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 <linux/seq_file.h>
  66 #include <linux/uio.h>
  67 #include <linux/skb_array.h>
  68 #include <linux/bpf.h>
  69 #include <linux/bpf_trace.h>
  70 #include <linux/mutex.h>
  71
  72 #include <linux/uaccess.h>
  73 #include <linux/proc_fs.h>
  74
  75 static void tun_default_link_ksettings(struct net_device *dev,
  76                                        struct ethtool_link_ksettings *cmd);
  77
  78 /* Uncomment to enable debugging */
  79 /* #define TUN_DEBUG 1 */
  80
  81 #ifdef TUN_DEBUG
  82 static int debug;
  83
  84 #define tun_debug(level, tun, fmt, args...)                     \
  85 do {                                                            \
  86         if (tun->debug)                                         \
  87                 netdev_printk(level, tun->dev, fmt, ##args);    \
  88 } while (0)
  89 #define DBG1(level, fmt, args...)                               \
  90 do {                                                            \
  91         if (debug == 2)                                         \
  92                 printk(level fmt, ##args);                      \
  93 } while (0)
  94 #else
  95 #define tun_debug(level, tun, fmt, args...)                     \
  96 do {                                                            \
  97         if (0)                                                  \
  98                 netdev_printk(level, tun->dev, fmt, ##args);    \
  99 } while (0)
 100 #define DBG1(level, fmt, args...)                               \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 printk(level fmt, ##args);                      \
 104 } while (0)
 105 #endif
 106
 107 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 108
 109 /* TUN device flags */
 110
 111 /* IFF_ATTACH_QUEUE is never stored in device flags,
 112  * overload it to mean fasync when stored there.
 113  */
 114 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 115 /* High bits in flags field are unused. */
 116 #define TUN_VNET_LE     0x80000000
 117 #define TUN_VNET_BE     0x40000000
 118
 119 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 120                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 121
 122 #define GOODCOPY_LEN 128
 123
 124 #define FLT_EXACT_COUNT 8
 125 struct tap_filter {
 126         unsigned int    count;    /* Number of addrs. Zero means disabled */
 127         u32             mask[2];  /* Mask of the hashed addrs */
 128         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 129 };
 130
 131 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 132  * to max number of VCPUs in guest. */
 133 #define MAX_TAP_QUEUES 256
 134 #define MAX_TAP_FLOWS  4096
 135
 136 #define TUN_FLOW_EXPIRE (3 * HZ)
 137
 138 struct tun_pcpu_stats {
 139         u64 rx_packets;
 140         u64 rx_bytes;
 141         u64 tx_packets;
 142         u64 tx_bytes;
 143         struct u64_stats_sync syncp;
 144         u32 rx_dropped;
 145         u32 tx_dropped;
 146         u32 rx_frame_errors;
 147 };
 148
 149 /* A tun_file connects an open character device to a tuntap netdevice. It
 150  * also contains all socket related structures (except sock_fprog and tap_filter)
 151  * to serve as one transmit queue for tuntap device. The sock_fprog and
 152  * tap_filter were kept in tun_struct since they were used for filtering for the
 153  * netdevice not for a specific queue (at least I didn't see the requirement for
 154  * this).
 155  *
 156  * RCU usage:
 157  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 158  * other can only be read while rcu_read_lock or rtnl_lock is held.
 159  */
 160 struct tun_file {
 161         struct sock sk;
 162         struct socket socket;
 163         struct socket_wq wq;
 164         struct tun_struct __rcu *tun;
 165         struct fasync_struct *fasync;
 166         /* only used for fasnyc */
 167         unsigned int flags;
 168         union {
 169                 u16 queue_index;
 170                 unsigned int ifindex;
 171         };
 172         struct napi_struct napi;
 173         bool napi_enabled;
 174         bool napi_frags_enabled;
 175         struct mutex napi_mutex;        /* Protects access to the above napi */
 176         struct list_head next;
 177         struct tun_struct *detached;
 178         struct ptr_ring tx_ring;
 179         struct xdp_rxq_info xdp_rxq;
 180 };
 181
 182 struct tun_page {
 183         struct page *page;
 184         int count;
 185 };
 186
 187 struct tun_flow_entry {
 188         struct hlist_node hash_link;
 189         struct rcu_head rcu;
 190         struct tun_struct *tun;
 191
 192         u32 rxhash;
 193         u32 rps_rxhash;
 194         int queue_index;
 195         unsigned long updated ____cacheline_aligned_in_smp;
 196 };
 197
 198 #define TUN_NUM_FLOW_ENTRIES 1024
 199 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 200
 201 struct tun_prog {
 202         struct rcu_head rcu;
 203         struct bpf_prog *prog;
 204 };
 205
 206 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 207  * device, socket filter, sndbuf and vnet header size were restore when the
 208  * file were attached to a persist device.
