drivers/net/tun.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  TUN - Universal TUN/TAP device driver.
   4  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5  *
   6  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7  */
   8
   9 /*
  10  *  Changes:
  11  *
  12  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13  *    Add TUNSETLINK ioctl to set the link encapsulation
  14  *
  15  *  Mark Smith <markzzzsmith@yahoo.com.au>
  16  *    Use eth_random_addr() for tap MAC address.
  17  *
  18  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19  *    Fixes in packet dropping, queue length setting and queue wakeup.
  20  *    Increased default tx queue length.
  21  *    Added ethtool API.
  22  *    Minor cleanups
  23  *
  24  *  Daniel Podlejski <underley@underley.eu.org>
  25  *    Modifications for 2.3.99-pre5 kernel.
  26  */
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #define DRV_NAME        "tun"
  31 #define DRV_VERSION     "1.6"
  32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  33 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35 #include <linux/module.h>
  36 #include <linux/errno.h>
  37 #include <linux/kernel.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/major.h>
  40 #include <linux/slab.h>
  41 #include <linux/poll.h>
  42 #include <linux/fcntl.h>
  43 #include <linux/init.h>
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/rtnetlink.h>
  50 #include <linux/compat.h>
  51 #include <linux/if.h>
  52 #include <linux/if_arp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_vlan.h>
  56 #include <linux/crc32.h>
  57 #include <linux/nsproxy.h>
  58 #include <linux/virtio_net.h>
  59 #include <linux/rcupdate.h>
  60 #include <net/net_namespace.h>
  61 #include <net/netns/generic.h>
  62 #include <net/rtnetlink.h>
  63 #include <net/sock.h>
  64 #include <net/xdp.h>
  65 #include <net/ip_tunnels.h>
  66 #include <linux/seq_file.h>
  67 #include <linux/uio.h>
  68 #include <linux/skb_array.h>
  69 #include <linux/bpf.h>
  70 #include <linux/bpf_trace.h>
  71 #include <linux/mutex.h>
  72 #include <linux/ieee802154.h>
  73 #include <linux/if_ltalk.h>
  74 #include <uapi/linux/if_fddi.h>
  75 #include <uapi/linux/if_hippi.h>
  76 #include <uapi/linux/if_fc.h>
  77 #include <net/ax25.h>
  78 #include <net/rose.h>
  79 #include <net/6lowpan.h>
  80
  81 #include <linux/uaccess.h>
  82 #include <linux/proc_fs.h>
  83
  84 static void tun_default_link_ksettings(struct net_device *dev,
  85                                        struct ethtool_link_ksettings *cmd);
  86
  87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  88
  89 /* TUN device flags */
  90
  91 /* IFF_ATTACH_QUEUE is never stored in device flags,
  92  * overload it to mean fasync when stored there.
  93  */
  94 #define TUN_FASYNC      IFF_ATTACH_QUEUE
  95 /* High bits in flags field are unused. */
  96 #define TUN_VNET_LE     0x80000000
  97 #define TUN_VNET_BE     0x40000000
  98
  99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 100                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 101
 102 #define GOODCOPY_LEN 128
 103
 104 #define FLT_EXACT_COUNT 8
 105 struct tap_filter {
 106         unsigned int    count;    /* Number of addrs. Zero means disabled */
 107         u32             mask[2];  /* Mask of the hashed addrs */
 108         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 109 };
 110
 111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 112  * to max number of VCPUs in guest. */
 113 #define MAX_TAP_QUEUES 256
 114 #define MAX_TAP_FLOWS  4096
 115
 116 #define TUN_FLOW_EXPIRE (3 * HZ)
 117
 118 /* A tun_file connects an open character device to a tuntap netdevice. It
 119  * also contains all socket related structures (except sock_fprog and tap_filter)
 120  * to serve as one transmit queue for tuntap device. The sock_fprog and
 121  * tap_filter were kept in tun_struct since they were used for filtering for the
 122  * netdevice not for a specific queue (at least I didn't see the requirement for
 123  * this).
 124  *
 125  * RCU usage:
 126  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 127  * other can only be read while rcu_read_lock or rtnl_lock is held.
 128  */
 129 struct tun_file {
 130         struct sock sk;
 131         struct socket socket;
 132         struct tun_struct __rcu *tun;
 133         struct fasync_struct *fasync;
 134         /* only used for fasnyc */
 135         unsigned int flags;
 136         union {
 137                 u16 queue_index;
 138                 unsigned int ifindex;
 139         };
 140         struct napi_struct napi;
 141         bool napi_enabled;
 142         bool napi_frags_enabled;
 143         struct mutex napi_mutex;        /* Protects access to the above napi */
 144         struct list_head next;
 145         struct tun_struct *detached;
 146         struct ptr_ring tx_ring;
 147         struct xdp_rxq_info xdp_rxq;
 148 };
 149
 150 struct tun_page {
 151         struct page *page;
 152         int count;
 153 };
 154
 155 struct tun_flow_entry {
 156         struct hlist_node hash_link;
 157         struct rcu_head rcu;
 158         struct tun_struct *tun;
 159
 160         u32 rxhash;
 161         u32 rps_rxhash;
 162         int queue_index;
 163         unsigned long updated ____cacheline_aligned_in_smp;
 164 };
 165
 166 #define TUN_NUM_FLOW_ENTRIES 1024
 167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 168
 169 struct tun_prog {
 170         struct rcu_head rcu;
 171         struct bpf_prog *prog;
 172 };
 173
 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 175  * device, socket filter, sndbuf and vnet header size were restore when the
 176  * file were attached to a persist device.
 177  */
 178 struct tun_struct {
 179         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 180         unsigned int            numqueues;
 181         unsigned int            flags;
 182         kuid_t                  owner;
 183         kgid_t                  group;
 184
 185         struct net_device       *dev;
 186         netdev_features_t       set_features;
 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 188                           NETIF_F_TSO6)
 189
 190         int                     align;
 191         int                     vnet_hdr_sz;
 192         int                     sndbuf;
 193         struct tap_filter       txflt;
 194         struct sock_fprog       fprog;
 195         /* protected by rtnl lock */
 196         bool                    filter_attached;
 197         u32                     msg_enable;
 198         spinlock_t lock;
 199         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 200         struct timer_list flow_gc_timer;
 201         unsigned long ageing_time;
 202         unsigned int numdisabled;
 203         struct list_head disabled;
 204         void *security;
 205         u32 flow_count;
 206         u32 rx_batched;
 207         atomic_long_t rx_frame_errors;
 208         struct bpf_prog __rcu *xdp_prog;
 209         struct tun_prog __rcu *steering_prog;
 210         struct tun_prog __rcu *filter_prog;
 211         struct ethtool_link_ksettings link_ksettings;
 212         /* init args */
 213         struct file *file;
 214         struct ifreq *ifr;
 215 };
 216
 217 struct veth {
 218         __be16 h_vlan_proto;
 219         __be16 h_vlan_TCI;
 220 };
 221
 222 static void tun_flow_init(struct tun_struct *tun);
 223 static void tun_flow_uninit(struct tun_struct *tun);
 224
 225 static int tun_napi_receive(struct napi_struct *napi, int budget)
 226 {
 227         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 228         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 229         struct sk_buff_head process_queue;
 230         struct sk_buff *skb;
 231         int received = 0;
 232
 233         __skb_queue_head_init(&process_queue);
 234
 235         spin_lock(&queue->lock);
 236         skb_queue_splice_tail_init(queue, &process_queue);
 237         spin_unlock(&queue->lock);
 238
 239         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 240                 napi_gro_receive(napi, skb);
 241                 ++received;
 242         }
 243
 244         if (!skb_queue_empty(&process_queue)) {
 245                 spin_lock(&queue->lock);
 246                 skb_queue_splice(&process_queue, queue);
 247                 spin_unlock(&queue->lock);
 248         }
 249
 250         return received;
 251 }
 252
 253 static int tun_napi_poll(struct napi_struct *napi, int budget)
 254 {
 255         unsigned int received;
 256
 257         received = tun_napi_receive(napi, budget);
 258
 259         if (received < budget)
 260                 napi_complete_done(napi, received);
 261
 262         return received;
 263 }
 264
 265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 266                           bool napi_en, bool napi_frags)
 267 {
 268         tfile->napi_enabled = napi_en;
 269         tfile->napi_frags_enabled = napi_en && napi_frags;
 270         if (napi_en) {
 271                 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 272                 napi_enable(&tfile->napi);
 273         }
 274 }
 275
 276 static void tun_napi_enable(struct tun_file *tfile)
 277 {
 278         if (tfile->napi_enabled)
 279                 napi_enable(&tfile->napi);
 280 }
 281
 282 static void tun_napi_disable(struct tun_file *tfile)
 283 {
 284         if (tfile->napi_enabled)
 285                 napi_disable(&tfile->napi);
 286 }
 287
 288 static void tun_napi_del(struct tun_file *tfile)
 289 {
 290         if (tfile->napi_enabled)
 291                 netif_napi_del(&tfile->napi);
 292 }
 293
 294 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 295 {
 296         return tfile->napi_frags_enabled;
 297 }
 298
 299 #ifdef CONFIG_TUN_VNET_CROSS_LE
 300 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 301 {
 302         return tun->flags & TUN_VNET_BE ? false :
 303                 virtio_legacy_is_little_endian();
 304 }
 305
 306 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 307 {
 308         int be = !!(tun->flags & TUN_VNET_BE);
 309
 310         if (put_user(be, argp))
 311                 return -EFAULT;
 312
 313         return 0;
 314 }
 315
 316 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 317 {
 318         int be;
 319
 320         if (get_user(be, argp))
 321                 return -EFAULT;
 322
 323         if (be)
 324                 tun->flags |= TUN_VNET_BE;
 325         else
 326                 tun->flags &= ~TUN_VNET_BE;
 327
 328         return 0;
 329 }
 330 #else
 331 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 332 {
 333         return virtio_legacy_is_little_endian();
 334 }
 335
 336 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 337 {
 338         return -EINVAL;
 339 }
 340
 341 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 342 {
 343         return -EINVAL;
 344 }
 345 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 346
 347 static inline bool tun_is_little_endian(struct tun_struct *tun)
 348 {
 349         return tun->flags & TUN_VNET_LE ||
 350                 tun_legacy_is_little_endian(tun);
 351 }
 352
 353 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 354 {
 355         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 356 }
 357
 358 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 359 {
 360         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 361 }
 362
 363 static inline u32 tun_hashfn(u32 rxhash)
 364 {
 365         return rxhash & TUN_MASK_FLOW_ENTRIES;
 366 }
 367
 368 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 369 {
 370         struct tun_flow_entry *e;
 371
 372         hlist_for_each_entry_rcu(e, head, hash_link) {
 373                 if (e->rxhash == rxhash)
 374                         return e;
 375         }
 376         return NULL;
 377 }
 378
 379 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 380                                               struct hlist_head *head,
 381                                               u32 rxhash, u16 queue_index)
 382 {
 383         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 384
 385         if (e) {
 386                 netif_info(tun, tx_queued, tun->dev,
 387                            "create flow: hash %u index %u\n",
 388                            rxhash, queue_index);
 389                 e->updated = jiffies;
 390                 e->rxhash = rxhash;
 391                 e->rps_rxhash = 0;
 392                 e->queue_index = queue_index;
 393                 e->tun = tun;
 394                 hlist_add_head_rcu(&e->hash_link, head);
 395                 ++tun->flow_count;
 396         }
 397         return e;
 398 }
 399
 400 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 401 {
 402         netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 403                    e->rxhash, e->queue_index);
 404         hlist_del_rcu(&e->hash_link);
 405         kfree_rcu(e, rcu);
 406         --tun->flow_count;
 407 }
 408
 409 static void tun_flow_flush(struct tun_struct *tun)
 410 {
 411         int i;
 412
 413         spin_lock_bh(&tun->lock);
 414         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 415                 struct tun_flow_entry *e;
 416                 struct hlist_node *n;
 417
 418                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 419                         tun_flow_delete(tun, e);
 420         }
 421         spin_unlock_bh(&tun->lock);
 422 }
 423
 424 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 425 {
 426         int i;
 427
 428         spin_lock_bh(&tun->lock);
 429         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 430                 struct tun_flow_entry *e;
 431                 struct hlist_node *n;
 432
 433                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 434                         if (e->queue_index == queue_index)
 435                                 tun_flow_delete(tun, e);
 436                 }
 437         }
 438         spin_unlock_bh(&tun->lock);
 439 }
 440
 441 static void tun_flow_cleanup(struct timer_list *t)
 442 {
 443         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 444         unsigned long delay = tun->ageing_time;
 445         unsigned long next_timer = jiffies + delay;
 446         unsigned long count = 0;
 447         int i;
 448
 449         spin_lock(&tun->lock);
 450         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 451                 struct tun_flow_entry *e;
 452                 struct hlist_node *n;
 453
 454                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 455                         unsigned long this_timer;
 456
 457                         this_timer = e->updated + delay;
 458                         if (time_before_eq(this_timer, jiffies)) {
 459                                 tun_flow_delete(tun, e);
 460                                 continue;
 461                         }
 462                         count++;
 463                         if (time_before(this_timer, next_timer))
 464                                 next_timer = this_timer;
 465                 }
 466         }
 467
 468         if (count)
 469                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 470         spin_unlock(&tun->lock);
 471 }
 472
 473 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 474                             struct tun_file *tfile)
 475 {
 476         struct hlist_head *head;
 477         struct tun_flow_entry *e;
 478         unsigned long delay = tun->ageing_time;
 479         u16 queue_index = tfile->queue_index;
 480
 481         head = &tun->flows[tun_hashfn(rxhash)];
 482
 483         rcu_read_lock();
 484
 485         e = tun_flow_find(head, rxhash);
 486         if (likely(e)) {
 487                 /* TODO: keep queueing to old queue until it's empty? */
 488                 if (READ_ONCE(e->queue_index) != queue_index)
 489                         WRITE_ONCE(e->queue_index, queue_index);
 490                 if (e->updated != jiffies)
 491                         e->updated = jiffies;
 492                 sock_rps_record_flow_hash(e->rps_rxhash);
 493         } else {
 494                 spin_lock_bh(&tun->lock);
 495                 if (!tun_flow_find(head, rxhash) &&
 496                     tun->flow_count < MAX_TAP_FLOWS)
 497                         tun_flow_create(tun, head, rxhash, queue_index);
 498
 499                 if (!timer_pending(&tun->flow_gc_timer))
 500                         mod_timer(&tun->flow_gc_timer,
 501                                   round_jiffies_up(jiffies + delay));
 502                 spin_unlock_bh(&tun->lock);
 503         }
 504
 505         rcu_read_unlock();
 506 }
 507
 508 /* Save the hash received in the stack receive path and update the
 509  * flow_hash table accordingly.
