_queue_splice_tail_init(queue, &process_queue);
 248         spin_unlock(&queue->lock);
 249
 250         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 251                 napi_gro_receive(napi, skb);
 252                 ++received;
 253         }
 254
 255         if (!skb_queue_empty(&process_queue)) {
 256                 spin_lock(&queue->lock);
 257                 skb_queue_splice(&process_queue, queue);
 258                 spin_unlock(&queue->lock);
 259         }
 260
 261         return received;
 262 }
 263
 264 static int tun_napi_poll(struct napi_struct *napi, int budget)
 265 {
 266         unsigned int received;
 267
 268         received = tun_napi_receive(napi, budget);
 269
 270         if (received < budget)
 271                 napi_complete_done(napi, received);
 272
 273         return received;
 274 }
 275
 276 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 277                           bool napi_en, bool napi_frags)
 278 {
 279         tfile->napi_enabled = napi_en;
 280         tfile->napi_frags_enabled = napi_en && napi_frags;
 281         if (napi_en) {
 282                 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 283                                   NAPI_POLL_WEIGHT);
 284                 napi_enable(&tfile->napi);
 285         }
 286 }
 287
 288 static void tun_napi_enable(struct tun_file *tfile)
 289 {
 290         if (tfile->napi_enabled)
 291                 napi_enable(&tfile->napi);
 292 }
 293
 294 static void tun_napi_disable(struct tun_file *tfile)
 295 {
 296         if (tfile->napi_enabled)
 297                 napi_disable(&tfile->napi);
 298 }
 299
 300 static void tun_napi_del(struct tun_file *tfile)
 301 {
 302         if (tfile->napi_enabled)
 303                 netif_napi_del(&tfile->napi);
 304 }
 305
 306 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 307 {
 308         return tfile->napi_frags_enabled;
 309 }
 310
 311 #ifdef CONFIG_TUN_VNET_CROSS_LE
 312 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 313 {
 314         return tun->flags & TUN_VNET_BE ? false :
 315                 virtio_legacy_is_little_endian();
 316 }
 317
 318 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 319 {
 320         int be = !!(tun->flags & TUN_VNET_BE);
 321
 322         if (put_user(be, argp))
 323                 return -EFAULT;
 324
 325         return 0;
 326 }
 327
 328 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 329 {
 330         int be;
 331
 332         if (get_user(be, argp))
 333                 return -EFAULT;
 334
 335         if (be)
 336                 tun->flags |= TUN_VNET_BE;
 337         else
 338                 tun->flags &= ~TUN_VNET_BE;
 339
 340         return 0;
 341 }
 342 #else
 343 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 344 {
 345         return virtio_legacy_is_little_endian();
 346 }
 347
 348 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 349 {
 350         return -EINVAL;
 351 }
 352
 353 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 354 {
 355         return -EINVAL;
 356 }
 357 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 358
 359 static inline bool tun_is_little_endian(struct tun_struct *tun)
 360 {
 361         return tun->flags & TUN_VNET_LE ||
 362                 tun_legacy_is_little_endian(tun);
 363 }
 364
 365 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 366 {
 367         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 368 }
 369
 370 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 371 {
 372         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 373 }
 374
 375 static inline u32 tun_hashfn(u32 rxhash)
 376 {
 377         return rxhash & TUN_MASK_FLOW_ENTRIES;
 378 }
 379
 380 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 381 {
 382         struct tun_flow_entry *e;
 383
 384         hlist_for_each_entry_rcu(e, head, hash_link) {
 385                 if (e->rxhash == rxhash)
 386                         return e;
 387         }
 388         return NULL;
 389 }
 390
 391 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 392                                               struct hlist_head *head,
 393                                               u32 rxhash, u16 queue_index)
 394 {
 395         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 396
 397         if (e) {
 398                 netif_info(tun, tx_queued, tun->dev,
 399                            "create flow: hash %u index %u\n",
 400                            rxhash, queue_index);
 401                 e->updated = jiffies;
 402                 e->rxhash = rxhash;
 403                 e->rps_rxhash = 0;
 404                 e->queue_index = queue_index;
 405                 e->tun = tun;
 406                 hlist_add_head_rcu(&e->hash_link, head);
 407                 ++tun->flow_count;
 408         }
 409         return e;
 410 }
 411
 412 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 413 {
 414         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 415                    e->rxhash, e->queue_index);
 416         hlist_del_rcu(&e->hash_link);
 417         kfree_rcu(e, rcu);
 418         --tun->flow_count;
 419 }
 420
 421 static void tun_flow_flush(struct tun_struct *tun)
 422 {
 423         int i;
 424
 425         spin_lock_bh(&tun->lock);
 426         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 427                 struct tun_flow_entry *e;
 428                 struct hlist_node *n;
 429
 430                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 431                         tun_flow_delete(tun, e);
 432         }
 433         spin_unlock_bh(&tun->lock);
 434 }
 435
 436 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 437 {
 438         int i;
 439
 440         spin_lock_bh(&tun->lock);
 441         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 442                 struct tun_flow_entry *e;
 443                 struct hlist_node *n;
 444
 445                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 446                         if (e->queue_index == queue_index)
 447                                 tun_flow_delete(tun, e);
 448                 }
 449         }
 450         spin_unlock_bh(&tun->lock);
 451 }
 452
 453 static void tun_flow_cleanup(struct timer_list *t)
 454 {
 455         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 456         unsigned long delay = tun->ageing_time;
 457         unsigned long next_timer = jiffies + delay;
 458         unsigned long count = 0;
 459         int i;
 460
 461         spin_lock(&tun->lock);
 462         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 463                 struct tun_flow_entry *e;
 464                 struct hlist_node *n;
 465
 466                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 467                         unsigned long this_timer;
 468
 469                         this_timer = e->updated + delay;
 470                         if (time_before_eq(this_timer, jiffies)) {
 471                                 tun_flow_delete(tun, e);
 472                                 continue;
 473                         }
 474                         count++;
 475                         if (time_before(this_timer, next_timer))
 476                                 next_timer = this_timer;
 477                 }
 478         }
 479
 480         if (count)
 481                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 482         spin_unlock(&tun->lock);
 483 }
 484
 485 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 486                             struct tun_file *tfile)
 487 {
 488         struct hlist_head *head;
 489         struct tun_flow_entry *e;
 490         unsigned long delay = tun->ageing_time;
 491         u16 queue_index = tfile->queue_index;
 492
 493         head = &tun->flows[tun_hashfn(rxhash)];
 494
 495         rcu_read_lock();
 496
 497         e = tun_flow_find(head, rxhash);
 498         if (likely(e)) {
 499                 /* TODO: keep queueing to old queue until it's empty? */
 500                 if (READ_ONCE(e->queue_index) != queue_index)
 501                         WRITE_ONCE(e->queue_index, queue_index);
 502                 if (e->updated != jiffies)
 503                         e->updated = jiffies;
 504                 sock_rps_record_flow_hash(e->rps_rxhash);
 505         } else {
 506                 spin_lock_bh(&tun->lock);
 507                 if (!tun_flow_find(head, rxhash) &&
 508                     tun->flow_count < MAX_TAP_FLOWS)
 509                         tun_flow_create(tun, head, rxhash, queue_index);
 510
 511                 if (!timer_pending(&tun->flow_gc_timer))
 512                         mod_timer(&tun->flow_gc_timer,
 513                                   round_jiffies_up(jiffies + delay));
 514                 spin_unlock_bh(&tun->lock);
 515         }
 516
 517         rcu_read_unlock();
 518 }
 519
 520 /* Save the hash received in the stack receive path and update the
 521  * flow_hash table accordingly.
 522  */
 523 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 524 {
 525         if (unlikely(e->rps_rxhash != hash))
 526                 e->rps_rxhash = hash;
 527 }
 528
 529 /* We try to identify a flow through its rxhash. The reason that
 530  * we do not check rxq no. is because some cards(e.g 82599), chooses
 531  * the rxq based on the txq where the last packet of the flow comes. As
 532  * the userspace application move between processors, we may get a
 533  * different rxq no. here.
 534  */
 535 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 536 {
 537         struct tun_flow_entry *e;
 538         u32 txq = 0;
 539         u32 numqueues = 0;
 540
 541         numqueues = READ_ONCE(tun->numqueues);
 542
 543         txq = __skb_get_hash_symmetric(skb);
 544         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 545         if (e) {
 546                 tun_flow_save_rps_rxhash(e, txq);
 547                 txq = e->queue_index;
 548         } else {
 549                 /* use multiply and shift instead of expensive divide */
 550                 txq = ((u64)txq * numqueues) >> 32;
 551         }
 552
 553         return txq;
 554 }
 555
 556 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 557 {
 558         struct tun_prog *prog;
 559         u32 numqueues;
 560         u16 ret = 0;
 561
 562         numqueues = READ_ONCE(tun->numqueues);
 563         if (!numqueues)
 564                 return 0;
 565
 566         prog = rcu_dereference(tun->steering_prog);
 567         if (prog)
 568                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 569
 570         return ret % numqueues;
 571 }
 572
 573 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 574                             struct net_device *sb_dev)
 575 {
 576         struct tun_struct *tun = netdev_priv(dev);
 577         u16 ret;
 578
 579         rcu_read_lock();
 580         if (rcu_dereference(tun->steering_prog))
 581                 ret = tun_ebpf_select_queue(tun, skb);
 582         else
 583                 ret = tun_automq_select_queue(tun, skb);
 584         rcu_read_unlock();
 585
 586         return ret;
 587 }
 588
 589 static inline bool tun_not_capable(struct tun_struct *tun)
 590 {
 591         const struct cred *cred = current_cred();
 592         struct net *net = dev_net(tun->dev);
 593
 594         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 595                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 596                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 597 }
 598
 599 static void tun_set_real_num_queues(struct tun_struct *tun)
 600 {
 601         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 602         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 603 }
 604
 605 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 606 {
 607         tfile->detached = tun;
 608         list_add_tail(&tfile->next, &tun->disabled);
 609         ++tun->numdisabled;
 610 }
 611
 612 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 613 {
 614         struct tun_struct *tun = tfile->detached;
 615
 616         tfile->detached = NULL;
 617         list_del_init(&tfile->next);
 618         --tun->numdisabled;
 619         return tun;
 620 }
 621
 622 void tun_ptr_free(void *ptr)
 623 {
 624         if (!ptr)
 625                 return;
 626         if (tun_is_xdp_frame(ptr)) {
 627                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 628
 629                 xdp_return_frame(xdpf);
 630         } else {
 631                 __skb_array_destroy_skb(ptr);
 632         }
 633 }
 634 EXPORT_SYMBOL_GPL(tun_ptr_free);
 635
 636 static void tun_queue_purge(struct tun_file *tfile)
 637 {
 638         void *ptr;
 639
 640         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 641                 tun_ptr_free(ptr);
 642
 643         skb_queue_purge(&tfile->sk.sk_write_queue);
 644         skb_queue_purge(&tfile->sk.sk_error_queue);
 645 }
 646
 647 static void __tun_detach(struct tun_file *tfile, bool clean)
 648 {
 649         struct tun_file *ntfile;
 650         struct tun_struct *tun;
 651
 652         tun = rtnl_dereference(tfile->tun);
 653
 654         if (tun && clean) {
 655                 if (!tfile->detached)
 656                         tun_napi_disable(tfile);
 657                 tun_napi_del(tfile);
 658         }
 659
 660         if (tun && !tfile->detached) {
 661                 u16 index = tfile->queue_index;
 662                 BUG_ON(index >= tun->numqueues);
 663
 664                 rcu_assign_pointer(tun->tfiles[index],
 665                                    tun->tfiles[tun->numqueues - 1]);
 666                 ntfile = rtnl_dereference(tun->tfiles[index]);
 667                 ntfile->queue_index = index;
 668                 ntfile->xdp_rxq.queue_index = index;
 669                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 670                                    NULL);
 671
 672                 --tun->numqueues;
 673                 if (clean) {
 674                         RCU_INIT_POINTER(tfile->tun, NULL);
 675                         sock_put(&tfile->sk);
 676                 } else {
 677                         tun_disable_queue(tun, tfile);
 678                         tun_napi_disable(tfile);
 679                 }
 680
 681                 synchronize_net();
 682                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 683                 /* Drop read queue */
 684                 tun_queue_purge(tfile);
 685                 tun_set_real_num_queues(tun);
 686         } else if (tfile->detached && clean) {
 687                 tun = tun_enable_queue(tfile);
 688                 sock_put(&tfile->sk);
 689         }
 690
 691         if (clean) {
 692                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 693                         netif_carrier_off(tun->dev);
 694
 695                         if (!(tun->flags & IFF_PERSIST) &&
 696                             tun->dev->reg_state == NETREG_REGISTERED)
 697                                 unregister_netdevice(tun->dev);
 698                 }
 699                 if (tun)
 700                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 701                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 702         }
 703 }
 704
 705 static void tun_detach(struct tun_file *tfile, bool clean)
 706 {
 707         struct tun_struct *tun;
 708         struct net_device *dev;
 709
 710         rtnl_lock();
 711         tun = rtnl_dereference(tfile->tun);