 209  */
 210 struct tun_struct {
 211         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 212         unsigned int            numqueues;
 213         unsigned int            flags;
 214         kuid_t                  owner;
 215         kgid_t                  group;
 216
 217         struct net_device       *dev;
 218         netdev_features_t       set_features;
 219 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 220                           NETIF_F_TSO6)
 221
 222         int                     align;
 223         int                     vnet_hdr_sz;
 224         int                     sndbuf;
 225         struct tap_filter       txflt;
 226         struct sock_fprog       fprog;
 227         /* protected by rtnl lock */
 228         bool                    filter_attached;
 229 #ifdef TUN_DEBUG
 230         int debug;
 231 #endif
 232         spinlock_t lock;
 233         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 234         struct timer_list flow_gc_timer;
 235         unsigned long ageing_time;
 236         unsigned int numdisabled;
 237         struct list_head disabled;
 238         void *security;
 239         u32 flow_count;
 240         u32 rx_batched;
 241         struct tun_pcpu_stats __percpu *pcpu_stats;
 242         struct bpf_prog __rcu *xdp_prog;
 243         struct tun_prog __rcu *steering_prog;
 244         struct tun_prog __rcu *filter_prog;
 245         struct ethtool_link_ksettings link_ksettings;
 246 };
 247
 248 struct veth {
 249         __be16 h_vlan_proto;
 250         __be16 h_vlan_TCI;
 251 };
 252
 253 bool tun_is_xdp_frame(void *ptr)
 254 {
 255         return (unsigned long)ptr & TUN_XDP_FLAG;
 256 }
 257 EXPORT_SYMBOL(tun_is_xdp_frame);
 258
 259 void *tun_xdp_to_ptr(void *ptr)
 260 {
 261         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 262 }
 263 EXPORT_SYMBOL(tun_xdp_to_ptr);
 264
 265 void *tun_ptr_to_xdp(void *ptr)
 266 {
 267         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 268 }
 269 EXPORT_SYMBOL(tun_ptr_to_xdp);
 270
 271 static int tun_napi_receive(struct napi_struct *napi, int budget)
 272 {
 273         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 274         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 275         struct sk_buff_head process_queue;
 276         struct sk_buff *skb;
 277         int received = 0;
 278
 279         __skb_queue_head_init(&process_queue);
 280
 281         spin_lock(&queue->lock);
 282         skb_queue_splice_tail_init(queue, &process_queue);
 283         spin_unlock(&queue->lock);
 284
 285         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 286                 napi_gro_receive(napi, skb);
 287                 ++received;
 288         }
 289
 290         if (!skb_queue_empty(&process_queue)) {
 291                 spin_lock(&queue->lock);
 292                 skb_queue_splice(&process_queue, queue);
 293                 spin_unlock(&queue->lock);
 294         }
 295
 296         return received;
 297 }
 298
 299 static int tun_napi_poll(struct napi_struct *napi, int budget)
 300 {
 301         unsigned int received;
 302
 303         received = tun_napi_receive(napi, budget);
 304
 305         if (received < budget)
 306                 napi_complete_done(napi, received);
 307
 308         return received;
 309 }
 310
 311 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 312                           bool napi_en, bool napi_frags)
 313 {
 314         tfile->napi_enabled = napi_en;
 315         tfile->napi_frags_enabled = napi_en && napi_frags;
 316         if (napi_en) {
 317                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 318                                NAPI_POLL_WEIGHT);
 319                 napi_enable(&tfile->napi);
 320         }
 321 }
 322
 323 static void tun_napi_disable(struct tun_file *tfile)
 324 {
 325         if (tfile->napi_enabled)
 326                 napi_disable(&tfile->napi);
 327 }
 328
 329 static void tun_napi_del(struct tun_file *tfile)
 330 {
 331         if (tfile->napi_enabled)
 332                 netif_napi_del(&tfile->napi);
 333 }
 334
 335 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 336 {
 337         return tfile->napi_frags_enabled;
 338 }
 339
 340 #ifdef CONFIG_TUN_VNET_CROSS_LE
 341 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 342 {
 343         return tun->flags & TUN_VNET_BE ? false :
 344                 virtio_legacy_is_little_endian();
 345 }
 346
 347 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 348 {
 349         int be = !!(tun->flags & TUN_VNET_BE);
 350
 351         if (put_user(be, argp))
 352                 return -EFAULT;
 353
 354         return 0;
 355 }
 356
 357 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 358 {
 359         int be;
 360
 361         if (get_user(be, argp))
 362                 return -EFAULT;
 363
 364         if (be)
 365                 tun->flags |= TUN_VNET_BE;
 366         else
 367                 tun->flags &= ~TUN_VNET_BE;
 368
 369         return 0;
 370 }
 371 #else
 372 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 373 {
 374         return virtio_legacy_is_little_endian();
 375 }
 376
 377 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 378 {
 379         return -EINVAL;
 380 }
 381
 382 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 383 {
 384         return -EINVAL;
 385 }
 386 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 387
 388 static inline bool tun_is_little_endian(struct tun_struct *tun)
 389 {
 390         return tun->flags & TUN_VNET_LE ||
 391                 tun_legacy_is_little_endian(tun);
 392 }
 393
 394 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 395 {
 396         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 397 }
 398
 399 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 400 {
 401         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 402 }
 403
 404 static inline u32 tun_hashfn(u32 rxhash)
 405 {
 406         return rxhash & TUN_MASK_FLOW_ENTRIES;
 407 }
 408
 409 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 410 {
 411         struct tun_flow_entry *e;
 412
 413         hlist_for_each_entry_rcu(e, head, hash_link) {
 414                 if (e->rxhash == rxhash)
 415                         return e;
 416         }
 417         return NULL;
 418 }
 419
 420 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 421                                               struct hlist_head *head,
 422                                               u32 rxhash, u16 queue_index)
 423 {
 424         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 425
 426         if (e) {
 427                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 428                           rxhash, queue_index);
 429                 e->updated = jiffies;
 430                 e->rxhash = rxhash;
 431                 e->rps_rxhash = 0;
 432                 e->queue_index = queue_index;
 433                 e->tun = tun;
 434                 hlist_add_head_rcu(&e->hash_link, head);
 435                 ++tun->flow_count;
 436         }
 437         return e;
 438 }
 439
 440 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 441 {
 442         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 443                   e->rxhash, e->queue_index);
 444         hlist_del_rcu(&e->hash_link);
 445         kfree_rcu(e, rcu);
 446         --tun->flow_count;
 447 }
 448
 449 static void tun_flow_flush(struct tun_struct *tun)
 450 {
 451         int i;
 452
 453         spin_lock_bh(&tun->lock);