 510  */
 511 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 512 {
 513         if (unlikely(e->rps_rxhash != hash))
 514                 e->rps_rxhash = hash;
 515 }
 516
 517 /* We try to identify a flow through its rxhash. The reason that
 518  * we do not check rxq no. is because some cards(e.g 82599), chooses
 519  * the rxq based on the txq where the last packet of the flow comes. As
 520  * the userspace application move between processors, we may get a
 521  * different rxq no. here.
 522  */
 523 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 524 {
 525         struct tun_flow_entry *e;
 526         u32 txq = 0;
 527         u32 numqueues = 0;
 528
 529         numqueues = READ_ONCE(tun->numqueues);
 530
 531         txq = __skb_get_hash_symmetric(skb);
 532         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 533         if (e) {
 534                 tun_flow_save_rps_rxhash(e, txq);
 535                 txq = e->queue_index;
 536         } else {
 537                 /* use multiply and shift instead of expensive divide */
 538                 txq = ((u64)txq * numqueues) >> 32;
 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                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 657                                    NULL);
 658
 659                 --tun->numqueues;
 660                 if (clean) {
 661                         RCU_INIT_POINTER(tfile->tun, NULL);
 662                         sock_put(&tfile->sk);
 663                 } else {
 664                         tun_disable_queue(tun, tfile);
 665                         tun_napi_disable(tfile);
 666                 }
 667
 668                 synchronize_net();
 669                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 670                 /* Drop read queue */
 671                 tun_queue_purge(tfile);
 672                 tun_set_real_num_queues(tun);
 673         } else if (tfile->detached && clean) {
 674                 tun = tun_enable_queue(tfile);
 675                 sock_put(&tfile->sk);
 676         }
 677
 678         if (clean) {
 679                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 680                         netif_carrier_off(tun->dev);
 681
 682                         if (!(tun->flags & IFF_PERSIST) &&
 683                             tun->dev->reg_state == NETREG_REGISTERED)
 684                                 unregister_netdevice(tun->dev);
 685                 }
 686                 if (tun)
 687                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 688                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 689                 sock_put(&tfile->sk);
 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
 707 static void tun_detach_all(struct net_device *dev)
 708 {
 709         struct tun_struct *tun = netdev_priv(dev);
 710         struct tun_file *tfile, *tmp;
 711         int i, n = tun->numqueues;
 712
 713         for (i = 0; i < n; i++) {
 714                 tfile = rtnl_dereference(tun->tfiles[i]);
 715                 BUG_ON(!tfile);
 716                 tun_napi_disable(tfile);
 717                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 718                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 719                 RCU_INIT_POINTER(tfile->tun, NULL);
 720                 --tun->numqueues;
 721         }
 722         list_for_each_entry(tfile, &tun->disabled, next) {
 723                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 724                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 725                 RCU_INIT_POINTER(tfile->tun, NULL);
 726         }
 727         BUG_ON(tun->numqueues != 0);
 728
 729         synchronize_net();
 730         for (i = 0; i < n; i++) {
 731                 tfile = rtnl_dereference(tun->tfiles[i]);
 732                 tun_napi_del(tfile);
 733                 /* Drop read queue */
 734                 tun_queue_purge(tfile);
 735                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 736                 sock_put(&tfile->sk);
 737         }
 738         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 739                 tun_napi_del(tfile);
 740                 tun_enable_queue(tfile);
 741                 tun_queue_purge(tfile);
 742                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 743                 sock_put(&tfile->sk);
 744         }
 745         BUG_ON(tun->numdisabled != 0);
 746
 747         if (tun->flags & IFF_PERSIST)
 748                 module_put(THIS_MODULE);
 749 }
 750
 751 static int tun_attach(struct tun_struct *tun, struct file *file,
 752                       bool skip_filter, bool napi, bool napi_frags,
 753                       bool publish_tun)
 754 {
 755         struct tun_file *tfile = file->private_data;
 756         struct net_device *dev = tun->dev;
 757         int err;
 758
 759         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 760         if (err < 0)
 761                 goto out;
 762
 763         err = -EINVAL;
 764         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 765                 goto out;
 766
 767         err = -EBUSY;
 768         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 769                 goto out;
 770
 771         err = -E2BIG;
 772         if (!tfile->detached &&
 773             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 774                 goto out;
 775
 776         err = 0;
 777
 778         /* Re-attach the filter to persist device */
 779         if (!skip_filter && (tun->filter_attached == true)) {
 780                 lock_sock(tfile->socket.sk);
 781                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 782                 release_sock(tfile->socket.sk);
 783                 if (!err)
 784                         goto out;
 785         }
 786
 787         if (!tfile->detached &&
 788             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 789                             GFP_KERNEL, tun_ptr_free)) {
 790                 err = -ENOMEM;
 791                 goto out;
 792         }
 793
 794         tfile->queue_index = tun->numqueues;
 795         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 796
 797         if (tfile->detached) {
 798                 /* Re-attach detached tfile, updating XDP queue_index */
 799                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 800
 801                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 802                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 803         } else {
 804                 /* Setup XDP RX-queue info, for new tfile getting attached */
 805                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 806                                        tun->dev, tfile->queue_index, 0);
 807                 if (err < 0)
 808                         goto out;
 809                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 810                                                  MEM_TYPE_PAGE_SHARED, NULL);
 811                 if (err < 0) {
 812                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 813                         goto out;
 814                 }
 815                 err = 0;
 816         }
 817
 818         if (tfile->detached) {
 819                 tun_enable_queue(tfile);
 820                 tun_napi_enable(tfile);
 821         } else {
 822                 sock_hold(&tfile->sk);
 823                 tun_napi_init(tun, tfile, napi, napi_frags);
 824         }
 825
 826         if (rtnl_dereference(tun->xdp_prog))
 827                 sock_set_flag(&tfile->sk, SOCK_XDP);
 828
 829         /* device is allowed to go away first, so no need to hold extra
 830          * refcnt.
 831          */
 832
 833         /* Publish tfile->tun and tun->tfiles only after we've fully
 834          * initialized tfile; otherwise we risk using half-initialized
 835          * object.
 836          */
 837         if (publish_tun)
 838                 rcu_assign_pointer(tfile->tun, tun);
 839         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 840         tun->numqueues++;
 841         tun_set_real_num_queues(tun);
 842 out:
 843         return err;
 844 }
 845
 846 static struct tun_struct *tun_get(struct tun_file *tfile)
 847 {
 848         struct tun_struct *tun;
 849
 850         rcu_read_lock();
 851         tun = rcu_dereference(tfile->tun);
 852         if (tun)
 853                 dev_hold(tun->dev);
 854         rcu_read_unlock();
 855
 856         return tun;
 857 }
 858
 859 static void tun_put(struct tun_struct *tun)
 860 {
 861         dev_put(tun->dev);
 862 }
 863
 864 /* TAP filtering */
 865 static void addr_hash_set(u32 *mask, const u8 *addr)
 866 {
 867         int n = ether_crc(ETH_ALEN, addr) >> 26;
 868         mask[n >> 5] |= (1 << (n & 31));
 869 }
 870
 871 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 872 {
 873         int n = ether_crc(ETH_ALEN, addr) >> 26;
 874         return mask[n >> 5] & (1 << (n & 31));
 875 }
 876
 877 static int update_filter(struct tap_filter *filter, void __user *arg)
 878 {
 879         struct { u8 u[ETH_ALEN]; } *addr;
 880         struct tun_filter uf;
 881         int err, alen, n, nexact;
 882
 883         if (copy_from_user(&uf, arg, sizeof(uf)))
 884                 return -EFAULT;
 885
 886         if (!uf.count) {
 887                 /* Disabled */
 888                 filter->count = 0;
 889                 return 0;
 890         }
 891
 892         alen = ETH_ALEN * uf.count;
 893         addr = memdup_user(arg + sizeof(uf), alen);
 894         if (IS_ERR(addr))
 895                 return PTR_ERR(addr);
 896
 897         /* The filter is updated without holding any locks. Which is
 898          * perfectly safe. We disable it first and in the worst
 899          * case we'll accept a few undesired packets. */
 900         filter->count = 0;
 901         wmb();
 902
 903         /* Use first set of addresses as an exact filter */
 904         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 905                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 906
 907         nexact = n;
 908
 909         /* Remaining multicast addresses are hashed,
 910          * unicast will leave the filter disabled. */
 911         memset(filter->mask, 0, sizeof(filter->mask));
 912         for (; n < uf.count; n++) {
 913                 if (!is_multicast_ether_addr(addr[n].u)) {
 914                         err = 0; /* no filter */
 915                         goto free_addr;
 916                 }
 917                 addr_hash_set(filter->mask, addr[n].u);
 918         }
 919
 920         /* For ALLMULTI just set the mask to all ones.
 921          * This overrides the mask populated above. */
 922         if ((uf.flags & TUN_FLT_ALLMULTI))
 923                 memset(filter->mask, ~0, sizeof(filter->mask));
 924
 925         /* Now enable the filter */
 926         wmb();
 927         filter->count = nexact;
 928
 929         /* Return the number of exact filters */
 930         err = nexact;
 931 free_addr:
 932         kfree(addr);
 933         return err;
 934 }
 935
 936 /* Returns: 0 - drop, !=0 - accept */
 937 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 938 {
 939         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 940          * at this point. */
 941         struct ethhdr *eh = (struct ethhdr *) skb->data;
 942         int i;
 943
 944         /* Exact match */
 945         for (i = 0; i < filter->count; i++)
 946                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 947                         return 1;
 948
 949         /* Inexact match (multicast only) */
 950         if (is_multicast_ether_addr(eh->h_dest))
 951                 return addr_hash_test(filter->mask, eh->h_dest);
 952
 953         return 0;
 954 }
 955
 956 /*
 957  * Checks whether the packet is accepted or not.
 958  * Returns: 0 - drop, !=0 - accept
 959  */
 960 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 961 {
 962         if (!filter->count)
 963                 return 1;
 964
 965         return run_filter(filter, skb);
 966 }
 967
 968 /* Network device part of the driver */
 969
 970 static const struct ethtool_ops tun_ethtool_ops;
 971
 972 static int tun_net_init(struct net_device *dev)
 973 {
 974         struct tun_struct *tun = netdev_priv(dev);
 975         struct ifreq *ifr = tun->ifr;
 976         int err;
 977
 978         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
 979         if (!dev->tstats)
 980                 return -ENOMEM;
 981
 982         spin_lock_init(&tun->lock);
 983
 984         err = security_tun_dev_alloc_security(&tun->security);
 985         if (err < 0) {
 986                 free_percpu(dev->tstats);
 987                 return err;
 988         }
 989
 990         tun_flow_init(tun);
 991
 992         dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 993                            TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 994                            NETIF_F_HW_VLAN_STAG_TX;
 995         dev->features = dev->hw_features | NETIF_F_LLTX;
 996         dev->vlan_features = dev->features &
 997                              ~(NETIF_F_HW_VLAN_CTAG_TX |
 998                                NETIF_F_HW_VLAN_STAG_TX);
 999
1000         tun->flags = (tun->flags & ~TUN_FEATURES) |
1001                       (ifr->ifr_flags & TUN_FEATURES);
1002
1003         INIT_LIST_HEAD(&tun->disabled);
1004         err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1005                          ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1006         if (err < 0) {
1007                 tun_flow_uninit(tun);
1008                 security_tun_dev_free_security(tun->security);
1009                 free_percpu(dev->tstats);
1010                 return err;
1011         }
1012         return 0;
1013 }
1014
1015 /* Net device detach from fd. */
1016 static void tun_net_uninit(struct net_device *dev)
1017 {
1018         tun_detach_all(dev);
1019 }
1020
1021 /* Net device open. */
1022 static int tun_net_open(struct net_device *dev)
1023 {
1024         netif_tx_start_all_queues(dev);
1025
1026         return 0;
1027 }
1028
1029 /* Net device close. */
1030 static int tun_net_close(struct net_device *dev)
1031 {
1032         netif_tx_stop_all_queues(dev);
1033         return 0;
1034 }
1035
1036 /* Net device start xmit */
1037 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1038 {
1039 #ifdef CONFIG_RPS
1040         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1041                 /* Select queue was not called for the skbuff, so we extract the
1042                  * RPS hash and save it into the flow_table here.