 712         dev = tun ? tun->dev : NULL;
 713         __tun_detach(tfile, clean);
 714         if (dev)
 715                 netdev_state_change(dev);
 716         rtnl_unlock();
 717
 718         if (clean)
 719                 sock_put(&tfile->sk);
 720 }
 721
 722 static void tun_detach_all(struct net_device *dev)
 723 {
 724         struct tun_struct *tun = netdev_priv(dev);
 725         struct tun_file *tfile, *tmp;
 726         int i, n = tun->numqueues;
 727
 728         for (i = 0; i < n; i++) {
 729                 tfile = rtnl_dereference(tun->tfiles[i]);
 730                 BUG_ON(!tfile);
 731                 tun_napi_disable(tfile);
 732                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 733                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 734                 RCU_INIT_POINTER(tfile->tun, NULL);
 735                 --tun->numqueues;
 736         }
 737         list_for_each_entry(tfile, &tun->disabled, next) {
 738                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 739                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 740                 RCU_INIT_POINTER(tfile->tun, NULL);
 741         }
 742         BUG_ON(tun->numqueues != 0);
 743
 744         synchronize_net();
 745         for (i = 0; i < n; i++) {
 746                 tfile = rtnl_dereference(tun->tfiles[i]);
 747                 tun_napi_del(tfile);
 748                 /* Drop read queue */
 749                 tun_queue_purge(tfile);
 750                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 751                 sock_put(&tfile->sk);
 752         }
 753         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 754                 tun_napi_del(tfile);
 755                 tun_enable_queue(tfile);
 756                 tun_queue_purge(tfile);
 757                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 758                 sock_put(&tfile->sk);
 759         }
 760         BUG_ON(tun->numdisabled != 0);
 761
 762         if (tun->flags & IFF_PERSIST)
 763                 module_put(THIS_MODULE);
 764 }
 765
 766 static int tun_attach(struct tun_struct *tun, struct file *file,
 767                       bool skip_filter, bool napi, bool napi_frags,
 768                       bool publish_tun)
 769 {
 770         struct tun_file *tfile = file->private_data;
 771         struct net_device *dev = tun->dev;
 772         int err;
 773
 774         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 775         if (err < 0)
 776                 goto out;
 777
 778         err = -EINVAL;
 779         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 780                 goto out;
 781
 782         err = -EBUSY;
 783         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 784                 goto out;
 785
 786         err = -E2BIG;
 787         if (!tfile->detached &&
 788             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 789                 goto out;
 790
 791         err = 0;
 792
 793         /* Re-attach the filter to persist device */
 794         if (!skip_filter && (tun->filter_attached == true)) {
 795                 lock_sock(tfile->socket.sk);
 796                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 797                 release_sock(tfile->socket.sk);
 798                 if (!err)
 799                         goto out;
 800         }
 801
 802         if (!tfile->detached &&
 803             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 804                             GFP_KERNEL, tun_ptr_free)) {
 805                 err = -ENOMEM;
 806                 goto out;
 807         }
 808
 809         tfile->queue_index = tun->numqueues;
 810         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 811
 812         if (tfile->detached) {
 813                 /* Re-attach detached tfile, updating XDP queue_index */
 814                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 815
 816                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 817                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 818         } else {
 819                 /* Setup XDP RX-queue info, for new tfile getting attached */
 820                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 821                                        tun->dev, tfile->queue_index);
 822                 if (err < 0)
 823                         goto out;
 824                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 825                                                  MEM_TYPE_PAGE_SHARED, NULL);
 826                 if (err < 0) {
 827                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 828                         goto out;
 829                 }
 830                 err = 0;
 831         }
 832
 833         if (tfile->detached) {
 834                 tun_enable_queue(tfile);
 835                 tun_napi_enable(tfile);
 836         } else {
 837                 sock_hold(&tfile->sk);
 838                 tun_napi_init(tun, tfile, napi, napi_frags);
 839         }
 840
 841         if (rtnl_dereference(tun->xdp_prog))
 842                 sock_set_flag(&tfile->sk, SOCK_XDP);
 843
 844         /* device is allowed to go away first, so no need to hold extra
 845          * refcnt.
 846          */
 847
 848         /* Publish tfile->tun and tun->tfiles only after we've fully
 849          * initialized tfile; otherwise we risk using half-initialized
 850          * object.
 851          */
 852         if (publish_tun)
 853                 rcu_assign_pointer(tfile->tun, tun);
 854         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 855         tun->numqueues++;
 856         tun_set_real_num_queues(tun);
 857 out:
 858         return err;
 859 }
 860
 861 static struct tun_struct *tun_get(struct tun_file *tfile)
 862 {
 863         struct tun_struct *tun;
 864
 865         rcu_read_lock();
 866         tun = rcu_dereference(tfile->tun);
 867         if (tun)
 868                 dev_hold(tun->dev);
 869         rcu_read_unlock();
 870
 871         return tun;
 872 }
 873
 874 static void tun_put(struct tun_struct *tun)
 875 {
 876         dev_put(tun->dev);
 877 }
 878
 879 /* TAP filtering */
 880 static void addr_hash_set(u32 *mask, const u8 *addr)
 881 {
 882         int n = ether_crc(ETH_ALEN, addr) >> 26;
 883         mask[n >> 5] |= (1 << (n & 31));
 884 }
 885
 886 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 887 {
 888         int n = ether_crc(ETH_ALEN, addr) >> 26;
 889         return mask[n >> 5] & (1 << (n & 31));
 890 }
 891
 892 static int update_filter(struct tap_filter *filter, void __user *arg)
 893 {
 894         struct { u8 u[ETH_ALEN]; } *addr;
 895         struct tun_filter uf;
 896         int err, alen, n, nexact;
 897
 898         if (copy_from_user(&uf, arg, sizeof(uf)))
 899                 return -EFAULT;
 900
 901         if (!uf.count) {
 902                 /* Disabled */
 903                 filter->count = 0;
 904                 return 0;
 905         }
 906
 907         alen = ETH_ALEN * uf.count;
 908         addr = memdup_user(arg + sizeof(uf), alen);
 909         if (IS_ERR(addr))
 910                 return PTR_ERR(addr);
 911
 912         /* The filter is updated without holding any locks. Which is
 913          * perfectly safe. We disable it first and in the worst
 914          * case we'll accept a few undesired packets. */
 915         filter->count = 0;
 916         wmb();
 917
 918         /* Use first set of addresses as an exact filter */
 919         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 920                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 921
 922         nexact = n;
 923
 924         /* Remaining multicast addresses are hashed,
 925          * unicast will leave the filter disabled. */
 926         memset(filter->mask, 0, sizeof(filter->mask));
 927         for (; n < uf.count; n++) {
 928                 if (!is_multicast_ether_addr(addr[n].u)) {
 929                         err = 0; /* no filter */
 930                         goto free_addr;
 931                 }
 932                 addr_hash_set(filter->mask, addr[n].u);
 933         }
 934
 935         /* For ALLMULTI just set the mask to all ones.
 936          * This overrides the mask populated above. */
 937         if ((uf.flags & TUN_FLT_ALLMULTI))
 938                 memset(filter->mask, ~0, sizeof(filter->mask));
 939
 940         /* Now enable the filter */
 941         wmb();
 942         filter->count = nexact;
 943
 944         /* Return the number of exact filters */
 945         err = nexact;
 946 free_addr:
 947         kfree(addr);
 948         return err;
 949 }
 950
 951 /* Returns: 0 - drop, !=0 - accept */
 952 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 953 {
 954         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 955          * at this point. */
 956         struct ethhdr *eh = (struct ethhdr *) skb->data;
 957         int i;
 958
 959         /* Exact match */
 960         for (i = 0; i < filter->count; i++)
 961                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 962                         return 1;
 963
 964         /* Inexact match (multicast only) */
 965         if (is_multicast_ether_addr(eh->h_dest))
 966                 return addr_hash_test(filter->mask, eh->h_dest);
 967
 968         return 0;
 969 }
 970
 971 /*
 972  * Checks whether the packet is accepted or not.
 973  * Returns: 0 - drop, !=0 - accept
 974  */
 975 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 976 {
 977         if (!filter->count)
 978                 return 1;
 979
 980         return run_filter(filter, skb);
 981 }
 982
 983 /* Network device part of the driver */
 984
 985 static const struct ethtool_ops tun_ethtool_ops;
 986
 987 static int tun_net_init(struct net_device *dev)
 988 {
 989         struct tun_struct *tun = netdev_priv(dev);
 990         struct ifreq *ifr = tun->ifr;
 991         int err;
 992
 993         tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
 994         if (!tun->pcpu_stats)
 995                 return -ENOMEM;
 996
 997         spin_lock_init(&tun->lock);
 998
 999         err = security_tun_dev_alloc_security(&tun->security);
1000         if (err < 0) {
1001                 free_percpu(tun->pcpu_stats);
1002                 return err;
1003         }
1004
1005         tun_flow_init(tun);
1006
1007         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1008                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1009                            NETIF_F_HW_VLAN_STAG_TX;
1010         dev->features = dev->hw_features | NETIF_F_LLTX;
1011         dev->vlan_features = dev->features &
1012                              ~(NETIF_F_HW_VLAN_CTAG_TX |
1013                                NETIF_F_HW_VLAN_STAG_TX);
1014
1015         tun->flags = (tun->flags & ~TUN_FEATURES) |
1016                       (ifr->ifr_flags & TUN_FEATURES);
1017
1018         INIT_LIST_HEAD(&tun->disabled);
1019         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1020                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1021         if (err < 0) {
1022                 tun_flow_uninit(tun);
1023                 security_tun_dev_free_security(tun->security);
1024                 free_percpu(tun->pcpu_stats);
1025                 return err;
1026         }
1027         return 0;
1028 }
1029
1030 /* Net device detach from fd. */
1031 static void tun_net_uninit(struct net_device *dev)
1032 {
1033         tun_detach_all(dev);
1034 }
1035
1036 /* Net device open. */
1037 static int tun_net_open(struct net_device *dev)
1038 {
1039         netif_tx_start_all_queues(dev);
1040
1041         return 0;
1042 }
1043
1044 /* Net device close. */
1045 static int tun_net_close(struct net_device *dev)
1046 {
1047         netif_tx_stop_all_queues(dev);
1048         return 0;
1049 }
1050
1051 /* Net device start xmit */
1052 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1053 {
1054 #ifdef CONFIG_RPS
1055         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1056                 /* Select queue was not called for the skbuff, so we extract the
1057                  * RPS hash and save it into the flow_table here.
1058                  */
1059                 struct tun_flow_entry *e;
1060                 __u32 rxhash;
1061
1062                 rxhash = __skb_get_hash_symmetric(skb);
1063                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1064                 if (e)
1065                         tun_flow_save_rps_rxhash(e, rxhash);
1066         }
1067 #endif
1068 }
1069
1070 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1071                                     struct sk_buff *skb,
1072                                     int len)
1073 {
1074         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1075
1076         if (prog)
1077                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1078
1079         return len;
1080 }
1081
1082 /* Net device start xmit */
1083 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1084 {
1085         struct tun_struct *tun = netdev_priv(dev);
1086         int txq = skb->queue_mapping;
1087         struct netdev_queue *queue;
1088         struct tun_file *tfile;
1089         int len = skb->len;
1090
1091         rcu_read_lock();
1092         tfile = rcu_dereference(tun->tfiles[txq]);
1093
1094         /* Drop packet if interface is not attached */
1095         if (!tfile)
1096                 goto drop;
1097
1098         if (!rcu_dereference(tun->steering_prog))
1099                 tun_automq_xmit(tun, skb);
1100
1101         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1102
1103         /* Drop if the filter does not like it.
1104          * This is a noop if the filter is disabled.
1105          * Filter can be enabled only for the TAP devices. */
1106         if (!check_filter(&tun->txflt, skb))
1107                 goto drop;
1108
1109         if (tfile->socket.sk->sk_filter &&
1110             sk_filter(tfile->socket.sk, skb))
1111                 goto drop;
1112
1113         len = run_ebpf_filter(tun, skb, len);
1114         if (len == 0 || pskb_trim(skb, len))
1115                 goto drop;
1116
1117         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1118                 goto drop;
1119
1120         skb_tx_timestamp(skb);
1121
1122         /* Orphan the skb - required as we might hang on to it
1123          * for indefinite time.
1124          */
1125         skb_orphan(skb);
1126
1127         nf_reset_ct(skb);
1128
1129         if (ptr_ring_produce(&tfile->tx_ring, skb))
1130                 goto drop;
1131
1132         /* NETIF_F_LLTX requires to do our own update of trans_start */
1133         queue = netdev_get_tx_queue(dev, txq);
1134         queue->trans_start = jiffies;
1135
1136         /* Notify and wake up reader process */
1137         if (tfile->flags & TUN_FASYNC)
1138                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1139         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1140
1141         rcu_read_unlock();
1142         return NETDEV_TX_OK;
1143
1144 drop:
1145         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1146         skb_tx_error(skb);
1147         kfree_skb(skb);
1148         rcu_read_unlock();
1149         return NET_XMIT_DROP;
1150 }
1151
1152 static void tun_net_mclist(struct net_device *dev)
1153 {
1154         /*
1155          * This callback is supposed to deal with mc filter in
1156          * _rx_ path and has nothing to do with the _tx_ path.