 454         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 455                 struct tun_flow_entry *e;
 456                 struct hlist_node *n;
 457
 458                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 459                         tun_flow_delete(tun, e);
 460         }
 461         spin_unlock_bh(&tun->lock);
 462 }
 463
 464 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 465 {
 466         int i;
 467
 468         spin_lock_bh(&tun->lock);
 469         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 470                 struct tun_flow_entry *e;
 471                 struct hlist_node *n;
 472
 473                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 474                         if (e->queue_index == queue_index)
 475                                 tun_flow_delete(tun, e);
 476                 }
 477         }
 478         spin_unlock_bh(&tun->lock);
 479 }
 480
 481 static void tun_flow_cleanup(struct timer_list *t)
 482 {
 483         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 484         unsigned long delay = tun->ageing_time;
 485         unsigned long next_timer = jiffies + delay;
 486         unsigned long count = 0;
 487         int i;
 488
 489         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 490
 491         spin_lock(&tun->lock);
 492         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 493                 struct tun_flow_entry *e;
 494                 struct hlist_node *n;
 495
 496                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 497                         unsigned long this_timer;
 498
 499                         this_timer = e->updated + delay;
 500                         if (time_before_eq(this_timer, jiffies)) {
 501                                 tun_flow_delete(tun, e);
 502                                 continue;
 503                         }
 504                         count++;
 505                         if (time_before(this_timer, next_timer))
 506                                 next_timer = this_timer;
 507                 }
 508         }
 509
 510         if (count)
 511                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 512         spin_unlock(&tun->lock);
 513 }
 514
 515 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 516                             struct tun_file *tfile)
 517 {
 518         struct hlist_head *head;
 519         struct tun_flow_entry *e;
 520         unsigned long delay = tun->ageing_time;
 521         u16 queue_index = tfile->queue_index;
 522
 523         head = &tun->flows[tun_hashfn(rxhash)];
 524
 525         rcu_read_lock();
 526
 527         e = tun_flow_find(head, rxhash);
 528         if (likely(e)) {
 529                 /* TODO: keep queueing to old queue until it's empty? */
 530                 if (e->queue_index != queue_index)
 531                         e->queue_index = queue_index;
 532                 if (e->updated != jiffies)
 533                         e->updated = jiffies;
 534                 sock_rps_record_flow_hash(e->rps_rxhash);
 535         } else {
 536                 spin_lock_bh(&tun->lock);
 537                 if (!tun_flow_find(head, rxhash) &&
 538                     tun->flow_count < MAX_TAP_FLOWS)
 539                         tun_flow_create(tun, head, rxhash, queue_index);
 540
 541                 if (!timer_pending(&tun->flow_gc_timer))
 542                         mod_timer(&tun->flow_gc_timer,
 543                                   round_jiffies_up(jiffies + delay));
 544                 spin_unlock_bh(&tun->lock);
 545         }
 546
 547         rcu_read_unlock();
 548 }
 549
 550 /**
 551  * Save the hash received in the stack receive path and update the
 552  * flow_hash table accordingly.
 553  */
 554 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 555 {
 556         if (unlikely(e->rps_rxhash != hash))
 557                 e->rps_rxhash = hash;
 558 }
 559
 560 /* We try to identify a flow through its rxhash. The reason that
 561  * we do not check rxq no. is because some cards(e.g 82599), chooses
 562  * the rxq based on the txq where the last packet of the flow comes. As
 563  * the userspace application move between processors, we may get a
 564  * different rxq no. here.
 565  */
 566 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 567 {
 568         struct tun_flow_entry *e;
 569         u32 txq = 0;
 570         u32 numqueues = 0;
 571
 572         numqueues = READ_ONCE(tun->numqueues);
 573
 574         txq = __skb_get_hash_symmetric(skb);
 575         e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 576         if (e) {
 577                 tun_flow_save_rps_rxhash(e, txq);
 578                 txq = e->queue_index;
 579         } else {
 580                 /* use multiply and shift instead of expensive divide */
 581                 txq = ((u64)txq * numqueues) >> 32;
 582         }
 583
 584         return txq;
 585 }
 586
 587 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 588 {
 589         struct tun_prog *prog;
 590         u32 numqueues;
 591         u16 ret = 0;
 592
 593         numqueues = READ_ONCE(tun->numqueues);
 594         if (!numqueues)
 595                 return 0;
 596
 597         prog = rcu_dereference(tun->steering_prog);
 598         if (prog)
 599                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 600
 601         return ret % numqueues;
 602 }
 603
 604 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 605                             struct net_device *sb_dev)
 606 {
 607         struct tun_struct *tun = netdev_priv(dev);
 608         u16 ret;
 609
 610         rcu_read_lock();
 611         if (rcu_dereference(tun->steering_prog))
 612                 ret = tun_ebpf_select_queue(tun, skb);
 613         else
 614                 ret = tun_automq_select_queue(tun, skb);
 615         rcu_read_unlock();
 616
 617         return ret;
 618 }
 619
 620 static inline bool tun_not_capable(struct tun_struct *tun)
 621 {
 622         const struct cred *cred = current_cred();
 623         struct net *net = dev_net(tun->dev);
 624
 625         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 626                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 627                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 628 }
 629
 630 static void tun_set_real_num_queues(struct tun_struct *tun)
 631 {
 632         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 633         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 634 }
 635
 636 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 637 {
 638         tfile->detached = tun;
 639         list_add_tail(&tfile->next, &tun->disabled);
 640         ++tun->numdisabled;
 641 }
 642
 643 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 644 {
 645         struct tun_struct *tun = tfile->detached;
 646
 647         tfile->detached = NULL;
 648         list_del_init(&tfile->next);
 649         --tun->numdisabled;
 650         return tun;
 651 }
 652
 653 void tun_ptr_free(void *ptr)
 654 {
 655         if (!ptr)
 656                 return;
 657         if (tun_is_xdp_frame(ptr)) {
 658                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 659
 660                 xdp_return_frame(xdpf);
 661         } else {
 662                 __skb_array_destroy_skb(ptr);
 663         }
 664 }
 665 EXPORT_SYMBOL_GPL(tun_ptr_free);
 666
 667 static void tun_queue_purge(struct tun_file *tfile)
 668 {
 669         void *ptr;
 670
 671         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 672                 tun_ptr_free(ptr);
 673
 674         skb_queue_purge(&tfile->sk.sk_write_queue);