1043                  */
1044                 struct tun_flow_entry *e;
1045                 __u32 rxhash;
1046
1047                 rxhash = __skb_get_hash_symmetric(skb);
1048                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1049                 if (e)
1050                         tun_flow_save_rps_rxhash(e, rxhash);
1051         }
1052 #endif
1053 }
1054
1055 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1056                                     struct sk_buff *skb,
1057                                     int len)
1058 {
1059         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1060
1061         if (prog)
1062                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1063
1064         return len;
1065 }
1066
1067 /* Net device start xmit */
1068 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1069 {
1070         struct tun_struct *tun = netdev_priv(dev);
1071         enum skb_drop_reason drop_reason;
1072         int txq = skb->queue_mapping;
1073         struct netdev_queue *queue;
1074         struct tun_file *tfile;
1075         int len = skb->len;
1076
1077         rcu_read_lock();
1078         tfile = rcu_dereference(tun->tfiles[txq]);
1079
1080         /* Drop packet if interface is not attached */
1081         if (!tfile) {
1082                 drop_reason = SKB_DROP_REASON_DEV_READY;
1083                 goto drop;
1084         }
1085
1086         if (!rcu_dereference(tun->steering_prog))
1087                 tun_automq_xmit(tun, skb);
1088
1089         netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1090
1091         /* Drop if the filter does not like it.
1092          * This is a noop if the filter is disabled.
1093          * Filter can be enabled only for the TAP devices. */
1094         if (!check_filter(&tun->txflt, skb)) {
1095                 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1096                 goto drop;
1097         }
1098
1099         if (tfile->socket.sk->sk_filter &&
1100             sk_filter(tfile->socket.sk, skb)) {
1101                 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1102                 goto drop;
1103         }
1104
1105         len = run_ebpf_filter(tun, skb, len);
1106         if (len == 0) {
1107                 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1108                 goto drop;
1109         }
1110
1111         if (pskb_trim(skb, len)) {
1112                 drop_reason = SKB_DROP_REASON_NOMEM;
1113                 goto drop;
1114         }
1115
1116         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1117                 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1118                 goto drop;
1119         }
1120
1121         skb_tx_timestamp(skb);
1122
1123         /* Orphan the skb - required as we might hang on to it
1124          * for indefinite time.
1125          */
1126         skb_orphan(skb);
1127
1128         nf_reset_ct(skb);
1129
1130         if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1131                 drop_reason = SKB_DROP_REASON_FULL_RING;
1132                 goto drop;
1133         }
1134
1135         /* NETIF_F_LLTX requires to do our own update of trans_start */
1136         queue = netdev_get_tx_queue(dev, txq);
1137         txq_trans_cond_update(queue);
1138
1139         /* Notify and wake up reader process */
1140         if (tfile->flags & TUN_FASYNC)
1141                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1142         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1143
1144         rcu_read_unlock();
1145         return NETDEV_TX_OK;
1146
1147 drop:
1148         dev_core_stats_tx_dropped_inc(dev);
1149         skb_tx_error(skb);
1150         kfree_skb_reason(skb, drop_reason);
1151         rcu_read_unlock();
1152         return NET_XMIT_DROP;
1153 }
1154
1155 static void tun_net_mclist(struct net_device *dev)
1156 {
1157         /*
1158          * This callback is supposed to deal with mc filter in
1159          * _rx_ path and has nothing to do with the _tx_ path.
1160          * In rx path we always accept everything userspace gives us.
1161          */
1162 }
1163
1164 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1165         netdev_features_t features)
1166 {
1167         struct tun_struct *tun = netdev_priv(dev);
1168
1169         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1170 }
1171
1172 static void tun_set_headroom(struct net_device *dev, int new_hr)
1173 {
1174         struct tun_struct *tun = netdev_priv(dev);
1175
1176         if (new_hr < NET_SKB_PAD)
1177                 new_hr = NET_SKB_PAD;
1178
1179         tun->align = new_hr;
1180 }
1181
1182 static void
1183 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1184 {
1185         struct tun_struct *tun = netdev_priv(dev);
1186
1187         dev_get_tstats64(dev, stats);
1188
1189         stats->rx_frame_errors +=
1190                 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1191 }
1192
1193 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1194                        struct netlink_ext_ack *extack)
1195 {
1196         struct tun_struct *tun = netdev_priv(dev);
1197         struct tun_file *tfile;
1198         struct bpf_prog *old_prog;
1199         int i;
1200
1201         old_prog = rtnl_dereference(tun->xdp_prog);
1202         rcu_assign_pointer(tun->xdp_prog, prog);
1203         if (old_prog)
1204                 bpf_prog_put(old_prog);
1205
1206         for (i = 0; i < tun->numqueues; i++) {
1207                 tfile = rtnl_dereference(tun->tfiles[i]);
1208                 if (prog)
1209                         sock_set_flag(&tfile->sk, SOCK_XDP);
1210                 else
1211                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1212         }
1213         list_for_each_entry(tfile, &tun->disabled, next) {
1214                 if (prog)
1215                         sock_set_flag(&tfile->sk, SOCK_XDP);
1216                 else
1217                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1218         }
1219
1220         return 0;
1221 }
1222
1223 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1224 {
1225         switch (xdp->command) {
1226         case XDP_SETUP_PROG:
1227                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1228         default:
1229                 return -EINVAL;
1230         }
1231 }
1232
1233 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1234 {
1235         if (new_carrier) {
1236                 struct tun_struct *tun = netdev_priv(dev);
1237
1238                 if (!tun->numqueues)
1239                         return -EPERM;
1240
1241                 netif_carrier_on(dev);
1242         } else {
1243                 netif_carrier_off(dev);
1244         }
1245         return 0;
1246 }
1247
1248 static const struct net_device_ops tun_netdev_ops = {
1249         .ndo_init               = tun_net_init,
1250         .ndo_uninit             = tun_net_uninit,
1251         .ndo_open               = tun_net_open,
1252         .ndo_stop               = tun_net_close,
1253         .ndo_start_xmit         = tun_net_xmit,
1254         .ndo_fix_features       = tun_net_fix_features,
1255         .ndo_select_queue       = tun_select_queue,
1256         .ndo_set_rx_headroom    = tun_set_headroom,
1257         .ndo_get_stats64        = tun_net_get_stats64,
1258         .ndo_change_carrier     = tun_net_change_carrier,
1259 };
1260
1261 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1262 {
1263         /* Notify and wake up reader process */
1264         if (tfile->flags & TUN_FASYNC)
1265                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1266         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1267 }
1268
1269 static int tun_xdp_xmit(struct net_device *dev, int n,
1270                         struct xdp_frame **frames, u32 flags)
1271 {
1272         struct tun_struct *tun = netdev_priv(dev);
1273         struct tun_file *tfile;
1274         u32 numqueues;
1275         int nxmit = 0;
1276         int i;
1277
1278         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1279                 return -EINVAL;
1280
1281         rcu_read_lock();
1282
1283 resample:
1284         numqueues = READ_ONCE(tun->numqueues);
1285         if (!numqueues) {
1286                 rcu_read_unlock();
1287                 return -ENXIO; /* Caller will free/return all frames */
1288         }
1289
1290         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1291                                             numqueues]);
1292         if (unlikely(!tfile))
1293                 goto resample;
1294
1295         spin_lock(&tfile->tx_ring.producer_lock);
1296         for (i = 0; i < n; i++) {
1297                 struct xdp_frame *xdp = frames[i];
1298                 /* Encode the XDP flag into lowest bit for consumer to differ
1299                  * XDP buffer from sk_buff.
1300                  */
1301                 void *frame = tun_xdp_to_ptr(xdp);
1302
1303                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1304                         dev_core_stats_tx_dropped_inc(dev);
1305                         break;
1306                 }
1307                 nxmit++;
1308         }
1309         spin_unlock(&tfile->tx_ring.producer_lock);
1310
1311         if (flags & XDP_XMIT_FLUSH)
1312                 __tun_xdp_flush_tfile(tfile);
1313
1314         rcu_read_unlock();
1315         return nxmit;
1316 }
1317
1318 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1319 {
1320         struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1321         int nxmit;
1322
1323         if (unlikely(!frame))
1324                 return -EOVERFLOW;
1325
1326         nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1327         if (!nxmit)
1328                 xdp_return_frame_rx_napi(frame);
1329         return nxmit;
1330 }
1331
1332 static const struct net_device_ops tap_netdev_ops = {
1333         .ndo_init               = tun_net_init,
1334         .ndo_uninit             = tun_net_uninit,
1335         .ndo_open               = tun_net_open,
1336         .ndo_stop               = tun_net_close,
1337         .ndo_start_xmit         = tun_net_xmit,
1338         .ndo_fix_features       = tun_net_fix_features,
1339         .ndo_set_rx_mode        = tun_net_mclist,
1340         .ndo_set_mac_address    = eth_mac_addr,
1341         .ndo_validate_addr      = eth_validate_addr,
1342         .ndo_select_queue       = tun_select_queue,
1343         .ndo_features_check     = passthru_features_check,
1344         .ndo_set_rx_headroom    = tun_set_headroom,
1345         .ndo_get_stats64        = dev_get_tstats64,
1346         .ndo_bpf                = tun_xdp,
1347         .ndo_xdp_xmit           = tun_xdp_xmit,
1348         .ndo_change_carrier     = tun_net_change_carrier,
1349 };
1350
1351 static void tun_flow_init(struct tun_struct *tun)
1352 {
1353         int i;
1354
1355         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1356                 INIT_HLIST_HEAD(&tun->flows[i]);
1357
1358         tun->ageing_time = TUN_FLOW_EXPIRE;
1359         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1360         mod_timer(&tun->flow_gc_timer,
1361                   round_jiffies_up(jiffies + tun->ageing_time));
1362 }
1363
1364 static void tun_flow_uninit(struct tun_struct *tun)
1365 {
1366         del_timer_sync(&tun->flow_gc_timer);
1367         tun_flow_flush(tun);
1368 }
1369
1370 #define MIN_MTU 68
1371 #define MAX_MTU 65535
1372
1373 /* Initialize net device. */
1374 static void tun_net_initialize(struct net_device *dev)
1375 {
1376         struct tun_struct *tun = netdev_priv(dev);
1377
1378         switch (tun->flags & TUN_TYPE_MASK) {
1379         case IFF_TUN:
1380                 dev->netdev_ops = &tun_netdev_ops;
1381                 dev->header_ops = &ip_tunnel_header_ops;
1382
1383                 /* Point-to-Point TUN Device */
1384                 dev->hard_header_len = 0;
1385                 dev->addr_len = 0;
1386                 dev->mtu = 1500;
1387
1388                 /* Zero header length */
1389                 dev->type = ARPHRD_NONE;
1390                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1391                 break;
1392
1393         case IFF_TAP:
1394                 dev->netdev_ops = &tap_netdev_ops;
1395                 /* Ethernet TAP Device */
1396                 ether_setup(dev);
1397                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1398                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1399
1400                 eth_hw_addr_random(dev);
1401
1402                 break;
1403         }
1404
1405         dev->min_mtu = MIN_MTU;
1406         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1407 }
1408
1409 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1410 {
1411         struct sock *sk = tfile->socket.sk;
1412
1413         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1414 }
1415
1416 /* Character device part */
1417
1418 /* Poll */
1419 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1420 {
1421         struct tun_file *tfile = file->private_data;
1422         struct tun_struct *tun = tun_get(tfile);
1423         struct sock *sk;
1424         __poll_t mask = 0;
1425
1426         if (!tun)
1427                 return EPOLLERR;
1428
1429         sk = tfile->socket.sk;
1430
1431         poll_wait(file, sk_sleep(sk), wait);
1432
1433         if (!ptr_ring_empty(&tfile->tx_ring))
1434                 mask |= EPOLLIN | EPOLLRDNORM;
1435
1436         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1437          * guarantee EPOLLOUT to be raised by either here or
1438          * tun_sock_write_space(). Then process could get notification
1439          * after it writes to a down device and meets -EIO.
1440          */
1441         if (tun_sock_writeable(tun, tfile) ||
1442             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1443              tun_sock_writeable(tun, tfile)))
1444                 mask |= EPOLLOUT | EPOLLWRNORM;
1445
1446         if (tun->dev->reg_state != NETREG_REGISTERED)
1447                 mask = EPOLLERR;
1448
1449         tun_put(tun);
1450         return mask;
1451 }
1452
1453 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1454                                             size_t len,
1455                                             const struct iov_iter *it)
1456 {
1457         struct sk_buff *skb;
1458         size_t linear;
1459         int err;
1460         int i;
1461
1462         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1463                 return ERR_PTR(-EMSGSIZE);
1464
1465         local_bh_disable();
1466         skb = napi_get_frags(&tfile->napi);
1467         local_bh_enable();
1468         if (!skb)
1469                 return ERR_PTR(-ENOMEM);
1470
1471         linear = iov_iter_single_seg_count(it);
1472         err = __skb_grow(skb, linear);
1473         if (err)
1474                 goto free;
1475
1476         skb->len = len;
1477         skb->data_len = len - linear;
1478         skb->truesize += skb->data_len;
1479
1480         for (i = 1; i < it->nr_segs; i++) {
1481                 size_t fragsz = it->iov[i].iov_len;
1482                 struct page *page;
1483                 void *frag;
1484
1485                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1486                         err = -EINVAL;
1487                         goto free;
1488                 }
1489                 frag = netdev_alloc_frag(fragsz);
1490                 if (!frag) {
1491                         err = -ENOMEM;
1492                         goto free;
1493                 }
1494                 page = virt_to_head_page(frag);
1495                 skb_fill_page_desc(skb, i - 1, page,
1496                                    frag - page_address(page), fragsz);
1497         }
1498
1499         return skb;
1500 free:
1501         /* frees skb and all frags allocated with napi_alloc_frag() */
1502         napi_free_frags(&tfile->napi);
1503         return ERR_PTR(err);
1504 }
1505
1506 /* prepad is the amount to reserve at front.  len is length after that.