1157          * In rx path we always accept everything userspace gives us.
1158          */
1159 }
1160
1161 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1162         netdev_features_t features)
1163 {
1164         struct tun_struct *tun = netdev_priv(dev);
1165
1166         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1167 }
1168
1169 static void tun_set_headroom(struct net_device *dev, int new_hr)
1170 {
1171         struct tun_struct *tun = netdev_priv(dev);
1172
1173         if (new_hr < NET_SKB_PAD)
1174                 new_hr = NET_SKB_PAD;
1175
1176         tun->align = new_hr;
1177 }
1178
1179 static void
1180 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1181 {
1182         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1183         struct tun_struct *tun = netdev_priv(dev);
1184         struct tun_pcpu_stats *p;
1185         int i;
1186
1187         for_each_possible_cpu(i) {
1188                 u64 rxpackets, rxbytes, txpackets, txbytes;
1189                 unsigned int start;
1190
1191                 p = per_cpu_ptr(tun->pcpu_stats, i);
1192                 do {
1193                         start = u64_stats_fetch_begin(&p->syncp);
1194                         rxpackets       = u64_stats_read(&p->rx_packets);
1195                         rxbytes         = u64_stats_read(&p->rx_bytes);
1196                         txpackets       = u64_stats_read(&p->tx_packets);
1197                         txbytes         = u64_stats_read(&p->tx_bytes);
1198                 } while (u64_stats_fetch_retry(&p->syncp, start));
1199
1200                 stats->rx_packets       += rxpackets;
1201                 stats->rx_bytes         += rxbytes;
1202                 stats->tx_packets       += txpackets;
1203                 stats->tx_bytes         += txbytes;
1204
1205                 /* u32 counters */
1206                 rx_dropped      += p->rx_dropped;
1207                 rx_frame_errors += p->rx_frame_errors;
1208                 tx_dropped      += p->tx_dropped;
1209         }
1210         stats->rx_dropped  = rx_dropped;
1211         stats->rx_frame_errors = rx_frame_errors;
1212         stats->tx_dropped = tx_dropped;
1213 }
1214
1215 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1216                        struct netlink_ext_ack *extack)
1217 {
1218         struct tun_struct *tun = netdev_priv(dev);
1219         struct tun_file *tfile;
1220         struct bpf_prog *old_prog;
1221         int i;
1222
1223         old_prog = rtnl_dereference(tun->xdp_prog);
1224         rcu_assign_pointer(tun->xdp_prog, prog);
1225         if (old_prog)
1226                 bpf_prog_put(old_prog);
1227
1228         for (i = 0; i < tun->numqueues; i++) {
1229                 tfile = rtnl_dereference(tun->tfiles[i]);
1230                 if (prog)
1231                         sock_set_flag(&tfile->sk, SOCK_XDP);
1232                 else
1233                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1234         }
1235         list_for_each_entry(tfile, &tun->disabled, next) {
1236                 if (prog)
1237                         sock_set_flag(&tfile->sk, SOCK_XDP);
1238                 else
1239                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1240         }
1241
1242         return 0;
1243 }
1244
1245 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1246 {
1247         switch (xdp->command) {
1248         case XDP_SETUP_PROG:
1249                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1250         default:
1251                 return -EINVAL;
1252         }
1253 }
1254
1255 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1256 {
1257         if (new_carrier) {
1258                 struct tun_struct *tun = netdev_priv(dev);
1259
1260                 if (!tun->numqueues)
1261                         return -EPERM;
1262
1263                 netif_carrier_on(dev);
1264         } else {
1265                 netif_carrier_off(dev);
1266         }
1267         return 0;
1268 }
1269
1270 static const struct net_device_ops tun_netdev_ops = {
1271         .ndo_init               = tun_net_init,
1272         .ndo_uninit             = tun_net_uninit,
1273         .ndo_open               = tun_net_open,
1274         .ndo_stop               = tun_net_close,
1275         .ndo_start_xmit         = tun_net_xmit,
1276         .ndo_fix_features       = tun_net_fix_features,
1277         .ndo_select_queue       = tun_select_queue,
1278         .ndo_set_rx_headroom    = tun_set_headroom,
1279         .ndo_get_stats64        = tun_net_get_stats64,
1280         .ndo_change_carrier     = tun_net_change_carrier,
1281 };
1282
1283 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1284 {
1285         /* Notify and wake up reader process */
1286         if (tfile->flags & TUN_FASYNC)
1287                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1288         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1289 }
1290
1291 static int tun_xdp_xmit(struct net_device *dev, int n,
1292                         struct xdp_frame **frames, u32 flags)
1293 {
1294         struct tun_struct *tun = netdev_priv(dev);
1295         struct tun_file *tfile;
1296         u32 numqueues;
1297         int drops = 0;
1298         int cnt = n;
1299         int i;
1300
1301         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1302                 return -EINVAL;
1303
1304         rcu_read_lock();
1305
1306 resample:
1307         numqueues = READ_ONCE(tun->numqueues);
1308         if (!numqueues) {
1309                 rcu_read_unlock();
1310                 return -ENXIO; /* Caller will free/return all frames */
1311         }
1312
1313         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1314                                             numqueues]);
1315         if (unlikely(!tfile))
1316                 goto resample;
1317
1318         spin_lock(&tfile->tx_ring.producer_lock);
1319         for (i = 0; i < n; i++) {
1320                 struct xdp_frame *xdp = frames[i];
1321                 /* Encode the XDP flag into lowest bit for consumer to differ
1322                  * XDP buffer from sk_buff.
1323                  */
1324                 void *frame = tun_xdp_to_ptr(xdp);
1325
1326                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1327                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1328                         xdp_return_frame_rx_napi(xdp);
1329                         drops++;
1330                 }
1331         }
1332         spin_unlock(&tfile->tx_ring.producer_lock);
1333
1334         if (flags & XDP_XMIT_FLUSH)
1335                 __tun_xdp_flush_tfile(tfile);
1336
1337         rcu_read_unlock();
1338         return cnt - drops;
1339 }
1340
1341 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1342 {
1343         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1344
1345         if (unlikely(!frame))
1346                 return -EOVERFLOW;
1347
1348         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1349 }
1350
1351 static const struct net_device_ops tap_netdev_ops = {
1352         .ndo_init               = tun_net_init,
1353         .ndo_uninit             = tun_net_uninit,
1354         .ndo_open               = tun_net_open,
1355         .ndo_stop               = tun_net_close,
1356         .ndo_start_xmit         = tun_net_xmit,
1357         .ndo_fix_features       = tun_net_fix_features,
1358         .ndo_set_rx_mode        = tun_net_mclist,
1359         .ndo_set_mac_address    = eth_mac_addr,
1360         .ndo_validate_addr      = eth_validate_addr,
1361         .ndo_select_queue       = tun_select_queue,
1362         .ndo_features_check     = passthru_features_check,
1363         .ndo_set_rx_headroom    = tun_set_headroom,
1364         .ndo_get_stats64        = tun_net_get_stats64,
1365         .ndo_bpf                = tun_xdp,
1366         .ndo_xdp_xmit           = tun_xdp_xmit,
1367         .ndo_change_carrier     = tun_net_change_carrier,
1368 };
1369
1370 static void tun_flow_init(struct tun_struct *tun)
1371 {
1372         int i;
1373
1374         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1375                 INIT_HLIST_HEAD(&tun->flows[i]);
1376
1377         tun->ageing_time = TUN_FLOW_EXPIRE;
1378         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1379         mod_timer(&tun->flow_gc_timer,
1380                   round_jiffies_up(jiffies + tun->ageing_time));
1381 }
1382
1383 static void tun_flow_uninit(struct tun_struct *tun)
1384 {
1385         del_timer_sync(&tun->flow_gc_timer);
1386         tun_flow_flush(tun);
1387 }
1388
1389 #define MIN_MTU 68
1390 #define MAX_MTU 65535
1391
1392 /* Initialize net device. */
1393 static void tun_net_initialize(struct net_device *dev)
1394 {
1395         struct tun_struct *tun = netdev_priv(dev);
1396
1397         switch (tun->flags & TUN_TYPE_MASK) {
1398         case IFF_TUN:
1399                 dev->netdev_ops = &tun_netdev_ops;
1400                 dev->header_ops = &ip_tunnel_header_ops;
1401
1402                 /* Point-to-Point TUN Device */
1403                 dev->hard_header_len = 0;
1404                 dev->addr_len = 0;
1405                 dev->mtu = 1500;
1406
1407                 /* Zero header length */
1408                 dev->type = ARPHRD_NONE;
1409                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1410                 break;
1411
1412         case IFF_TAP:
1413                 dev->netdev_ops = &tap_netdev_ops;
1414                 /* Ethernet TAP Device */
1415                 ether_setup(dev);
1416                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1417                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1418
1419                 eth_hw_addr_random(dev);
1420
1421                 break;
1422         }
1423
1424         dev->min_mtu = MIN_MTU;
1425         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1426 }
1427
1428 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1429 {
1430         struct sock *sk = tfile->socket.sk;
1431
1432         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1433 }
1434
1435 /* Character device part */
1436
1437 /* Poll */
1438 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1439 {
1440         struct tun_file *tfile = file->private_data;
1441         struct tun_struct *tun = tun_get(tfile);
1442         struct sock *sk;
1443         __poll_t mask = 0;
1444
1445         if (!tun)
1446                 return EPOLLERR;
1447
1448         sk = tfile->socket.sk;
1449
1450         poll_wait(file, sk_sleep(sk), wait);
1451
1452         if (!ptr_ring_empty(&tfile->tx_ring))
1453                 mask |= EPOLLIN | EPOLLRDNORM;
1454
1455         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1456          * guarantee EPOLLOUT to be raised by either here or
1457          * tun_sock_write_space(). Then process could get notification
1458          * after it writes to a down device and meets -EIO.
1459          */
1460         if (tun_sock_writeable(tun, tfile) ||
1461             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1462              tun_sock_writeable(tun, tfile)))
1463                 mask |= EPOLLOUT | EPOLLWRNORM;
1464
1465         if (tun->dev->reg_state != NETREG_REGISTERED)
1466                 mask = EPOLLERR;
1467
1468         tun_put(tun);
1469         return mask;
1470 }
1471
1472 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1473                                             size_t len,
1474                                             const struct iov_iter *it)
1475 {
1476         struct sk_buff *skb;
1477         size_t linear;
1478         int err;
1479         int i;
1480
1481         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1482             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1483                 return ERR_PTR(-EMSGSIZE);
1484
1485         local_bh_disable();
1486         skb = napi_get_frags(&tfile->napi);
1487         local_bh_enable();
1488         if (!skb)
1489                 return ERR_PTR(-ENOMEM);
1490
1491         linear = iov_iter_single_seg_count(it);
1492         err = __skb_grow(skb, linear);
1493         if (err)
1494                 goto free;
1495
1496         skb->len = len;
1497         skb->data_len = len - linear;
1498         skb->truesize += skb->data_len;
1499
1500         for (i = 1; i < it->nr_segs; i++) {
1501                 size_t fragsz = it->iov[i].iov_len;
1502                 struct page *page;
1503                 void *frag;
1504
1505                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1506                         err = -EINVAL;
1507                         goto free;
1508                 }
1509                 frag = netdev_alloc_frag(fragsz);
1510                 if (!frag) {
1511                         err = -ENOMEM;
1512                         goto free;
1513                 }
1514                 page = virt_to_head_page(frag);
1515                 skb_fill_page_desc(skb, i - 1, page,
1516                                    frag - page_address(page), fragsz);
1517         }
1518
1519         return skb;
1520 free:
1521         /* frees skb and all frags allocated with napi_alloc_frag() */
1522         napi_free_frags(&tfile->napi);
1523         return ERR_PTR(err);
1524 }
1525
1526 /* prepad is the amount to reserve at front.  len is length after that.