 675         skb_queue_purge(&tfile->sk.sk_error_queue);
 676 }
 677
 678 static void __tun_detach(struct tun_file *tfile, bool clean)
 679 {
 680         struct tun_file *ntfile;
 681         struct tun_struct *tun;
 682
 683         tun = rtnl_dereference(tfile->tun);
 684
 685         if (tun && clean) {
 686                 tun_napi_disable(tfile);
 687                 tun_napi_del(tfile);
 688         }
 689
 690         if (tun && !tfile->detached) {
 691                 u16 index = tfile->queue_index;
 692                 BUG_ON(index >= tun->numqueues);
 693
 694                 rcu_assign_pointer(tun->tfiles[index],
 695                                    tun->tfiles[tun->numqueues - 1]);
 696                 ntfile = rtnl_dereference(tun->tfiles[index]);
 697                 ntfile->queue_index = index;
 698                 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 699                                    NULL);
 700
 701                 --tun->numqueues;
 702                 if (clean) {
 703                         RCU_INIT_POINTER(tfile->tun, NULL);
 704                         sock_put(&tfile->sk);
 705                 } else
 706                         tun_disable_queue(tun, tfile);
 707
 708                 synchronize_net();
 709                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 710                 /* Drop read queue */
 711                 tun_queue_purge(tfile);
 712                 tun_set_real_num_queues(tun);
 713         } else if (tfile->detached && clean) {
 714                 tun = tun_enable_queue(tfile);
 715                 sock_put(&tfile->sk);
 716         }
 717
 718         if (clean) {
 719                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 720                         netif_carrier_off(tun->dev);
 721
 722                         if (!(tun->flags & IFF_PERSIST) &&
 723                             tun->dev->reg_state == NETREG_REGISTERED)
 724                                 unregister_netdevice(tun->dev);
 725                 }
 726                 if (tun)
 727                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 728                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 729                 sock_put(&tfile->sk);
 730         }
 731 }
 732
 733 static void tun_detach(struct tun_file *tfile, bool clean)
 734 {
 735         struct tun_struct *tun;
 736         struct net_device *dev;
 737
 738         rtnl_lock();
 739         tun = rtnl_dereference(tfile->tun);
 740         dev = tun ? tun->dev : NULL;
 741         __tun_detach(tfile, clean);
 742         if (dev)
 743                 netdev_state_change(dev);
 744         rtnl_unlock();
 745 }
 746
 747 static void tun_detach_all(struct net_device *dev)
 748 {
 749         struct tun_struct *tun = netdev_priv(dev);
 750         struct tun_file *tfile, *tmp;
 751         int i, n = tun->numqueues;
 752
 753         for (i = 0; i < n; i++) {
 754                 tfile = rtnl_dereference(tun->tfiles[i]);
 755                 BUG_ON(!tfile);
 756                 tun_napi_disable(tfile);
 757                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 758                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 759                 RCU_INIT_POINTER(tfile->tun, NULL);
 760                 --tun->numqueues;
 761         }
 762         list_for_each_entry(tfile, &tun->disabled, next) {
 763                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 764                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 765                 RCU_INIT_POINTER(tfile->tun, NULL);
 766         }
 767         BUG_ON(tun->numqueues != 0);
 768
 769         synchronize_net();
 770         for (i = 0; i < n; i++) {
 771                 tfile = rtnl_dereference(tun->tfiles[i]);
 772                 tun_napi_del(tfile);
 773                 /* Drop read queue */
 774                 tun_queue_purge(tfile);
 775                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 776                 sock_put(&tfile->sk);
 777         }
 778         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 779                 tun_enable_queue(tfile);
 780                 tun_queue_purge(tfile);
 781                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 782                 sock_put(&tfile->sk);
 783         }
 784         BUG_ON(tun->numdisabled != 0);
 785
 786         if (tun->flags & IFF_PERSIST)
 787                 module_put(THIS_MODULE);
 788 }
 789
 790 static int tun_attach(struct tun_struct *tun, struct file *file,
 791                       bool skip_filter, bool napi, bool napi_frags)
 792 {
 793         struct tun_file *tfile = file->private_data;
 794         struct net_device *dev = tun->dev;
 795         int err;
 796
 797         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 798         if (err < 0)
 799                 goto out;
 800
 801         err = -EINVAL;
 802         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 803                 goto out;
 804
 805         err = -EBUSY;
 806         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 807                 goto out;
 808
 809         err = -E2BIG;
 810         if (!tfile->detached &&
 811             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 812                 goto out;
 813
 814         err = 0;
 815
 816         /* Re-attach the filter to persist device */
 817         if (!skip_filter && (tun->filter_attached == true)) {
 818                 lock_sock(tfile->socket.sk);
 819                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 820                 release_sock(tfile->socket.sk);
 821                 if (!err)
 822                         goto out;
 823         }
 824
 825         if (!tfile->detached &&
 826             ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 827                             GFP_KERNEL, tun_ptr_free)) {
 828                 err = -ENOMEM;
 829                 goto out;
 830         }
 831
 832         tfile->queue_index = tun->numqueues;
 833         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 834
 835         if (tfile->detached) {
 836                 /* Re-attach detached tfile, updating XDP queue_index */
 837                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 838
 839                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 840                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 841         } else {
 842                 /* Setup XDP RX-queue info, for new tfile getting attached */
 843                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 844                                        tun->dev, tfile->queue_index);
 845                 if (err < 0)
 846                         goto out;
 847                 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 848                                                  MEM_TYPE_PAGE_SHARED, NULL);
 849                 if (err < 0) {
 850                         xdp_rxq_info_unreg(&tfile->xdp_rxq);
 851                         goto out;
 852                 }
 853                 err = 0;
 854         }
 855
 856         if (tfile->detached) {
 857                 tun_enable_queue(tfile);
 858         } else {
 859                 sock_hold(&tfile->sk);
 860                 tun_napi_init(tun, tfile, napi, napi_frags);
 861         }
 862
 863         if (rtnl_dereference(tun->xdp_prog))
 864                 sock_set_flag(&tfile->sk, SOCK_XDP);
 865
 866         /* device is allowed to go away first, so no need to hold extra
 867          * refcnt.
