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/math.h>
  58 #include <linux/nsproxy.h>
  59 #include <linux/virtio_net.h>
  60 #include <linux/rcupdate.h>
  61 #include <net/net_namespace.h>
  62 #include <net/netns/generic.h>
  63 #include <net/rtnetlink.h>
  64 #include <net/sock.h>
  65 #include <net/xdp.h>
  66 #include <net/ip_tunnels.h>
  67 #include <linux/seq_file.h>
  68 #include <linux/uio.h>
  69 #include <linux/skb_array.h>
  70 #include <linux/bpf.h>
  71 #include <linux/bpf_trace.h>
  72 #include <linux/mutex.h>
  73 #include <linux/ieee802154.h>
  74 #include <linux/if_ltalk.h>
  75 #include <uapi/linux/if_fddi.h>
  76 #include <uapi/linux/if_hippi.h>
  77 #include <uapi/linux/if_fc.h>
  78 #include <net/ax25.h>
  79 #include <net/rose.h>
  80 #include <net/6lowpan.h>
  81 #include <net/rps.h>
  82
  83 #include <linux/uaccess.h>
  84 #include <linux/proc_fs.h>
  85
  86 static void tun_default_link_ksettings(struct net_device *dev,
  87                                        struct ethtool_link_ksettings *cmd);
  88
  89 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  90
  91 /* TUN device flags */
  92
  93 /* IFF_ATTACH_QUEUE is never stored in device flags,
  94  * overload it to mean fasync when stored there.
  95  */
  96 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  97 /* High bits in flags field are unused. */
  98 #define TUN_VNET_LE     0x80000000
  99 #define TUN_VNET_BE     0x40000000
 100
 101 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 102                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 103
 104 #define GOODCOPY_LEN 128
 105
 106 #define FLT_EXACT_COUNT 8
 107 struct tap_filter {
 108         unsigned int    count;    /* Number of addrs. Zero means disabled */
 109         u32             mask[2];  /* Mask of the hashed addrs */
 110         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 111 };
 112
 113 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 114  * to max number of VCPUs in guest. */
 115 #define MAX_TAP_QUEUES 256
 116 #define MAX_TAP_FLOWS  4096
 117
 118 #define TUN_FLOW_EXPIRE (3 * HZ)
 119
 120 /* A tun_file connects an open character device to a tuntap netdevice. It
 121  * also contains all socket related structures (except sock_fprog and tap_filter)
 122  * to serve as one transmit queue for tuntap device. The sock_fprog and
 123  * tap_filter were kept in tun_struct since they were used for filtering for the
 124  * netdevice not for a specific queue (at least I didn't see the requirement for
 125  * this).
 126  *
 127  * RCU usage:
 128  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 129  * other can only be read while rcu_read_lock or rtnl_lock is held.
 130  */
 131 struct tun_file {
 132         struct sock sk;
 133         struct socket socket;
 134         struct tun_struct __rcu *tun;
 135         struct fasync_struct *fasync;
 136         /* only used for fasnyc */
 137         unsigned int flags;
 138         union {
 139                 u16 queue_index;
 140                 unsigned int ifindex;
 141         };
 142         struct napi_struct napi;
 143         bool napi_enabled;
 144         bool napi_frags_enabled;
 145         struct mutex napi_mutex;        /* Protects access to the above napi */
 146         struct list_head next;
 147         struct tun_struct *detached;
 148         struct ptr_ring tx_ring;
 149         struct xdp_rxq_info xdp_rxq;
 150 };
 151
 152 struct tun_page {
 153         struct page *page;
 154         int count;
 155 };
 156
 157 struct tun_flow_entry {
 158         struct hlist_node hash_link;
 159         struct rcu_head rcu;
 160         struct tun_struct *tun;
 161
 162         u32 rxhash;
 163         u32 rps_rxhash;
 164         int queue_index;
 165         unsigned long updated ____cacheline_aligned_in_smp;
 166 };
 167
 168 #define TUN_NUM_FLOW_ENTRIES 1024
 169 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 170
 171 struct tun_prog {
 172         struct rcu_head rcu;
 173         struct bpf_prog *prog;
 174 };
 175
 176 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 177  * device, socket filter, sndbuf and vnet header size were restore when the
 178  * file were attached to a persist device.
 179  */
 180 struct tun_struct {
 181         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 182         unsigned int            numqueues;
 183         unsigned int            flags;
 184         kuid_t                  owner;
 185         kgid_t                  group;
 186
 187         struct net_device       *dev;
 188         netdev_features_t       set_features;
 189 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 190                           NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)
 191
 192         int                     align;
 193         int                     vnet_hdr_sz;
 194         int                     sndbuf;
 195         struct tap_filter       txflt;
 196         struct sock_fprog       fprog;
 197         /* protected by rtnl lock */
 198         bool                    filter_attached;
 199         u32                     msg_enable;
 200         spinlock_t lock;
 201         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202         struct timer_list flow_gc_timer;
 203         unsigned long ageing_time;
 204         unsigned int numdisabled;
 205         struct list_head disabled;
 206         void *security;
 207         u32 flow_count;
 208         u32 rx_batched;
 209         atomic_long_t rx_frame_errors;
 210         struct bpf_prog __rcu *xdp_prog;
 211         struct tun_prog __rcu *steering_prog;
 212         struct tun_prog __rcu *filter_prog;
 213         struct ethtool_link_ksettings link_ksettings;
 214         /* init args */
 215         struct file *file;
 216         struct ifreq *ifr;
 217 };
 218
 219 struct veth {
 220         __be16 h_vlan_proto;
 221         __be16 h_vlan_TCI;
 222 };
 223
 224 static void tun_flow_init(struct tun_struct *tun);
 225 static void tun_flow_uninit(struct tun_struct *tun);
 226
 227 static int tun_napi_receive(struct napi_struct *napi, int budget)
 228 {
 229         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 230         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 231         struct sk_buff_head process_queue;
 232         struct sk_buff *skb;
 233         int received = 0;
 234
 235         __skb_queue_head_init(&process_queue);
 236
 237         spin_lock(&queue->lock);
 238         skb_queue_splice_tail_init(queue, &process_queue);
 239         spin_unlock(&queue->lock);
 240
 241         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 242                 napi_gro_receive(napi, skb);
 243                 ++received;
 244         }
 245
 246         if (!skb_queue_empty(&process_queue)) {
 247                 spin_lock(&queue->lock);
 248                 skb_queue_splice(&process_queue, queue);
 249                 spin_unlock(&queue->lock);
 250         }
 251
 252         return received;
 253 }
 254
 255 static int tun_napi_poll(struct napi_struct *napi, int budget)
 256 {
 257         unsigned int received;
 258
 259         received = tun_napi_receive(napi, budget);
 260
 261         if (received < budget)
 262                 napi_complete_done(napi, received);
 263
 264         return received;
 265 }
 266
 267 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 268                           bool napi_en, bool napi_frags)
 269 {
 270         tfile->napi_enabled = napi_en;
 271         tfile->napi_frags_enabled = napi_en && napi_frags;
 272         if (napi_en) {
 273                 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 274                 napi_enable(&tfile->napi);
 275         }
 276 }
 277
 278 static void tun_napi_enable(struct tun_file *tfile)
 279 {
 280         if (tfile->napi_enabled)
 281                 napi_enable(&tfile->napi);
 282 }
 283
 284 static void tun_napi_disable(struct tun_file *tfile)
 285 {
 286         if (tfile->napi_enabled)
 287                 napi_disable(&tfile->napi);
 288 }
 289
 290 static void tun_napi_del(struct tun_file *tfile)
 291 {
 292         if (tfile->napi_enabled)
 293                 netif_napi_del(&tfile->napi);
 294 }
 295
 296 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 297 {
 298         return tfile->napi_frags_enabled;
 299 }
 300
 301 #ifdef CONFIG_TUN_VNET_CROSS_LE
 302 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 303 {
 304         return tun->flags & TUN_VNET_BE ? false :
 305                 virtio_legacy_is_little_endian();
 306 }
 307
 308 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 309 {
 310         int be = !!(tun->flags & TUN_VNET_BE);
 311
 312         if (put_user(be, argp))
 313                 return -EFAULT;
 314
 315         return 0;
 316 }
 317
 318 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 319 {
 320         int be;
 321
 322         if (get_user(be, argp))
 323                 return -EFAULT;
 324
 325         if (be)
 326                 tun->flags |= TUN_VNET_BE;
 327         else
 328                 tun->flags &= ~TUN_VNET_BE;
 329
 330         return 0;
 331 }
 332 #else
 333 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 334 {
 335         return virtio_legacy_is_little_endian();
 336 }
 337
 338 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 339 {
 340         return -EINVAL;
 341 }
 342
 343 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 344 {
 345         return -EINVAL;
 346 }
 347 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 348
 349 static inline bool tun_is_little_endian(struct tun_struct *tun)
 350 {
 351         return tun->flags & TUN_VNET_LE ||
 352                 tun_legacy_is_little_endian(tun);
 353 }
 354
 355 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 356 {
 357         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 358 }
 359
 360 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 361 {
 362         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 363 }
 364
 365 static inline u32 tun_hashfn(u32 rxhash)
 366 {
 367         return rxhash & TUN_MASK_FLOW_ENTRIES;
 368 }
 369
 370 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 371 {
 372         struct tun_flow_entry *e;
 373
 374         hlist_for_each_entry_rcu(e, head, hash_link) {
 375                 if (e->rxhash == rxhash)
 376                         return e;
 377         }
 378         return NULL;
 379 }
 380
 381 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 382                                               struct hlist_head *head,
 383                                               u32 rxhash, u16 queue_index)
 384 {
 385         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 386
 387         if (e) {
 388                 netif_info(tun, tx_queued, tun->dev,
 389                            "create flow: hash %u index %u\n",
 390                            rxhash, queue_index);
 391                 e->updated = jiffies;
 392                 e->rxhash = rxhash;
 393                 e->rps_rxhash = 0;
 394                 e->queue_index = queue_index;
 395                 e->tun = tun;
 396                 hlist_add_head_rcu(&e->hash_link, head);
 397                 ++tun->flow_count;
 398         }
 399         return e;
 400 }
 401
 402 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 403 {
 404         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 405                    e->rxhash, e->queue_index);
 406         hlist_del_rcu(&e->hash_link);
 407         kfree_rcu(e, rcu);
 408         --tun->flow_count;
 409 }
 410
 411 static void tun_flow_flush(struct tun_struct *tun)
 412 {
 413         int i;
 414
 415         spin_lock_bh(&tun->lock);
 416         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 417                 struct tun_flow_entry *e;
 418                 struct hlist_node *n;
 419
 420                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 421                         tun_flow_delete(tun, e);
 422         }
 423         spin_unlock_bh(&tun->lock);
 424 }
 425
 426 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 427 {
 428         int i;
 429
 430         spin_lock_bh(&tun->lock);
 431         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 432                 struct tun_flow_entry *e;
 433                 struct hlist_node *n;
 434
 435                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 436                         if (e->queue_index == queue_index)
 437                                 tun_flow_delete(tun, e);
 438                 }
 439         }
 440         spin_unlock_bh(&tun->lock);
 441 }
 442
 443 static void tun_flow_cleanup(struct timer_list *t)
 444 {
 445         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 446         unsigned long delay = tun->ageing_time;
 447         unsigned long next_timer = jiffies + delay;
 448         unsigned long count = 0;
 449         int i;
 450
 451         spin_lock(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 457                         unsigned long this_timer;
 458
 459                         this_timer = e->updated + delay;
 460                         if (time_before_eq(this_timer, jiffies)) {
 461                                 tun_flow_delete(tun, e);
 462                                 continue;
 463                         }
 464                         count++;
 465                         if (time_before(this_timer, next_timer))
 466                                 next_timer = this_timer;
 467                 }
 468         }
 469
 470         if (count)
 471                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 472         spin_unlock(&tun->lock);
 473 }
 474
 475 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 476                             struct tun_file *tfile)
 477 {
 478         struct hlist_head *head;
 479         struct tun_flow_entry *e;
 480         unsigned long delay = tun->ageing_time;
 481         u16 queue_index = tfile->queue_index;
 482
 483         head = &tun->flows[tun_hashfn(rxhash)];
 484
 485         rcu_read_lock();
 486
 487         e = tun_flow_find(head, rxhash);
 488         if (likely(e)) {
 489                 /* TODO: keep queueing to old queue until it's empty? */
 490                 if (READ_ONCE(e->queue_index) != queue_index)
 491                         WRITE_ONCE(e->queue_index, queue_index);
 492                 if (e->updated != jiffies)
 493                         e->updated = jiffies;
 494                 sock_rps_record_flow_hash(e->rps_rxhash);
 495         } else {
 496                 spin_lock_bh(&tun->lock);
 497                 if (!tun_flow_find(head, rxhash) &&
 498                     tun->flow_count < MAX_TAP_FLOWS)
 499                         tun_flow_create(tun, head, rxhash, queue_index);
 500
 501                 if (!timer_pending(&tun->flow_gc_timer))
 502                         mod_timer(&tun->flow_gc_timer,
 503                                   round_jiffies_up(jiffies + delay));
 504                 spin_unlock_bh(&tun->lock);
 505         }
 506
 507         rcu_read_unlock();
 508 }
 509
 510 /* Save the hash received in the stack receive path and update the
 511  * flow_hash table accordingly.
 512  */
 513 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 514 {
 515         if (unlikely(e->rps_rxhash != hash))
 516                 e->rps_rxhash = hash;
 517 }
 518
 519 /* We try to identify a flow through its rxhash. The reason that
 520  * we do not check rxq no. is because some cards(e.g 82599), chooses
 521  * the rxq based on the txq where the last packet of the flow comes. As
 522  * the userspace application move between processors, we may get a
 523  * different rxq no. here.