1507  * linear is a hint as to how much to copy (usually headers). */
1508 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1509                                      size_t prepad, size_t len,
1510                                      size_t linear, int noblock)
1511 {
1512         struct sock *sk = tfile->socket.sk;
1513         struct sk_buff *skb;
1514         int err;
1515
1516         /* Under a page?  Don't bother with paged skb. */
1517         if (prepad + len < PAGE_SIZE || !linear)
1518                 linear = len;
1519
1520         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1521                                    &err, 0);
1522         if (!skb)
1523                 return ERR_PTR(err);
1524
1525         skb_reserve(skb, prepad);
1526         skb_put(skb, linear);
1527         skb->data_len = len - linear;
1528         skb->len += len - linear;
1529
1530         return skb;
1531 }
1532
1533 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1534                            struct sk_buff *skb, int more)
1535 {
1536         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1537         struct sk_buff_head process_queue;
1538         u32 rx_batched = tun->rx_batched;
1539         bool rcv = false;
1540
1541         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1542                 local_bh_disable();
1543                 skb_record_rx_queue(skb, tfile->queue_index);
1544                 netif_receive_skb(skb);
1545                 local_bh_enable();
1546                 return;
1547         }
1548
1549         spin_lock(&queue->lock);
1550         if (!more || skb_queue_len(queue) == rx_batched) {
1551                 __skb_queue_head_init(&process_queue);
1552                 skb_queue_splice_tail_init(queue, &process_queue);
1553                 rcv = true;
1554         } else {
1555                 __skb_queue_tail(queue, skb);
1556         }
1557         spin_unlock(&queue->lock);
1558
1559         if (rcv) {
1560                 struct sk_buff *nskb;
1561
1562                 local_bh_disable();
1563                 while ((nskb = __skb_dequeue(&process_queue))) {
1564                         skb_record_rx_queue(nskb, tfile->queue_index);
1565                         netif_receive_skb(nskb);
1566                 }
1567                 skb_record_rx_queue(skb, tfile->queue_index);
1568                 netif_receive_skb(skb);
1569                 local_bh_enable();
1570         }
1571 }
1572
1573 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1574                               int len, int noblock, bool zerocopy)
1575 {
1576         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1577                 return false;
1578
1579         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1580                 return false;
1581
1582         if (!noblock)
1583                 return false;
1584
1585         if (zerocopy)
1586                 return false;
1587
1588         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1589             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1590                 return false;
1591
1592         return true;
1593 }
1594
1595 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1596                                        struct page_frag *alloc_frag, char *buf,
1597                                        int buflen, int len, int pad)
1598 {
1599         struct sk_buff *skb = build_skb(buf, buflen);
1600
1601         if (!skb)
1602                 return ERR_PTR(-ENOMEM);
1603
1604         skb_reserve(skb, pad);
1605         skb_put(skb, len);
1606         skb_set_owner_w(skb, tfile->socket.sk);
1607
1608         get_page(alloc_frag->page);
1609         alloc_frag->offset += buflen;
1610
1611         return skb;
1612 }
1613
1614 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1615                        struct xdp_buff *xdp, u32 act)
1616 {
1617         int err;
1618
1619         switch (act) {
1620         case XDP_REDIRECT:
1621                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1622                 if (err)
1623                         return err;
1624                 break;
1625         case XDP_TX:
1626                 err = tun_xdp_tx(tun->dev, xdp);
1627                 if (err < 0)
1628                         return err;
1629                 break;
1630         case XDP_PASS:
1631                 break;
1632         default:
1633                 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1634                 fallthrough;
1635         case XDP_ABORTED:
1636                 trace_xdp_exception(tun->dev, xdp_prog, act);
1637                 fallthrough;
1638         case XDP_DROP:
1639                 dev_core_stats_rx_dropped_inc(tun->dev);
1640                 break;
1641         }
1642
1643         return act;
1644 }
1645
1646 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1647                                      struct tun_file *tfile,
1648                                      struct iov_iter *from,
1649                                      struct virtio_net_hdr *hdr,
1650                                      int len, int *skb_xdp)
1651 {
1652         struct page_frag *alloc_frag = &current->task_frag;
1653         struct bpf_prog *xdp_prog;
1654         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1655         char *buf;
1656         size_t copied;
1657         int pad = TUN_RX_PAD;
1658         int err = 0;
1659
1660         rcu_read_lock();
1661         xdp_prog = rcu_dereference(tun->xdp_prog);
1662         if (xdp_prog)
1663                 pad += XDP_PACKET_HEADROOM;
1664         buflen += SKB_DATA_ALIGN(len + pad);
1665         rcu_read_unlock();
1666
1667         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1668         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1669                 return ERR_PTR(-ENOMEM);
1670
1671         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1672         copied = copy_page_from_iter(alloc_frag->page,
1673                                      alloc_frag->offset + pad,
1674                                      len, from);
1675         if (copied != len)
1676                 return ERR_PTR(-EFAULT);
1677
1678         /* There's a small window that XDP may be set after the check
1679          * of xdp_prog above, this should be rare and for simplicity
1680          * we do XDP on skb in case the headroom is not enough.
1681          */
1682         if (hdr->gso_type || !xdp_prog) {
1683                 *skb_xdp = 1;
1684                 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1685                                        pad);
1686         }
1687
1688         *skb_xdp = 0;
1689
1690         local_bh_disable();
1691         rcu_read_lock();
1692         xdp_prog = rcu_dereference(tun->xdp_prog);
1693         if (xdp_prog) {
1694                 struct xdp_buff xdp;
1695                 u32 act;
1696
1697                 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1698                 xdp_prepare_buff(&xdp, buf, pad, len, false);
1699
1700                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1701                 if (act == XDP_REDIRECT || act == XDP_TX) {
1702                         get_page(alloc_frag->page);
1703                         alloc_frag->offset += buflen;
1704                 }
1705                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1706                 if (err < 0) {
1707                         if (act == XDP_REDIRECT || act == XDP_TX)
1708                                 put_page(alloc_frag->page);
1709                         goto out;
1710                 }
1711
1712                 if (err == XDP_REDIRECT)
1713                         xdp_do_flush();
1714                 if (err != XDP_PASS)
1715                         goto out;
1716
1717                 pad = xdp.data - xdp.data_hard_start;
1718                 len = xdp.data_end - xdp.data;
1719         }
1720         rcu_read_unlock();
1721         local_bh_enable();
1722
1723         return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1724
1725 out:
1726         rcu_read_unlock();
1727         local_bh_enable();
1728         return NULL;
1729 }
1730
1731 /* Get packet from user space buffer */
1732 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1733                             void *msg_control, struct iov_iter *from,
1734                             int noblock, bool more)
1735 {
1736         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1737         struct sk_buff *skb;
1738         size_t total_len = iov_iter_count(from);
1739         size_t len = total_len, align = tun->align, linear;
1740         struct virtio_net_hdr gso = { 0 };
1741         int good_linear;
1742         int copylen;
1743         bool zerocopy = false;
1744         int err;
1745         u32 rxhash = 0;
1746         int skb_xdp = 1;
1747         bool frags = tun_napi_frags_enabled(tfile);
1748         enum skb_drop_reason drop_reason;
1749
1750         if (!(tun->flags & IFF_NO_PI)) {
1751                 if (len < sizeof(pi))
1752                         return -EINVAL;
1753                 len -= sizeof(pi);
1754
1755                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1756                         return -EFAULT;
1757         }
1758
1759         if (tun->flags & IFF_VNET_HDR) {
1760                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1761
1762                 if (len < vnet_hdr_sz)
1763                         return -EINVAL;
1764                 len -= vnet_hdr_sz;
1765
1766                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1767                         return -EFAULT;
1768
1769                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1770                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1771                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1772
1773                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1774                         return -EINVAL;
1775                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1776         }
1777
1778         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1779                 align += NET_IP_ALIGN;
1780                 if (unlikely(len < ETH_HLEN ||
1781                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1782                         return -EINVAL;
1783         }
1784
1785         good_linear = SKB_MAX_HEAD(align);
1786
1787         if (msg_control) {
1788                 struct iov_iter i = *from;
1789
1790                 /* There are 256 bytes to be copied in skb, so there is
1791                  * enough room for skb expand head in case it is used.
1792                  * The rest of the buffer is mapped from userspace.
1793                  */
1794                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1795                 if (copylen > good_linear)
1796                         copylen = good_linear;
1797                 linear = copylen;
1798                 iov_iter_advance(&i, copylen);
1799                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1800                         zerocopy = true;
1801         }
1802
1803         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1804                 /* For the packet that is not easy to be processed
1805                  * (e.g gso or jumbo packet), we will do it at after
1806                  * skb was created with generic XDP routine.
1807                  */
1808                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1809                 if (IS_ERR(skb)) {
1810                         dev_core_stats_rx_dropped_inc(tun->dev);
1811                         return PTR_ERR(skb);
1812                 }
1813                 if (!skb)
1814                         return total_len;
1815         } else {
1816                 if (!zerocopy) {
1817                         copylen = len;
1818                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1819                                 linear = good_linear;
1820                         else
1821                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1822                 }
1823
1824                 if (frags) {
1825                         mutex_lock(&tfile->napi_mutex);
1826                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1827                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1828                          * If zerocopy is enabled, then this layout will be
1829                          * overwritten by zerocopy_sg_from_iter().
1830                          */
1831                         zerocopy = false;
1832                 } else {
1833                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1834                                             noblock);
1835                 }
1836
1837                 if (IS_ERR(skb)) {
1838                         if (PTR_ERR(skb) != -EAGAIN)
1839                                 dev_core_stats_rx_dropped_inc(tun->dev);
1840                         if (frags)
1841                                 mutex_unlock(&tfile->napi_mutex);
1842                         return PTR_ERR(skb);
1843                 }
1844
1845                 if (zerocopy)
1846                         err = zerocopy_sg_from_iter(skb, from);
1847                 else
1848                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1849
1850                 if (err) {
1851                         err = -EFAULT;
1852                         drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1853 drop:
1854                         dev_core_stats_rx_dropped_inc(tun->dev);
1855                         kfree_skb_reason(skb, drop_reason);
1856                         if (frags) {
1857                                 tfile->napi.skb = NULL;
1858                                 mutex_unlock(&tfile->napi_mutex);
1859                         }
1860
1861                         return err;
1862                 }
1863         }
1864
1865         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1866                 atomic_long_inc(&tun->rx_frame_errors);
1867                 kfree_skb(skb);
1868                 if (frags) {
1869                         tfile->napi.skb = NULL;
1870                         mutex_unlock(&tfile->napi_mutex);
1871                 }
1872
1873                 return -EINVAL;
1874         }
1875
1876         switch (tun->flags & TUN_TYPE_MASK) {
1877         case IFF_TUN:
1878                 if (tun->flags & IFF_NO_PI) {
1879                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1880
1881                         switch (ip_version) {
1882                         case 4:
1883                                 pi.proto = htons(ETH_P_IP);
1884                                 break;
1885                         case 6:
1886                                 pi.proto = htons(ETH_P_IPV6);
1887                                 break;
1888                         default:
1889                                 dev_core_stats_rx_dropped_inc(tun->dev);
1890                                 kfree_skb(skb);
1891                                 return -EINVAL;
1892                         }
1893                 }
1894
1895                 skb_reset_mac_header(skb);
1896                 skb->protocol = pi.proto;
1897                 skb->dev = tun->dev;
1898                 break;
1899         case IFF_TAP:
1900                 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1901                         err = -ENOMEM;
1902                         drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1903                         goto drop;
1904                 }
1905                 skb->protocol = eth_type_trans(skb, tun->dev);
1906                 break;
1907         }
1908
1909         /* copy skb_ubuf_info for callback when skb has no error */
1910         if (zerocopy) {
1911                 skb_zcopy_init(skb, msg_control);
1912         } else if (msg_control) {
1913                 struct ubuf_info *uarg = msg_control;
1914                 uarg->callback(NULL, uarg, false);
1915         }
1916
1917         skb_reset_network_header(skb);
1918         skb_probe_transport_header(skb);
1919         skb_record_rx_queue(skb, tfile->queue_index);
1920
1921         if (skb_xdp) {
1922                 struct bpf_prog *xdp_prog;
1923                 int ret;
1924
1925                 local_bh_disable();
1926                 rcu_read_lock();
1927                 xdp_prog = rcu_dereference(tun->xdp_prog);
1928                 if (xdp_prog) {
1929                         ret = do_xdp_generic(xdp_prog, skb);
1930                         if (ret != XDP_PASS) {
1931                                 rcu_read_unlock();
1932                                 local_bh_enable();
1933                                 if (frags) {
1934                                         tfile->napi.skb = NULL;
1935                                         mutex_unlock(&tfile->napi_mutex);
1936                                 }
1937                                 return total_len;
1938                         }
1939                 }
1940                 rcu_read_unlock();
1941                 local_bh_enable();
1942         }
1943
1944         /* Compute the costly rx hash only if needed for flow updates.