 868          */
 869
 870         /* Publish tfile->tun and tun->tfiles only after we've fully
 871          * initialized tfile; otherwise we risk using half-initialized
 872          * object.
 873          */
 874         rcu_assign_pointer(tfile->tun, tun);
 875         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 876         tun->numqueues++;
 877         tun_set_real_num_queues(tun);
 878 out:
 879         return err;
 880 }
 881
 882 static struct tun_struct *tun_get(struct tun_file *tfile)
 883 {
 884         struct tun_struct *tun;
 885
 886         rcu_read_lock();
 887         tun = rcu_dereference(tfile->tun);
 888         if (tun)
 889                 dev_hold(tun->dev);
 890         rcu_read_unlock();
 891
 892         return tun;
 893 }
 894
 895 static void tun_put(struct tun_struct *tun)
 896 {
 897         dev_put(tun->dev);
 898 }
 899
 900 /* TAP filtering */
 901 static void addr_hash_set(u32 *mask, const u8 *addr)
 902 {
 903         int n = ether_crc(ETH_ALEN, addr) >> 26;
 904         mask[n >> 5] |= (1 << (n & 31));
 905 }
 906
 907 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 908 {
 909         int n = ether_crc(ETH_ALEN, addr) >> 26;
 910         return mask[n >> 5] & (1 << (n & 31));
 911 }
 912
 913 static int update_filter(struct tap_filter *filter, void __user *arg)
 914 {
 915         struct { u8 u[ETH_ALEN]; } *addr;
 916         struct tun_filter uf;
 917         int err, alen, n, nexact;
 918
 919         if (copy_from_user(&uf, arg, sizeof(uf)))
 920                 return -EFAULT;
 921
 922         if (!uf.count) {
 923                 /* Disabled */
 924                 filter->count = 0;
 925                 return 0;
 926         }
 927
 928         alen = ETH_ALEN * uf.count;
 929         addr = memdup_user(arg + sizeof(uf), alen);
 930         if (IS_ERR(addr))
 931                 return PTR_ERR(addr);
 932
 933         /* The filter is updated without holding any locks. Which is
 934          * perfectly safe. We disable it first and in the worst
 935          * case we'll accept a few undesired packets. */
 936         filter->count = 0;
 937         wmb();
 938
 939         /* Use first set of addresses as an exact filter */
 940         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 941                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 942
 943         nexact = n;
 944
 945         /* Remaining multicast addresses are hashed,
 946          * unicast will leave the filter disabled. */
 947         memset(filter->mask, 0, sizeof(filter->mask));
 948         for (; n < uf.count; n++) {
 949                 if (!is_multicast_ether_addr(addr[n].u)) {
 950                         err = 0; /* no filter */
 951                         goto free_addr;
 952                 }
 953                 addr_hash_set(filter->mask, addr[n].u);
 954         }
 955
 956         /* For ALLMULTI just set the mask to all ones.
 957          * This overrides the mask populated above. */
 958         if ((uf.flags & TUN_FLT_ALLMULTI))
 959                 memset(filter->mask, ~0, sizeof(filter->mask));
 960
 961         /* Now enable the filter */
 962         wmb();
 963         filter->count = nexact;
 964
 965         /* Return the number of exact filters */
 966         err = nexact;
 967 free_addr:
 968         kfree(addr);
 969         return err;
 970 }
 971
 972 /* Returns: 0 - drop, !=0 - accept */
 973 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 974 {
 975         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 976          * at this point. */
 977         struct ethhdr *eh = (struct ethhdr *) skb->data;
 978         int i;
 979
 980         /* Exact match */
 981         for (i = 0; i < filter->count; i++)
 982                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 983                         return 1;
 984
 985         /* Inexact match (multicast only) */
 986         if (is_multicast_ether_addr(eh->h_dest))
 987                 return addr_hash_test(filter->mask, eh->h_dest);
 988
 989         return 0;
 990 }
 991
 992 /*
 993  * Checks whether the packet is accepted or not.
 994  * Returns: 0 - drop, !=0 - accept
 995  */
 996 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 997 {
 998         if (!filter->count)
 999                 return 1;
1000
1001         return run_filter(filter, skb);
1002 }
1003
1004 /* Network device part of the driver */
1005
1006 static const struct ethtool_ops tun_ethtool_ops;
1007
1008 /* Net device detach from fd. */
1009 static void tun_net_uninit(struct net_device *dev)
1010 {
1011         tun_detach_all(dev);
1012 }
1013
1014 /* Net device open. */
1015 static int tun_net_open(struct net_device *dev)
1016 {
1017         netif_tx_start_all_queues(dev);
1018
1019         return 0;
1020 }
1021
1022 /* Net device close. */
1023 static int tun_net_close(struct net_device *dev)
1024 {
1025         netif_tx_stop_all_queues(dev);
1026         return 0;
1027 }
1028
1029 /* Net device start xmit */
1030 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1031 {
1032 #ifdef CONFIG_RPS
1033         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1034                 /* Select queue was not called for the skbuff, so we extract the
1035                  * RPS hash and save it into the flow_table here.