 524  */
 525 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 526 {
 527         struct tun_flow_entry *e;
 528         u32 txq, numqueues;
 529
 530         numqueues = READ_ONCE(tun->numqueues);
 531
 532         txq = __skb_get_hash_symmetric(skb);
 533         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 534         if (e) {
 535                 tun_flow_save_rps_rxhash(e, txq);
 536                 txq = e->queue_index;
 537         } else {
 538                 txq = reciprocal_scale(txq, numqueues);
 539         }
 540
 541         return txq;
 542 }
 543
 544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 545 {
 546         struct tun_prog *prog;
 547         u32 numqueues;
 548         u16 ret = 0;
 549
 550         numqueues = READ_ONCE(tun->numqueues);
 551         if (!numqueues)
 552                 return 0;
 553
 554         prog = rcu_dereference(tun->steering_prog);
 555         if (prog)
 556                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 557
 558         return ret % numqueues;
 559 }
 560
 561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 562                             struct net_device *sb_dev)
 563 {
 564         struct tun_struct *tun = netdev_priv(dev);
 565         u16 ret;
 566
 567         rcu_read_lock();
 568         if (rcu_dereference(tun->steering_prog))
 569                 ret = tun_ebpf_select_queue(tun, skb);
 570         else
 571                 ret = tun_automq_select_queue(tun, skb);
 572         rcu_read_unlock();
 573
 574         return ret;
 575 }
 576
 577 static inline bool tun_not_capable(struct tun_struct *tun)
 578 {
 579         const struct cred *cred = current_cred();
 580         struct net *net = dev_net(tun->dev);
 581
 582         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 583                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 584                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 585 }
 586
 587 static void tun_set_real_num_queues(struct tun_struct *tun)
 588 {
 589         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 590         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 591 }
 592
 593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 594 {
 595         tfile->detached = tun;
 596         list_add_tail(&tfile->next, &tun->disabled);
 597         ++tun->numdisabled;
 598 }
 599
 600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 601 {
 602         struct tun_struct *tun = tfile->detached;
 603
 604         tfile->detached = NULL;
 605         list_del_init(&tfile->next);
 606         --tun->numdisabled;
 607         return tun;
 608 }
 609
 610 void tun_ptr_free(void *ptr)
 611 {
 612         if (!ptr)
 613                 return;
 614         if (tun_is_xdp_frame(ptr)) {
 615                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 616
 617                 xdp_return_frame(xdpf);
 618         } else {
 619                 __skb_array_destroy_skb(ptr);
 620         }
 621 }
 622 EXPORT_SYMBOL_GPL(tun_ptr_free);
 623
 624 static void tun_queue_purge(struct tun_file *tfile)
 625 {
 626         void *ptr;
 627
 628         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 629                 tun_ptr_free(ptr);
 630
 631         skb_queue_purge(&tfile->sk.sk_write_queue);
 632         skb_queue_purge(&tfile->sk.sk_error_queue);
 633 }
 634
 635 static void __tun_detach(struct tun_file *tfile, bool clean)
 636 {
 637         struct tun_file *ntfile;
 638         struct tun_struct *tun;
 639
 640         tun = rtnl_dereference(tfile->tun);
 641
 642         if (tun && clean) {
 643                 if (!tfile->detached)
 644                         tun_napi_disable(tfile);
 645                 tun_napi_del(tfile);
 646         }
 647
 648         if (tun && !tfile->detached) {
 649                 u16 index = tfile->queue_index;
 650                 BUG_ON(index >= tun->numqueues);
 651
 652                 rcu_assign_pointer(tun->tfiles[index],
 653                                    tun->tfiles[tun->numqueues - 1]);
 654                 ntfile = rtnl_dereference(tun->tfiles[index]);
 655                 ntfile->queue_index = index;
 656                 ntfile->xdp_rxq.queue_index = index;
 657                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 658                                    NULL);
 659
 660                 --tun->numqueues;
 661                 if (clean) {
 662                         RCU_INIT_POINTER(tfile->tun, NULL);
 663                         sock_put(&tfile->sk);
 664                 } else {
 665                         tun_disable_queue(tun, tfile);
 666                         tun_napi_disable(tfile);
 667                 }
 668
 669                 synchronize_net();
 670                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 671                 /* Drop read queue */
 672                 tun_queue_purge(tfile);
 673                 tun_set_real_num_queues(tun);
 674         } else if (tfile->detached && clean) {
 675                 tun = tun_enable_queue(tfile);
 676                 sock_put(&tfile->sk);
 677         }
 678
 679         if (clean) {
 680                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 681                         netif_carrier_off(tun->dev);
 682
 683                         if (!(tun->flags & IFF_PERSIST) &&
 684                             tun->dev->reg_state == NETREG_REGISTERED)
 685                                 unregister_netdevice(tun->dev);
 686                 }
 687                 if (tun)
 688                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 689                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 690         }
 691 }
 692
 693 static void tun_detach(struct tun_file *tfile, bool clean)
 694 {
 695         struct tun_struct *tun;
 696         struct net_device *dev;
 697
 698         rtnl_lock();
 699         tun = rtnl_dereference(tfile->tun);
 700         dev = tun ? tun->dev : NULL;
 701         __tun_detach(tfile, clean);
 702         if (dev)
 703                 netdev_state_change(dev);
 704         rtnl_unlock();
 705
 706         if (clean)
 707                 sock_put(&tfile->sk);
 708 }
 709
 710 static void tun_detach_all(struct net_device *dev)
 711 {
 712         struct tun_struct *tun = netdev_priv(dev);
 713         struct tun_file *tfile, *tmp;
 714         int i, n = tun->numqueues;
 715
 716         for (i = 0; i < n; i++) {
 717                 tfile = rtnl_dereference(tun->tfiles[i]);
 718                 BUG_ON(!tfile);
 719                 tun_napi_disable(tfile);
 720                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 721                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 722                 RCU_INIT_POINTER(tfile->tun, NULL);
 723                 --tun->numqueues;
 724         }
 725         list_for_each_entry(tfile, &tun->disabled, next) {
 726                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 727                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 728                 RCU_INIT_POINTER(tfile->tun, NULL);
 729         }
 730         BUG_ON(tun->numqueues != 0);
 731
 732         synchronize_net();
 733         for (i = 0; i < n; i++) {
 734                 tfile = rtnl_dereference(tun->tfiles[i]);
 735                 tun_napi_del(tfile);
 736                 /* Drop read queue */
 737                 tun_queue_purge(tfile);
 738                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 739                 sock_put(&tfile->sk);
 740         }
 741         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 742                 tun_napi_del(tfile);
 743                 tun_enable_queue(tfile);
 744                 tun_queue_purge(tfile);
 745                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 746                 sock_put(&tfile->sk);
 747         }
 748         BUG_ON(tun->numdisabled != 0);
 749
 750         if (tun->flags & IFF_PERSIST)
 751                 module_put(THIS_MODULE);
 752 }
 753
 754 static int tun_attach(struct tun_struct *tun, struct file *file,
 755                       bool skip_filter, bool napi, bool napi_frags,
 756                       bool publish_tun)
 757 {
 758         struct tun_file *tfile = file->private_data;
 759         struct net_device *dev = tun->dev;
 760         int err;
 761
 762         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 763         if (err < 0)
 764                 goto out;
 765
 766         err = -EINVAL;
 767         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 768                 goto out;
 769
 770         err = -EBUSY;
 771         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 772                 goto out;
 773
 774         err = -E2BIG;
 775         if (!tfile->detached &&
 776             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 777                 goto out;
 778
 779         err = 0;
 780
 781         /* Re-attach the filter to persist device */
 782         if (!skip_filter && (tun->filter_attached == true)) {
 783                 lock_sock(tfile->socket.sk);
 784                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 785                 release_sock(tfile->socket.sk);
 786                 if (!err)
 787                         goto out;
 788         }
 789
 790         if (!tfile->detached &&
 791             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 792                             GFP_KERNEL, tun_ptr_free)) {
 793                 err = -ENOMEM;
 794                 goto out;
 795         }
 796
 797         tfile->queue_index = tun->numqueues;
 798         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 799
 800         if (tfile->detached) {
 801                 /* Re-attach detached tfile, updating XDP queue_index */
 802                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 803
 804                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 805                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 806         } else {
 807                 /* Setup XDP RX-queue info, for new tfile getting attached */
 808                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 809                                        tun->dev, tfile->queue_index, 0);
 810                 if (err < 0)
 811                         goto out;
 812                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 813                                                  MEM_TYPE_PAGE_SHARED, NULL);
 814                 if (err < 0) {
 815                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 816                         goto out;
 817                 }
 818                 err = 0;
 819         }
 820
 821         if (tfile->detached) {
 822                 tun_enable_queue(tfile);
 823                 tun_napi_enable(tfile);
 824         } else {
 825                 sock_hold(&tfile->sk);
 826                 tun_napi_init(tun, tfile, napi, napi_frags);
 827         }
 828
 829         if (rtnl_dereference(tun->xdp_prog))
 830                 sock_set_flag(&tfile->sk, SOCK_XDP);
 831
 832         /* device is allowed to go away first, so no need to hold extra
 833          * refcnt.
 834          */
 835
 836         /* Publish tfile->tun and tun->tfiles only after we've fully
 837          * initialized tfile; otherwise we risk using half-initialized
 838          * object.
 839          */
 840         if (publish_tun)
 841                 rcu_assign_pointer(tfile->tun, tun);
 842         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 843         tun->numqueues++;
 844         tun_set_real_num_queues(tun);
 845 out:
 846         return err;
 847 }
 848
 849 static struct tun_struct *tun_get(struct tun_file *tfile)
 850 {
 851         struct tun_struct *tun;
 852
 853         rcu_read_lock();
 854         tun = rcu_dereference(tfile->tun);
 855         if (tun)
 856                 dev_hold(tun->dev);
 857         rcu_read_unlock();
 858
 859         return tun;
 860 }
 861
 862 static void tun_put(struct tun_struct *tun)
 863 {
 864         dev_put(tun->dev);
 865 }
 866
 867 /* TAP filtering */
 868 static void addr_hash_set(u32 *mask, const u8 *addr)
 869 {
 870         int n = ether_crc(ETH_ALEN, addr) >> 26;
 871         mask[n >> 5] |= (1 << (n & 31));
 872 }
 873
 874 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 875 {
 876         int n = ether_crc(ETH_ALEN, addr) >> 26;
 877         return mask[n >> 5] & (1 << (n & 31));
 878 }
 879
 880 static int update_filter(struct tap_filter *filter, void __user *arg)
 881 {
 882         struct { u8 u[ETH_ALEN]; } *addr;
 883         struct tun_filter uf;
 884         int err, alen, n, nexact;
 885
 886         if (copy_from_user(&uf, arg, sizeof(uf)))
 887                 return -EFAULT;
 888
 889         if (!uf.count) {
 890                 /* Disabled */
 891                 filter->count = 0;
 892                 return 0;
 893         }
 894
 895         alen = ETH_ALEN * uf.count;
 896         addr = memdup_user(arg + sizeof(uf), alen);
 897         if (IS_ERR(addr))
 898                 return PTR_ERR(addr);
 899
 900         /* The filter is updated without holding any locks. Which is
 901          * perfectly safe. We disable it first and in the worst
 902          * case we'll accept a few undesired packets. */
 903         filter->count = 0;
 904         wmb();
 905
 906         /* Use first set of addresses as an exact filter */
 907         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 908                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 909
 910         nexact = n;
 911
 912         /* Remaining multicast addresses are hashed,
 913          * unicast will leave the filter disabled. */
 914         memset(filter->mask, 0, sizeof(filter->mask));
 915         for (; n < uf.count; n++) {
 916                 if (!is_multicast_ether_addr(addr[n].u)) {
 917                         err = 0; /* no filter */
 918                         goto free_addr;
 919                 }
 920                 addr_hash_set(filter->mask, addr[n].u);
 921         }
 922
 923         /* For ALLMULTI just set the mask to all ones.
 924          * This overrides the mask populated above. */
 925         if ((uf.flags & TUN_FLT_ALLMULTI))
 926                 memset(filter->mask, ~0, sizeof(filter->mask));
 927
 928         /* Now enable the filter */
 929         wmb();
 930         filter->count = nexact;
 931
 932         /* Return the number of exact filters */
 933         err = nexact;
 934 free_addr:
 935         kfree(addr);
 936         return err;
 937 }
 938
 939 /* Returns: 0 - drop, !=0 - accept */
 940 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 941 {
 942         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 943          * at this point. */
 944         struct ethhdr *eh = (struct ethhdr *) skb->data;
 945         int i;
 946
 947         /* Exact match */
 948         for (i = 0; i < filter->count; i++)
 949                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 950                         return 1;
 951
 952         /* Inexact match (multicast only) */
 953         if (is_multicast_ether_addr(eh->h_dest))
 954                 return addr_hash_test(filter->mask, eh->h_dest);
 955
 956         return 0;
 957 }
 958
 959 /*
 960  * Checks whether the packet is accepted or not.