1945          * We may get a very small possibility of OOO during switching, not
1946          * worth to optimize.
1947          */
1948         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1949             !tfile->detached)
1950                 rxhash = __skb_get_hash_symmetric(skb);
1951
1952         rcu_read_lock();
1953         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1954                 err = -EIO;
1955                 rcu_read_unlock();
1956                 drop_reason = SKB_DROP_REASON_DEV_READY;
1957                 goto drop;
1958         }
1959
1960         if (frags) {
1961                 u32 headlen;
1962
1963                 /* Exercise flow dissector code path. */
1964                 skb_push(skb, ETH_HLEN);
1965                 headlen = eth_get_headlen(tun->dev, skb->data,
1966                                           skb_headlen(skb));
1967
1968                 if (unlikely(headlen > skb_headlen(skb))) {
1969                         dev_core_stats_rx_dropped_inc(tun->dev);
1970                         napi_free_frags(&tfile->napi);
1971                         rcu_read_unlock();
1972                         mutex_unlock(&tfile->napi_mutex);
1973                         WARN_ON(1);
1974                         return -ENOMEM;
1975                 }
1976
1977                 local_bh_disable();
1978                 napi_gro_frags(&tfile->napi);
1979                 local_bh_enable();
1980                 mutex_unlock(&tfile->napi_mutex);
1981         } else if (tfile->napi_enabled) {
1982                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1983                 int queue_len;
1984
1985                 spin_lock_bh(&queue->lock);
1986                 __skb_queue_tail(queue, skb);
1987                 queue_len = skb_queue_len(queue);
1988                 spin_unlock(&queue->lock);
1989
1990                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1991                         napi_schedule(&tfile->napi);
1992
1993                 local_bh_enable();
1994         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1995                 tun_rx_batched(tun, tfile, skb, more);
1996         } else {
1997                 netif_rx(skb);
1998         }
1999         rcu_read_unlock();
2000
2001         preempt_disable();
2002         dev_sw_netstats_rx_add(tun->dev, len);
2003         preempt_enable();
2004
2005         if (rxhash)
2006                 tun_flow_update(tun, rxhash, tfile);
2007
2008         return total_len;
2009 }
2010
2011 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2012 {
2013         struct file *file = iocb->ki_filp;
2014         struct tun_file *tfile = file->private_data;
2015         struct tun_struct *tun = tun_get(tfile);
2016         ssize_t result;
2017         int noblock = 0;
2018
2019         if (!tun)
2020                 return -EBADFD;
2021
2022         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2023                 noblock = 1;
2024
2025         result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2026
2027         tun_put(tun);
2028         return result;
2029 }
2030
2031 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2032                                 struct tun_file *tfile,
2033                                 struct xdp_frame *xdp_frame,
2034                                 struct iov_iter *iter)
2035 {
2036         int vnet_hdr_sz = 0;
2037         size_t size = xdp_frame->len;
2038         size_t ret;
2039
2040         if (tun->flags & IFF_VNET_HDR) {
2041                 struct virtio_net_hdr gso = { 0 };
2042
2043                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2044                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2045                         return -EINVAL;
2046                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2047                              sizeof(gso)))
2048                         return -EFAULT;
2049                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2050         }
2051
2052         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2053
2054         preempt_disable();
2055         dev_sw_netstats_tx_add(tun->dev, 1, ret);
2056         preempt_enable();
2057
2058         return ret;
2059 }
2060
2061 /* Put packet to the user space buffer */
2062 static ssize_t tun_put_user(struct tun_struct *tun,
2063                             struct tun_file *tfile,
2064                             struct sk_buff *skb,
2065                             struct iov_iter *iter)
2066 {
2067         struct tun_pi pi = { 0, skb->protocol };
2068         ssize_t total;
2069         int vlan_offset = 0;
2070         int vlan_hlen = 0;
2071         int vnet_hdr_sz = 0;
2072
2073         if (skb_vlan_tag_present(skb))
2074                 vlan_hlen = VLAN_HLEN;
2075
2076         if (tun->flags & IFF_VNET_HDR)
2077                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2078
2079         total = skb->len + vlan_hlen + vnet_hdr_sz;
2080
2081         if (!(tun->flags & IFF_NO_PI)) {
2082                 if (iov_iter_count(iter) < sizeof(pi))
2083                         return -EINVAL;
2084
2085                 total += sizeof(pi);
2086                 if (iov_iter_count(iter) < total) {
2087                         /* Packet will be striped */
2088                         pi.flags |= TUN_PKT_STRIP;
2089                 }
2090
2091                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2092                         return -EFAULT;
2093         }
2094
2095         if (vnet_hdr_sz) {
2096                 struct virtio_net_hdr gso;
2097
2098                 if (iov_iter_count(iter) < vnet_hdr_sz)
2099                         return -EINVAL;
2100
2101                 if (virtio_net_hdr_from_skb(skb, &gso,
2102                                             tun_is_little_endian(tun), true,
2103                                             vlan_hlen)) {
2104                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2105                         pr_err("unexpected GSO type: "
2106                                "0x%x, gso_size %d, hdr_len %d\n",
2107                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2108                                tun16_to_cpu(tun, gso.hdr_len));
2109                         print_hex_dump(KERN_ERR, "tun: ",
2110                                        DUMP_PREFIX_NONE,
2111                                        16, 1, skb->head,
2112                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2113                         WARN_ON_ONCE(1);
2114                         return -EINVAL;
2115                 }
2116
2117                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2118                         return -EFAULT;
2119
2120                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2121         }
2122
2123         if (vlan_hlen) {
2124                 int ret;
2125                 struct veth veth;
2126
2127                 veth.h_vlan_proto = skb->vlan_proto;
2128                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2129
2130                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2131
2132                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2133                 if (ret || !iov_iter_count(iter))
2134                         goto done;
2135
2136                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2137                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2138                         goto done;
2139         }
2140
2141         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2142
2143 done:
2144         /* caller is in process context, */
2145         preempt_disable();
2146         dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2147         preempt_enable();
2148
2149         return total;
2150 }
2151
2152 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2153 {
2154         DECLARE_WAITQUEUE(wait, current);
2155         void *ptr = NULL;
2156         int error = 0;
2157
2158         ptr = ptr_ring_consume(&tfile->tx_ring);
2159         if (ptr)
2160                 goto out;
2161         if (noblock) {
2162                 error = -EAGAIN;
2163                 goto out;
2164         }
2165
2166         add_wait_queue(&tfile->socket.wq.wait, &wait);
2167
2168         while (1) {
2169                 set_current_state(TASK_INTERRUPTIBLE);
2170                 ptr = ptr_ring_consume(&tfile->tx_ring);
2171                 if (ptr)
2172                         break;
2173                 if (signal_pending(current)) {
2174                         error = -ERESTARTSYS;
2175                         break;
2176                 }
2177                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2178                         error = -EFAULT;
2179                         break;
2180                 }
2181
2182                 schedule();
2183         }
2184
2185         __set_current_state(TASK_RUNNING);
2186         remove_wait_queue(&tfile->socket.wq.wait, &wait);
2187
2188 out:
2189         *err = error;
2190         return ptr;
2191 }
2192
2193 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2194                            struct iov_iter *to,
2195                            int noblock, void *ptr)
2196 {
2197         ssize_t ret;
2198         int err;
2199
2200         if (!iov_iter_count(to)) {
2201                 tun_ptr_free(ptr);
2202                 return 0;
2203         }
2204
2205         if (!ptr) {
2206                 /* Read frames from ring */
2207                 ptr = tun_ring_recv(tfile, noblock, &err);
2208                 if (!ptr)
2209                         return err;
2210         }
2211
2212         if (tun_is_xdp_frame(ptr)) {
2213                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2214
2215                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2216                 xdp_return_frame(xdpf);
2217         } else {
2218                 struct sk_buff *skb = ptr;
2219
2220                 ret = tun_put_user(tun, tfile, skb, to);
2221                 if (unlikely(ret < 0))
2222                         kfree_skb(skb);
2223                 else
2224                         consume_skb(skb);
2225         }
2226
2227         return ret;
2228 }
2229
2230 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2231 {
2232         struct file *file = iocb->ki_filp;
2233         struct tun_file *tfile = file->private_data;
2234         struct tun_struct *tun = tun_get(tfile);
2235         ssize_t len = iov_iter_count(to), ret;
2236         int noblock = 0;
2237
2238         if (!tun)
2239                 return -EBADFD;
2240
2241         if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2242                 noblock = 1;
2243
2244         ret = tun_do_read(tun, tfile, to, noblock, NULL);
2245         ret = min_t(ssize_t, ret, len);
2246         if (ret > 0)
2247                 iocb->ki_pos = ret;
2248         tun_put(tun);
2249         return ret;
2250 }
2251
2252 static void tun_prog_free(struct rcu_head *rcu)
2253 {
2254         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2255
2256         bpf_prog_destroy(prog->prog);
2257         kfree(prog);
2258 }
2259
2260 static int __tun_set_ebpf(struct tun_struct *tun,
2261                           struct tun_prog __rcu **prog_p,
2262                           struct bpf_prog *prog)
2263 {
2264         struct tun_prog *old, *new = NULL;
2265
2266         if (prog) {
2267                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2268                 if (!new)
2269                         return -ENOMEM;
2270                 new->prog = prog;
2271         }
2272
2273         spin_lock_bh(&tun->lock);
2274         old = rcu_dereference_protected(*prog_p,
2275                                         lockdep_is_held(&tun->lock));
2276         rcu_assign_pointer(*prog_p, new);
2277         spin_unlock_bh(&tun->lock);
2278
2279         if (old)
2280                 call_rcu(&old->rcu, tun_prog_free);
2281
2282         return 0;
2283 }
2284
2285 static void tun_free_netdev(struct net_device *dev)
2286 {
2287         struct tun_struct *tun = netdev_priv(dev);
2288
2289         BUG_ON(!(list_empty(&tun->disabled)));
2290
2291         free_percpu(dev->tstats);
2292         tun_flow_uninit(tun);
2293         security_tun_dev_free_security(tun->security);
2294         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2295         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2296 }
2297
2298 static void tun_setup(struct net_device *dev)
2299 {
2300         struct tun_struct *tun = netdev_priv(dev);
2301
2302         tun->owner = INVALID_UID;
2303         tun->group = INVALID_GID;
2304         tun_default_link_ksettings(dev, &tun->link_ksettings);
2305
2306         dev->ethtool_ops = &tun_ethtool_ops;
2307         dev->needs_free_netdev = true;
2308         dev->priv_destructor = tun_free_netdev;
2309         /* We prefer our own queue length */
2310         dev->tx_queue_len = TUN_READQ_SIZE;
2311 }
2312
2313 /* Trivial set of netlink ops to allow deleting tun or tap
2314  * device with netlink.