1527  * linear is a hint as to how much to copy (usually headers). */
1528 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1529                                      size_t prepad, size_t len,
1530                                      size_t linear, int noblock)
1531 {
1532         struct sock *sk = tfile->socket.sk;
1533         struct sk_buff *skb;
1534         int err;
1535
1536         /* Under a page?  Don't bother with paged skb. */
1537         if (prepad + len < PAGE_SIZE || !linear)
1538                 linear = len;
1539
1540         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1541                                    &err, 0);
1542         if (!skb)
1543                 return ERR_PTR(err);
1544
1545         skb_reserve(skb, prepad);
1546         skb_put(skb, linear);
1547         skb->data_len = len - linear;
1548         skb->len += len - linear;
1549
1550         return skb;
1551 }
1552
1553 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1554                            struct sk_buff *skb, int more)
1555 {
1556         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1557         struct sk_buff_head process_queue;
1558         u32 rx_batched = tun->rx_batched;
1559         bool rcv = false;
1560
1561         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1562                 local_bh_disable();
1563                 skb_record_rx_queue(skb, tfile->queue_index);
1564                 netif_receive_skb(skb);
1565                 local_bh_enable();
1566                 return;
1567         }
1568
1569         spin_lock(&queue->lock);
1570         if (!more || skb_queue_len(queue) == rx_batched) {
1571                 __skb_queue_head_init(&process_queue);
1572                 skb_queue_splice_tail_init(queue, &process_queue);
1573                 rcv = true;
1574         } else {
1575                 __skb_queue_tail(queue, skb);
1576         }
1577         spin_unlock(&queue->lock);
1578
1579         if (rcv) {
1580                 struct sk_buff *nskb;
1581
1582                 local_bh_disable();
1583                 while ((nskb = __skb_dequeue(&process_queue))) {
1584                         skb_record_rx_queue(nskb, tfile->queue_index);
1585                         netif_receive_skb(nskb);
1586                 }
1587                 skb_record_rx_queue(skb, tfile->queue_index);
1588                 netif_receive_skb(skb);
1589                 local_bh_enable();
1590         }
1591 }
1592
1593 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1594                               int len, int noblock, bool zerocopy)
1595 {
1596         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1597                 return false;
1598
1599         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1600                 return false;
1601
1602         if (!noblock)
1603                 return false;
1604
1605         if (zerocopy)
1606                 return false;
1607
1608         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1609             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1610                 return false;
1611
1612         return true;
1613 }
1614
1615 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1616                                        struct page_frag *alloc_frag, char *buf,
1617                                        int buflen, int len, int pad)
1618 {
1619         struct sk_buff *skb = build_skb(buf, buflen);
1620
1621         if (!skb)
1622                 return ERR_PTR(-ENOMEM);
1623
1624         skb_reserve(skb, pad);
1625         skb_put(skb, len);
1626         skb_set_owner_w(skb, tfile->socket.sk);
1627
1628         get_page(alloc_frag->page);
1629         alloc_frag->offset += buflen;
1630
1631         return skb;
1632 }
1633
1634 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1635                        struct xdp_buff *xdp, u32 act)
1636 {
1637         int err;
1638
1639         switch (act) {
1640         case XDP_REDIRECT:
1641                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1642                 if (err)
1643                         return err;
1644                 break;
1645         case XDP_TX:
1646                 err = tun_xdp_tx(tun->dev, xdp);
1647                 if (err < 0)
1648                         return err;
1649                 break;
1650         case XDP_PASS:
1651                 break;
1652         default:
1653                 bpf_warn_invalid_xdp_action(act);
1654                 fallthrough;
1655         case XDP_ABORTED:
1656                 trace_xdp_exception(tun->dev, xdp_prog, act);
1657                 fallthrough;
1658         case XDP_DROP:
1659                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1660                 break;
1661         }
1662
1663         return act;
1664 }
1665
1666 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1667                                      struct tun_file *tfile,
1668                                      struct iov_iter *from,
1669                                      struct virtio_net_hdr *hdr,
1670                                      int len, int *skb_xdp)
1671 {
1672         struct page_frag *alloc_frag = &current->task_frag;
1673         struct bpf_prog *xdp_prog;
1674         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1675         char *buf;
1676         size_t copied;
1677         int pad = TUN_RX_PAD;
1678         int err = 0;
1679
1680         rcu_read_lock();
1681         xdp_prog = rcu_dereference(tun->xdp_prog);
1682         if (xdp_prog)
1683                 pad += XDP_PACKET_HEADROOM;
1684         buflen += SKB_DATA_ALIGN(len + pad);
1685         rcu_read_unlock();
1686
1687         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1688         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1689                 return ERR_PTR(-ENOMEM);
1690
1691         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1692         copied = copy_page_from_iter(alloc_frag->page,
1693                                      alloc_frag->offset + pad,
1694                                      len, from);
1695         if (copied != len)
1696                 return ERR_PTR(-EFAULT);
1697
1698         /* There's a small window that XDP may be set after the check
1699          * of xdp_prog above, this should be rare and for simplicity
1700          * we do XDP on skb in case the headroom is not enough.
1701          */
1702         if (hdr->gso_type || !xdp_prog) {
1703                 *skb_xdp = 1;
1704                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1705                                        pad);
1706         }
1707
1708         *skb_xdp = 0;
1709
1710         local_bh_disable();
1711         rcu_read_lock();
1712         xdp_prog = rcu_dereference(tun->xdp_prog);
1713         if (xdp_prog) {
1714                 struct xdp_buff xdp;
1715                 u32 act;
1716
1717                 xdp.data_hard_start = buf;
1718                 xdp.data = buf + pad;
1719                 xdp_set_data_meta_invalid(&xdp);
1720                 xdp.data_end = xdp.data + len;
1721                 xdp.rxq = &tfile->xdp_rxq;
1722                 xdp.frame_sz = buflen;
1723
1724                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1725                 if (act == XDP_REDIRECT || act == XDP_TX) {
1726                         get_page(alloc_frag->page);
1727                         alloc_frag->offset += buflen;
1728                 }
1729                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1730                 if (err < 0) {
1731                         if (act == XDP_REDIRECT || act == XDP_TX)
1732                                 put_page(alloc_frag->page);
1733                         goto out;
1734                 }
1735
1736                 if (err == XDP_REDIRECT)
1737                         xdp_do_flush();
1738                 if (err != XDP_PASS)
1739                         goto out;
1740
1741                 pad = xdp.data - xdp.data_hard_start;
1742                 len = xdp.data_end - xdp.data;
1743         }
1744         rcu_read_unlock();
1745         local_bh_enable();
1746
1747         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1748
1749 out:
1750         rcu_read_unlock();
1751         local_bh_enable();
1752         return NULL;
1753 }
1754
1755 /* Get packet from user space buffer */
1756 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1757                             void *msg_control, struct iov_iter *from,
1758                             int noblock, bool more)
1759 {
1760         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1761         struct sk_buff *skb;
1762         size_t total_len = iov_iter_count(from);
1763         size_t len = total_len, align = tun->align, linear;
1764         struct virtio_net_hdr gso = { 0 };
1765         struct tun_pcpu_stats *stats;
1766         int good_linear;
1767         int copylen;
1768         bool zerocopy = false;
1769         int err;
1770         u32 rxhash = 0;
1771         int skb_xdp = 1;
1772         bool frags = tun_napi_frags_enabled(tfile);
1773
1774         if (!(tun->flags & IFF_NO_PI)) {
1775                 if (len < sizeof(pi))
1776                         return -EINVAL;
1777                 len -= sizeof(pi);
1778
1779                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1780                         return -EFAULT;
1781         }
1782
1783         if (tun->flags & IFF_VNET_HDR) {
1784                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1785
1786                 if (len < vnet_hdr_sz)
1787                         return -EINVAL;
1788                 len -= vnet_hdr_sz;
1789
1790                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1791                         return -EFAULT;
1792
1793                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1794                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1795                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1796
1797                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1798                         return -EINVAL;
1799                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1800         }
1801
1802         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1803                 align += NET_IP_ALIGN;
1804                 if (unlikely(len < ETH_HLEN ||
1805                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1806                         return -EINVAL;
1807         }
1808
1809         good_linear = SKB_MAX_HEAD(align);
1810
1811         if (msg_control) {
1812                 struct iov_iter i = *from;
1813
1814                 /* There are 256 bytes to be copied in skb, so there is
1815                  * enough room for skb expand head in case it is used.
1816                  * The rest of the buffer is mapped from userspace.
1817                  */
1818                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1819                 if (copylen > good_linear)
1820                         copylen = good_linear;
1821                 linear = copylen;
1822                 iov_iter_advance(&i, copylen);
1823                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1824                         zerocopy = true;
1825         }
1826
1827         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1828                 /* For the packet that is not easy to be processed
1829                  * (e.g gso or jumbo packet), we will do it at after
1830                  * skb was created with generic XDP routine.
1831                  */
1832                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1833                 if (IS_ERR(skb)) {
1834                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1835                         return PTR_ERR(skb);
1836                 }
1837                 if (!skb)
1838                         return total_len;
1839         } else {
1840                 if (!zerocopy) {
1841                         copylen = len;
1842                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1843                                 linear = good_linear;
1844                         else
1845                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1846                 }
1847
1848                 if (frags) {
1849                         mutex_lock(&tfile->napi_mutex);
1850                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1851                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1852                          * If zerocopy is enabled, then this layout will be
1853                          * overwritten by zerocopy_sg_from_iter().
1854                          */
1855                         zerocopy = false;
1856                 } else {
1857                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1858                                             noblock);
1859                 }
1860
1861                 if (IS_ERR(skb)) {
1862                         if (PTR_ERR(skb) != -EAGAIN)
1863                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1864                         if (frags)
1865                                 mutex_unlock(&tfile->napi_mutex);
1866                         return PTR_ERR(skb);
1867                 }
1868
1869                 if (zerocopy)
1870                         err = zerocopy_sg_from_iter(skb, from);
1871                 else
1872                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1873
1874                 if (err) {
1875                         err = -EFAULT;
1876 drop:
1877                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1878                         kfree_skb(skb);
1879                         if (frags) {
1880                                 tfile->napi.skb = NULL;
1881                                 mutex_unlock(&tfile->napi_mutex);
1882                         }
1883
1884                         return err;
1885                 }
1886         }
1887
1888         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1889                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1890                 kfree_skb(skb);
1891                 if (frags) {
1892                         tfile->napi.skb = NULL;
1893                         mutex_unlock(&tfile->napi_mutex);
1894                 }
1895
1896                 return -EINVAL;
1897         }
1898
1899         switch (tun->flags & TUN_TYPE_MASK) {
1900         case IFF_TUN:
1901                 if (tun->flags & IFF_NO_PI) {
1902                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1903
1904                         switch (ip_version) {
1905                         case 4:
1906                                 pi.proto = htons(ETH_P_IP);
1907                                 break;
1908                         case 6:
1909                                 pi.proto = htons(ETH_P_IPV6);
1910                                 break;
1911                         default:
1912                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1913                                 kfree_skb(skb);
1914                                 return -EINVAL;
1915                         }
1916                 }
1917
1918                 skb_reset_mac_header(skb);
1919                 skb->protocol = pi.proto;
1920                 skb->dev = tun->dev;
1921                 break;
1922         case IFF_TAP:
1923                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1924                         err = -ENOMEM;
1925                         goto drop;
1926                 }
1927                 skb->protocol = eth_type_trans(skb, tun->dev);
1928                 break;
1929         }
1930
1931         /* copy skb_ubuf_info for callback when skb has no error */
1932         if (zerocopy) {
1933                 skb_shinfo(skb)->destructor_arg = msg_control;
1934                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1935                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1936         } else if (msg_control) {
1937                 struct ubuf_info *uarg = msg_control;
1938                 uarg->callback(uarg, false);
1939         }
1940
1941         skb_reset_network_header(skb);
1942         skb_probe_transport_header(skb);
1943         skb_record_rx_queue(skb, tfile->queue_index);
1944
1945         if (skb_xdp) {
1946                 struct bpf_prog *xdp_prog;
1947                 int ret;
1948
1949                 local_bh_disable();
1950                 rcu_read_lock();
1951                 xdp_prog = rcu_dereference(tun->xdp_prog);
1952                 if (xdp_prog) {
1953                         ret = do_xdp_generic(xdp_prog, skb);
1954                         if (ret != XDP_PASS) {
1955                                 rcu_read_unlock();
1956                                 local_bh_enable();
1957                                 if (frags) {
1958                                         tfile->napi.skb = NULL;
1959                                         mutex_unlock(&tfile->napi_mutex);
1960                                 }
1961                                 return total_len;
1962                         }
1963                 }
1964                 rcu_read_unlock();
1965                 local_bh_enable();
1966         }
1967
1968         /* Compute the costly rx hash only if needed for flow updates.
1969          * We may get a very small possibility of OOO during switching, not
1970          * worth to optimize.