 961  * Returns: 0 - drop, !=0 - accept
 962  */
 963 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 964 {
 965         if (!filter->count)
 966                 return 1;
 967
 968         return run_filter(filter, skb);
 969 }
 970
 971 /* Network device part of the driver */
 972
 973 static const struct ethtool_ops tun_ethtool_ops;
 974
 975 static int tun_net_init(struct net_device *dev)
 976 {
 977         struct tun_struct *tun = netdev_priv(dev);
 978         struct ifreq *ifr = tun->ifr;
 979         int err;
 980
 981         spin_lock_init(&tun->lock);
 982
 983         err = security_tun_dev_alloc_security(&tun->security);
 984         if (err < 0)
 985                 return err;
 986
 987         tun_flow_init(tun);
 988
 989         dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
 990         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 991                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 992                            NETIF_F_HW_VLAN_STAG_TX;
 993         dev->features = dev->hw_features | NETIF_F_LLTX;
 994         dev->vlan_features = dev->features &
 995                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 996                                NETIF_F_HW_VLAN_STAG_TX);
 997
 998         tun->flags = (tun->flags & ~TUN_FEATURES) |
 999                       (ifr->ifr_flags & TUN_FEATURES);
1000
1001         INIT_LIST_HEAD(&tun->disabled);
1002         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1003                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1004         if (err < 0) {
1005                 tun_flow_uninit(tun);
1006                 security_tun_dev_free_security(tun->security);
1007                 return err;
1008         }
1009         return 0;
1010 }
1011
1012 /* Net device detach from fd. */
1013 static void tun_net_uninit(struct net_device *dev)
1014 {
1015         tun_detach_all(dev);
1016 }
1017
1018 /* Net device open. */
1019 static int tun_net_open(struct net_device *dev)
1020 {
1021         netif_tx_start_all_queues(dev);
1022
1023         return 0;
1024 }
1025
1026 /* Net device close. */
1027 static int tun_net_close(struct net_device *dev)
1028 {
1029         netif_tx_stop_all_queues(dev);
1030         return 0;
1031 }
1032
1033 /* Net device start xmit */
1034 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1035 {
1036 #ifdef CONFIG_RPS
1037         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1038                 /* Select queue was not called for the skbuff, so we extract the
1039                  * RPS hash and save it into the flow_table here.
1040                  */
1041                 struct tun_flow_entry *e;
1042                 __u32 rxhash;
1043
1044                 rxhash = __skb_get_hash_symmetric(skb);
1045                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1046                 if (e)
1047                         tun_flow_save_rps_rxhash(e, rxhash);
1048         }
1049 #endif
1050 }
1051
1052 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1053                                     struct sk_buff *skb,
1054                                     int len)
1055 {
1056         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1057
1058         if (prog)
1059                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1060
1061         return len;
1062 }
1063
1064 /* Net device start xmit */
1065 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1066 {
1067         struct tun_struct *tun = netdev_priv(dev);
1068         enum skb_drop_reason drop_reason;
1069         int txq = skb->queue_mapping;
1070         struct netdev_queue *queue;
1071         struct tun_file *tfile;
1072         int len = skb->len;
1073
1074         rcu_read_lock();
1075         tfile = rcu_dereference(tun->tfiles[txq]);
1076
1077         /* Drop packet if interface is not attached */
1078         if (!tfile) {
1079                 drop_reason = SKB_DROP_REASON_DEV_READY;
1080                 goto drop;
1081         }
1082
1083         if (!rcu_dereference(tun->steering_prog))
1084                 tun_automq_xmit(tun, skb);
1085
1086         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1087
1088         /* Drop if the filter does not like it.
1089          * This is a noop if the filter is disabled.
1090          * Filter can be enabled only for the TAP devices. */
1091         if (!check_filter(&tun->txflt, skb)) {
1092                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1093                 goto drop;
1094         }
1095
1096         if (tfile->socket.sk->sk_filter &&
1097             sk_filter(tfile->socket.sk, skb)) {
1098                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1099                 goto drop;
1100         }
1101
1102         len = run_ebpf_filter(tun, skb, len);
1103         if (len == 0) {
1104                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1105                 goto drop;
1106         }
1107
1108         if (pskb_trim(skb, len)) {
1109                 drop_reason = SKB_DROP_REASON_NOMEM;
1110                 goto drop;
1111         }
1112
1113         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1114                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1115                 goto drop;
1116         }
1117
1118         skb_tx_timestamp(skb);
1119
1120         /* Orphan the skb - required as we might hang on to it
1121          * for indefinite time.
1122          */
1123         skb_orphan(skb);
1124
1125         nf_reset_ct(skb);
1126
1127         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1128                 drop_reason = SKB_DROP_REASON_FULL_RING;
1129                 goto drop;
1130         }
1131
1132         /* NETIF_F_LLTX requires to do our own update of trans_start */
1133         queue = netdev_get_tx_queue(dev, txq);
1134         txq_trans_cond_update(queue);
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         dev_core_stats_tx_dropped_inc(dev);
1146         skb_tx_error(skb);
1147         kfree_skb_reason(skb, drop_reason);
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         struct tun_struct *tun = netdev_priv(dev);
1183
1184         dev_get_tstats64(dev, stats);
1185
1186         stats->rx_frame_errors +=
1187                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1188 }
1189
1190 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1191                        struct netlink_ext_ack *extack)
1192 {
1193         struct tun_struct *tun = netdev_priv(dev);
1194         struct tun_file *tfile;
1195         struct bpf_prog *old_prog;
1196         int i;
1197
1198         old_prog = rtnl_dereference(tun->xdp_prog);
1199         rcu_assign_pointer(tun->xdp_prog, prog);
1200         if (old_prog)
1201                 bpf_prog_put(old_prog);
1202
1203         for (i = 0; i < tun->numqueues; i++) {
1204                 tfile = rtnl_dereference(tun->tfiles[i]);
1205                 if (prog)
1206                         sock_set_flag(&tfile->sk, SOCK_XDP);
1207                 else
1208                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1209         }
1210         list_for_each_entry(tfile, &tun->disabled, next) {
1211                 if (prog)
1212                         sock_set_flag(&tfile->sk, SOCK_XDP);
1213                 else
1214                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1215         }
1216
1217         return 0;
1218 }
1219
1220 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1221 {
1222         switch (xdp->command) {
1223         case XDP_SETUP_PROG:
1224                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1225         default:
1226                 return -EINVAL;
1227         }
1228 }
1229
1230 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1231 {
1232         if (new_carrier) {
1233                 struct tun_struct *tun = netdev_priv(dev);
1234
1235                 if (!tun->numqueues)
1236                         return -EPERM;
1237
1238                 netif_carrier_on(dev);
1239         } else {
1240                 netif_carrier_off(dev);
1241         }
1242         return 0;
1243 }
1244
1245 static const struct net_device_ops tun_netdev_ops = {
1246         .ndo_init               = tun_net_init,
1247         .ndo_uninit             = tun_net_uninit,
1248         .ndo_open               = tun_net_open,
1249         .ndo_stop               = tun_net_close,
1250         .ndo_start_xmit         = tun_net_xmit,
1251         .ndo_fix_features       = tun_net_fix_features,
1252         .ndo_select_queue       = tun_select_queue,
1253         .ndo_set_rx_headroom    = tun_set_headroom,
1254         .ndo_get_stats64        = tun_net_get_stats64,
1255         .ndo_change_carrier     = tun_net_change_carrier,
1256 };
1257
1258 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1259 {
1260         /* Notify and wake up reader process */
1261         if (tfile->flags & TUN_FASYNC)
1262                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1263         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1264 }
1265
1266 static int tun_xdp_xmit(struct net_device *dev, int n,
1267                         struct xdp_frame **frames, u32 flags)
1268 {
1269         struct tun_struct *tun = netdev_priv(dev);
1270         struct tun_file *tfile;
1271         u32 numqueues;
1272         int nxmit = 0;
1273         int i;
1274
1275         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1276                 return -EINVAL;
1277
1278         rcu_read_lock();
1279
1280 resample:
1281         numqueues = READ_ONCE(tun->numqueues);
1282         if (!numqueues) {
1283                 rcu_read_unlock();
1284                 return -ENXIO; /* Caller will free/return all frames */
1285         }
1286
1287         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1288                                             numqueues]);
1289         if (unlikely(!tfile))
1290                 goto resample;
1291
1292         spin_lock(&tfile->tx_ring.producer_lock);
1293         for (i = 0; i < n; i++) {
1294                 struct xdp_frame *xdp = frames[i];
1295                 /* Encode the XDP flag into lowest bit for consumer to differ
1296                  * XDP buffer from sk_buff.
1297                  */
1298                 void *frame = tun_xdp_to_ptr(xdp);
1299
1300                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1301                         dev_core_stats_tx_dropped_inc(dev);
1302                         break;
1303                 }
1304                 nxmit++;
1305         }
1306         spin_unlock(&tfile->tx_ring.producer_lock);
1307
1308         if (flags & XDP_XMIT_FLUSH)
1309                 __tun_xdp_flush_tfile(tfile);
1310
1311         rcu_read_unlock();
1312         return nxmit;
1313 }
1314
1315 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1316 {
1317         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1318         int nxmit;
1319
1320         if (unlikely(!frame))
1321                 return -EOVERFLOW;
1322
1323         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1324         if (!nxmit)
1325                 xdp_return_frame_rx_napi(frame);
1326         return nxmit;
1327 }
1328
1329 static const struct net_device_ops tap_netdev_ops = {
1330         .ndo_init               = tun_net_init,
1331         .ndo_uninit             = tun_net_uninit,
1332         .ndo_open               = tun_net_open,
1333         .ndo_stop               = tun_net_close,
1334         .ndo_start_xmit         = tun_net_xmit,
1335         .ndo_fix_features       = tun_net_fix_features,
1336         .ndo_set_rx_mode        = tun_net_mclist,
1337         .ndo_set_mac_address    = eth_mac_addr,
1338         .ndo_validate_addr      = eth_validate_addr,
1339         .ndo_select_queue       = tun_select_queue,
1340         .ndo_features_check     = passthru_features_check,
1341         .ndo_set_rx_headroom    = tun_set_headroom,
1342         .ndo_bpf                = tun_xdp,
1343         .ndo_xdp_xmit           = tun_xdp_xmit,
1344         .ndo_change_carrier     = tun_net_change_carrier,
1345 };
1346
1347 static void tun_flow_init(struct tun_struct *tun)
1348 {
1349         int i;
1350
1351         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1352                 INIT_HLIST_HEAD(&tun->flows[i]);
1353
1354         tun->ageing_time = TUN_FLOW_EXPIRE;
1355         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1356         mod_timer(&tun->flow_gc_timer,
1357                   round_jiffies_up(jiffies + tun->ageing_time));
1358 }
1359
1360 static void tun_flow_uninit(struct tun_struct *tun)
1361 {
1362         del_timer_sync(&tun->flow_gc_timer);
1363         tun_flow_flush(tun);
1364 }
1365
1366 #define MIN_MTU 68
1367 #define MAX_MTU 65535
1368
1369 /* Initialize net device. */
1370 static void tun_net_initialize(struct net_device *dev)
1371 {
1372         struct tun_struct *tun = netdev_priv(dev);
1373
1374         switch (tun->flags & TUN_TYPE_MASK) {
1375         case IFF_TUN:
1376                 dev->netdev_ops = &tun_netdev_ops;
1377                 dev->header_ops = &ip_tunnel_header_ops;
1378
1379                 /* Point-to-Point TUN Device */
1380                 dev->hard_header_len = 0;
1381                 dev->addr_len = 0;
1382                 dev->mtu = 1500;
1383
1384                 /* Zero header length */
1385                 dev->type = ARPHRD_NONE;
1386                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1387                 break;
1388
1389         case IFF_TAP:
1390                 dev->netdev_ops = &tap_netdev_ops;
1391                 /* Ethernet TAP Device */
1392                 ether_setup(dev);
1393                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1394                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1395
1396                 eth_hw_addr_random(dev);
1397
1398                 /* Currently tun does not support XDP, only tap does. */
1399                 dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1400                                     NETDEV_XDP_ACT_REDIRECT |
1401                                     NETDEV_XDP_ACT_NDO_XMIT;
1402
1403                 break;
1404         }
1405
1406         dev->min_mtu = MIN_MTU;
1407         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1408 }
1409
1410 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1411 {
1412         struct sock *sk = tfile->socket.sk;
1413
1414         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1415 }
1416
1417 /* Character device part */
1418
1419 /* Poll */
1420 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1421 {
1422         struct tun_file *tfile = file->private_data;
1423         struct tun_struct *tun = tun_get(tfile);
1424         struct sock *sk;
1425         __poll_t mask = 0;
1426
1427         if (!tun)
1428                 return EPOLLERR;
1429
1430         sk = tfile->socket.sk;
1431
1432         poll_wait(file, sk_sleep(sk), wait);
1433
1434         if (!ptr_ring_empty(&tfile->tx_ring))
1435                 mask |= EPOLLIN | EPOLLRDNORM;
1436
1437         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1438          * guarantee EPOLLOUT to be raised by either here or
1439          * tun_sock_write_space(). Then process could get notification
1440          * after it writes to a down device and meets -EIO.
1441          */
1442         if (tun_sock_writeable(tun, tfile) ||
1443             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1444              tun_sock_writeable(tun, tfile)))
1445                 mask |= EPOLLOUT | EPOLLWRNORM;
1446
1447         if (tun->dev->reg_state != NETREG_REGISTERED)
1448                 mask = EPOLLERR;
1449
1450         tun_put(tun);
1451         return mask;
1452 }
1453
1454 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1455                                             size_t len,
1456                                             const struct iov_iter *it)
1457 {
1458         struct sk_buff *skb;
1459         size_t linear;
1460         int err;
1461         int i;
1462
1463         if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1464             len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1465                 return ERR_PTR(-EMSGSIZE);
1466
1467         local_bh_disable();
1468         skb = napi_get_frags(&tfile->napi);
1469         local_bh_enable();
1470         if (!skb)
1471                 return ERR_PTR(-ENOMEM);
1472
1473         linear = iov_iter_single_seg_count(it);
1474         err = __skb_grow(skb, linear);
1475         if (err)
1476                 goto free;
1477
1478         skb->len = len;
1479         skb->data_len = len - linear;
1480         skb->truesize += skb->data_len;
1481
1482         for (i = 1; i < it->nr_segs; i++) {
1483                 const struct iovec *iov = iter_iov(it);
1484                 size_t fragsz = iov->iov_len;
1485                 struct page *page;
1486                 void *frag;
1487
1488                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1489                         err = -EINVAL;
1490                         goto free;
1491                 }
1492                 frag = netdev_alloc_frag(fragsz);
1493                 if (!frag) {
1494                         err = -ENOMEM;
1495                         goto free;
1496                 }
1497                 page = virt_to_head_page(frag);
1498                 skb_fill_page_desc(skb, i - 1, page,
1499                                    frag - page_address(page), fragsz);
1500         }
1501
1502         return skb;
1503 free:
1504         /* frees skb and all frags allocated with napi_alloc_frag() */
1505         napi_free_frags(&tfile->napi);
1506         return ERR_PTR(err);
1507 }
1508
1509 /* prepad is the amount to reserve at front.  len is length after that.