1036                  */
1037                 struct tun_flow_entry *e;
1038                 __u32 rxhash;
1039
1040                 rxhash = __skb_get_hash_symmetric(skb);
1041                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1042                 if (e)
1043                         tun_flow_save_rps_rxhash(e, rxhash);
1044         }
1045 #endif
1046 }
1047
1048 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1049                                     struct sk_buff *skb,
1050                                     int len)
1051 {
1052         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1053
1054         if (prog)
1055                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1056
1057         return len;
1058 }
1059
1060 /* Net device start xmit */
1061 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1062 {
1063         struct tun_struct *tun = netdev_priv(dev);
1064         int txq = skb->queue_mapping;
1065         struct tun_file *tfile;
1066         int len = skb->len;
1067
1068         rcu_read_lock();
1069         tfile = rcu_dereference(tun->tfiles[txq]);
1070
1071         /* Drop packet if interface is not attached */
1072         if (!tfile)
1073                 goto drop;
1074
1075         if (!rcu_dereference(tun->steering_prog))
1076                 tun_automq_xmit(tun, skb);
1077
1078         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1079
1080         BUG_ON(!tfile);
1081
1082         /* Drop if the filter does not like it.
1083          * This is a noop if the filter is disabled.
1084          * Filter can be enabled only for the TAP devices. */
1085         if (!check_filter(&tun->txflt, skb))
1086                 goto drop;
1087
1088         if (tfile->socket.sk->sk_filter &&
1089             sk_filter(tfile->socket.sk, skb))
1090                 goto drop;
1091
1092         len = run_ebpf_filter(tun, skb, len);
1093         if (len == 0 || pskb_trim(skb, len))
1094                 goto drop;
1095
1096         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1097                 goto drop;
1098
1099         skb_tx_timestamp(skb);
1100
1101         /* Orphan the skb - required as we might hang on to it
1102          * for indefinite time.
1103          */
1104         skb_orphan(skb);
1105
1106         nf_reset(skb);
1107
1108         if (ptr_ring_produce(&tfile->tx_ring, skb))
1109                 goto drop;
1110
1111         /* Notify and wake up reader process */
1112         if (tfile->flags & TUN_FASYNC)
1113                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1114         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1115
1116         rcu_read_unlock();
1117         return NETDEV_TX_OK;
1118
1119 drop:
1120         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1121         skb_tx_error(skb);
1122         kfree_skb(skb);
1123         rcu_read_unlock();
1124         return NET_XMIT_DROP;
1125 }
1126
1127 static void tun_net_mclist(struct net_device *dev)
1128 {
1129         /*
1130          * This callback is supposed to deal with mc filter in
1131          * _rx_ path and has nothing to do with the _tx_ path.
1132          * In rx path we always accept everything userspace gives us.
1133          */
1134 }
1135
1136 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1137         netdev_features_t features)
1138 {
1139         struct tun_struct *tun = netdev_priv(dev);
1140
1141         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1142 }
1143
1144 static void tun_set_headroom(struct net_device *dev, int new_hr)
1145 {
1146         struct tun_struct *tun = netdev_priv(dev);
1147
1148         if (new_hr < NET_SKB_PAD)
1149                 new_hr = NET_SKB_PAD;
1150
1151         tun->align = new_hr;
1152 }
1153
1154 static void
1155 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1156 {
1157         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1158         struct tun_struct *tun = netdev_priv(dev);
1159         struct tun_pcpu_stats *p;
1160         int i;
1161
1162         for_each_possible_cpu(i) {
1163                 u64 rxpackets, rxbytes, txpackets, txbytes;
1164                 unsigned int start;
1165
1166                 p = per_cpu_ptr(tun->pcpu_stats, i);
1167                 do {
1168                         start = u64_stats_fetch_begin(&p->syncp);
1169                         rxpackets       = p->rx_packets;
1170                         rxbytes         = p->rx_bytes;
1171                         txpackets       = p->tx_packets;
1172                         txbytes         = p->tx_bytes;
1173                 } while (u64_stats_fetch_retry(&p->syncp, start));
1174
1175                 stats->rx_packets       += rxpackets;
1176                 stats->rx_bytes         += rxbytes;
1177                 stats->tx_packets       += txpackets;
1178                 stats->tx_bytes         += txbytes;
1179
1180                 /* u32 counters */
1181                 rx_dropped      += p->rx_dropped;
1182                 rx_frame_errors += p->rx_frame_errors;
1183                 tx_dropped      += p->tx_dropped;
1184         }
1185         stats->rx_dropped  = rx_dropped;
1186         stats->rx_frame_errors = rx_frame_errors;
1187         stats->tx_dropped = tx_dropped;
1188 }
1189
1190 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1191                        struct netlink_ext_ack *extack)
1192 {
1193         struct tun_struct *tun = netdev_priv(dev);
1194         struct tun_file *tfile;
1195         struct bpf_prog *old_prog;
1196         int i;
1197
1198         old_prog = rtnl_dereference(tun->xdp_prog);
1199         rcu_assign_pointer(tun->xdp_prog, prog);
1200         if (old_prog)
1201                 bpf_prog_put(old_prog);
1202
1203         for (i = 0; i < tun->numqueues; i++) {
1204                 tfile = rtnl_dereference(tun->tfiles[i]);
1205                 if (prog)
1206                         sock_set_flag(&tfile->sk, SOCK_XDP);
1207                 else
1208                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1209         }
1210         list_for_each_entry(tfile, &tun->disabled, next) {
1211                 if (prog)
1212                         sock_set_flag(&tfile->sk, SOCK_XDP);
1213                 else
1214                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1215         }
1216
1217         return 0;
1218 }
1219
1220 static u32 tun_xdp_query(struct net_device *dev)
1221 {
1222         struct tun_struct *tun = netdev_priv(dev);
1223         const struct bpf_prog *xdp_prog;
1224
1225         xdp_prog = rtnl_dereference(tun->xdp_prog);
1226         if (xdp_prog)
1227                 return xdp_prog->aux->id;
1228
1229         return 0;
1230 }
1231
1232 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1233 {
1234         switch (xdp->command) {
1235         case XDP_SETUP_PROG:
1236                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1237         case XDP_QUERY_PROG:
1238                 xdp->prog_id = tun_xdp_query(dev);
1239                 return 0;
1240         default:
1241                 return -EINVAL;
1242         }
1243 }
1244
1245 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1246 {