2315  */
2316 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2317                         struct netlink_ext_ack *extack)
2318 {
2319         NL_SET_ERR_MSG(extack,
2320                        "tun/tap creation via rtnetlink is not supported.");
2321         return -EOPNOTSUPP;
2322 }
2323
2324 static size_t tun_get_size(const struct net_device *dev)
2325 {
2326         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2327         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2328
2329         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2330                nla_total_size(sizeof(gid_t)) + /* GROUP */
2331                nla_total_size(sizeof(u8)) + /* TYPE */
2332                nla_total_size(sizeof(u8)) + /* PI */
2333                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2334                nla_total_size(sizeof(u8)) + /* PERSIST */
2335                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2336                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2337                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2338                0;
2339 }
2340
2341 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2342 {
2343         struct tun_struct *tun = netdev_priv(dev);
2344
2345         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2346                 goto nla_put_failure;
2347         if (uid_valid(tun->owner) &&
2348             nla_put_u32(skb, IFLA_TUN_OWNER,
2349                         from_kuid_munged(current_user_ns(), tun->owner)))
2350                 goto nla_put_failure;
2351         if (gid_valid(tun->group) &&
2352             nla_put_u32(skb, IFLA_TUN_GROUP,
2353                         from_kgid_munged(current_user_ns(), tun->group)))
2354                 goto nla_put_failure;
2355         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2356                 goto nla_put_failure;
2357         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2358                 goto nla_put_failure;
2359         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2360                 goto nla_put_failure;
2361         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2362                        !!(tun->flags & IFF_MULTI_QUEUE)))
2363                 goto nla_put_failure;
2364         if (tun->flags & IFF_MULTI_QUEUE) {
2365                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2366                         goto nla_put_failure;
2367                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2368                                 tun->numdisabled))
2369                         goto nla_put_failure;
2370         }
2371
2372         return 0;
2373
2374 nla_put_failure:
2375         return -EMSGSIZE;
2376 }
2377
2378 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2379         .kind           = DRV_NAME,
2380         .priv_size      = sizeof(struct tun_struct),
2381         .setup          = tun_setup,
2382         .validate       = tun_validate,
2383         .get_size       = tun_get_size,
2384         .fill_info      = tun_fill_info,
2385 };
2386
2387 static void tun_sock_write_space(struct sock *sk)
2388 {
2389         struct tun_file *tfile;
2390         wait_queue_head_t *wqueue;
2391
2392         if (!sock_writeable(sk))
2393                 return;
2394
2395         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2396                 return;
2397
2398         wqueue = sk_sleep(sk);
2399         if (wqueue && waitqueue_active(wqueue))
2400                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2401                                                 EPOLLWRNORM | EPOLLWRBAND);
2402
2403         tfile = container_of(sk, struct tun_file, sk);
2404         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2405 }
2406
2407 static void tun_put_page(struct tun_page *tpage)
2408 {
2409         if (tpage->page)
2410                 __page_frag_cache_drain(tpage->page, tpage->count);
2411 }
2412
2413 static int tun_xdp_one(struct tun_struct *tun,
2414                        struct tun_file *tfile,
2415                        struct xdp_buff *xdp, int *flush,
2416                        struct tun_page *tpage)
2417 {
2418         unsigned int datasize = xdp->data_end - xdp->data;
2419         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2420         struct virtio_net_hdr *gso = &hdr->gso;
2421         struct bpf_prog *xdp_prog;
2422         struct sk_buff *skb = NULL;
2423         struct sk_buff_head *queue;
2424         u32 rxhash = 0, act;
2425         int buflen = hdr->buflen;
2426         int ret = 0;
2427         bool skb_xdp = false;
2428         struct page *page;
2429
2430         xdp_prog = rcu_dereference(tun->xdp_prog);
2431         if (xdp_prog) {
2432                 if (gso->gso_type) {
2433                         skb_xdp = true;
2434                         goto build;
2435                 }
2436
2437                 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2438                 xdp_set_data_meta_invalid(xdp);
2439
2440                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2441                 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2442                 if (ret < 0) {
2443                         put_page(virt_to_head_page(xdp->data));
2444                         return ret;
2445                 }
2446
2447                 switch (ret) {
2448                 case XDP_REDIRECT:
2449                         *flush = true;
2450                         fallthrough;
2451                 case XDP_TX:
2452                         return 0;
2453                 case XDP_PASS:
2454                         break;
2455                 default:
2456                         page = virt_to_head_page(xdp->data);
2457                         if (tpage->page == page) {
2458                                 ++tpage->count;
2459                         } else {
2460                                 tun_put_page(tpage);
2461                                 tpage->page = page;
2462                                 tpage->count = 1;
2463                         }
2464                         return 0;
2465                 }
2466         }
2467
2468 build:
2469         skb = build_skb(xdp->data_hard_start, buflen);
2470         if (!skb) {
2471                 ret = -ENOMEM;
2472                 goto out;
2473         }
2474
2475         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2476         skb_put(skb, xdp->data_end - xdp->data);
2477
2478         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2479                 atomic_long_inc(&tun->rx_frame_errors);
2480                 kfree_skb(skb);
2481                 ret = -EINVAL;
2482                 goto out;
2483         }
2484
2485         skb->protocol = eth_type_trans(skb, tun->dev);
2486         skb_reset_network_header(skb);
2487         skb_probe_transport_header(skb);
2488         skb_record_rx_queue(skb, tfile->queue_index);
2489
2490         if (skb_xdp) {
2491                 ret = do_xdp_generic(xdp_prog, skb);
2492                 if (ret != XDP_PASS) {
2493                         ret = 0;
2494                         goto out;
2495                 }
2496         }
2497
2498         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2499             !tfile->detached)
2500                 rxhash = __skb_get_hash_symmetric(skb);
2501
2502         if (tfile->napi_enabled) {
2503                 queue = &tfile->sk.sk_write_queue;
2504                 spin_lock(&queue->lock);
2505                 __skb_queue_tail(queue, skb);
2506                 spin_unlock(&queue->lock);
2507                 ret = 1;
2508         } else {
2509                 netif_receive_skb(skb);
2510                 ret = 0;
2511         }
2512
2513         /* No need to disable preemption here since this function is
2514          * always called with bh disabled
2515          */
2516         dev_sw_netstats_rx_add(tun->dev, datasize);
2517
2518         if (rxhash)
2519                 tun_flow_update(tun, rxhash, tfile);
2520
2521 out:
2522         return ret;
2523 }
2524
2525 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2526 {
2527         int ret, i;
2528         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2529         struct tun_struct *tun = tun_get(tfile);
2530         struct tun_msg_ctl *ctl = m->msg_control;
2531         struct xdp_buff *xdp;
2532
2533         if (!tun)
2534                 return -EBADFD;
2535
2536         if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2537             ctl && ctl->type == TUN_MSG_PTR) {
2538                 struct tun_page tpage;
2539                 int n = ctl->num;
2540                 int flush = 0, queued = 0;
2541
2542                 memset(&tpage, 0, sizeof(tpage));
2543
2544                 local_bh_disable();
2545                 rcu_read_lock();
2546
2547                 for (i = 0; i < n; i++) {
2548                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2549                         ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2550                         if (ret > 0)
2551                                 queued += ret;
2552                 }
2553
2554                 if (flush)
2555                         xdp_do_flush();
2556
2557                 if (tfile->napi_enabled && queued > 0)
2558                         napi_schedule(&tfile->napi);
2559
2560                 rcu_read_unlock();
2561                 local_bh_enable();
2562
2563                 tun_put_page(&tpage);
2564
2565                 ret = total_len;
2566                 goto out;
2567         }
2568
2569         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2570                            m->msg_flags & MSG_DONTWAIT,
2571                            m->msg_flags & MSG_MORE);
2572 out:
2573         tun_put(tun);
2574         return ret;
2575 }
2576
2577 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2578                        int flags)
2579 {
2580         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2581         struct tun_struct *tun = tun_get(tfile);
2582         void *ptr = m->msg_control;
2583         int ret;
2584
2585         if (!tun) {
2586                 ret = -EBADFD;
2587                 goto out_free;
2588         }
2589
2590         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2591                 ret = -EINVAL;
2592                 goto out_put_tun;
2593         }
2594         if (flags & MSG_ERRQUEUE) {
2595                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2596                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2597                 goto out;
2598         }
2599         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2600         if (ret > (ssize_t)total_len) {
2601                 m->msg_flags |= MSG_TRUNC;
2602                 ret = flags & MSG_TRUNC ? ret : total_len;
2603         }
2604 out:
2605         tun_put(tun);
2606         return ret;
2607
2608 out_put_tun:
2609         tun_put(tun);
2610 out_free:
2611         tun_ptr_free(ptr);
2612         return ret;
2613 }
2614
2615 static int tun_ptr_peek_len(void *ptr)
2616 {
2617         if (likely(ptr)) {
2618                 if (tun_is_xdp_frame(ptr)) {
2619                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2620
2621                         return xdpf->len;
2622                 }
2623                 return __skb_array_len_with_tag(ptr);
2624         } else {
2625                 return 0;
2626         }
2627 }
2628
2629 static int tun_peek_len(struct socket *sock)
2630 {
2631         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2632         struct tun_struct *tun;
2633         int ret = 0;
2634
2635         tun = tun_get(tfile);
2636         if (!tun)
2637                 return 0;
2638
2639         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2640         tun_put(tun);
2641
2642         return ret;
2643 }
2644
2645 /* Ops structure to mimic raw sockets with tun */
2646 static const struct proto_ops tun_socket_ops = {
2647         .peek_len = tun_peek_len,
2648         .sendmsg = tun_sendmsg,
2649         .recvmsg = tun_recvmsg,
2650 };
2651
2652 static struct proto tun_proto = {
2653         .name           = "tun",
2654         .owner          = THIS_MODULE,
2655         .obj_size       = sizeof(struct tun_file),
2656 };
2657
2658 static int tun_flags(struct tun_struct *tun)
2659 {
2660         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2661 }
2662
2663 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2664                               char *buf)
2665 {
2666         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2667         return sprintf(buf, "0x%x\n", tun_flags(tun));
2668 }
2669
2670 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2671                           char *buf)
2672 {
2673         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2674         return uid_valid(tun->owner)?
2675                 sprintf(buf, "%u\n",
2676                         from_kuid_munged(current_user_ns(), tun->owner)):
2677                 sprintf(buf, "-1\n");
2678 }
2679
2680 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2681                           char *buf)
2682 {
2683         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2684         return gid_valid(tun->group) ?
2685                 sprintf(buf, "%u\n",
2686                         from_kgid_munged(current_user_ns(), tun->group)):
2687                 sprintf(buf, "-1\n");
2688 }
2689
2690 static DEVICE_ATTR_RO(tun_flags);
2691 static DEVICE_ATTR_RO(owner);
2692 static DEVICE_ATTR_RO(group);
2693
2694 static struct attribute *tun_dev_attrs[] = {
2695         &dev_attr_tun_flags.attr,
2696         &dev_attr_owner.attr,
2697         &dev_attr_group.attr,
2698         NULL
2699 };
2700
2701 static const struct attribute_group tun_attr_group = {
2702         .attrs = tun_dev_attrs
2703 };
2704
2705 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2706 {
2707         struct tun_struct *tun;
2708         struct tun_file *tfile = file->private_data;
2709         struct net_device *dev;
2710         int err;
2711
2712         if (tfile->detached)
2713                 return -EINVAL;
2714
2715         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2716                 if (!capable(CAP_NET_ADMIN))
2717                         return -EPERM;
2718
2719                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2720                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2721                         return -EINVAL;
2722         }
2723
2724         dev = __dev_get_by_name(net, ifr->ifr_name);
2725         if (dev) {
2726                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2727                         return -EBUSY;
2728                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2729                         tun = netdev_priv(dev);
2730                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2731                         tun = netdev_priv(dev);
2732                 else
2733                         return -EINVAL;
2734
2735                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2736                     !!(tun->flags & IFF_MULTI_QUEUE))
2737                         return -EINVAL;
2738
2739                 if (tun_not_capable(tun))
2740                         return -EPERM;
2741                 err = security_tun_dev_open(tun->security);
2742                 if (err < 0)
2743                         return err;
2744
2745                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2746                                  ifr->ifr_flags & IFF_NAPI,
2747                                  ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2748                 if (err < 0)
2749                         return err;
2750
2751                 if (tun->flags & IFF_MULTI_QUEUE &&
2752                     (tun->numqueues + tun->numdisabled > 1)) {
2753                         /* One or more queue has already been attached, no need
2754                          * to initialize the device again.
2755                          */
2756                         netdev_state_change(dev);
2757                         return 0;
2758                 }
2759
2760                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2761                               (ifr->ifr_flags & TUN_FEATURES);
2762
2763                 netdev_state_change(dev);
2764         } else {
2765                 char *name;
2766                 unsigned long flags = 0;
2767                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2768                              MAX_TAP_QUEUES : 1;
2769
2770                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2771                         return -EPERM;
2772                 err = security_tun_dev_create();
2773                 if (err < 0)
2774                         return err;
2775
2776                 /* Set dev type */
2777                 if (ifr->ifr_flags & IFF_TUN) {
2778                         /* TUN device */
2779                         flags |= IFF_TUN;
2780                         name = "tun%d";
2781                 } else if (ifr->ifr_flags & IFF_TAP) {
2782                         /* TAP device */
2783                         flags |= IFF_TAP;
2784                         name = "tap%d";
2785                 } else
2786                         return -EINVAL;
2787
2788                 if (*ifr->ifr_name)
2789                         name = ifr->ifr_name;
2790
2791                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2792                                        NET_NAME_UNKNOWN, tun_setup, queues,
2793                                        queues);
2794
2795                 if (!dev)
2796                         return -ENOMEM;
2797
2798                 dev_net_set(dev, net);
2799                 dev->rtnl_link_ops = &tun_link_ops;
2800                 dev->ifindex = tfile->ifindex;
2801                 dev->sysfs_groups[0] = &tun_attr_group;
2802
2803                 tun = netdev_priv(dev);
2804                 tun->dev = dev;
2805                 tun->flags = flags;
2806                 tun->txflt.count = 0;
2807                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2808
2809                 tun->align = NET_SKB_PAD;
2810                 tun->filter_attached = false;
2811                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2812                 tun->rx_batched = 0;
2813                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2814
2815                 tun->ifr = ifr;
2816                 tun->file = file;
2817
2818                 tun_net_initialize(dev);
2819
2820                 err = register_netdevice(tun->dev);
2821                 if (err < 0) {
2822                         free_netdev(dev);
2823                         return err;
2824                 }
2825                 /* free_netdev() won't check refcnt, to avoid race
2826                  * with dev_put() we need publish tun after registration.
2827                  */
2828                 rcu_assign_pointer(tfile->tun, tun);
2829         }
2830
2831         netif_carrier_on(tun->dev);
2832
2833         /* Make sure persistent devices do not get stuck in
2834          * xoff state.