1510  * linear is a hint as to how much to copy (usually headers). */
1511 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1512                                      size_t prepad, size_t len,
1513                                      size_t linear, int noblock)
1514 {
1515         struct sock *sk = tfile->socket.sk;
1516         struct sk_buff *skb;
1517         int err;
1518
1519         /* Under a page?  Don't bother with paged skb. */
1520         if (prepad + len < PAGE_SIZE)
1521                 linear = len;
1522
1523         if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1524                 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1525         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1526                                    &err, PAGE_ALLOC_COSTLY_ORDER);
1527         if (!skb)
1528                 return ERR_PTR(err);
1529
1530         skb_reserve(skb, prepad);
1531         skb_put(skb, linear);
1532         skb->data_len = len - linear;
1533         skb->len += len - linear;
1534
1535         return skb;
1536 }
1537
1538 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1539                            struct sk_buff *skb, int more)
1540 {
1541         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1542         struct sk_buff_head process_queue;
1543         u32 rx_batched = tun->rx_batched;
1544         bool rcv = false;
1545
1546         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1547                 local_bh_disable();
1548                 skb_record_rx_queue(skb, tfile->queue_index);
1549                 netif_receive_skb(skb);
1550                 local_bh_enable();
1551                 return;
1552         }
1553
1554         spin_lock(&queue->lock);
1555         if (!more || skb_queue_len(queue) == rx_batched) {
1556                 __skb_queue_head_init(&process_queue);
1557                 skb_queue_splice_tail_init(queue, &process_queue);
1558                 rcv = true;
1559         } else {
1560                 __skb_queue_tail(queue, skb);
1561         }
1562         spin_unlock(&queue->lock);
1563
1564         if (rcv) {
1565                 struct sk_buff *nskb;
1566
1567                 local_bh_disable();
1568                 while ((nskb = __skb_dequeue(&process_queue))) {
1569                         skb_record_rx_queue(nskb, tfile->queue_index);
1570                         netif_receive_skb(nskb);
1571                 }
1572                 skb_record_rx_queue(skb, tfile->queue_index);
1573                 netif_receive_skb(skb);
1574                 local_bh_enable();
1575         }
1576 }
1577
1578 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1579                               int len, int noblock, bool zerocopy)
1580 {
1581         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1582                 return false;
1583
1584         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1585                 return false;
1586
1587         if (!noblock)
1588                 return false;
1589
1590         if (zerocopy)
1591                 return false;
1592
1593         if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1594             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1595                 return false;
1596
1597         return true;
1598 }
1599
1600 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1601                                        struct page_frag *alloc_frag, char *buf,
1602                                        int buflen, int len, int pad)
1603 {
1604         struct sk_buff *skb = build_skb(buf, buflen);
1605
1606         if (!skb)
1607                 return ERR_PTR(-ENOMEM);
1608
1609         skb_reserve(skb, pad);
1610         skb_put(skb, len);
1611         skb_set_owner_w(skb, tfile->socket.sk);
1612
1613         get_page(alloc_frag->page);
1614         alloc_frag->offset += buflen;
1615
1616         return skb;
1617 }
1618
1619 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1620                        struct xdp_buff *xdp, u32 act)
1621 {
1622         int err;
1623
1624         switch (act) {
1625         case XDP_REDIRECT:
1626                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1627                 if (err) {
1628                         dev_core_stats_rx_dropped_inc(tun->dev);
1629                         return err;
1630                 }
1631                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1632                 break;
1633         case XDP_TX:
1634                 err = tun_xdp_tx(tun->dev, xdp);
1635                 if (err < 0) {
1636                         dev_core_stats_rx_dropped_inc(tun->dev);
1637                         return err;
1638                 }
1639                 dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1640                 break;
1641         case XDP_PASS:
1642                 break;
1643         default:
1644                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1645                 fallthrough;
1646         case XDP_ABORTED:
1647                 trace_xdp_exception(tun->dev, xdp_prog, act);
1648                 fallthrough;
1649         case XDP_DROP:
1650                 dev_core_stats_rx_dropped_inc(tun->dev);
1651                 break;
1652         }
1653
1654         return act;
1655 }
1656
1657 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1658                                      struct tun_file *tfile,
1659                                      struct iov_iter *from,
1660                                      struct virtio_net_hdr *hdr,
1661                                      int len, int *skb_xdp)
1662 {
1663         struct page_frag *alloc_frag = &current->task_frag;
1664         struct bpf_prog *xdp_prog;
1665         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1666         char *buf;
1667         size_t copied;
1668         int pad = TUN_RX_PAD;
1669         int err = 0;
1670
1671         rcu_read_lock();
1672         xdp_prog = rcu_dereference(tun->xdp_prog);
1673         if (xdp_prog)
1674                 pad += XDP_PACKET_HEADROOM;
1675         buflen += SKB_DATA_ALIGN(len + pad);
1676         rcu_read_unlock();
1677
1678         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1679         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1680                 return ERR_PTR(-ENOMEM);
1681
1682         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1683         copied = copy_page_from_iter(alloc_frag->page,
1684                                      alloc_frag->offset + pad,
1685                                      len, from);
1686         if (copied != len)
1687                 return ERR_PTR(-EFAULT);
1688
1689         /* There's a small window that XDP may be set after the check
1690          * of xdp_prog above, this should be rare and for simplicity
1691          * we do XDP on skb in case the headroom is not enough.
1692          */
1693         if (hdr->gso_type || !xdp_prog) {
1694                 *skb_xdp = 1;
1695                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1696                                        pad);
1697         }
1698
1699         *skb_xdp = 0;
1700
1701         local_bh_disable();
1702         rcu_read_lock();
1703         xdp_prog = rcu_dereference(tun->xdp_prog);
1704         if (xdp_prog) {
1705                 struct xdp_buff xdp;
1706                 u32 act;
1707
1708                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1709                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1710
1711                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1712                 if (act == XDP_REDIRECT || act == XDP_TX) {
1713                         get_page(alloc_frag->page);
1714                         alloc_frag->offset += buflen;
1715                 }
1716                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1717                 if (err < 0) {
1718                         if (act == XDP_REDIRECT || act == XDP_TX)
1719                                 put_page(alloc_frag->page);
1720                         goto out;
1721                 }
1722
1723                 if (err == XDP_REDIRECT)
1724                         xdp_do_flush();
1725                 if (err != XDP_PASS)
1726                         goto out;
1727
1728                 pad = xdp.data - xdp.data_hard_start;
1729                 len = xdp.data_end - xdp.data;
1730         }
1731         rcu_read_unlock();
1732         local_bh_enable();
1733
1734         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1735
1736 out:
1737         rcu_read_unlock();
1738         local_bh_enable();
1739         return NULL;
1740 }
1741
1742 /* Get packet from user space buffer */
1743 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1744                             void *msg_control, struct iov_iter *from,
1745                             int noblock, bool more)
1746 {
1747         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1748         struct sk_buff *skb;
1749         size_t total_len = iov_iter_count(from);
1750         size_t len = total_len, align = tun->align, linear;
1751         struct virtio_net_hdr gso = { 0 };
1752         int good_linear;
1753         int copylen;
1754         bool zerocopy = false;
1755         int err;
1756         u32 rxhash = 0;
1757         int skb_xdp = 1;
1758         bool frags = tun_napi_frags_enabled(tfile);
1759         enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1760
1761         if (!(tun->flags & IFF_NO_PI)) {
1762                 if (len < sizeof(pi))
1763                         return -EINVAL;
1764                 len -= sizeof(pi);
1765
1766                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1767                         return -EFAULT;
1768         }
1769
1770         if (tun->flags & IFF_VNET_HDR) {
1771                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1772
1773                 if (len < vnet_hdr_sz)
1774                         return -EINVAL;
1775                 len -= vnet_hdr_sz;
1776
1777                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1778                         return -EFAULT;
1779
1780                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1781                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1782                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1783
1784                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1785                         return -EINVAL;
1786                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1787         }
1788
1789         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1790                 align += NET_IP_ALIGN;
1791                 if (unlikely(len < ETH_HLEN ||
1792                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1793                         return -EINVAL;
1794         }
1795
1796         good_linear = SKB_MAX_HEAD(align);
1797
1798         if (msg_control) {
1799                 struct iov_iter i = *from;
1800
1801                 /* There are 256 bytes to be copied in skb, so there is
1802                  * enough room for skb expand head in case it is used.
1803                  * The rest of the buffer is mapped from userspace.
1804                  */
1805                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1806                 if (copylen > good_linear)
1807                         copylen = good_linear;
1808                 linear = copylen;
1809                 iov_iter_advance(&i, copylen);
1810                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1811                         zerocopy = true;
1812         }
1813
1814         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1815                 /* For the packet that is not easy to be processed
1816                  * (e.g gso or jumbo packet), we will do it at after
1817                  * skb was created with generic XDP routine.
1818                  */
1819                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1820                 err = PTR_ERR_OR_ZERO(skb);
1821                 if (err)
1822                         goto drop;
1823                 if (!skb)
1824                         return total_len;
1825         } else {
1826                 if (!zerocopy) {
1827                         copylen = len;
1828                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1829                                 linear = good_linear;
1830                         else
1831                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1832                 }
1833
1834                 if (frags) {
1835                         mutex_lock(&tfile->napi_mutex);
1836                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1837                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1838                          * If zerocopy is enabled, then this layout will be
1839                          * overwritten by zerocopy_sg_from_iter().
1840                          */
1841                         zerocopy = false;
1842                 } else {
1843                         if (!linear)
1844                                 linear = min_t(size_t, good_linear, copylen);
1845
1846                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1847                                             noblock);
1848                 }
1849
1850                 err = PTR_ERR_OR_ZERO(skb);
1851                 if (err)
1852                         goto drop;
1853
1854                 if (zerocopy)
1855                         err = zerocopy_sg_from_iter(skb, from);
1856                 else
1857                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1858
1859                 if (err) {
1860                         err = -EFAULT;
1861                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1862                         goto drop;
1863                 }
1864         }
1865
1866         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1867                 atomic_long_inc(&tun->rx_frame_errors);
1868                 err = -EINVAL;
1869                 goto free_skb;
1870         }
1871
1872         switch (tun->flags & TUN_TYPE_MASK) {
1873         case IFF_TUN:
1874                 if (tun->flags & IFF_NO_PI) {
1875                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1876
1877                         switch (ip_version) {
1878                         case 4:
1879                                 pi.proto = htons(ETH_P_IP);
1880                                 break;
1881                         case 6:
1882                                 pi.proto = htons(ETH_P_IPV6);
1883                                 break;
1884                         default:
1885                                 err = -EINVAL;
1886                                 goto drop;
1887                         }
1888                 }
1889
1890                 skb_reset_mac_header(skb);
1891                 skb->protocol = pi.proto;
1892                 skb->dev = tun->dev;
1893                 break;
1894         case IFF_TAP:
1895                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1896                         err = -ENOMEM;
1897                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1898                         goto drop;
1899                 }
1900                 skb->protocol = eth_type_trans(skb, tun->dev);
1901                 break;
1902         }
1903
1904         /* copy skb_ubuf_info for callback when skb has no error */
1905         if (zerocopy) {
1906                 skb_zcopy_init(skb, msg_control);
1907         } else if (msg_control) {
1908                 struct ubuf_info *uarg = msg_control;
1909                 uarg->callback(NULL, uarg, false);
1910         }
1911
1912         skb_reset_network_header(skb);
1913         skb_probe_transport_header(skb);
1914         skb_record_rx_queue(skb, tfile->queue_index);
1915
1916         if (skb_xdp) {
1917                 struct bpf_prog *xdp_prog;
1918                 int ret;
1919
1920                 local_bh_disable();
1921                 rcu_read_lock();
1922                 xdp_prog = rcu_dereference(tun->xdp_prog);
1923                 if (xdp_prog) {
1924                         ret = do_xdp_generic(xdp_prog, &skb);
1925                         if (ret != XDP_PASS) {
1926                                 rcu_read_unlock();
1927                                 local_bh_enable();
1928                                 goto unlock_frags;
1929                         }
1930                 }
1931                 rcu_read_unlock();
1932                 local_bh_enable();
1933         }
1934
1935         /* Compute the costly rx hash only if needed for flow updates.
1936          * We may get a very small possibility of OOO during switching, not
1937          * worth to optimize.