1247         if (new_carrier) {
1248                 struct tun_struct *tun = netdev_priv(dev);
1249
1250                 if (!tun->numqueues)
1251                         return -EPERM;
1252
1253                 netif_carrier_on(dev);
1254         } else {
1255                 netif_carrier_off(dev);
1256         }
1257         return 0;
1258 }
1259
1260 static const struct net_device_ops tun_netdev_ops = {
1261         .ndo_uninit             = tun_net_uninit,
1262         .ndo_open               = tun_net_open,
1263         .ndo_stop               = tun_net_close,
1264         .ndo_start_xmit         = tun_net_xmit,
1265         .ndo_fix_features       = tun_net_fix_features,
1266         .ndo_select_queue       = tun_select_queue,
1267         .ndo_set_rx_headroom    = tun_set_headroom,
1268         .ndo_get_stats64        = tun_net_get_stats64,
1269         .ndo_change_carrier     = tun_net_change_carrier,
1270 };
1271
1272 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1273 {
1274         /* Notify and wake up reader process */
1275         if (tfile->flags & TUN_FASYNC)
1276                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1277         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1278 }
1279
1280 static int tun_xdp_xmit(struct net_device *dev, int n,
1281                         struct xdp_frame **frames, u32 flags)
1282 {
1283         struct tun_struct *tun = netdev_priv(dev);
1284         struct tun_file *tfile;
1285         u32 numqueues;
1286         int drops = 0;
1287         int cnt = n;
1288         int i;
1289
1290         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1291                 return -EINVAL;
1292
1293         rcu_read_lock();
1294
1295 resample:
1296         numqueues = READ_ONCE(tun->numqueues);
1297         if (!numqueues) {
1298                 rcu_read_unlock();
1299                 return -ENXIO; /* Caller will free/return all frames */
1300         }
1301
1302         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1303                                             numqueues]);
1304         if (unlikely(!tfile))
1305                 goto resample;
1306
1307         spin_lock(&tfile->tx_ring.producer_lock);
1308         for (i = 0; i < n; i++) {
1309                 struct xdp_frame *xdp = frames[i];
1310                 /* Encode the XDP flag into lowest bit for consumer to differ
1311                  * XDP buffer from sk_buff.
1312                  */
1313                 void *frame = tun_xdp_to_ptr(xdp);
1314
1315                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1316                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1317                         xdp_return_frame_rx_napi(xdp);
1318                         drops++;
1319                 }
1320         }
1321         spin_unlock(&tfile->tx_ring.producer_lock);
1322
1323         if (flags & XDP_XMIT_FLUSH)
1324                 __tun_xdp_flush_tfile(tfile);
1325
1326         rcu_read_unlock();
1327         return cnt - drops;
1328 }
1329
1330 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1331 {
1332         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1333
1334         if (unlikely(!frame))
1335                 return -EOVERFLOW;
1336
1337         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1338 }
1339
1340 static const struct net_device_ops tap_netdev_ops = {
1341         .ndo_uninit             = tun_net_uninit,
1342         .ndo_open               = tun_net_open,
1343         .ndo_stop               = tun_net_close,
1344         .ndo_start_xmit         = tun_net_xmit,
1345         .ndo_fix_features       = tun_net_fix_features,
1346         .ndo_set_rx_mode        = tun_net_mclist,
1347         .ndo_set_mac_address    = eth_mac_addr,
1348         .ndo_validate_addr      = eth_validate_addr,
1349         .ndo_select_queue       = tun_select_queue,
1350         .ndo_features_check     = passthru_features_check,
1351         .ndo_set_rx_headroom    = tun_set_headroom,
1352         .ndo_get_stats64        = tun_net_get_stats64,
1353         .ndo_bpf                = tun_xdp,
1354         .ndo_xdp_xmit           = tun_xdp_xmit,
1355         .ndo_change_carrier     = tun_net_change_carrier,
1356 };
1357
1358 static void tun_flow_init(struct tun_struct *tun)
1359 {
1360         int i;
1361
1362         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1363                 INIT_HLIST_HEAD(&tun->flows[i]);
1364
1365         tun->ageing_time = TUN_FLOW_EXPIRE;
1366         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1367         mod_timer(&tun->flow_gc_timer,
1368                   round_jiffies_up(jiffies + tun->ageing_time));
1369 }
1370
1371 static void tun_flow_uninit(struct tun_struct *tun)
1372 {
1373         del_timer_sync(&tun->flow_gc_timer);
1374         tun_flow_flush(tun);
1375 }
1376
1377 #define MIN_MTU 68
1378 #define MAX_MTU 65535
1379
1380 /* Initialize net device. */
1381 static void tun_net_init(struct net_device *dev)
1382 {
1383         struct tun_struct *tun = netdev_priv(dev);
1384
1385         switch (tun->flags & TUN_TYPE_MASK) {
1386         case IFF_TUN:
1387                 dev->netdev_ops = &tun_netdev_ops;
1388
1389                 /* Point-to-Point TUN Device */
1390                 dev->hard_header_len = 0;
1391                 dev->addr_len = 0;
1392                 dev->mtu = 1500;
1393
1394                 /* Zero header length */
1395                 dev->type = ARPHRD_NONE;
1396                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1397                 break;
1398
1399         case IFF_TAP:
1400                 dev->netdev_ops = &tap_netdev_ops;
1401                 /* Ethernet TAP Device */
1402                 ether_setup(dev);
1403                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1404                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1405
1406                 eth_hw_addr_random(dev);
1407
1408                 break;
1409         }
1410
1411         dev->min_mtu = MIN_MTU;
1412         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1413 }
1414
1415 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1416 {
1417         struct sock *sk = tfile->socket.sk;
1418
1419         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1420 }
1421
1422 /* Character device part */
1423
1424 /* Poll */
1425 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1426 {
1427         struct tun_file *tfile = file->private_data;
1428         struct tun_struct *tun = tun_get(tfile);
1429         struct sock *sk;
1430         __poll_t mask = 0;
1431
1432         if (!tun)
1433                 return EPOLLERR;
1434
1435         sk = tfile->socket.sk;
1436
1437         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1438
1439         poll_wait(file, sk_sleep(sk), wait);
1440
1441         if (!ptr_ring_empty(&tfile->tx_ring))
1442                 mask |= EPOLLIN | EPOLLRDNORM;
1443
1444         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1445          * guarantee EPOLLOUT to be raised by either here or
1446          * tun_sock_write_space(). Then process could get notification
1447          * after it writes to a down device and meets -EIO.