1971          */
1972         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1973             !tfile->detached)
1974                 rxhash = __skb_get_hash_symmetric(skb);
1975
1976         rcu_read_lock();
1977         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1978                 err = -EIO;
1979                 rcu_read_unlock();
1980                 goto drop;
1981         }
1982
1983         if (frags) {
1984                 u32 headlen;
1985
1986                 /* Exercise flow dissector code path. */
1987                 skb_push(skb, ETH_HLEN);
1988                 headlen = eth_get_headlen(tun->dev, skb->data,
1989                                           skb_headlen(skb));
1990
1991                 if (unlikely(headlen > skb_headlen(skb))) {
1992                         WARN_ON_ONCE(1);
1993                         err = -ENOMEM;
1994                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1995 napi_busy:
1996                         napi_free_frags(&tfile->napi);
1997                         rcu_read_unlock();
1998                         mutex_unlock(&tfile->napi_mutex);
1999                         return err;
2000                 }
2001
2002                 if (likely(napi_schedule_prep(&tfile->napi))) {
2003                         local_bh_disable();
2004                         napi_gro_frags(&tfile->napi);
2005                         napi_complete(&tfile->napi);
2006                         local_bh_enable();
2007                 } else {
2008                         err = -EBUSY;
2009                         goto napi_busy;
2010                 }
2011                 mutex_unlock(&tfile->napi_mutex);
2012         } else if (tfile->napi_enabled) {
2013                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
2014                 int queue_len;
2015
2016                 spin_lock_bh(&queue->lock);
2017                 __skb_queue_tail(queue, skb);
2018                 queue_len = skb_queue_len(queue);
2019                 spin_unlock(&queue->lock);
2020
2021                 if (!more || queue_len > NAPI_POLL_WEIGHT)
2022                         napi_schedule(&tfile->napi);
2023
2024                 local_bh_enable();
2025         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2026                 tun_rx_batched(tun, tfile, skb, more);
2027         } else {
2028                 netif_rx_ni(skb);
2029         }
2030         rcu_read_unlock();
2031
2032         stats = get_cpu_ptr(tun->pcpu_stats);
2033         u64_stats_update_begin(&stats->syncp);
2034         u64_stats_inc(&stats->rx_packets);
2035         u64_stats_add(&stats->rx_bytes, len);
2036         u64_stats_update_end(&stats->syncp);
2037         put_cpu_ptr(stats);
2038
2039         if (rxhash)
2040                 tun_flow_update(tun, rxhash, tfile);
2041
2042         return total_len;
2043 }
2044
2045 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2046 {
2047         struct file *file = iocb->ki_filp;
2048         struct tun_file *tfile = file->private_data;
2049         struct tun_struct *tun = tun_get(tfile);
2050         ssize_t result;
2051         int noblock = 0;
2052
2053         if (!tun)
2054                 return -EBADFD;
2055
2056         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2057                 noblock = 1;
2058
2059         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2060
2061         tun_put(tun);
2062         return result;
2063 }
2064
2065 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2066                                 struct tun_file *tfile,
2067                                 struct xdp_frame *xdp_frame,
2068                                 struct iov_iter *iter)
2069 {
2070         int vnet_hdr_sz = 0;
2071         size_t size = xdp_frame->len;
2072         struct tun_pcpu_stats *stats;
2073         size_t ret;
2074
2075         if (tun->flags & IFF_VNET_HDR) {
2076                 struct virtio_net_hdr gso = { 0 };
2077
2078                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2079                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2080                         return -EINVAL;
2081                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2082                              sizeof(gso)))
2083                         return -EFAULT;
2084                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2085         }
2086
2087         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2088
2089         stats = get_cpu_ptr(tun->pcpu_stats);
2090         u64_stats_update_begin(&stats->syncp);
2091         u64_stats_inc(&stats->tx_packets);
2092         u64_stats_add(&stats->tx_bytes, ret);
2093         u64_stats_update_end(&stats->syncp);
2094         put_cpu_ptr(tun->pcpu_stats);
2095
2096         return ret;
2097 }
2098
2099 /* Put packet to the user space buffer */
2100 static ssize_t tun_put_user(struct tun_struct *tun,
2101                             struct tun_file *tfile,
2102                             struct sk_buff *skb,
2103                             struct iov_iter *iter)
2104 {
2105         struct tun_pi pi = { 0, skb->protocol };
2106         struct tun_pcpu_stats *stats;
2107         ssize_t total;
2108         int vlan_offset = 0;
2109         int vlan_hlen = 0;
2110         int vnet_hdr_sz = 0;
2111
2112         if (skb_vlan_tag_present(skb))
2113                 vlan_hlen = VLAN_HLEN;
2114
2115         if (tun->flags & IFF_VNET_HDR)
2116                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2117
2118         total = skb->len + vlan_hlen + vnet_hdr_sz;
2119
2120         if (!(tun->flags & IFF_NO_PI)) {
2121                 if (iov_iter_count(iter) < sizeof(pi))
2122                         return -EINVAL;
2123
2124                 total += sizeof(pi);
2125                 if (iov_iter_count(iter) < total) {
2126                         /* Packet will be striped */
2127                         pi.flags |= TUN_PKT_STRIP;
2128                 }
2129
2130                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2131                         return -EFAULT;
2132         }
2133
2134         if (vnet_hdr_sz) {
2135                 struct virtio_net_hdr gso;
2136
2137                 if (iov_iter_count(iter) < vnet_hdr_sz)
2138                         return -EINVAL;
2139
2140                 if (virtio_net_hdr_from_skb(skb, &gso,
2141                                             tun_is_little_endian(tun), true,
2142                                             vlan_hlen)) {
2143                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2144
2145                         if (net_ratelimit()) {
2146                                 netdev_err(tun->dev, "unexpected GSO type: 0x%x, gso_size %d, hdr_len %d\n",
2147                                            sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2148                                            tun16_to_cpu(tun, gso.hdr_len));
2149                                 print_hex_dump(KERN_ERR, "tun: ",
2150                                                DUMP_PREFIX_NONE,
2151                                                16, 1, skb->head,
2152                                                min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2153                         }
2154                         WARN_ON_ONCE(1);
2155                         return -EINVAL;
2156                 }
2157
2158                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2159                         return -EFAULT;
2160
2161                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2162         }
2163
2164         if (vlan_hlen) {
2165                 int ret;
2166                 struct veth veth;
2167
2168                 veth.h_vlan_proto = skb->vlan_proto;
2169                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2170
2171                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2172
2173                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2174                 if (ret || !iov_iter_count(iter))
2175                         goto done;
2176
2177                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2178                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2179                         goto done;
2180         }
2181
2182         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2183
2184 done:
2185         /* caller is in process context, */
2186         stats = get_cpu_ptr(tun->pcpu_stats);
2187         u64_stats_update_begin(&stats->syncp);
2188         u64_stats_inc(&stats->tx_packets);
2189         u64_stats_add(&stats->tx_bytes, skb->len + vlan_hlen);
2190         u64_stats_update_end(&stats->syncp);
2191         put_cpu_ptr(tun->pcpu_stats);
2192
2193         return total;
2194 }
2195
2196 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2197 {
2198         DECLARE_WAITQUEUE(wait, current);
2199         void *ptr = NULL;
2200         int error = 0;
2201
2202         ptr = ptr_ring_consume(&tfile->tx_ring);
2203         if (ptr)
2204                 goto out;
2205         if (noblock) {
2206                 error = -EAGAIN;
2207                 goto out;
2208         }
2209
2210         add_wait_queue(&tfile->socket.wq.wait, &wait);
2211
2212         while (1) {
2213                 set_current_state(TASK_INTERRUPTIBLE);
2214                 ptr = ptr_ring_consume(&tfile->tx_ring);
2215                 if (ptr)
2216                         break;
2217                 if (signal_pending(current)) {
2218                         error = -ERESTARTSYS;
2219                         break;
2220                 }
2221                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2222                         error = -EFAULT;
2223                         break;
2224                 }
2225
2226                 schedule();
2227         }
2228
2229         __set_current_state(TASK_RUNNING);
2230         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2231
2232 out:
2233         *err = error;
2234         return ptr;
2235 }
2236
2237 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2238                            struct iov_iter *to,
2239                            int noblock, void *ptr)
2240 {
2241         ssize_t ret;
2242         int err;
2243
2244         if (!iov_iter_count(to)) {
2245                 tun_ptr_free(ptr);
2246                 return 0;
2247         }
2248
2249         if (!ptr) {
2250                 /* Read frames from ring */
2251                 ptr = tun_ring_recv(tfile, noblock, &err);
2252                 if (!ptr)
2253                         return err;
2254         }
2255
2256         if (tun_is_xdp_frame(ptr)) {
2257                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2258
2259                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2260                 xdp_return_frame(xdpf);
2261         } else {
2262                 struct sk_buff *skb = ptr;
2263
2264                 ret = tun_put_user(tun, tfile, skb, to);
2265                 if (unlikely(ret < 0))
2266                         kfree_skb(skb);
2267                 else
2268                         consume_skb(skb);
2269         }
2270
2271         return ret;
2272 }
2273
2274 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2275 {
2276         struct file *file = iocb->ki_filp;
2277         struct tun_file *tfile = file->private_data;
2278         struct tun_struct *tun = tun_get(tfile);
2279         ssize_t len = iov_iter_count(to), ret;
2280         int noblock = 0;
2281
2282         if (!tun)
2283                 return -EBADFD;
2284
2285         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2286                 noblock = 1;
2287
2288         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2289         ret = min_t(ssize_t, ret, len);
2290         if (ret > 0)
2291                 iocb->ki_pos = ret;
2292         tun_put(tun);
2293         return ret;
2294 }
2295
2296 static void tun_prog_free(struct rcu_head *rcu)
2297 {
2298         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2299
2300         bpf_prog_destroy(prog->prog);
2301         kfree(prog);
2302 }
2303
2304 static int __tun_set_ebpf(struct tun_struct *tun,
2305                           struct tun_prog __rcu **prog_p,
2306                           struct bpf_prog *prog)
2307 {
2308         struct tun_prog *old, *new = NULL;
2309
2310         if (prog) {
2311                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2312                 if (!new)
2313                         return -ENOMEM;
2314                 new->prog = prog;
2315         }
2316
2317         spin_lock_bh(&tun->lock);
2318         old = rcu_dereference_protected(*prog_p,
2319                                         lockdep_is_held(&tun->lock));
2320         rcu_assign_pointer(*prog_p, new);
2321         spin_unlock_bh(&tun->lock);
2322
2323         if (old)
2324                 call_rcu(&old->rcu, tun_prog_free);
2325
2326         return 0;
2327 }
2328
2329 static void tun_free_netdev(struct net_device *dev)
2330 {
2331         struct tun_struct *tun = netdev_priv(dev);
2332
2333         BUG_ON(!(list_empty(&tun->disabled)));
2334
2335         free_percpu(tun->pcpu_stats);
2336
2337         tun_flow_uninit(tun);
2338         security_tun_dev_free_security(tun->security);
2339         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2340         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2341 }
2342
2343 static void tun_setup(struct net_device *dev)
2344 {
2345         struct tun_struct *tun = netdev_priv(dev);
2346
2347         tun->owner = INVALID_UID;
2348         tun->group = INVALID_GID;
2349         tun_default_link_ksettings(dev, &tun->link_ksettings);
2350
2351         dev->ethtool_ops = &tun_ethtool_ops;
2352         dev->needs_free_netdev = true;
2353         dev->priv_destructor = tun_free_netdev;
2354         /* We prefer our own queue length */
2355         dev->tx_queue_len = TUN_READQ_SIZE;
2356 }
2357
2358 /* Trivial set of netlink ops to allow deleting tun or tap
2359  * device with netlink.