1938          */
1939         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1940             !tfile->detached)
1941                 rxhash = __skb_get_hash_symmetric(skb);
1942
1943         rcu_read_lock();
1944         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1945                 err = -EIO;
1946                 rcu_read_unlock();
1947                 drop_reason = SKB_DROP_REASON_DEV_READY;
1948                 goto drop;
1949         }
1950
1951         if (frags) {
1952                 u32 headlen;
1953
1954                 /* Exercise flow dissector code path. */
1955                 skb_push(skb, ETH_HLEN);
1956                 headlen = eth_get_headlen(tun->dev, skb->data,
1957                                           skb_headlen(skb));
1958
1959                 if (unlikely(headlen > skb_headlen(skb))) {
1960                         WARN_ON_ONCE(1);
1961                         err = -ENOMEM;
1962                         dev_core_stats_rx_dropped_inc(tun->dev);
1963 napi_busy:
1964                         napi_free_frags(&tfile->napi);
1965                         rcu_read_unlock();
1966                         mutex_unlock(&tfile->napi_mutex);
1967                         return err;
1968                 }
1969
1970                 if (likely(napi_schedule_prep(&tfile->napi))) {
1971                         local_bh_disable();
1972                         napi_gro_frags(&tfile->napi);
1973                         napi_complete(&tfile->napi);
1974                         local_bh_enable();
1975                 } else {
1976                         err = -EBUSY;
1977                         goto napi_busy;
1978                 }
1979                 mutex_unlock(&tfile->napi_mutex);
1980         } else if (tfile->napi_enabled) {
1981                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1982                 int queue_len;
1983
1984                 spin_lock_bh(&queue->lock);
1985
1986                 if (unlikely(tfile->detached)) {
1987                         spin_unlock_bh(&queue->lock);
1988                         rcu_read_unlock();
1989                         err = -EBUSY;
1990                         goto free_skb;
1991                 }
1992
1993                 __skb_queue_tail(queue, skb);
1994                 queue_len = skb_queue_len(queue);
1995                 spin_unlock(&queue->lock);
1996
1997                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1998                         napi_schedule(&tfile->napi);
1999
2000                 local_bh_enable();
2001         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2002                 tun_rx_batched(tun, tfile, skb, more);
2003         } else {
2004                 netif_rx(skb);
2005         }
2006         rcu_read_unlock();
2007
2008         preempt_disable();
2009         dev_sw_netstats_rx_add(tun->dev, len);
2010         preempt_enable();
2011
2012         if (rxhash)
2013                 tun_flow_update(tun, rxhash, tfile);
2014
2015         return total_len;
2016
2017 drop:
2018         if (err != -EAGAIN)
2019                 dev_core_stats_rx_dropped_inc(tun->dev);
2020
2021 free_skb:
2022         if (!IS_ERR_OR_NULL(skb))
2023                 kfree_skb_reason(skb, drop_reason);
2024
2025 unlock_frags:
2026         if (frags) {
2027                 tfile->napi.skb = NULL;
2028                 mutex_unlock(&tfile->napi_mutex);
2029         }
2030
2031         return err ?: total_len;
2032 }
2033
2034 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2035 {
2036         struct file *file = iocb->ki_filp;
2037         struct tun_file *tfile = file->private_data;
2038         struct tun_struct *tun = tun_get(tfile);
2039         ssize_t result;
2040         int noblock = 0;
2041
2042         if (!tun)
2043                 return -EBADFD;
2044
2045         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2046                 noblock = 1;
2047
2048         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2049
2050         tun_put(tun);
2051         return result;
2052 }
2053
2054 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2055                                 struct tun_file *tfile,
2056                                 struct xdp_frame *xdp_frame,
2057                                 struct iov_iter *iter)
2058 {
2059         int vnet_hdr_sz = 0;
2060         size_t size = xdp_frame->len;
2061         size_t ret;
2062
2063         if (tun->flags & IFF_VNET_HDR) {
2064                 struct virtio_net_hdr gso = { 0 };
2065
2066                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2067                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2068                         return -EINVAL;
2069                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2070                              sizeof(gso)))
2071                         return -EFAULT;
2072                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2073         }
2074
2075         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2076
2077         preempt_disable();
2078         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2079         preempt_enable();
2080
2081         return ret;
2082 }
2083
2084 /* Put packet to the user space buffer */
2085 static ssize_t tun_put_user(struct tun_struct *tun,
2086                             struct tun_file *tfile,
2087                             struct sk_buff *skb,
2088                             struct iov_iter *iter)
2089 {
2090         struct tun_pi pi = { 0, skb->protocol };
2091         ssize_t total;
2092         int vlan_offset = 0;
2093         int vlan_hlen = 0;
2094         int vnet_hdr_sz = 0;
2095
2096         if (skb_vlan_tag_present(skb))
2097                 vlan_hlen = VLAN_HLEN;
2098
2099         if (tun->flags & IFF_VNET_HDR)
2100                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2101
2102         total = skb->len + vlan_hlen + vnet_hdr_sz;
2103
2104         if (!(tun->flags & IFF_NO_PI)) {
2105                 if (iov_iter_count(iter) < sizeof(pi))
2106                         return -EINVAL;
2107
2108                 total += sizeof(pi);
2109                 if (iov_iter_count(iter) < total) {
2110                         /* Packet will be striped */
2111                         pi.flags |= TUN_PKT_STRIP;
2112                 }
2113
2114                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2115                         return -EFAULT;
2116         }
2117
2118         if (vnet_hdr_sz) {
2119                 struct virtio_net_hdr gso;
2120
2121                 if (iov_iter_count(iter) < vnet_hdr_sz)
2122                         return -EINVAL;
2123
2124                 if (virtio_net_hdr_from_skb(skb, &gso,
2125                                             tun_is_little_endian(tun), true,
2126                                             vlan_hlen)) {
2127                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2128                         pr_err("unexpected GSO type: "
2129                                "0x%x, gso_size %d, hdr_len %d\n",
2130                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2131                                tun16_to_cpu(tun, gso.hdr_len));
2132                         print_hex_dump(KERN_ERR, "tun: ",
2133                                        DUMP_PREFIX_NONE,
2134                                        16, 1, skb->head,
2135                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2136                         WARN_ON_ONCE(1);
2137                         return -EINVAL;
2138                 }
2139
2140                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2141                         return -EFAULT;
2142
2143                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2144         }
2145
2146         if (vlan_hlen) {
2147                 int ret;
2148                 struct veth veth;
2149
2150                 veth.h_vlan_proto = skb->vlan_proto;
2151                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2152
2153                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2154
2155                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2156                 if (ret || !iov_iter_count(iter))
2157                         goto done;
2158
2159                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2160                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2161                         goto done;
2162         }
2163
2164         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2165
2166 done:
2167         /* caller is in process context, */
2168         preempt_disable();
2169         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2170         preempt_enable();
2171
2172         return total;
2173 }
2174
2175 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2176 {
2177         DECLARE_WAITQUEUE(wait, current);
2178         void *ptr = NULL;
2179         int error = 0;
2180
2181         ptr = ptr_ring_consume(&tfile->tx_ring);
2182         if (ptr)
2183                 goto out;
2184         if (noblock) {
2185                 error = -EAGAIN;
2186                 goto out;
2187         }
2188
2189         add_wait_queue(&tfile->socket.wq.wait, &wait);
2190
2191         while (1) {
2192                 set_current_state(TASK_INTERRUPTIBLE);
2193                 ptr = ptr_ring_consume(&tfile->tx_ring);
2194                 if (ptr)
2195                         break;
2196                 if (signal_pending(current)) {
2197                         error = -ERESTARTSYS;
2198                         break;
2199                 }
2200                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2201                         error = -EFAULT;
2202                         break;
2203                 }
2204
2205                 schedule();
2206         }
2207
2208         __set_current_state(TASK_RUNNING);
2209         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2210
2211 out:
2212         *err = error;
2213         return ptr;
2214 }
2215
2216 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2217                            struct iov_iter *to,
2218                            int noblock, void *ptr)
2219 {
2220         ssize_t ret;
2221         int err;
2222
2223         if (!iov_iter_count(to)) {
2224                 tun_ptr_free(ptr);
2225                 return 0;
2226         }
2227
2228         if (!ptr) {
2229                 /* Read frames from ring */
2230                 ptr = tun_ring_recv(tfile, noblock, &err);
2231                 if (!ptr)
2232                         return err;
2233         }
2234
2235         if (tun_is_xdp_frame(ptr)) {
2236                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2237
2238                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2239                 xdp_return_frame(xdpf);
2240         } else {
2241                 struct sk_buff *skb = ptr;
2242
2243                 ret = tun_put_user(tun, tfile, skb, to);
2244                 if (unlikely(ret < 0))
2245                         kfree_skb(skb);
2246                 else
2247                         consume_skb(skb);
2248         }
2249
2250         return ret;
2251 }
2252
2253 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2254 {
2255         struct file *file = iocb->ki_filp;
2256         struct tun_file *tfile = file->private_data;
2257         struct tun_struct *tun = tun_get(tfile);
2258         ssize_t len = iov_iter_count(to), ret;
2259         int noblock = 0;
2260
2261         if (!tun)
2262                 return -EBADFD;
2263
2264         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2265                 noblock = 1;
2266
2267         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2268         ret = min_t(ssize_t, ret, len);
2269         if (ret > 0)
2270                 iocb->ki_pos = ret;
2271         tun_put(tun);
2272         return ret;
2273 }
2274
2275 static void tun_prog_free(struct rcu_head *rcu)
2276 {
2277         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2278
2279         bpf_prog_destroy(prog->prog);
2280         kfree(prog);
2281 }
2282
2283 static int __tun_set_ebpf(struct tun_struct *tun,
2284                           struct tun_prog __rcu **prog_p,
2285                           struct bpf_prog *prog)
2286 {
2287         struct tun_prog *old, *new = NULL;
2288
2289         if (prog) {
2290                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2291                 if (!new)
2292                         return -ENOMEM;
2293                 new->prog = prog;
2294         }
2295
2296         spin_lock_bh(&tun->lock);
2297         old = rcu_dereference_protected(*prog_p,
2298                                         lockdep_is_held(&tun->lock));
2299         rcu_assign_pointer(*prog_p, new);
2300         spin_unlock_bh(&tun->lock);
2301
2302         if (old)
2303                 call_rcu(&old->rcu, tun_prog_free);
2304
2305         return 0;
2306 }
2307
2308 static void tun_free_netdev(struct net_device *dev)
2309 {
2310         struct tun_struct *tun = netdev_priv(dev);
2311
2312         BUG_ON(!(list_empty(&tun->disabled)));
2313
2314         tun_flow_uninit(tun);
2315         security_tun_dev_free_security(tun->security);
2316         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2317         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2318 }
2319
2320 static void tun_setup(struct net_device *dev)
2321 {
2322         struct tun_struct *tun = netdev_priv(dev);
2323
2324         tun->owner = INVALID_UID;
2325         tun->group = INVALID_GID;
2326         tun_default_link_ksettings(dev, &tun->link_ksettings);
2327
2328         dev->ethtool_ops = &tun_ethtool_ops;
2329         dev->needs_free_netdev = true;
2330         dev->priv_destructor = tun_free_netdev;
2331         /* We prefer our own queue length */
2332         dev->tx_queue_len = TUN_READQ_SIZE;
2333 }
2334
2335 /* Trivial set of netlink ops to allow deleting tun or tap
2336  * device with netlink.