2835          */
2836         if (netif_running(tun->dev))
2837                 netif_tx_wake_all_queues(tun->dev);
2838
2839         strcpy(ifr->ifr_name, tun->dev->name);
2840         return 0;
2841 }
2842
2843 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2844 {
2845         strcpy(ifr->ifr_name, tun->dev->name);
2846
2847         ifr->ifr_flags = tun_flags(tun);
2848
2849 }
2850
2851 /* This is like a cut-down ethtool ops, except done via tun fd so no
2852  * privs required. */
2853 static int set_offload(struct tun_struct *tun, unsigned long arg)
2854 {
2855         netdev_features_t features = 0;
2856
2857         if (arg & TUN_F_CSUM) {
2858                 features |= NETIF_F_HW_CSUM;
2859                 arg &= ~TUN_F_CSUM;
2860
2861                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2862                         if (arg & TUN_F_TSO_ECN) {
2863                                 features |= NETIF_F_TSO_ECN;
2864                                 arg &= ~TUN_F_TSO_ECN;
2865                         }
2866                         if (arg & TUN_F_TSO4)
2867                                 features |= NETIF_F_TSO;
2868                         if (arg & TUN_F_TSO6)
2869                                 features |= NETIF_F_TSO6;
2870                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2871                 }
2872
2873                 arg &= ~TUN_F_UFO;
2874         }
2875
2876         /* This gives the user a way to test for new features in future by
2877          * trying to set them. */
2878         if (arg)
2879                 return -EINVAL;
2880
2881         tun->set_features = features;
2882         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2883         tun->dev->wanted_features |= features;
2884         netdev_update_features(tun->dev);
2885
2886         return 0;
2887 }
2888
2889 static void tun_detach_filter(struct tun_struct *tun, int n)
2890 {
2891         int i;
2892         struct tun_file *tfile;
2893
2894         for (i = 0; i < n; i++) {
2895                 tfile = rtnl_dereference(tun->tfiles[i]);
2896                 lock_sock(tfile->socket.sk);
2897                 sk_detach_filter(tfile->socket.sk);
2898                 release_sock(tfile->socket.sk);
2899         }
2900
2901         tun->filter_attached = false;
2902 }
2903
2904 static int tun_attach_filter(struct tun_struct *tun)
2905 {
2906         int i, ret = 0;
2907         struct tun_file *tfile;
2908
2909         for (i = 0; i < tun->numqueues; i++) {
2910                 tfile = rtnl_dereference(tun->tfiles[i]);
2911                 lock_sock(tfile->socket.sk);
2912                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2913                 release_sock(tfile->socket.sk);
2914                 if (ret) {
2915                         tun_detach_filter(tun, i);
2916                         return ret;
2917                 }
2918         }
2919
2920         tun->filter_attached = true;
2921         return ret;
2922 }
2923
2924 static void tun_set_sndbuf(struct tun_struct *tun)
2925 {
2926         struct tun_file *tfile;
2927         int i;
2928
2929         for (i = 0; i < tun->numqueues; i++) {
2930                 tfile = rtnl_dereference(tun->tfiles[i]);
2931                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2932         }
2933 }
2934
2935 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2936 {
2937         struct tun_file *tfile = file->private_data;
2938         struct tun_struct *tun;
2939         int ret = 0;
2940
2941         rtnl_lock();
2942
2943         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2944                 tun = tfile->detached;
2945                 if (!tun) {
2946                         ret = -EINVAL;
2947                         goto unlock;
2948                 }
2949                 ret = security_tun_dev_attach_queue(tun->security);
2950                 if (ret < 0)
2951                         goto unlock;
2952                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2953                                  tun->flags & IFF_NAPI_FRAGS, true);
2954         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2955                 tun = rtnl_dereference(tfile->tun);
2956                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2957                         ret = -EINVAL;
2958                 else
2959                         __tun_detach(tfile, false);
2960         } else
2961                 ret = -EINVAL;
2962
2963         if (ret >= 0)
2964                 netdev_state_change(tun->dev);
2965
2966 unlock:
2967         rtnl_unlock();
2968         return ret;
2969 }
2970
2971 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2972                         void __user *data)
2973 {
2974         struct bpf_prog *prog;
2975         int fd;
2976
2977         if (copy_from_user(&fd, data, sizeof(fd)))
2978                 return -EFAULT;
2979
2980         if (fd == -1) {
2981                 prog = NULL;
2982         } else {
2983                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2984                 if (IS_ERR(prog))
2985                         return PTR_ERR(prog);
2986         }
2987
2988         return __tun_set_ebpf(tun, prog_p, prog);
2989 }
2990
2991 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2992 static unsigned char tun_get_addr_len(unsigned short type)
2993 {
2994         switch (type) {
2995         case ARPHRD_IP6GRE:
2996         case ARPHRD_TUNNEL6:
2997                 return sizeof(struct in6_addr);
2998         case ARPHRD_IPGRE:
2999         case ARPHRD_TUNNEL:
3000         case ARPHRD_SIT:
3001                 return 4;
3002         case ARPHRD_ETHER:
3003                 return ETH_ALEN;
3004         case ARPHRD_IEEE802154:
3005         case ARPHRD_IEEE802154_MONITOR:
3006                 return IEEE802154_EXTENDED_ADDR_LEN;
3007         case ARPHRD_PHONET_PIPE:
3008         case ARPHRD_PPP:
3009         case ARPHRD_NONE:
3010                 return 0;
3011         case ARPHRD_6LOWPAN:
3012                 return EUI64_ADDR_LEN;
3013         case ARPHRD_FDDI:
3014                 return FDDI_K_ALEN;
3015         case ARPHRD_HIPPI:
3016                 return HIPPI_ALEN;
3017         case ARPHRD_IEEE802:
3018                 return FC_ALEN;
3019         case ARPHRD_ROSE:
3020                 return ROSE_ADDR_LEN;
3021         case ARPHRD_NETROM:
3022                 return AX25_ADDR_LEN;
3023         case ARPHRD_LOCALTLK:
3024                 return LTALK_ALEN;
3025         default:
3026                 return 0;
3027         }
3028 }
3029
3030 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3031                             unsigned long arg, int ifreq_len)
3032 {
3033         struct tun_file *tfile = file->private_data;
3034         struct net *net = sock_net(&tfile->sk);
3035         struct tun_struct *tun;
3036         void __user* argp = (void __user*)arg;
3037         unsigned int ifindex, carrier;
3038         struct ifreq ifr;
3039         kuid_t owner;
3040         kgid_t group;
3041         int sndbuf;
3042         int vnet_hdr_sz;
3043         int le;
3044         int ret;
3045         bool do_notify = false;
3046
3047         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3048             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3049                 if (copy_from_user(&ifr, argp, ifreq_len))
3050                         return -EFAULT;
3051         } else {
3052                 memset(&ifr, 0, sizeof(ifr));
3053         }
3054         if (cmd == TUNGETFEATURES) {
3055                 /* Currently this just means: "what IFF flags are valid?".
3056                  * This is needed because we never checked for invalid flags on
3057                  * TUNSETIFF.
3058                  */
3059                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3060                                 (unsigned int __user*)argp);
3061         } else if (cmd == TUNSETQUEUE) {
3062                 return tun_set_queue(file, &ifr);
3063         } else if (cmd == SIOCGSKNS) {
3064                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3065                         return -EPERM;
3066                 return open_related_ns(&net->ns, get_net_ns);
3067         }
3068
3069         rtnl_lock();
3070
3071         tun = tun_get(tfile);
3072         if (cmd == TUNSETIFF) {
3073                 ret = -EEXIST;
3074                 if (tun)
3075                         goto unlock;
3076
3077                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3078
3079                 ret = tun_set_iff(net, file, &ifr);
3080
3081                 if (ret)
3082                         goto unlock;
3083
3084                 if (copy_to_user(argp, &ifr, ifreq_len))
3085                         ret = -EFAULT;
3086                 goto unlock;
3087         }
3088         if (cmd == TUNSETIFINDEX) {
3089                 ret = -EPERM;
3090                 if (tun)
3091                         goto unlock;
3092
3093                 ret = -EFAULT;
3094                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3095                         goto unlock;
3096
3097                 ret = 0;
3098                 tfile->ifindex = ifindex;
3099                 goto unlock;
3100         }
3101
3102         ret = -EBADFD;
3103         if (!tun)
3104                 goto unlock;
3105
3106         netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3107
3108         net = dev_net(tun->dev);
3109         ret = 0;
3110         switch (cmd) {
3111         case TUNGETIFF:
3112                 tun_get_iff(tun, &ifr);
3113
3114                 if (tfile->detached)
3115                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3116                 if (!tfile->socket.sk->sk_filter)
3117                         ifr.ifr_flags |= IFF_NOFILTER;
3118
3119                 if (copy_to_user(argp, &ifr, ifreq_len))
3120                         ret = -EFAULT;
3121                 break;
3122
3123         case TUNSETNOCSUM:
3124                 /* Disable/Enable checksum */
3125
3126                 /* [unimplemented] */
3127                 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3128                            arg ? "disabled" : "enabled");
3129                 break;
3130
3131         case TUNSETPERSIST:
3132                 /* Disable/Enable persist mode. Keep an extra reference to the
3133                  * module to prevent the module being unprobed.
3134                  */
3135                 if (arg && !(tun->flags & IFF_PERSIST)) {
3136                         tun->flags |= IFF_PERSIST;
3137                         __module_get(THIS_MODULE);
3138                         do_notify = true;
3139                 }
3140                 if (!arg && (tun->flags & IFF_PERSIST)) {
3141                         tun->flags &= ~IFF_PERSIST;
3142                         module_put(THIS_MODULE);
3143                         do_notify = true;
3144                 }
3145
3146                 netif_info(tun, drv, tun->dev, "persist %s\n",
3147                            arg ? "enabled" : "disabled");
3148                 break;
3149
3150         case TUNSETOWNER:
3151                 /* Set owner of the device */
3152                 owner = make_kuid(current_user_ns(), arg);
3153                 if (!uid_valid(owner)) {
3154                         ret = -EINVAL;
3155                         break;
3156                 }
3157                 tun->owner = owner;
3158                 do_notify = true;
3159                 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3160                            from_kuid(&init_user_ns, tun->owner));
3161                 break;
3162
3163         case TUNSETGROUP:
3164                 /* Set group of the device */
3165                 group = make_kgid(current_user_ns(), arg);
3166                 if (!gid_valid(group)) {
3167                         ret = -EINVAL;
3168                         break;
3169                 }
3170                 tun->group = group;
3171                 do_notify = true;
3172                 netif_info(tun, drv, tun->dev, "group set to %u\n",
3173                            from_kgid(&init_user_ns, tun->group));
3174                 break;
3175
3176         case TUNSETLINK:
3177                 /* Only allow setting the type when the interface is down */
3178                 if (tun->dev->flags & IFF_UP) {
3179                         netif_info(tun, drv, tun->dev,
3180                                    "Linktype set failed because interface is up\n");
3181                         ret = -EBUSY;
3182                 } else {
3183                         ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3184                                                        tun->dev);
3185                         ret = notifier_to_errno(ret);
3186                         if (ret) {
3187                                 netif_info(tun, drv, tun->dev,
3188                                            "Refused to change device type\n");
3189                                 break;
3190                         }
3191                         tun->dev->type = (int) arg;
3192                         tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3193                         netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3194                                    tun->dev->type);
3195                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3196                                                  tun->dev);
3197                 }
3198                 break;
3199
3200         case TUNSETDEBUG:
3201                 tun->msg_enable = (u32)arg;
3202                 break;
3203
3204         case TUNSETOFFLOAD:
3205                 ret = set_offload(tun, arg);
3206                 break;
3207
3208         case TUNSETTXFILTER:
3209                 /* Can be set only for TAPs */
3210                 ret = -EINVAL;
3211                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3212                         break;
3213                 ret = update_filter(&tun->txflt, (void __user *)arg);
3214                 break;
3215
3216         case SIOCGIFHWADDR:
3217                 /* Get hw address */
3218                 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3219                 if (copy_to_user(argp, &ifr, ifreq_len))
3220                         ret = -EFAULT;
3221                 break;
3222
3223         case SIOCSIFHWADDR:
3224                 /* Set hw address */
3225                 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3226                 break;
3227
3228         case TUNGETSNDBUF:
3229                 sndbuf = tfile->socket.sk->sk_sndbuf;
3230                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3231                         ret = -EFAULT;
3232                 break;
3233
3234         case TUNSETSNDBUF:
3235                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3236                         ret = -EFAULT;
3237                         break;
3238                 }
3239                 if (sndbuf <= 0) {
3240                         ret = -EINVAL;
3241                         break;
3242                 }
3243
3244                 tun->sndbuf = sndbuf;
3245                 tun_set_sndbuf(tun);
3246                 break;
3247
3248         case TUNGETVNETHDRSZ:
3249                 vnet_hdr_sz = tun->vnet_hdr_sz;
3250                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3251                         ret = -EFAULT;
3252                 break;
3253
3254         case TUNSETVNETHDRSZ:
3255                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3256                         ret = -EFAULT;
3257                         break;
3258                 }
3259                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3260                         ret = -EINVAL;
3261                         break;
3262                 }
3263
3264                 tun->vnet_hdr_sz = vnet_hdr_sz;
3265                 break;
3266
3267         case TUNGETVNETLE:
3268                 le = !!(tun->flags & TUN_VNET_LE);
3269                 if (put_user(le, (int __user *)argp))
3270                         ret = -EFAULT;
3271                 break;
3272
3273         case TUNSETVNETLE:
3274                 if (get_user(le, (int __user *)argp)) {
3275                         ret = -EFAULT;
3276                         break;
3277                 }
3278                 if (le)
3279                         tun->flags |= TUN_VNET_LE;
3280                 else
3281                         tun->flags &= ~TUN_VNET_LE;
3282                 break;
3283
3284         case TUNGETVNETBE:
3285                 ret = tun_get_vnet_be(tun, argp);
3286                 break;
3287
3288         case TUNSETVNETBE:
3289                 ret = tun_set_vnet_be(tun, argp);
3290                 break;
3291
3292         case TUNATTACHFILTER:
3293                 /* Can be set only for TAPs */
3294                 ret = -EINVAL;
3295                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3296                         break;
3297                 ret = -EFAULT;
3298                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3299                         break;
3300
3301                 ret = tun_attach_filter(tun);
3302                 break;
3303
3304         case TUNDETACHFILTER:
3305                 /* Can be set only for TAPs */
3306                 ret = -EINVAL;
3307                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3308                         break;
3309                 ret = 0;
3310                 tun_detach_filter(tun, tun->numqueues);
3311                 break;
3312
3313         case TUNGETFILTER:
3314                 ret = -EINVAL;
3315                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3316                         break;
3317                 ret = -EFAULT;
3318                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3319                         break;
3320                 ret = 0;
3321                 break;
3322
3323         case TUNSETSTEERINGEBPF:
3324                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3325                 break;
3326
3327         case TUNSETFILTEREBPF:
3328                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3329                 break;
3330
3331         case TUNSETCARRIER:
3332                 ret = -EFAULT;
3333                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3334                         goto unlock;
3335
3336                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3337                 break;
3338
3339         case TUNGETDEVNETNS:
3340                 ret = -EPERM;
3341                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3342                         goto unlock;
3343                 ret = open_related_ns(&net->ns, get_net_ns);
3344                 break;
3345
3346         default:
3347                 ret = -EINVAL;
3348                 break;
3349         }
3350
3351         if (do_notify)
3352                 netdev_state_change(tun->dev);
3353
3354 unlock:
3355         rtnl_unlock();
3356         if (tun)
3357                 tun_put(tun);
3358         return ret;
3359 }
3360
3361 static long tun_chr_ioctl(struct file *file,
3362                           unsigned int cmd, unsigned long arg)
3363 {
3364         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3365 }
3366
3367 #ifdef CONFIG_COMPAT
3368 static long tun_chr_compat_ioctl(struct file *file,
3369                          unsigned int cmd, unsigned long arg)
3370 {
3371         switch (cmd) {
3372         case TUNSETIFF:
3373         case TUNGETIFF:
3374         case TUNSETTXFILTER:
3375         case TUNGETSNDBUF:
3376         case TUNSETSNDBUF:
3377         case SIOCGIFHWADDR:
3378         case SIOCSIFHWADDR:
3379                 arg = (unsigned long)compat_ptr(arg);
3380                 break;
3381         default:
3382                 arg = (compat_ulong_t)arg;
3383                 break;
3384         }
3385
3386         /*
3387          * compat_ifreq is shorter than ifreq, so we must not access beyond
3388          * the end of that structure. All fields that are used in this
3389          * driver are compatible though, we don't need to convert the
3390          * contents.