2360  */
2361 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2362                         struct netlink_ext_ack *extack)
2363 {
2364         NL_SET_ERR_MSG(extack,
2365                        "tun/tap creation via rtnetlink is not supported.");
2366         return -EOPNOTSUPP;
2367 }
2368
2369 static size_t tun_get_size(const struct net_device *dev)
2370 {
2371         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2372         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2373
2374         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2375                nla_total_size(sizeof(gid_t)) + /* GROUP */
2376                nla_total_size(sizeof(u8)) + /* TYPE */
2377                nla_total_size(sizeof(u8)) + /* PI */
2378                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2379                nla_total_size(sizeof(u8)) + /* PERSIST */
2380                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2381                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2382                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2383                0;
2384 }
2385
2386 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2387 {
2388         struct tun_struct *tun = netdev_priv(dev);
2389
2390         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2391                 goto nla_put_failure;
2392         if (uid_valid(tun->owner) &&
2393             nla_put_u32(skb, IFLA_TUN_OWNER,
2394                         from_kuid_munged(current_user_ns(), tun->owner)))
2395                 goto nla_put_failure;
2396         if (gid_valid(tun->group) &&
2397             nla_put_u32(skb, IFLA_TUN_GROUP,
2398                         from_kgid_munged(current_user_ns(), tun->group)))
2399                 goto nla_put_failure;
2400         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2401                 goto nla_put_failure;
2402         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2403                 goto nla_put_failure;
2404         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2405                 goto nla_put_failure;
2406         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2407                        !!(tun->flags & IFF_MULTI_QUEUE)))
2408                 goto nla_put_failure;
2409         if (tun->flags & IFF_MULTI_QUEUE) {
2410                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2411                         goto nla_put_failure;
2412                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2413                                 tun->numdisabled))
2414                         goto nla_put_failure;
2415         }
2416
2417         return 0;
2418
2419 nla_put_failure:
2420         return -EMSGSIZE;
2421 }
2422
2423 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2424         .kind           = DRV_NAME,
2425         .priv_size      = sizeof(struct tun_struct),
2426         .setup          = tun_setup,
2427         .validate       = tun_validate,
2428         .get_size       = tun_get_size,
2429         .fill_info      = tun_fill_info,
2430 };
2431
2432 static void tun_sock_write_space(struct sock *sk)
2433 {
2434         struct tun_file *tfile;
2435         wait_queue_head_t *wqueue;
2436
2437         if (!sock_writeable(sk))
2438                 return;
2439
2440         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2441                 return;
2442
2443         wqueue = sk_sleep(sk);
2444         if (wqueue && waitqueue_active(wqueue))
2445                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2446                                                 EPOLLWRNORM | EPOLLWRBAND);
2447
2448         tfile = container_of(sk, struct tun_file, sk);
2449         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2450 }
2451
2452 static void tun_put_page(struct tun_page *tpage)
2453 {
2454         if (tpage->page)
2455                 __page_frag_cache_drain(tpage->page, tpage->count);
2456 }
2457
2458 static int tun_xdp_one(struct tun_struct *tun,
2459                        struct tun_file *tfile,
2460                        struct xdp_buff *xdp, int *flush,
2461                        struct tun_page *tpage)
2462 {
2463         unsigned int datasize = xdp->data_end - xdp->data;
2464         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2465         struct virtio_net_hdr *gso = &hdr->gso;
2466         struct tun_pcpu_stats *stats;
2467         struct bpf_prog *xdp_prog;
2468         struct sk_buff *skb = NULL;
2469         u32 rxhash = 0, act;
2470         int buflen = hdr->buflen;
2471         int err = 0;
2472         bool skb_xdp = false;
2473         struct page *page;
2474
2475         xdp_prog = rcu_dereference(tun->xdp_prog);
2476         if (xdp_prog) {
2477                 if (gso->gso_type) {
2478                         skb_xdp = true;
2479                         goto build;
2480                 }
2481                 xdp_set_data_meta_invalid(xdp);
2482                 xdp->rxq = &tfile->xdp_rxq;
2483                 xdp->frame_sz = buflen;
2484
2485                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2486                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2487                 if (err < 0) {
2488                         put_page(virt_to_head_page(xdp->data));
2489                         return err;
2490                 }
2491
2492                 switch (err) {
2493                 case XDP_REDIRECT:
2494                         *flush = true;
2495                         fallthrough;
2496                 case XDP_TX:
2497                         return 0;
2498                 case XDP_PASS:
2499                         break;
2500                 default:
2501                         page = virt_to_head_page(xdp->data);
2502                         if (tpage->page == page) {
2503                                 ++tpage->count;
2504                         } else {
2505                                 tun_put_page(tpage);
2506                                 tpage->page = page;
2507                                 tpage->count = 1;
2508                         }
2509                         return 0;
2510                 }
2511         }
2512
2513 build:
2514         skb = build_skb(xdp->data_hard_start, buflen);
2515         if (!skb) {
2516                 err = -ENOMEM;
2517                 goto out;
2518         }
2519
2520         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2521         skb_put(skb, xdp->data_end - xdp->data);
2522
2523         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2524                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2525                 kfree_skb(skb);
2526                 err = -EINVAL;
2527                 goto out;
2528         }
2529
2530         skb->protocol = eth_type_trans(skb, tun->dev);
2531         skb_reset_network_header(skb);
2532         skb_probe_transport_header(skb);
2533         skb_record_rx_queue(skb, tfile->queue_index);
2534
2535         if (skb_xdp) {
2536                 err = do_xdp_generic(xdp_prog, skb);
2537                 if (err != XDP_PASS)
2538                         goto out;
2539         }
2540
2541         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2542             !tfile->detached)
2543                 rxhash = __skb_get_hash_symmetric(skb);
2544
2545         netif_receive_skb(skb);
2546
2547         /* No need for get_cpu_ptr() here since this function is
2548          * always called with bh disabled
2549          */
2550         stats = this_cpu_ptr(tun->pcpu_stats);
2551         u64_stats_update_begin(&stats->syncp);
2552         u64_stats_inc(&stats->rx_packets);
2553         u64_stats_add(&stats->rx_bytes, datasize);
2554         u64_stats_update_end(&stats->syncp);
2555
2556         if (rxhash)
2557                 tun_flow_update(tun, rxhash, tfile);
2558
2559 out:
2560         return err;
2561 }
2562
2563 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2564 {
2565         int ret, i;
2566         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2567         struct tun_struct *tun = tun_get(tfile);
2568         struct tun_msg_ctl *ctl = m->msg_control;
2569         struct xdp_buff *xdp;
2570
2571         if (!tun)
2572                 return -EBADFD;
2573
2574         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2575             ctl && ctl->type == TUN_MSG_PTR) {
2576                 struct tun_page tpage;
2577                 int n = ctl->num;
2578                 int flush = 0;
2579
2580                 memset(&tpage, 0, sizeof(tpage));
2581
2582                 local_bh_disable();
2583                 rcu_read_lock();
2584
2585                 for (i = 0; i < n; i++) {
2586                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2587                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2588                 }
2589
2590                 if (flush)
2591                         xdp_do_flush();
2592
2593                 rcu_read_unlock();
2594                 local_bh_enable();
2595
2596                 tun_put_page(&tpage);
2597
2598                 ret = total_len;
2599                 goto out;
2600         }
2601
2602         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2603                            m->msg_flags & MSG_DONTWAIT,
2604                            m->msg_flags & MSG_MORE);
2605 out:
2606         tun_put(tun);
2607         return ret;
2608 }
2609
2610 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2611                        int flags)
2612 {
2613         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2614         struct tun_struct *tun = tun_get(tfile);
2615         void *ptr = m->msg_control;
2616         int ret;
2617
2618         if (!tun) {
2619                 ret = -EBADFD;
2620                 goto out_free;
2621         }
2622
2623         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2624                 ret = -EINVAL;
2625                 goto out_put_tun;
2626         }
2627         if (flags & MSG_ERRQUEUE) {
2628                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2629                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2630                 goto out;
2631         }
2632         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2633         if (ret > (ssize_t)total_len) {
2634                 m->msg_flags |= MSG_TRUNC;
2635                 ret = flags & MSG_TRUNC ? ret : total_len;
2636         }
2637 out:
2638         tun_put(tun);
2639         return ret;
2640
2641 out_put_tun:
2642         tun_put(tun);
2643 out_free:
2644         tun_ptr_free(ptr);
2645         return ret;
2646 }
2647
2648 static int tun_ptr_peek_len(void *ptr)
2649 {
2650         if (likely(ptr)) {
2651                 if (tun_is_xdp_frame(ptr)) {
2652                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2653
2654                         return xdpf->len;
2655                 }
2656                 return __skb_array_len_with_tag(ptr);
2657         } else {
2658                 return 0;
2659         }
2660 }
2661
2662 static int tun_peek_len(struct socket *sock)
2663 {
2664         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2665         struct tun_struct *tun;
2666         int ret = 0;
2667
2668         tun = tun_get(tfile);
2669         if (!tun)
2670                 return 0;
2671
2672         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2673         tun_put(tun);
2674
2675         return ret;
2676 }
2677
2678 /* Ops structure to mimic raw sockets with tun */
2679 static const struct proto_ops tun_socket_ops = {
2680         .peek_len = tun_peek_len,
2681         .sendmsg = tun_sendmsg,
2682         .recvmsg = tun_recvmsg,
2683 };
2684
2685 static struct proto tun_proto = {
2686         .name           = "tun",
2687         .owner          = THIS_MODULE,
2688         .obj_size       = sizeof(struct tun_file),
2689 };
2690
2691 static int tun_flags(struct tun_struct *tun)
2692 {
2693         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2694 }
2695
2696 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2697                               char *buf)
2698 {
2699         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2700         return sprintf(buf, "0x%x\n", tun_flags(tun));
2701 }
2702
2703 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2704                               char *buf)
2705 {
2706         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2707         return uid_valid(tun->owner)?
2708                 sprintf(buf, "%u\n",
2709                         from_kuid_munged(current_user_ns(), tun->owner)):
2710                 sprintf(buf, "-1\n");
2711 }
2712
2713 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2714                               char *buf)
2715 {
2716         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2717         return gid_valid(tun->group) ?
2718                 sprintf(buf, "%u\n",
2719                         from_kgid_munged(current_user_ns(), tun->group)):
2720                 sprintf(buf, "-1\n");
2721 }
2722
2723 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2724 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2725 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2726
2727 static struct attribute *tun_dev_attrs[] = {
2728         &dev_attr_tun_flags.attr,
2729         &dev_attr_owner.attr,
2730         &dev_attr_group.attr,
2731         NULL
2732 };
2733
2734 static const struct attribute_group tun_attr_group = {
2735         .attrs = tun_dev_attrs
2736 };
2737
2738 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2739 {
2740         struct tun_struct *tun;
2741         struct tun_file *tfile = file->private_data;
2742         struct net_device *dev;
2743         int err;
2744
2745         if (tfile->detached)
2746                 return -EINVAL;
2747
2748         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2749                 if (!capable(CAP_NET_ADMIN))
2750                         return -EPERM;
2751
2752                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2753                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2754                         return -EINVAL;
2755         }
2756
2757         dev = __dev_get_by_name(net, ifr->ifr_name);
2758         if (dev) {
2759                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2760                         return -EBUSY;
2761                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2762                         tun = netdev_priv(dev);
2763                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2764                         tun = netdev_priv(dev);
2765                 else
2766                         return -EINVAL;
2767
2768                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2769                     !!(tun->flags & IFF_MULTI_QUEUE))
2770                         return -EINVAL;
2771
2772                 if (tun_not_capable(tun))
2773                         return -EPERM;
2774                 err = security_tun_dev_open(tun->security);
2775                 if (err < 0)
2776                         return err;
2777
2778                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2779                                  ifr->ifr_flags & IFF_NAPI,
2780                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2781                 if (err < 0)
2782                         return err;
2783
2784                 if (tun->flags & IFF_MULTI_QUEUE &&
2785                     (tun->numqueues + tun->numdisabled > 1)) {
2786                         /* One or more queue has already been attached, no need
2787                          * to initialize the device again.
2788                          */
2789                         netdev_state_change(dev);
2790                         return 0;
2791                 }
2792
2793                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2794                               (ifr->ifr_flags & TUN_FEATURES);
2795
2796                 netdev_state_change(dev);
2797         } else {
2798                 char *name;
2799                 unsigned long flags = 0;
2800                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2801                              MAX_TAP_QUEUES : 1;
2802
2803                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2804                         return -EPERM;
2805                 err = security_tun_dev_create();
2806                 if (err < 0)
2807                         return err;
2808
2809                 /* Set dev type */
2810                 if (ifr->ifr_flags & IFF_TUN) {
2811                         /* TUN device */
2812                         flags |= IFF_TUN;
2813                         name = "tun%d";
2814                 } else if (ifr->ifr_flags & IFF_TAP) {
2815                         /* TAP device */
2816                         flags |= IFF_TAP;
2817                         name = "tap%d";
2818                 } else
2819                         return -EINVAL;
2820
2821                 if (*ifr->ifr_name)
2822                         name = ifr->ifr_name;
2823
2824                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2825                                        NET_NAME_UNKNOWN, tun_setup, queues,
2826                                        queues);
2827
2828                 if (!dev)
2829                         return -ENOMEM;
2830
2831                 dev_net_set(dev, net);
2832                 dev->rtnl_link_ops = &tun_link_ops;
2833                 dev->ifindex = tfile->ifindex;
2834                 dev->sysfs_groups[0] = &tun_attr_group;
2835
2836                 tun = netdev_priv(dev);
2837                 tun->dev = dev;
2838                 tun->flags = flags;
2839                 tun->txflt.count = 0;
2840                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2841
2842                 tun->align = NET_SKB_PAD;
2843                 tun->filter_attached = false;
2844                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2845                 tun->rx_batched = 0;
2846                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2847
2848                 tun->ifr = ifr;
2849                 tun->file = file;
2850
2851                 tun_net_initialize(dev);
2852
2853                 err = register_netdevice(tun->dev);
2854                 if (err < 0) {
2855                         free_netdev(dev);
2856                         return err;
2857                 }
2858                 /* free_netdev() won't check refcnt, to aovid race
2859                  * with dev_put() we need publish tun after registration.
2860                  */
2861                 rcu_assign_pointer(tfile->tun, tun);
2862         }
2863
2864         netif_carrier_on(tun->dev);
2865
2866         /* Make sure persistent devices do not get stuck in
2867          * xoff state.