2337  */
2338 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2339                         struct netlink_ext_ack *extack)
2340 {
2341         NL_SET_ERR_MSG(extack,
2342                        "tun/tap creation via rtnetlink is not supported.");
2343         return -EOPNOTSUPP;
2344 }
2345
2346 static size_t tun_get_size(const struct net_device *dev)
2347 {
2348         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2349         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2350
2351         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2352                nla_total_size(sizeof(gid_t)) + /* GROUP */
2353                nla_total_size(sizeof(u8)) + /* TYPE */
2354                nla_total_size(sizeof(u8)) + /* PI */
2355                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2356                nla_total_size(sizeof(u8)) + /* PERSIST */
2357                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2358                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2359                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2360                0;
2361 }
2362
2363 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2364 {
2365         struct tun_struct *tun = netdev_priv(dev);
2366
2367         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2368                 goto nla_put_failure;
2369         if (uid_valid(tun->owner) &&
2370             nla_put_u32(skb, IFLA_TUN_OWNER,
2371                         from_kuid_munged(current_user_ns(), tun->owner)))
2372                 goto nla_put_failure;
2373         if (gid_valid(tun->group) &&
2374             nla_put_u32(skb, IFLA_TUN_GROUP,
2375                         from_kgid_munged(current_user_ns(), tun->group)))
2376                 goto nla_put_failure;
2377         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2378                 goto nla_put_failure;
2379         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2380                 goto nla_put_failure;
2381         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2382                 goto nla_put_failure;
2383         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2384                        !!(tun->flags & IFF_MULTI_QUEUE)))
2385                 goto nla_put_failure;
2386         if (tun->flags & IFF_MULTI_QUEUE) {
2387                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2388                         goto nla_put_failure;
2389                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2390                                 tun->numdisabled))
2391                         goto nla_put_failure;
2392         }
2393
2394         return 0;
2395
2396 nla_put_failure:
2397         return -EMSGSIZE;
2398 }
2399
2400 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2401         .kind           = DRV_NAME,
2402         .priv_size      = sizeof(struct tun_struct),
2403         .setup          = tun_setup,
2404         .validate       = tun_validate,
2405         .get_size       = tun_get_size,
2406         .fill_info      = tun_fill_info,
2407 };
2408
2409 static void tun_sock_write_space(struct sock *sk)
2410 {
2411         struct tun_file *tfile;
2412         wait_queue_head_t *wqueue;
2413
2414         if (!sock_writeable(sk))
2415                 return;
2416
2417         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2418                 return;
2419
2420         wqueue = sk_sleep(sk);
2421         if (wqueue && waitqueue_active(wqueue))
2422                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2423                                                 EPOLLWRNORM | EPOLLWRBAND);
2424
2425         tfile = container_of(sk, struct tun_file, sk);
2426         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2427 }
2428
2429 static void tun_put_page(struct tun_page *tpage)
2430 {
2431         if (tpage->page)
2432                 __page_frag_cache_drain(tpage->page, tpage->count);
2433 }
2434
2435 static int tun_xdp_one(struct tun_struct *tun,
2436                        struct tun_file *tfile,
2437                        struct xdp_buff *xdp, int *flush,
2438                        struct tun_page *tpage)
2439 {
2440         unsigned int datasize = xdp->data_end - xdp->data;
2441         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2442         struct virtio_net_hdr *gso = &hdr->gso;
2443         struct bpf_prog *xdp_prog;
2444         struct sk_buff *skb = NULL;
2445         struct sk_buff_head *queue;
2446         u32 rxhash = 0, act;
2447         int buflen = hdr->buflen;
2448         int ret = 0;
2449         bool skb_xdp = false;
2450         struct page *page;
2451
2452         xdp_prog = rcu_dereference(tun->xdp_prog);
2453         if (xdp_prog) {
2454                 if (gso->gso_type) {
2455                         skb_xdp = true;
2456                         goto build;
2457                 }
2458
2459                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2460                 xdp_set_data_meta_invalid(xdp);
2461
2462                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2463                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2464                 if (ret < 0) {
2465                         put_page(virt_to_head_page(xdp->data));
2466                         return ret;
2467                 }
2468
2469                 switch (ret) {
2470                 case XDP_REDIRECT:
2471                         *flush = true;
2472                         fallthrough;
2473                 case XDP_TX:
2474                         return 0;
2475                 case XDP_PASS:
2476                         break;
2477                 default:
2478                         page = virt_to_head_page(xdp->data);
2479                         if (tpage->page == page) {
2480                                 ++tpage->count;
2481                         } else {
2482                                 tun_put_page(tpage);
2483                                 tpage->page = page;
2484                                 tpage->count = 1;
2485                         }
2486                         return 0;
2487                 }
2488         }
2489
2490 build:
2491         skb = build_skb(xdp->data_hard_start, buflen);
2492         if (!skb) {
2493                 ret = -ENOMEM;
2494                 goto out;
2495         }
2496
2497         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2498         skb_put(skb, xdp->data_end - xdp->data);
2499
2500         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2501                 atomic_long_inc(&tun->rx_frame_errors);
2502                 kfree_skb(skb);
2503                 ret = -EINVAL;
2504                 goto out;
2505         }
2506
2507         skb->protocol = eth_type_trans(skb, tun->dev);
2508         skb_reset_network_header(skb);
2509         skb_probe_transport_header(skb);
2510         skb_record_rx_queue(skb, tfile->queue_index);
2511
2512         if (skb_xdp) {
2513                 ret = do_xdp_generic(xdp_prog, &skb);
2514                 if (ret != XDP_PASS) {
2515                         ret = 0;
2516                         goto out;
2517                 }
2518         }
2519
2520         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2521             !tfile->detached)
2522                 rxhash = __skb_get_hash_symmetric(skb);
2523
2524         if (tfile->napi_enabled) {
2525                 queue = &tfile->sk.sk_write_queue;
2526                 spin_lock(&queue->lock);
2527
2528                 if (unlikely(tfile->detached)) {
2529                         spin_unlock(&queue->lock);
2530                         kfree_skb(skb);
2531                         return -EBUSY;
2532                 }
2533
2534                 __skb_queue_tail(queue, skb);
2535                 spin_unlock(&queue->lock);
2536                 ret = 1;
2537         } else {
2538                 netif_receive_skb(skb);
2539                 ret = 0;
2540         }
2541
2542         /* No need to disable preemption here since this function is
2543          * always called with bh disabled
2544          */
2545         dev_sw_netstats_rx_add(tun->dev, datasize);
2546
2547         if (rxhash)
2548                 tun_flow_update(tun, rxhash, tfile);
2549
2550 out:
2551         return ret;
2552 }
2553
2554 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2555 {
2556         int ret, i;
2557         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2558         struct tun_struct *tun = tun_get(tfile);
2559         struct tun_msg_ctl *ctl = m->msg_control;
2560         struct xdp_buff *xdp;
2561
2562         if (!tun)
2563                 return -EBADFD;
2564
2565         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2566             ctl && ctl->type == TUN_MSG_PTR) {
2567                 struct tun_page tpage;
2568                 int n = ctl->num;
2569                 int flush = 0, queued = 0;
2570
2571                 memset(&tpage, 0, sizeof(tpage));
2572
2573                 local_bh_disable();
2574                 rcu_read_lock();
2575
2576                 for (i = 0; i < n; i++) {
2577                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2578                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2579                         if (ret > 0)
2580                                 queued += ret;
2581                 }
2582
2583                 if (flush)
2584                         xdp_do_flush();
2585
2586                 if (tfile->napi_enabled && queued > 0)
2587                         napi_schedule(&tfile->napi);
2588
2589                 rcu_read_unlock();
2590                 local_bh_enable();
2591
2592                 tun_put_page(&tpage);
2593
2594                 ret = total_len;
2595                 goto out;
2596         }
2597
2598         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2599                            m->msg_flags & MSG_DONTWAIT,
2600                            m->msg_flags & MSG_MORE);
2601 out:
2602         tun_put(tun);
2603         return ret;
2604 }
2605
2606 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2607                        int flags)
2608 {
2609         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2610         struct tun_struct *tun = tun_get(tfile);
2611         void *ptr = m->msg_control;
2612         int ret;
2613
2614         if (!tun) {
2615                 ret = -EBADFD;
2616                 goto out_free;
2617         }
2618
2619         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2620                 ret = -EINVAL;
2621                 goto out_put_tun;
2622         }
2623         if (flags & MSG_ERRQUEUE) {
2624                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2625                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2626                 goto out;
2627         }
2628         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2629         if (ret > (ssize_t)total_len) {
2630                 m->msg_flags |= MSG_TRUNC;
2631                 ret = flags & MSG_TRUNC ? ret : total_len;
2632         }
2633 out:
2634         tun_put(tun);
2635         return ret;
2636
2637 out_put_tun:
2638         tun_put(tun);
2639 out_free:
2640         tun_ptr_free(ptr);
2641         return ret;
2642 }
2643
2644 static int tun_ptr_peek_len(void *ptr)
2645 {
2646         if (likely(ptr)) {
2647                 if (tun_is_xdp_frame(ptr)) {
2648                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2649
2650                         return xdpf->len;
2651                 }
2652                 return __skb_array_len_with_tag(ptr);
2653         } else {
2654                 return 0;
2655         }
2656 }
2657
2658 static int tun_peek_len(struct socket *sock)
2659 {
2660         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2661         struct tun_struct *tun;
2662         int ret = 0;
2663
2664         tun = tun_get(tfile);
2665         if (!tun)
2666                 return 0;
2667
2668         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2669         tun_put(tun);
2670
2671         return ret;
2672 }
2673
2674 /* Ops structure to mimic raw sockets with tun */
2675 static const struct proto_ops tun_socket_ops = {
2676         .peek_len = tun_peek_len,
2677         .sendmsg = tun_sendmsg,
2678         .recvmsg = tun_recvmsg,
2679 };
2680
2681 static struct proto tun_proto = {
2682         .name           = "tun",
2683         .owner          = THIS_MODULE,
2684         .obj_size       = sizeof(struct tun_file),
2685 };
2686
2687 static int tun_flags(struct tun_struct *tun)
2688 {
2689         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2690 }
2691
2692 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2693                               char *buf)
2694 {
2695         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2696         return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2697 }
2698
2699 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2700                           char *buf)
2701 {
2702         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2703         return uid_valid(tun->owner)?
2704                 sysfs_emit(buf, "%u\n",
2705                            from_kuid_munged(current_user_ns(), tun->owner)) :
2706                 sysfs_emit(buf, "-1\n");
2707 }
2708
2709 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2710                           char *buf)
2711 {
2712         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2713         return gid_valid(tun->group) ?
2714                 sysfs_emit(buf, "%u\n",
2715                            from_kgid_munged(current_user_ns(), tun->group)) :
2716                 sysfs_emit(buf, "-1\n");
2717 }
2718
2719 static DEVICE_ATTR_RO(tun_flags);
2720 static DEVICE_ATTR_RO(owner);
2721 static DEVICE_ATTR_RO(group);
2722
2723 static struct attribute *tun_dev_attrs[] = {
2724         &dev_attr_tun_flags.attr,
2725         &dev_attr_owner.attr,
2726         &dev_attr_group.attr,
2727         NULL
2728 };
2729
2730 static const struct attribute_group tun_attr_group = {
2731         .attrs = tun_dev_attrs
2732 };
2733
2734 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2735 {
2736         struct tun_struct *tun;
2737         struct tun_file *tfile = file->private_data;
2738         struct net_device *dev;
2739         int err;
2740
2741         if (tfile->detached)
2742                 return -EINVAL;
2743
2744         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2745                 if (!capable(CAP_NET_ADMIN))
2746                         return -EPERM;
2747
2748                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2749                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2750                         return -EINVAL;
2751         }
2752
2753         dev = __dev_get_by_name(net, ifr->ifr_name);
2754         if (dev) {
2755                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2756                         return -EBUSY;
2757                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2758                         tun = netdev_priv(dev);
2759                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2760                         tun = netdev_priv(dev);
2761                 else
2762                         return -EINVAL;
2763
2764                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2765                     !!(tun->flags & IFF_MULTI_QUEUE))
2766                         return -EINVAL;
2767
2768                 if (tun_not_capable(tun))
2769                         return -EPERM;
2770                 err = security_tun_dev_open(tun->security);
2771                 if (err < 0)
2772                         return err;
2773
2774                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2775                                  ifr->ifr_flags & IFF_NAPI,
2776                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2777                 if (err < 0)
2778                         return err;
2779
2780                 if (tun->flags & IFF_MULTI_QUEUE &&
2781                     (tun->numqueues + tun->numdisabled > 1)) {
2782                         /* One or more queue has already been attached, no need
2783                          * to initialize the device again.
2784                          */
2785                         netdev_state_change(dev);
2786                         return 0;
2787                 }
2788
2789                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2790                               (ifr->ifr_flags & TUN_FEATURES);
2791
2792                 netdev_state_change(dev);
2793         } else {
2794                 char *name;
2795                 unsigned long flags = 0;
2796                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2797                              MAX_TAP_QUEUES : 1;
2798
2799                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2800                         return -EPERM;
2801                 err = security_tun_dev_create();
2802                 if (err < 0)
2803                         return err;
2804
2805                 /* Set dev type */
2806                 if (ifr->ifr_flags & IFF_TUN) {
2807                         /* TUN device */
2808                         flags |= IFF_TUN;
2809                         name = "tun%d";
2810                 } else if (ifr->ifr_flags & IFF_TAP) {
2811                         /* TAP device */
2812                         flags |= IFF_TAP;
2813                         name = "tap%d";
2814                 } else
2815                         return -EINVAL;
2816
2817                 if (*ifr->ifr_name)
2818                         name = ifr->ifr_name;
2819
2820                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2821                                        NET_NAME_UNKNOWN, tun_setup, queues,
2822                                        queues);
2823
2824                 if (!dev)
2825                         return -ENOMEM;
2826
2827                 dev_net_set(dev, net);
2828                 dev->rtnl_link_ops = &tun_link_ops;
2829                 dev->ifindex = tfile->ifindex;
2830                 dev->sysfs_groups[0] = &tun_attr_group;
2831
2832                 tun = netdev_priv(dev);
2833                 tun->dev = dev;
2834                 tun->flags = flags;
2835                 tun->txflt.count = 0;
2836                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2837
2838                 tun->align = NET_SKB_PAD;
2839                 tun->filter_attached = false;
2840                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2841                 tun->rx_batched = 0;
2842                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2843
2844                 tun->ifr = ifr;
2845                 tun->file = file;
2846
2847                 tun_net_initialize(dev);
2848
2849                 err = register_netdevice(tun->dev);
2850                 if (err < 0) {
2851                         free_netdev(dev);
2852                         return err;
2853                 }
2854                 /* free_netdev() won't check refcnt, to avoid race
2855                  * with dev_put() we need publish tun after registration.
2856                  */
2857                 rcu_assign_pointer(tfile->tun, tun);
2858         }
2859
2860         if (ifr->ifr_flags & IFF_NO_CARRIER)
2861                 netif_carrier_off(tun->dev);
2862         else
2863                 netif_carrier_on(tun->dev);
2864
2865         /* Make sure persistent devices do not get stuck in
2866          * xoff state.