3391          */
3392         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3393 }
3394 #endif /* CONFIG_COMPAT */
3395
3396 static int tun_chr_fasync(int fd, struct file *file, int on)
3397 {
3398         struct tun_file *tfile = file->private_data;
3399         int ret;
3400
3401         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3402                 goto out;
3403
3404         if (on) {
3405                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3406                 tfile->flags |= TUN_FASYNC;
3407         } else
3408                 tfile->flags &= ~TUN_FASYNC;
3409         ret = 0;
3410 out:
3411         return ret;
3412 }
3413
3414 static int tun_chr_open(struct inode *inode, struct file * file)
3415 {
3416         struct net *net = current->nsproxy->net_ns;
3417         struct tun_file *tfile;
3418
3419         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3420                                             &tun_proto, 0);
3421         if (!tfile)
3422                 return -ENOMEM;
3423         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3424                 sk_free(&tfile->sk);
3425                 return -ENOMEM;
3426         }
3427
3428         mutex_init(&tfile->napi_mutex);
3429         RCU_INIT_POINTER(tfile->tun, NULL);
3430         tfile->flags = 0;
3431         tfile->ifindex = 0;
3432
3433         init_waitqueue_head(&tfile->socket.wq.wait);
3434
3435         tfile->socket.file = file;
3436         tfile->socket.ops = &tun_socket_ops;
3437
3438         sock_init_data(&tfile->socket, &tfile->sk);
3439
3440         tfile->sk.sk_write_space = tun_sock_write_space;
3441         tfile->sk.sk_sndbuf = INT_MAX;
3442
3443         file->private_data = tfile;
3444         INIT_LIST_HEAD(&tfile->next);
3445
3446         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3447
3448         return 0;
3449 }
3450
3451 static int tun_chr_close(struct inode *inode, struct file *file)
3452 {
3453         struct tun_file *tfile = file->private_data;
3454
3455         tun_detach(tfile, true);
3456
3457         return 0;
3458 }
3459
3460 #ifdef CONFIG_PROC_FS
3461 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3462 {
3463         struct tun_file *tfile = file->private_data;
3464         struct tun_struct *tun;
3465         struct ifreq ifr;
3466
3467         memset(&ifr, 0, sizeof(ifr));
3468
3469         rtnl_lock();
3470         tun = tun_get(tfile);
3471         if (tun)
3472                 tun_get_iff(tun, &ifr);
3473         rtnl_unlock();
3474
3475         if (tun)
3476                 tun_put(tun);
3477
3478         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3479 }
3480 #endif
3481
3482 static const struct file_operations tun_fops = {
3483         .owner  = THIS_MODULE,
3484         .llseek = no_llseek,
3485         .read_iter  = tun_chr_read_iter,
3486         .write_iter = tun_chr_write_iter,
3487         .poll   = tun_chr_poll,
3488         .unlocked_ioctl = tun_chr_ioctl,
3489 #ifdef CONFIG_COMPAT
3490         .compat_ioctl = tun_chr_compat_ioctl,
3491 #endif
3492         .open   = tun_chr_open,
3493         .release = tun_chr_close,
3494         .fasync = tun_chr_fasync,
3495 #ifdef CONFIG_PROC_FS
3496         .show_fdinfo = tun_chr_show_fdinfo,
3497 #endif
3498 };
3499
3500 static struct miscdevice tun_miscdev = {
3501         .minor = TUN_MINOR,
3502         .name = "tun",
3503         .nodename = "net/tun",
3504         .fops = &tun_fops,
3505 };
3506
3507 /* ethtool interface */
3508
3509 static void tun_default_link_ksettings(struct net_device *dev,
3510                                        struct ethtool_link_ksettings *cmd)
3511 {
3512         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3513         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3514         cmd->base.speed         = SPEED_10;
3515         cmd->base.duplex        = DUPLEX_FULL;
3516         cmd->base.port          = PORT_TP;
3517         cmd->base.phy_address   = 0;
3518         cmd->base.autoneg       = AUTONEG_DISABLE;
3519 }
3520
3521 static int tun_get_link_ksettings(struct net_device *dev,
3522                                   struct ethtool_link_ksettings *cmd)
3523 {
3524         struct tun_struct *tun = netdev_priv(dev);
3525
3526         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3527         return 0;
3528 }
3529
3530 static int tun_set_link_ksettings(struct net_device *dev,
3531                                   const struct ethtool_link_ksettings *cmd)
3532 {
3533         struct tun_struct *tun = netdev_priv(dev);
3534
3535         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3536         return 0;
3537 }
3538
3539 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3540 {
3541         struct tun_struct *tun = netdev_priv(dev);
3542
3543         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3544         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3545
3546         switch (tun->flags & TUN_TYPE_MASK) {
3547         case IFF_TUN:
3548                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3549                 break;
3550         case IFF_TAP:
3551                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3552                 break;
3553         }
3554 }
3555
3556 static u32 tun_get_msglevel(struct net_device *dev)
3557 {
3558         struct tun_struct *tun = netdev_priv(dev);
3559
3560         return tun->msg_enable;
3561 }
3562
3563 static void tun_set_msglevel(struct net_device *dev, u32 value)
3564 {
3565         struct tun_struct *tun = netdev_priv(dev);
3566
3567         tun->msg_enable = value;
3568 }
3569
3570 static int tun_get_coalesce(struct net_device *dev,
3571                             struct ethtool_coalesce *ec,
3572                             struct kernel_ethtool_coalesce *kernel_coal,
3573                             struct netlink_ext_ack *extack)
3574 {
3575         struct tun_struct *tun = netdev_priv(dev);
3576
3577         ec->rx_max_coalesced_frames = tun->rx_batched;
3578
3579         return 0;
3580 }
3581
3582 static int tun_set_coalesce(struct net_device *dev,
3583                             struct ethtool_coalesce *ec,
3584                             struct kernel_ethtool_coalesce *kernel_coal,
3585                             struct netlink_ext_ack *extack)
3586 {
3587         struct tun_struct *tun = netdev_priv(dev);
3588
3589         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3590                 tun->rx_batched = NAPI_POLL_WEIGHT;
3591         else
3592                 tun->rx_batched = ec->rx_max_coalesced_frames;
3593
3594         return 0;
3595 }
3596
3597 static const struct ethtool_ops tun_ethtool_ops = {
3598         .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3599         .get_drvinfo    = tun_get_drvinfo,
3600         .get_msglevel   = tun_get_msglevel,
3601         .set_msglevel   = tun_set_msglevel,
3602         .get_link       = ethtool_op_get_link,
3603         .get_ts_info    = ethtool_op_get_ts_info,
3604         .get_coalesce   = tun_get_coalesce,
3605         .set_coalesce   = tun_set_coalesce,
3606         .get_link_ksettings = tun_get_link_ksettings,
3607         .set_link_ksettings = tun_set_link_ksettings,
3608 };
3609
3610 static int tun_queue_resize(struct tun_struct *tun)
3611 {
3612         struct net_device *dev = tun->dev;
3613         struct tun_file *tfile;
3614         struct ptr_ring **rings;
3615         int n = tun->numqueues + tun->numdisabled;
3616         int ret, i;
3617
3618         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3619         if (!rings)
3620                 return -ENOMEM;
3621
3622         for (i = 0; i < tun->numqueues; i++) {
3623                 tfile = rtnl_dereference(tun->tfiles[i]);
3624                 rings[i] = &tfile->tx_ring;
3625         }
3626         list_for_each_entry(tfile, &tun->disabled, next)
3627                 rings[i++] = &tfile->tx_ring;
3628
3629         ret = ptr_ring_resize_multiple(rings, n,
3630                                        dev->tx_queue_len, GFP_KERNEL,
3631                                        tun_ptr_free);
3632
3633         kfree(rings);
3634         return ret;
3635 }
3636
3637 static int tun_device_event(struct notifier_block *unused,
3638                             unsigned long event, void *ptr)
3639 {
3640         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3641         struct tun_struct *tun = netdev_priv(dev);
3642         int i;
3643
3644         if (dev->rtnl_link_ops != &tun_link_ops)
3645                 return NOTIFY_DONE;
3646
3647         switch (event) {
3648         case NETDEV_CHANGE_TX_QUEUE_LEN:
3649                 if (tun_queue_resize(tun))
3650                         return NOTIFY_BAD;
3651                 break;
3652         case NETDEV_UP:
3653                 for (i = 0; i < tun->numqueues; i++) {
3654                         struct tun_file *tfile;
3655
3656                         tfile = rtnl_dereference(tun->tfiles[i]);
3657                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3658                 }
3659                 break;
3660         default:
3661                 break;
3662         }
3663
3664         return NOTIFY_DONE;
3665 }
3666
3667 static struct notifier_block tun_notifier_block __read_mostly = {
3668         .notifier_call  = tun_device_event,
3669 };
3670
3671 static int __init tun_init(void)
3672 {
3673         int ret = 0;
3674
3675         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3676
3677         ret = rtnl_link_register(&tun_link_ops);
3678         if (ret) {
3679                 pr_err("Can't register link_ops\n");
3680                 goto err_linkops;
3681         }
3682
3683         ret = misc_register(&tun_miscdev);
3684         if (ret) {
3685                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3686                 goto err_misc;
3687         }
3688
3689         ret = register_netdevice_notifier(&tun_notifier_block);
3690         if (ret) {
3691                 pr_err("Can't register netdevice notifier\n");
3692                 goto err_notifier;
3693         }
3694
3695         return  0;
3696
3697 err_notifier:
3698         misc_deregister(&tun_miscdev);
3699 err_misc:
3700         rtnl_link_unregister(&tun_link_ops);
3701 err_linkops:
3702         return ret;
3703 }
3704
3705 static void tun_cleanup(void)
3706 {
3707         misc_deregister(&tun_miscdev);
3708         rtnl_link_unregister(&tun_link_ops);
3709         unregister_netdevice_notifier(&tun_notifier_block);
3710 }
3711
3712 /* Get an underlying socket object from tun file.  Returns error unless file is
3713  * attached to a device.  The returned object works like a packet socket, it
3714  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3715  * holding a reference to the file for as long as the socket is in use. */
3716 struct socket *tun_get_socket(struct file *file)
3717 {
3718         struct tun_file *tfile;
3719         if (file->f_op != &tun_fops)
3720                 return ERR_PTR(-EINVAL);
3721         tfile = file->private_data;
3722         if (!tfile)
3723                 return ERR_PTR(-EBADFD);
3724         return &tfile->socket;
3725 }
3726 EXPORT_SYMBOL_GPL(tun_get_socket);
3727
3728 struct ptr_ring *tun_get_tx_ring(struct file *file)
3729 {
3730         struct tun_file *tfile;
3731
3732         if (file->f_op != &tun_fops)
3733                 return ERR_PTR(-EINVAL);
3734         tfile = file->private_data;
3735         if (!tfile)
3736                 return ERR_PTR(-EBADFD);
3737         return &tfile->tx_ring;
3738 }
3739 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3740
3741 module_init(tun_init);
3742 module_exit(tun_cleanup);
3743 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3744 MODULE_AUTHOR(DRV_COPYRIGHT);
3745 MODULE_LICENSE("GPL");
3746 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3747 MODULE_ALIAS("devname:net/tun");