2868          */
2869         if (netif_running(tun->dev))
2870                 netif_tx_wake_all_queues(tun->dev);
2871
2872         strcpy(ifr->ifr_name, tun->dev->name);
2873         return 0;
2874 }
2875
2876 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2877 {
2878         strcpy(ifr->ifr_name, tun->dev->name);
2879
2880         ifr->ifr_flags = tun_flags(tun);
2881
2882 }
2883
2884 /* This is like a cut-down ethtool ops, except done via tun fd so no
2885  * privs required. */
2886 static int set_offload(struct tun_struct *tun, unsigned long arg)
2887 {
2888         netdev_features_t features = 0;
2889
2890         if (arg & TUN_F_CSUM) {
2891                 features |= NETIF_F_HW_CSUM;
2892                 arg &= ~TUN_F_CSUM;
2893
2894                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2895                         if (arg & TUN_F_TSO_ECN) {
2896                                 features |= NETIF_F_TSO_ECN;
2897                                 arg &= ~TUN_F_TSO_ECN;
2898                         }
2899                         if (arg & TUN_F_TSO4)
2900                                 features |= NETIF_F_TSO;
2901                         if (arg & TUN_F_TSO6)
2902                                 features |= NETIF_F_TSO6;
2903                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2904                 }
2905
2906                 arg &= ~TUN_F_UFO;
2907         }
2908
2909         /* This gives the user a way to test for new features in future by
2910          * trying to set them. */
2911         if (arg)
2912                 return -EINVAL;
2913
2914         tun->set_features = features;
2915         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2916         tun->dev->wanted_features |= features;
2917         netdev_update_features(tun->dev);
2918
2919         return 0;
2920 }
2921
2922 static void tun_detach_filter(struct tun_struct *tun, int n)
2923 {
2924         int i;
2925         struct tun_file *tfile;
2926
2927         for (i = 0; i < n; i++) {
2928                 tfile = rtnl_dereference(tun->tfiles[i]);
2929                 lock_sock(tfile->socket.sk);
2930                 sk_detach_filter(tfile->socket.sk);
2931                 release_sock(tfile->socket.sk);
2932         }
2933
2934         tun->filter_attached = false;
2935 }
2936
2937 static int tun_attach_filter(struct tun_struct *tun)
2938 {
2939         int i, ret = 0;
2940         struct tun_file *tfile;
2941
2942         for (i = 0; i < tun->numqueues; i++) {
2943                 tfile = rtnl_dereference(tun->tfiles[i]);
2944                 lock_sock(tfile->socket.sk);
2945                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2946                 release_sock(tfile->socket.sk);
2947                 if (ret) {
2948                         tun_detach_filter(tun, i);
2949                         return ret;
2950                 }
2951         }
2952
2953         tun->filter_attached = true;
2954         return ret;
2955 }
2956
2957 static void tun_set_sndbuf(struct tun_struct *tun)
2958 {
2959         struct tun_file *tfile;
2960         int i;
2961
2962         for (i = 0; i < tun->numqueues; i++) {
2963                 tfile = rtnl_dereference(tun->tfiles[i]);
2964                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2965         }
2966 }
2967
2968 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2969 {
2970         struct tun_file *tfile = file->private_data;
2971         struct tun_struct *tun;
2972         int ret = 0;
2973
2974         rtnl_lock();
2975
2976         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2977                 tun = tfile->detached;
2978                 if (!tun) {
2979                         ret = -EINVAL;
2980                         goto unlock;
2981                 }
2982                 ret = security_tun_dev_attach_queue(tun->security);
2983                 if (ret < 0)
2984                         goto unlock;
2985                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2986                                  tun->flags & IFF_NAPI_FRAGS, true);
2987         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2988                 tun = rtnl_dereference(tfile->tun);
2989                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2990                         ret = -EINVAL;
2991                 else
2992                         __tun_detach(tfile, false);
2993         } else
2994                 ret = -EINVAL;
2995
2996         if (ret >= 0)
2997                 netdev_state_change(tun->dev);
2998
2999 unlock:
3000         rtnl_unlock();
3001         return ret;
3002 }
3003
3004 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3005                         void __user *data)
3006 {
3007         struct bpf_prog *prog;
3008         int fd;
3009
3010         if (copy_from_user(&fd, data, sizeof(fd)))
3011                 return -EFAULT;
3012
3013         if (fd == -1) {
3014                 prog = NULL;
3015         } else {
3016                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3017                 if (IS_ERR(prog))
3018                         return PTR_ERR(prog);
3019         }
3020
3021         return __tun_set_ebpf(tun, prog_p, prog);
3022 }
3023
3024 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3025 static unsigned char tun_get_addr_len(unsigned short type)
3026 {
3027         switch (type) {
3028         case ARPHRD_IP6GRE:
3029         case ARPHRD_TUNNEL6:
3030                 return sizeof(struct in6_addr);
3031         case ARPHRD_IPGRE:
3032         case ARPHRD_TUNNEL:
3033         case ARPHRD_SIT:
3034                 return 4;
3035         case ARPHRD_ETHER:
3036                 return ETH_ALEN;
3037         case ARPHRD_IEEE802154:
3038         case ARPHRD_IEEE802154_MONITOR:
3039                 return IEEE802154_EXTENDED_ADDR_LEN;
3040         case ARPHRD_PHONET_PIPE:
3041         case ARPHRD_PPP:
3042         case ARPHRD_NONE:
3043                 return 0;
3044         case ARPHRD_6LOWPAN:
3045                 return EUI64_ADDR_LEN;
3046         case ARPHRD_FDDI:
3047                 return FDDI_K_ALEN;
3048         case ARPHRD_HIPPI:
3049                 return HIPPI_ALEN;
3050         case ARPHRD_IEEE802:
3051                 return FC_ALEN;
3052         case ARPHRD_ROSE:
3053                 return ROSE_ADDR_LEN;
3054         case ARPHRD_NETROM:
3055                 return AX25_ADDR_LEN;
3056         case ARPHRD_LOCALTLK:
3057                 return LTALK_ALEN;
3058         default:
3059                 return 0;
3060         }
3061 }
3062
3063 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3064                             unsigned long arg, int ifreq_len)
3065 {
3066         struct tun_file *tfile = file->private_data;
3067         struct net *net = sock_net(&tfile->sk);
3068         struct tun_struct *tun;
3069         void __user* argp = (void __user*)arg;
3070         unsigned int carrier;
3071         struct ifreq ifr;
3072         kuid_t owner;
3073         kgid_t group;
3074         int ifindex;
3075         int sndbuf;
3076         int vnet_hdr_sz;
3077         int le;
3078         int ret;
3079         bool do_notify = false;
3080
3081         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3082             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3083                 if (copy_from_user(&ifr, argp, ifreq_len))
3084                         return -EFAULT;
3085         } else {
3086                 memset(&ifr, 0, sizeof(ifr));
3087         }
3088         if (cmd == TUNGETFEATURES) {
3089                 /* Currently this just means: "what IFF flags are valid?".
3090                  * This is needed because we never checked for invalid flags on
3091                  * TUNSETIFF.
3092                  */
3093                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3094                                 (unsigned int __user*)argp);
3095         } else if (cmd == TUNSETQUEUE) {
3096                 return tun_set_queue(file, &ifr);
3097         } else if (cmd == SIOCGSKNS) {
3098                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3099                         return -EPERM;
3100                 return open_related_ns(&net->ns, get_net_ns);
3101         }
3102
3103         ret = 0;
3104         rtnl_lock();
3105
3106         tun = tun_get(tfile);
3107         if (cmd == TUNSETIFF) {
3108                 ret = -EEXIST;
3109                 if (tun)
3110                         goto unlock;
3111
3112                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3113
3114                 ret = tun_set_iff(net, file, &ifr);
3115
3116                 if (ret)
3117                         goto unlock;
3118
3119                 if (copy_to_user(argp, &ifr, ifreq_len))
3120                         ret = -EFAULT;
3121                 goto unlock;
3122         }
3123         if (cmd == TUNSETIFINDEX) {
3124                 ret = -EPERM;
3125                 if (tun)
3126                         goto unlock;
3127
3128                 ret = -EFAULT;
3129                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3130                         goto unlock;
3131                 ret = -EINVAL;
3132                 if (ifindex < 0)
3133                         goto unlock;
3134                 ret = 0;
3135                 tfile->ifindex = ifindex;
3136                 goto unlock;
3137         }
3138
3139         ret = -EBADFD;
3140         if (!tun)
3141                 goto unlock;
3142
3143         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3144
3145         net = dev_net(tun->dev);
3146         ret = 0;
3147         switch (cmd) {
3148         case TUNGETIFF:
3149                 tun_get_iff(tun, &ifr);
3150
3151                 if (tfile->detached)
3152                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3153                 if (!tfile->socket.sk->sk_filter)
3154                         ifr.ifr_flags |= IFF_NOFILTER;
3155
3156                 if (copy_to_user(argp, &ifr, ifreq_len))
3157                         ret = -EFAULT;
3158                 break;
3159
3160         case TUNSETNOCSUM:
3161                 /* Disable/Enable checksum */
3162
3163                 /* [unimplemented] */
3164                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3165                            arg ? "disabled" : "enabled");
3166                 break;
3167
3168         case TUNSETPERSIST:
3169                 /* Disable/Enable persist mode. Keep an extra reference to the
3170                  * module to prevent the module being unprobed.
3171                  */
3172                 if (arg && !(tun->flags & IFF_PERSIST)) {
3173                         tun->flags |= IFF_PERSIST;
3174                         __module_get(THIS_MODULE);
3175                         do_notify = true;
3176                 }
3177                 if (!arg && (tun->flags & IFF_PERSIST)) {
3178                         tun->flags &= ~IFF_PERSIST;
3179                         module_put(THIS_MODULE);
3180                         do_notify = true;
3181                 }
3182
3183                 netif_info(tun, drv, tun->dev, "persist %s\n",
3184                            arg ? "enabled" : "disabled");
3185                 break;
3186
3187         case TUNSETOWNER:
3188                 /* Set owner of the device */
3189                 owner = make_kuid(current_user_ns(), arg);
3190                 if (!uid_valid(owner)) {
3191                         ret = -EINVAL;
3192                         break;
3193                 }
3194                 tun->owner = owner;
3195                 do_notify = true;
3196                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3197                            from_kuid(&init_user_ns, tun->owner));
3198                 break;
3199
3200         case TUNSETGROUP:
3201                 /* Set group of the device */
3202                 group = make_kgid(current_user_ns(), arg);
3203                 if (!gid_valid(group)) {
3204                         ret = -EINVAL;
3205                         break;
3206                 }
3207                 tun->group = group;
3208                 do_notify = true;
3209                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3210                            from_kgid(&init_user_ns, tun->group));
3211                 break;
3212
3213         case TUNSETLINK:
3214                 /* Only allow setting the type when the interface is down */
3215                 if (tun->dev->flags & IFF_UP) {
3216                         netif_info(tun, drv, tun->dev,
3217                                    "Linktype set failed because interface is up\n");
3218                         ret = -EBUSY;
3219                 } else {
3220                         tun->dev->type = (int) arg;
3221                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3222                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3223                                    tun->dev->type);
3224                         ret = 0;
3225                 }
3226                 break;
3227
3228         case TUNSETDEBUG:
3229                 tun->msg_enable = (u32)arg;
3230                 break;
3231
3232         case TUNSETOFFLOAD:
3233                 ret = set_offload(tun, arg);
3234                 break;
3235
3236         case TUNSETTXFILTER:
3237                 /* Can be set only for TAPs */
3238                 ret = -EINVAL;
3239                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3240                         break;
3241                 ret = update_filter(&tun->txflt, (void __user *)arg);
3242                 break;
3243
3244         case SIOCGIFHWADDR:
3245                 /* Get hw address */
3246                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3247                 if (copy_to_user(argp, &ifr, ifreq_len))
3248                         ret = -EFAULT;
3249                 break;
3250
3251         case SIOCSIFHWADDR:
3252                 /* Set hw address */
3253                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3254                 break;
3255
3256         case TUNGETSNDBUF:
3257                 sndbuf = tfile->socket.sk->sk_sndbuf;
3258                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3259                         ret = -EFAULT;
3260                 break;
3261
3262         case TUNSETSNDBUF:
3263                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3264                         ret = -EFAULT;
3265                         break;
3266                 }
3267                 if (sndbuf <= 0) {
3268                         ret = -EINVAL;
3269                         break;
3270                 }
3271
3272                 tun->sndbuf = sndbuf;
3273                 tun_set_sndbuf(tun);
3274                 break;
3275
3276         case TUNGETVNETHDRSZ:
3277                 vnet_hdr_sz = tun->vnet_hdr_sz;
3278                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3279                         ret = -EFAULT;
3280                 break;
3281
3282         case TUNSETVNETHDRSZ:
3283                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3284                         ret = -EFAULT;
3285                         break;
3286                 }
3287                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3288                         ret = -EINVAL;
3289                         break;
3290                 }
3291
3292                 tun->vnet_hdr_sz = vnet_hdr_sz;
3293                 break;
3294
3295         case TUNGETVNETLE:
3296                 le = !!(tun->flags & TUN_VNET_LE);
3297                 if (put_user(le, (int __user *)argp))
3298                         ret = -EFAULT;
3299                 break;
3300
3301         case TUNSETVNETLE:
3302                 if (get_user(le, (int __user *)argp)) {
3303                         ret = -EFAULT;
3304                         break;
3305                 }
3306                 if (le)
3307                         tun->flags |= TUN_VNET_LE;
3308                 else
3309                         tun->flags &= ~TUN_VNET_LE;
3310                 break;
3311
3312         case TUNGETVNETBE:
3313                 ret = tun_get_vnet_be(tun, argp);
3314                 break;
3315
3316         case TUNSETVNETBE:
3317                 ret = tun_set_vnet_be(tun, argp);
3318                 break;
3319
3320         case TUNATTACHFILTER:
3321                 /* Can be set only for TAPs */
3322                 ret = -EINVAL;
3323                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3324                         break;
3325                 ret = -EFAULT;
3326                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3327                         break;
3328
3329                 ret = tun_attach_filter(tun);
3330                 break;
3331
3332         case TUNDETACHFILTER:
3333                 /* Can be set only for TAPs */
3334                 ret = -EINVAL;
3335                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3336                         break;
3337                 ret = 0;
3338                 tun_detach_filter(tun, tun->numqueues);
3339                 break;
3340
3341         case TUNGETFILTER:
3342                 ret = -EINVAL;
3343                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3344                         break;
3345                 ret = -EFAULT;
3346                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3347                         break;
3348                 ret = 0;
3349                 break;
3350
3351         case TUNSETSTEERINGEBPF:
3352                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3353                 break;
3354
3355         case TUNSETFILTEREBPF:
3356                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3357                 break;
3358
3359         case TUNSETCARRIER:
3360                 ret = -EFAULT;
3361                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3362                         goto unlock;
3363
3364                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3365                 break;
3366
3367         case TUNGETDEVNETNS:
3368                 ret = -EPERM;
3369                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3370                         goto unlock;
3371                 ret = open_related_ns(&net->ns, get_net_ns);
3372                 break;
3373
3374         default:
3375                 ret = -EINVAL;
3376                 break;
3377         }
3378
3379         if (do_notify)
3380                 netdev_state_change(tun->dev);
3381
3382 unlock:
3383         rtnl_unlock();
3384         if (tun)
3385                 tun_put(tun);
3386         return ret;
3387 }
3388
3389 static long tun_chr_ioctl(struct file *file,
3390                           unsigned int cmd, unsigned long arg)
3391 {
3392         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3393 }
3394
3395 #ifdef CONFIG_COMPAT
3396 static long tun_chr_compat_ioctl(struct file *file,
3397                          unsigned int cmd, unsigned long arg)
3398 {
3399         switch (cmd) {
3400         case TUNSETIFF:
3401         case TUNGETIFF:
3402         case TUNSETTXFILTER:
3403         case TUNGETSNDBUF:
3404         case TUNSETSNDBUF:
3405         case SIOCGIFHWADDR:
3406         case SIOCSIFHWADDR:
3407                 arg = (unsigned long)compat_ptr(arg);
3408                 break;
3409         default:
3410                 arg = (compat_ulong_t)arg;
3411                 break;
3412         }
3413
3414         /*
3415          * compat_ifreq is shorter than ifreq, so we must not access beyond
3416          * the end of that structure. All fields that are used in this
3417          * driver are compatible though, we don't need to convert the
3418          * contents.