2867          */
2868         if (netif_running(tun->dev))
2869                 netif_tx_wake_all_queues(tun->dev);
2870
2871         strcpy(ifr->ifr_name, tun->dev->name);
2872         return 0;
2873 }
2874
2875 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2876 {
2877         strcpy(ifr->ifr_name, tun->dev->name);
2878
2879         ifr->ifr_flags = tun_flags(tun);
2880
2881 }
2882
2883 /* This is like a cut-down ethtool ops, except done via tun fd so no
2884  * privs required. */
2885 static int set_offload(struct tun_struct *tun, unsigned long arg)
2886 {
2887         netdev_features_t features = 0;
2888
2889         if (arg & TUN_F_CSUM) {
2890                 features |= NETIF_F_HW_CSUM;
2891                 arg &= ~TUN_F_CSUM;
2892
2893                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2894                         if (arg & TUN_F_TSO_ECN) {
2895                                 features |= NETIF_F_TSO_ECN;
2896                                 arg &= ~TUN_F_TSO_ECN;
2897                         }
2898                         if (arg & TUN_F_TSO4)
2899                                 features |= NETIF_F_TSO;
2900                         if (arg & TUN_F_TSO6)
2901                                 features |= NETIF_F_TSO6;
2902                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2903                 }
2904
2905                 arg &= ~TUN_F_UFO;
2906
2907                 /* TODO: for now USO4 and USO6 should work simultaneously */
2908                 if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2909                         features |= NETIF_F_GSO_UDP_L4;
2910                         arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2911                 }
2912         }
2913
2914         /* This gives the user a way to test for new features in future by
2915          * trying to set them. */
2916         if (arg)
2917                 return -EINVAL;
2918
2919         tun->set_features = features;
2920         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2921         tun->dev->wanted_features |= features;
2922         netdev_update_features(tun->dev);
2923
2924         return 0;
2925 }
2926
2927 static void tun_detach_filter(struct tun_struct *tun, int n)
2928 {
2929         int i;
2930         struct tun_file *tfile;
2931
2932         for (i = 0; i < n; i++) {
2933                 tfile = rtnl_dereference(tun->tfiles[i]);
2934                 lock_sock(tfile->socket.sk);
2935                 sk_detach_filter(tfile->socket.sk);
2936                 release_sock(tfile->socket.sk);
2937         }
2938
2939         tun->filter_attached = false;
2940 }
2941
2942 static int tun_attach_filter(struct tun_struct *tun)
2943 {
2944         int i, ret = 0;
2945         struct tun_file *tfile;
2946
2947         for (i = 0; i < tun->numqueues; i++) {
2948                 tfile = rtnl_dereference(tun->tfiles[i]);
2949                 lock_sock(tfile->socket.sk);
2950                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2951                 release_sock(tfile->socket.sk);
2952                 if (ret) {
2953                         tun_detach_filter(tun, i);
2954                         return ret;
2955                 }
2956         }
2957
2958         tun->filter_attached = true;
2959         return ret;
2960 }
2961
2962 static void tun_set_sndbuf(struct tun_struct *tun)
2963 {
2964         struct tun_file *tfile;
2965         int i;
2966
2967         for (i = 0; i < tun->numqueues; i++) {
2968                 tfile = rtnl_dereference(tun->tfiles[i]);
2969                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2970         }
2971 }
2972
2973 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2974 {
2975         struct tun_file *tfile = file->private_data;
2976         struct tun_struct *tun;
2977         int ret = 0;
2978
2979         rtnl_lock();
2980
2981         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2982                 tun = tfile->detached;
2983                 if (!tun) {
2984                         ret = -EINVAL;
2985                         goto unlock;
2986                 }
2987                 ret = security_tun_dev_attach_queue(tun->security);
2988                 if (ret < 0)
2989                         goto unlock;
2990                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2991                                  tun->flags & IFF_NAPI_FRAGS, true);
2992         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2993                 tun = rtnl_dereference(tfile->tun);
2994                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2995                         ret = -EINVAL;
2996                 else
2997                         __tun_detach(tfile, false);
2998         } else
2999                 ret = -EINVAL;
3000
3001         if (ret >= 0)
3002                 netdev_state_change(tun->dev);
3003
3004 unlock:
3005         rtnl_unlock();
3006         return ret;
3007 }
3008
3009 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
3010                         void __user *data)
3011 {
3012         struct bpf_prog *prog;
3013         int fd;
3014
3015         if (copy_from_user(&fd, data, sizeof(fd)))
3016                 return -EFAULT;
3017
3018         if (fd == -1) {
3019                 prog = NULL;
3020         } else {
3021                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3022                 if (IS_ERR(prog))
3023                         return PTR_ERR(prog);
3024         }
3025
3026         return __tun_set_ebpf(tun, prog_p, prog);
3027 }
3028
3029 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3030 static unsigned char tun_get_addr_len(unsigned short type)
3031 {
3032         switch (type) {
3033         case ARPHRD_IP6GRE:
3034         case ARPHRD_TUNNEL6:
3035                 return sizeof(struct in6_addr);
3036         case ARPHRD_IPGRE:
3037         case ARPHRD_TUNNEL:
3038         case ARPHRD_SIT:
3039                 return 4;
3040         case ARPHRD_ETHER:
3041                 return ETH_ALEN;
3042         case ARPHRD_IEEE802154:
3043         case ARPHRD_IEEE802154_MONITOR:
3044                 return IEEE802154_EXTENDED_ADDR_LEN;
3045         case ARPHRD_PHONET_PIPE:
3046         case ARPHRD_PPP:
3047         case ARPHRD_NONE:
3048                 return 0;
3049         case ARPHRD_6LOWPAN:
3050                 return EUI64_ADDR_LEN;
3051         case ARPHRD_FDDI:
3052                 return FDDI_K_ALEN;
3053         case ARPHRD_HIPPI:
3054                 return HIPPI_ALEN;
3055         case ARPHRD_IEEE802:
3056                 return FC_ALEN;
3057         case ARPHRD_ROSE:
3058                 return ROSE_ADDR_LEN;
3059         case ARPHRD_NETROM:
3060                 return AX25_ADDR_LEN;
3061         case ARPHRD_LOCALTLK:
3062                 return LTALK_ALEN;
3063         default:
3064                 return 0;
3065         }
3066 }
3067
3068 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3069                             unsigned long arg, int ifreq_len)
3070 {
3071         struct tun_file *tfile = file->private_data;
3072         struct net *net = sock_net(&tfile->sk);
3073         struct tun_struct *tun;
3074         void __user* argp = (void __user*)arg;
3075         unsigned int carrier;
3076         struct ifreq ifr;
3077         kuid_t owner;
3078         kgid_t group;
3079         int ifindex;
3080         int sndbuf;
3081         int vnet_hdr_sz;
3082         int le;
3083         int ret;
3084         bool do_notify = false;
3085
3086         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3087             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3088                 if (copy_from_user(&ifr, argp, ifreq_len))
3089                         return -EFAULT;
3090         } else {
3091                 memset(&ifr, 0, sizeof(ifr));
3092         }
3093         if (cmd == TUNGETFEATURES) {
3094                 /* Currently this just means: "what IFF flags are valid?".
3095                  * This is needed because we never checked for invalid flags on
3096                  * TUNSETIFF.
3097                  */
3098                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3099                                 TUN_FEATURES, (unsigned int __user*)argp);
3100         } else if (cmd == TUNSETQUEUE) {
3101                 return tun_set_queue(file, &ifr);
3102         } else if (cmd == SIOCGSKNS) {
3103                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3104                         return -EPERM;
3105                 return open_related_ns(&net->ns, get_net_ns);
3106         }
3107
3108         rtnl_lock();
3109
3110         tun = tun_get(tfile);
3111         if (cmd == TUNSETIFF) {
3112                 ret = -EEXIST;
3113                 if (tun)
3114                         goto unlock;
3115
3116                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3117
3118                 ret = tun_set_iff(net, file, &ifr);
3119
3120                 if (ret)
3121                         goto unlock;
3122
3123                 if (copy_to_user(argp, &ifr, ifreq_len))
3124                         ret = -EFAULT;
3125                 goto unlock;
3126         }
3127         if (cmd == TUNSETIFINDEX) {
3128                 ret = -EPERM;
3129                 if (tun)
3130                         goto unlock;
3131
3132                 ret = -EFAULT;
3133                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3134                         goto unlock;
3135                 ret = -EINVAL;
3136                 if (ifindex < 0)
3137                         goto unlock;
3138                 ret = 0;
3139                 tfile->ifindex = ifindex;
3140                 goto unlock;
3141         }
3142
3143         ret = -EBADFD;
3144         if (!tun)
3145                 goto unlock;
3146
3147         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3148
3149         net = dev_net(tun->dev);
3150         ret = 0;
3151         switch (cmd) {
3152         case TUNGETIFF:
3153                 tun_get_iff(tun, &ifr);
3154
3155                 if (tfile->detached)
3156                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3157                 if (!tfile->socket.sk->sk_filter)
3158                         ifr.ifr_flags |= IFF_NOFILTER;
3159
3160                 if (copy_to_user(argp, &ifr, ifreq_len))
3161                         ret = -EFAULT;
3162                 break;
3163
3164         case TUNSETNOCSUM:
3165                 /* Disable/Enable checksum */
3166
3167                 /* [unimplemented] */
3168                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3169                            arg ? "disabled" : "enabled");
3170                 break;
3171
3172         case TUNSETPERSIST:
3173                 /* Disable/Enable persist mode. Keep an extra reference to the
3174                  * module to prevent the module being unprobed.
3175                  */
3176                 if (arg && !(tun->flags & IFF_PERSIST)) {
3177                         tun->flags |= IFF_PERSIST;
3178                         __module_get(THIS_MODULE);
3179                         do_notify = true;
3180                 }
3181                 if (!arg && (tun->flags & IFF_PERSIST)) {
3182                         tun->flags &= ~IFF_PERSIST;
3183                         module_put(THIS_MODULE);
3184                         do_notify = true;
3185                 }
3186
3187                 netif_info(tun, drv, tun->dev, "persist %s\n",
3188                            arg ? "enabled" : "disabled");
3189                 break;
3190
3191         case TUNSETOWNER:
3192                 /* Set owner of the device */
3193                 owner = make_kuid(current_user_ns(), arg);
3194                 if (!uid_valid(owner)) {
3195                         ret = -EINVAL;
3196                         break;
3197                 }
3198                 tun->owner = owner;
3199                 do_notify = true;
3200                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3201                            from_kuid(&init_user_ns, tun->owner));
3202                 break;
3203
3204         case TUNSETGROUP:
3205                 /* Set group of the device */
3206                 group = make_kgid(current_user_ns(), arg);
3207                 if (!gid_valid(group)) {
3208                         ret = -EINVAL;
3209                         break;
3210                 }
3211                 tun->group = group;
3212                 do_notify = true;
3213                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3214                            from_kgid(&init_user_ns, tun->group));
3215                 break;
3216
3217         case TUNSETLINK:
3218                 /* Only allow setting the type when the interface is down */
3219                 if (tun->dev->flags & IFF_UP) {
3220                         netif_info(tun, drv, tun->dev,
3221                                    "Linktype set failed because interface is up\n");
3222                         ret = -EBUSY;
3223                 } else {
3224                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3225                                                        tun->dev);
3226                         ret = notifier_to_errno(ret);
3227                         if (ret) {
3228                                 netif_info(tun, drv, tun->dev,
3229                                            "Refused to change device type\n");
3230                                 break;
3231                         }
3232                         tun->dev->type = (int) arg;
3233                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3234                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3235                                    tun->dev->type);
3236                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3237                                                  tun->dev);
3238                 }
3239                 break;
3240
3241         case TUNSETDEBUG:
3242                 tun->msg_enable = (u32)arg;
3243                 break;
3244
3245         case TUNSETOFFLOAD:
3246                 ret = set_offload(tun, arg);
3247                 break;
3248
3249         case TUNSETTXFILTER:
3250                 /* Can be set only for TAPs */
3251                 ret = -EINVAL;
3252                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3253                         break;
3254                 ret = update_filter(&tun->txflt, (void __user *)arg);
3255                 break;
3256
3257         case SIOCGIFHWADDR:
3258                 /* Get hw address */
3259                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3260                 if (copy_to_user(argp, &ifr, ifreq_len))
3261                         ret = -EFAULT;
3262                 break;
3263
3264         case SIOCSIFHWADDR:
3265                 /* Set hw address */
3266                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3267                 break;
3268
3269         case TUNGETSNDBUF:
3270                 sndbuf = tfile->socket.sk->sk_sndbuf;
3271                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3272                         ret = -EFAULT;
3273                 break;
3274
3275         case TUNSETSNDBUF:
3276                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3277                         ret = -EFAULT;
3278                         break;
3279                 }
3280                 if (sndbuf <= 0) {
3281                         ret = -EINVAL;
3282                         break;
3283                 }
3284
3285                 tun->sndbuf = sndbuf;
3286                 tun_set_sndbuf(tun);
3287                 break;
3288
3289         case TUNGETVNETHDRSZ:
3290                 vnet_hdr_sz = tun->vnet_hdr_sz;
3291                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3292                         ret = -EFAULT;
3293                 break;
3294
3295         case TUNSETVNETHDRSZ:
3296                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3297                         ret = -EFAULT;
3298                         break;
3299                 }
3300                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3301                         ret = -EINVAL;
3302                         break;
3303                 }
3304
3305                 tun->vnet_hdr_sz = vnet_hdr_sz;
3306                 break;
3307
3308         case TUNGETVNETLE:
3309                 le = !!(tun->flags & TUN_VNET_LE);
3310                 if (put_user(le, (int __user *)argp))
3311                         ret = -EFAULT;
3312                 break;
3313
3314         case TUNSETVNETLE:
3315                 if (get_user(le, (int __user *)argp)) {
3316                         ret = -EFAULT;
3317                         break;
3318                 }
3319                 if (le)
3320                         tun->flags |= TUN_VNET_LE;
3321                 else
3322                         tun->flags &= ~TUN_VNET_LE;
3323                 break;
3324
3325         case TUNGETVNETBE:
3326                 ret = tun_get_vnet_be(tun, argp);
3327                 break;
3328
3329         case TUNSETVNETBE:
3330                 ret = tun_set_vnet_be(tun, argp);
3331                 break;
3332
3333         case TUNATTACHFILTER:
3334                 /* Can be set only for TAPs */
3335                 ret = -EINVAL;
3336                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3337                         break;
3338                 ret = -EFAULT;
3339                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3340                         break;
3341
3342                 ret = tun_attach_filter(tun);
3343                 break;
3344
3345         case TUNDETACHFILTER:
3346                 /* Can be set only for TAPs */
3347                 ret = -EINVAL;
3348                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3349                         break;
3350                 ret = 0;
3351                 tun_detach_filter(tun, tun->numqueues);
3352                 break;
3353
3354         case TUNGETFILTER:
3355                 ret = -EINVAL;
3356                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3357                         break;
3358                 ret = -EFAULT;
3359                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3360                         break;
3361                 ret = 0;
3362                 break;
3363
3364         case TUNSETSTEERINGEBPF:
3365                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3366                 break;
3367
3368         case TUNSETFILTEREBPF:
3369                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3370                 break;
3371
3372         case TUNSETCARRIER:
3373                 ret = -EFAULT;
3374                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3375                         goto unlock;
3376
3377                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3378                 break;
3379
3380         case TUNGETDEVNETNS:
3381                 ret = -EPERM;
3382                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3383                         goto unlock;
3384                 ret = open_related_ns(&net->ns, get_net_ns);
3385                 break;
3386
3387         default:
3388                 ret = -EINVAL;
3389                 break;
3390         }
3391
3392         if (do_notify)
3393                 netdev_state_change(tun->dev);
3394
3395 unlock:
3396         rtnl_unlock();
3397         if (tun)
3398                 tun_put(tun);
3399         return ret;
3400 }
3401
3402 static long tun_chr_ioctl(struct file *file,
3403                           unsigned int cmd, unsigned long arg)
3404 {
3405         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3406 }
3407
3408 #ifdef CONFIG_COMPAT
3409 static long tun_chr_compat_ioctl(struct file *file,
3410                          unsigned int cmd, unsigned long arg)
3411 {
3412         switch (cmd) {
3413         case TUNSETIFF:
3414         case TUNGETIFF:
3415         case TUNSETTXFILTER:
3416         case TUNGETSNDBUF:
3417         case TUNSETSNDBUF:
3418         case SIOCGIFHWADDR:
3419         case SIOCSIFHWADDR:
3420                 arg = (unsigned long)compat_ptr(arg);
3421                 break;
3422         default:
3423                 arg = (compat_ulong_t)arg;
3424                 break;
3425         }
3426
3427         /*
3428          * compat_ifreq is shorter than ifreq, so we must not access beyond
3429          * the end of that structure. All fields that are used in this
3430          * driver are compatible though, we don't need to convert the
3431          * contents.
3432          */
3433         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3434 }
3435 #endif /* CONFIG_COMPAT */
3436
3437 static int tun_chr_fasync(int fd, struct file *file, int on)
3438 {
3439         struct tun_file *tfile = file->private_data;
3440         int ret;
3441
3442         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3443                 goto out;
3444
3445         if (on) {
3446                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3447                 tfile->flags |= TUN_FASYNC;
3448         } else
3449                 tfile->flags &= ~TUN_FASYNC;
3450         ret = 0;
3451 out:
3452         return ret;
3453 }
3454
3455 static int tun_chr_open(struct inode *inode, struct file * file)
3456 {
3457         struct net *net = current->nsproxy->net_ns;
3458         struct tun_file *tfile;
3459
3460         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3461                                             &tun_proto, 0);
3462         if (!tfile)
3463                 return -ENOMEM;
3464         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3465                 sk_free(&tfile->sk);
3466                 return -ENOMEM;
3467         }
3468
3469         mutex_init(&tfile->napi_mutex);
3470         RCU_INIT_POINTER(tfile->tun, NULL);
3471         tfile->flags = 0;
3472         tfile->ifindex = 0;
3473
3474         init_waitqueue_head(&tfile->socket.wq.wait);
3475
3476         tfile->socket.file = file;
3477         tfile->socket.ops = &tun_socket_ops;
3478
3479         sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3480
3481         tfile->sk.sk_write_space = tun_sock_write_space;
3482         tfile->sk.sk_sndbuf = INT_MAX;
3483
3484         file->private_data = tfile;
3485         INIT_LIST_HEAD(&tfile->next);
3486
3487         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3488
3489         /* tun groks IOCB_NOWAIT just fine, mark it as such */
3490         file->f_mode |= FMODE_NOWAIT;
3491         return 0;
3492 }
3493
3494 static int tun_chr_close(struct inode *inode, struct file *file)
3495 {
3496         struct tun_file *tfile = file->private_data;
3497
3498         tun_detach(tfile, true);
3499
3500         return 0;
3501 }
3502
3503 #ifdef CONFIG_PROC_FS
3504 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3505 {
3506         struct tun_file *tfile = file->private_data;
3507         struct tun_struct *tun;
3508         struct ifreq ifr;
3509
3510         memset(&ifr, 0, sizeof(ifr));
3511
3512         rtnl_lock();
3513         tun = tun_get(tfile);
3514         if (tun)
3515                 tun_get_iff(tun, &ifr);
3516         rtnl_unlock();
3517
3518         if (tun)
3519                 tun_put(tun);
3520
3521         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3522 }
3523 #endif
3524
3525 static const struct file_operations tun_fops = {
3526         .owner  = THIS_MODULE,
3527         .llseek = no_llseek,
3528         .read_iter  = tun_chr_read_iter,
3529         .write_iter = tun_chr_write_iter,
3530         .poll   = tun_chr_poll,
3531         .unlocked_ioctl = tun_chr_ioctl,
3532 #ifdef CONFIG_COMPAT
3533         .compat_ioctl = tun_chr_compat_ioctl,
3534 #endif
3535         .open   = tun_chr_open,
3536         .release = tun_chr_close,
3537         .fasync = tun_chr_fasync,
3538 #ifdef CONFIG_PROC_FS
3539         .show_fdinfo = tun_chr_show_fdinfo,
3540 #endif
3541 };
3542
3543 static struct miscdevice tun_miscdev = {
3544         .minor = TUN_MINOR,
3545         .name = "tun",
3546         .nodename = "net/tun",
3547         .fops = &tun_fops,
3548 };
3549
3550 /* ethtool interface */
3551
3552 static void tun_default_link_ksettings(struct net_device *dev,
3553                                        struct ethtool_link_ksettings *cmd)
3554 {
3555         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3556         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3557         cmd->base.speed         = SPEED_10000;
3558         cmd->base.duplex        = DUPLEX_FULL;
3559         cmd->base.port          = PORT_TP;
3560         cmd->base.phy_address   = 0;
3561         cmd->base.autoneg       = AUTONEG_DISABLE;
3562 }
3563
3564 static int tun_get_link_ksettings(struct net_device *dev,
3565                                   struct ethtool_link_ksettings *cmd)
3566 {
3567         struct tun_struct *tun = netdev_priv(dev);
3568
3569         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3570         return 0;
3571 }
3572
3573 static int tun_set_link_ksettings(struct net_device *dev,
3574                                   const struct ethtool_link_ksettings *cmd)
3575 {
3576         struct tun_struct *tun = netdev_priv(dev);
3577
3578         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3579         return 0;
3580 }
3581
3582 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3583 {
3584         struct tun_struct *tun = netdev_priv(dev);
3585
3586         strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3587         strscpy(info->version, DRV_VERSION, sizeof(info->version));
3588
3589         switch (tun->flags & TUN_TYPE_MASK) {
3590         case IFF_TUN:
3591                 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3592                 break;
3593         case IFF_TAP:
3594                 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3595                 break;
3596         }
3597 }
3598
3599 static u32 tun_get_msglevel(struct net_device *dev)
3600 {
3601         struct tun_struct *tun = netdev_priv(dev);
3602
3603         return tun->msg_enable;
3604 }
3605
3606 static void tun_set_msglevel(struct net_device *dev, u32 value)
3607 {
3608         struct tun_struct *tun = netdev_priv(dev);
3609
3610         tun->msg_enable = value;
3611 }
3612
3613 static int tun_get_coalesce(struct net_device *dev,
3614                             struct ethtool_coalesce *ec,
3615                             struct kernel_ethtool_coalesce *kernel_coal,
3616                             struct netlink_ext_ack *extack)
3617 {
3618         struct tun_struct *tun = netdev_priv(dev);
3619
3620         ec->rx_max_coalesced_frames = tun->rx_batched;
3621
3622         return 0;
3623 }
3624
3625 static int tun_set_coalesce(struct net_device *dev,
3626                             struct ethtool_coalesce *ec,
3627                             struct kernel_ethtool_coalesce *kernel_coal,
3628                             struct netlink_ext_ack *extack)
3629 {
3630         struct tun_struct *tun = netdev_priv(dev);
3631
3632         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3633                 tun->rx_batched = NAPI_POLL_WEIGHT;
3634         else
3635                 tun->rx_batched = ec->rx_max_coalesced_frames;
3636
3637         return 0;
3638 }
3639
3640 static void tun_get_channels(struct net_device *dev,
3641                              struct ethtool_channels *channels)
3642 {
3643         struct tun_struct *tun = netdev_priv(dev);
3644
3645         channels->combined_count = tun->numqueues;
3646         channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3647 }
3648
3649 static const struct ethtool_ops tun_ethtool_ops = {
3650         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3651         .get_drvinfo    = tun_get_drvinfo,
3652         .get_msglevel   = tun_get_msglevel,
3653         .set_msglevel   = tun_set_msglevel,
3654         .get_link       = ethtool_op_get_link,
3655         .get_channels   = tun_get_channels,
3656         .get_ts_info    = ethtool_op_get_ts_info,
3657         .get_coalesce   = tun_get_coalesce,
3658         .set_coalesce   = tun_set_coalesce,
3659         .get_link_ksettings = tun_get_link_ksettings,
3660         .set_link_ksettings = tun_set_link_ksettings,
3661 };
3662
3663 static int tun_queue_resize(struct tun_struct *tun)
3664 {
3665         struct net_device *dev = tun->dev;
3666         struct tun_file *tfile;
3667         struct ptr_ring **rings;
3668         int n = tun->numqueues + tun->numdisabled;
3669         int ret, i;
3670
3671         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3672         if (!rings)
3673                 return -ENOMEM;
3674
3675         for (i = 0; i < tun->numqueues; i++) {
3676                 tfile = rtnl_dereference(tun->tfiles[i]);
3677                 rings[i] = &tfile->tx_ring;
3678         }
3679         list_for_each_entry(tfile, &tun->disabled, next)
3680                 rings[i++] = &tfile->tx_ring;
3681
3682         ret = ptr_ring_resize_multiple(rings, n,
3683                                        dev->tx_queue_len, GFP_KERNEL,
3684                                        tun_ptr_free);
3685
3686         kfree(rings);
3687         return ret;
3688 }
3689
3690 static int tun_device_event(struct notifier_block *unused,
3691                             unsigned long event, void *ptr)
3692 {
3693         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3694         struct tun_struct *tun = netdev_priv(dev);
3695         int i;
3696
3697         if (dev->rtnl_link_ops != &tun_link_ops)
3698                 return NOTIFY_DONE;
3699
3700         switch (event) {
3701         case NETDEV_CHANGE_TX_QUEUE_LEN:
3702                 if (tun_queue_resize(tun))
3703                         return NOTIFY_BAD;
3704                 break;
3705         case NETDEV_UP:
3706                 for (i = 0; i < tun->numqueues; i++) {
3707                         struct tun_file *tfile;
3708
3709                         tfile = rtnl_dereference(tun->tfiles[i]);
3710                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3711                 }
3712                 break;
3713         default:
3714                 break;
3715         }
3716
3717         return NOTIFY_DONE;
3718 }
3719
3720 static struct notifier_block tun_notifier_block __read_mostly = {
3721         .notifier_call  = tun_device_event,
3722 };
3723
3724 static int __init tun_init(void)
3725 {
3726         int ret = 0;
3727
3728         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3729
3730         ret = rtnl_link_register(&tun_link_ops);
3731         if (ret) {
3732                 pr_err("Can't register link_ops\n");
3733                 goto err_linkops;
3734         }
3735
3736         ret = misc_register(&tun_miscdev);
3737         if (ret) {
3738                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3739                 goto err_misc;
3740         }
3741
3742         ret = register_netdevice_notifier(&tun_notifier_block);
3743         if (ret) {
3744                 pr_err("Can't register netdevice notifier\n");
3745                 goto err_notifier;
3746         }
3747
3748         return  0;
3749
3750 err_notifier:
3751         misc_deregister(&tun_miscdev);
3752 err_misc:
3753         rtnl_link_unregister(&tun_link_ops);
3754 err_linkops:
3755         return ret;
3756 }
3757
3758 static void __exit tun_cleanup(void)
3759 {
3760         misc_deregister(&tun_miscdev);
3761         rtnl_link_unregister(&tun_link_ops);
3762         unregister_netdevice_notifier(&tun_notifier_block);
3763 }
3764
3765 /* Get an underlying socket object from tun file.  Returns error unless file is
3766  * attached to a device.  The returned object works like a packet socket, it
3767  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3768  * holding a reference to the file for as long as the socket is in use. */
3769 struct socket *tun_get_socket(struct file *file)
3770 {
3771         struct tun_file *tfile;
3772         if (file->f_op != &tun_fops)
3773                 return ERR_PTR(-EINVAL);
3774         tfile = file->private_data;
3775         if (!tfile)
3776                 return ERR_PTR(-EBADFD);
3777         return &tfile->socket;
3778 }
3779 EXPORT_SYMBOL_GPL(tun_get_socket);
3780
3781 struct ptr_ring *tun_get_tx_ring(struct file *file)
3782 {
3783         struct tun_file *tfile;
3784
3785         if (file->f_op != &tun_fops)
3786                 return ERR_PTR(-EINVAL);
3787         tfile = file->private_data;
3788         if (!tfile)
3789                 return ERR_PTR(-EBADFD);
3790         return &tfile->tx_ring;
3791 }
3792 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3793
3794 module_init(tun_init);
3795 module_exit(tun_cleanup);
3796 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3797 MODULE_AUTHOR(DRV_COPYRIGHT);
3798 MODULE_LICENSE("GPL");
3799 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3800 MODULE_ALIAS("devname:net/tun");