3419          */
3420         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3421 }
3422 #endif /* CONFIG_COMPAT */
3423
3424 static int tun_chr_fasync(int fd, struct file *file, int on)
3425 {
3426         struct tun_file *tfile = file->private_data;
3427         int ret;
3428
3429         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3430                 goto out;
3431
3432         if (on) {
3433                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3434                 tfile->flags |= TUN_FASYNC;
3435         } else
3436                 tfile->flags &= ~TUN_FASYNC;
3437         ret = 0;
3438 out:
3439         return ret;
3440 }
3441
3442 static int tun_chr_open(struct inode *inode, struct file * file)
3443 {
3444         struct net *net = current->nsproxy->net_ns;
3445         struct tun_file *tfile;
3446
3447         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3448                                             &tun_proto, 0);
3449         if (!tfile)
3450                 return -ENOMEM;
3451         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3452                 sk_free(&tfile->sk);
3453                 return -ENOMEM;
3454         }
3455
3456         mutex_init(&tfile->napi_mutex);
3457         RCU_INIT_POINTER(tfile->tun, NULL);
3458         tfile->flags = 0;
3459         tfile->ifindex = 0;
3460
3461         init_waitqueue_head(&tfile->socket.wq.wait);
3462
3463         tfile->socket.file = file;
3464         tfile->socket.ops = &tun_socket_ops;
3465
3466         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3467
3468         tfile->sk.sk_write_space = tun_sock_write_space;
3469         tfile->sk.sk_sndbuf = INT_MAX;
3470
3471         file->private_data = tfile;
3472         INIT_LIST_HEAD(&tfile->next);
3473
3474         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3475
3476         return 0;
3477 }
3478
3479 static int tun_chr_close(struct inode *inode, struct file *file)
3480 {
3481         struct tun_file *tfile = file->private_data;
3482
3483         tun_detach(tfile, true);
3484
3485         return 0;
3486 }
3487
3488 #ifdef CONFIG_PROC_FS
3489 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3490 {
3491         struct tun_file *tfile = file->private_data;
3492         struct tun_struct *tun;
3493         struct ifreq ifr;
3494
3495         memset(&ifr, 0, sizeof(ifr));
3496
3497         rtnl_lock();
3498         tun = tun_get(tfile);
3499         if (tun)
3500                 tun_get_iff(tun, &ifr);
3501         rtnl_unlock();
3502
3503         if (tun)
3504                 tun_put(tun);
3505
3506         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3507 }
3508 #endif
3509
3510 static const struct file_operations tun_fops = {
3511         .owner  = THIS_MODULE,
3512         .llseek = no_llseek,
3513         .read_iter  = tun_chr_read_iter,
3514         .write_iter = tun_chr_write_iter,
3515         .poll   = tun_chr_poll,
3516         .unlocked_ioctl = tun_chr_ioctl,
3517 #ifdef CONFIG_COMPAT
3518         .compat_ioctl = tun_chr_compat_ioctl,
3519 #endif
3520         .open   = tun_chr_open,
3521         .release = tun_chr_close,
3522         .fasync = tun_chr_fasync,
3523 #ifdef CONFIG_PROC_FS
3524         .show_fdinfo = tun_chr_show_fdinfo,
3525 #endif
3526 };
3527
3528 static struct miscdevice tun_miscdev = {
3529         .minor = TUN_MINOR,
3530         .name = "tun",
3531         .nodename = "net/tun",
3532         .fops = &tun_fops,
3533 };
3534
3535 /* ethtool interface */
3536
3537 static void tun_default_link_ksettings(struct net_device *dev,
3538                                        struct ethtool_link_ksettings *cmd)
3539 {
3540         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3541         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3542         cmd->base.speed         = SPEED_10;
3543         cmd->base.duplex        = DUPLEX_FULL;
3544         cmd->base.port          = PORT_TP;
3545         cmd->base.phy_address   = 0;
3546         cmd->base.autoneg       = AUTONEG_DISABLE;
3547 }
3548
3549 static int tun_get_link_ksettings(struct net_device *dev,
3550                                   struct ethtool_link_ksettings *cmd)
3551 {
3552         struct tun_struct *tun = netdev_priv(dev);
3553
3554         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3555         return 0;
3556 }
3557
3558 static int tun_set_link_ksettings(struct net_device *dev,
3559                                   const struct ethtool_link_ksettings *cmd)
3560 {
3561         struct tun_struct *tun = netdev_priv(dev);
3562
3563         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3564         return 0;
3565 }
3566
3567 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3568 {
3569         struct tun_struct *tun = netdev_priv(dev);
3570
3571         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3572         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3573
3574         switch (tun->flags & TUN_TYPE_MASK) {
3575         case IFF_TUN:
3576                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3577                 break;
3578         case IFF_TAP:
3579                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3580                 break;
3581         }
3582 }
3583
3584 static u32 tun_get_msglevel(struct net_device *dev)
3585 {
3586         struct tun_struct *tun = netdev_priv(dev);
3587
3588         return tun->msg_enable;
3589 }
3590
3591 static void tun_set_msglevel(struct net_device *dev, u32 value)
3592 {
3593         struct tun_struct *tun = netdev_priv(dev);
3594
3595         tun->msg_enable = value;
3596 }
3597
3598 static int tun_get_coalesce(struct net_device *dev,
3599                             struct ethtool_coalesce *ec)
3600 {
3601         struct tun_struct *tun = netdev_priv(dev);
3602
3603         ec->rx_max_coalesced_frames = tun->rx_batched;
3604
3605         return 0;
3606 }
3607
3608 static int tun_set_coalesce(struct net_device *dev,
3609                             struct ethtool_coalesce *ec)
3610 {
3611         struct tun_struct *tun = netdev_priv(dev);
3612
3613         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3614                 tun->rx_batched = NAPI_POLL_WEIGHT;
3615         else
3616                 tun->rx_batched = ec->rx_max_coalesced_frames;
3617
3618         return 0;
3619 }
3620
3621 static const struct ethtool_ops tun_ethtool_ops = {
3622         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3623         .get_drvinfo    = tun_get_drvinfo,
3624         .get_msglevel   = tun_get_msglevel,
3625         .set_msglevel   = tun_set_msglevel,
3626         .get_link       = ethtool_op_get_link,
3627         .get_ts_info    = ethtool_op_get_ts_info,
3628         .get_coalesce   = tun_get_coalesce,
3629         .set_coalesce   = tun_set_coalesce,
3630         .get_link_ksettings = tun_get_link_ksettings,
3631         .set_link_ksettings = tun_set_link_ksettings,
3632 };
3633
3634 static int tun_queue_resize(struct tun_struct *tun)
3635 {
3636         struct net_device *dev = tun->dev;
3637         struct tun_file *tfile;
3638         struct ptr_ring **rings;
3639         int n = tun->numqueues + tun->numdisabled;
3640         int ret, i;
3641
3642         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3643         if (!rings)
3644                 return -ENOMEM;
3645
3646         for (i = 0; i < tun->numqueues; i++) {
3647                 tfile = rtnl_dereference(tun->tfiles[i]);
3648                 rings[i] = &tfile->tx_ring;
3649         }
3650         list_for_each_entry(tfile, &tun->disabled, next)
3651                 rings[i++] = &tfile->tx_ring;
3652
3653         ret = ptr_ring_resize_multiple(rings, n,
3654                                        dev->tx_queue_len, GFP_KERNEL,
3655                                        tun_ptr_free);
3656
3657         kfree(rings);
3658         return ret;
3659 }
3660
3661 static int tun_device_event(struct notifier_block *unused,
3662                             unsigned long event, void *ptr)
3663 {
3664         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3665         struct tun_struct *tun = netdev_priv(dev);
3666         int i;
3667
3668         if (dev->rtnl_link_ops != &tun_link_ops)
3669                 return NOTIFY_DONE;
3670
3671         switch (event) {
3672         case NETDEV_CHANGE_TX_QUEUE_LEN:
3673                 if (tun_queue_resize(tun))
3674                         return NOTIFY_BAD;
3675                 break;
3676         case NETDEV_UP:
3677                 for (i = 0; i < tun->numqueues; i++) {
3678                         struct tun_file *tfile;
3679
3680                         tfile = rtnl_dereference(tun->tfiles[i]);
3681                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3682                 }
3683                 break;
3684         default:
3685                 break;
3686         }
3687
3688         return NOTIFY_DONE;
3689 }
3690
3691 static struct notifier_block tun_notifier_block __read_mostly = {
3692         .notifier_call  = tun_device_event,
3693 };
3694
3695 static int __init tun_init(void)
3696 {
3697         int ret = 0;
3698
3699         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3700
3701         ret = rtnl_link_register(&tun_link_ops);
3702         if (ret) {
3703                 pr_err("Can't register link_ops\n");
3704                 goto err_linkops;
3705         }
3706
3707         ret = misc_register(&tun_miscdev);
3708         if (ret) {
3709                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3710                 goto err_misc;
3711         }
3712
3713         ret = register_netdevice_notifier(&tun_notifier_block);
3714         if (ret) {
3715                 pr_err("Can't register netdevice notifier\n");
3716                 goto err_notifier;
3717         }
3718
3719         return  0;
3720
3721 err_notifier:
3722         misc_deregister(&tun_miscdev);
3723 err_misc:
3724         rtnl_link_unregister(&tun_link_ops);
3725 err_linkops:
3726         return ret;
3727 }
3728
3729 static void tun_cleanup(void)
3730 {
3731         misc_deregister(&tun_miscdev);
3732         rtnl_link_unregister(&tun_link_ops);
3733         unregister_netdevice_notifier(&tun_notifier_block);
3734 }
3735
3736 /* Get an underlying socket object from tun file.  Returns error unless file is
3737  * attached to a device.  The returned object works like a packet socket, it
3738  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3739  * holding a reference to the file for as long as the socket is in use. */
3740 struct socket *tun_get_socket(struct file *file)
3741 {
3742         struct tun_file *tfile;
3743         if (file->f_op != &tun_fops)
3744                 return ERR_PTR(-EINVAL);
3745         tfile = file->private_data;
3746         if (!tfile)
3747                 return ERR_PTR(-EBADFD);
3748         return &tfile->socket;
3749 }
3750 EXPORT_SYMBOL_GPL(tun_get_socket);
3751
3752 struct ptr_ring *tun_get_tx_ring(struct file *file)
3753 {
3754         struct tun_file *tfile;
3755
3756         if (file->f_op != &tun_fops)
3757                 return ERR_PTR(-EINVAL);
3758         tfile = file->private_data;
3759         if (!tfile)
3760                 return ERR_PTR(-EBADFD);
3761         return &tfile->tx_ring;
3762 }
3763 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3764
3765 module_init(tun_init);
3766 module_exit(tun_cleanup);
3767 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3768 MODULE_AUTHOR(DRV_COPYRIGHT);
3769 MODULE_LICENSE("GPL");
3770 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3771 MODULE_ALIAS("devname:net/tun");