1448          */
1449         if (tun_sock_writeable(tun, tfile) ||
1450             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1451              tun_sock_writeable(tun, tfile)))
1452                 mask |= EPOLLOUT | EPOLLWRNORM;
1453
1454         if (tun->dev->reg_state != NETREG_REGISTERED)
1455                 mask = EPOLLERR;
1456
1457         tun_put(tun);
1458         return mask;
1459 }
1460
1461 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1462                                             size_t len,
1463                                             const struct iov_iter *it)
1464 {
1465         struct sk_buff *skb;
1466         size_t linear;
1467         int err;
1468         int i;
1469
1470         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1471                 return ERR_PTR(-ENOMEM);
1472
1473         local_bh_disable();
1474         skb = napi_get_frags(&tfile->napi);
1475         local_bh_enable();
1476         if (!skb)
1477                 return ERR_PTR(-ENOMEM);
1478
1479         linear = iov_iter_single_seg_count(it);
1480         err = __skb_grow(skb, linear);
1481         if (err)
1482                 goto free;
1483
1484         skb->len = len;
1485         skb->data_len = len - linear;
1486         skb->truesize += skb->data_len;
1487
1488         for (i = 1; i < it->nr_segs; i++) {
1489                 size_t fragsz = it->iov[i].iov_len;
1490                 struct page *page;
1491                 void *frag;
1492
1493                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1494                         err = -EINVAL;
1495                         goto free;
1496                 }
1497                 frag = netdev_alloc_frag(fragsz);
1498                 if (!frag) {
1499                         err = -ENOMEM;
1500                         goto free;
1501                 }
1502                 page = virt_to_head_page(frag);
1503                 skb_fill_page_desc(skb, i - 1, page,
1504                                    frag - page_address(page), fragsz);
1505         }
1506
1507         return skb;
1508 free:
1509         /* frees skb and all frags allocated with napi_alloc_frag() */
1510         napi_free_frags(&tfile->napi);
1511         return ERR_PTR(err);
1512 }
1513
1514 /* prepad is the amount to reserve at front.  len is length after that.
1515  * linear is a hint as to how much to copy (usually headers). */
1516 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1517                                      size_t prepad, size_t len,
1518                                      size_t linear, int noblock)
1519 {
1520         struct sock *sk = tfile->socket.sk;
1521         struct sk_buff *skb;
1522         int err;
1523
1524         /* Under a page?  Don't bother with paged skb. */
1525         if (prepad + len < PAGE_SIZE || !linear)
1526                 linear = len;
1527
1528         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1529                                    &err, 0);
1530         if (!skb)
1531                 return ERR_PTR(err);
1532
1533         skb_reserve(skb, prepad);
1534         skb_put(skb, linear);
1535         skb->data_len = len - linear;
1536         skb->len += len - linear;
1537
1538         return skb;
1539 }
1540
1541 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1542                            struct sk_buff *skb, int more)
1543 {
1544         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1545         struct sk_buff_head process_queue;
1546         u32 rx_batched = tun->rx_batched;
1547         bool rcv = false;
1548
1549         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1550                 local_bh_disable();
1551                 skb_record_rx_queue(skb, tfile->queue_index);
1552                 netif_receive_skb(skb);
1553                 local_bh_enable();
1554                 return;
1555         }
1556
1557         spin_lock(&queue->lock);
1558         if (!more || skb_queue_len(queue) == rx_batched) {
1559                 __skb_queue_head_init(&process_queue);
1560                 skb_queue_splice_tail_init(queue, &process_queue);
1561                 rcv = true;
1562         } else {
1563                 __skb_queue_tail(queue, skb);
1564         }
1565         spin_unlock(&queue->lock);
1566
1567         if (rcv) {
1568                 struct sk_buff *nskb;
1569
1570                 local_bh_disable();
1571                 while ((nskb = __skb_dequeue(&process_queue))) {
1572                         skb_record_rx_queue(nskb, tfile->queue_index);
1573                         netif_receive_skb(nskb);
1574                 }
1575                 skb_record_rx_queue(skb, tfile->queue_index);
1576                 netif_receive_skb(skb);
1577                 local_bh_enable();
1578         }
1579 }
1580
1581 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1582                               int len, int noblock, bool zerocopy)
1583 {
1584         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1585                 return false;
1586
1587         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1588                 return false;
1589
1590         if (!noblock)
1591                 return false;
1592
1593         if (zerocopy)
1594                 return false;
1595
1596         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1597             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1598                 return false;
1599
1600         return true;
1601 }
1602
1603 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1604                                        int buflen, int len, int pad)
1605 {
1606         struct sk_buff *skb = build_skb(buf, buflen);
1607
1608         if (!skb)
1609                 return ERR_PTR(-ENOMEM);
1610
1611         skb_reserve(skb, pad);
1612         skb_put(skb, len);
1613
1614         get_page(alloc_frag->page);
1615         alloc_frag->offset += buflen;
1616
1617         return skb;
1618 }
1619
1620 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1621                        struct xdp_buff *xdp, u32 act)
1622 {
1623         int err;
1624
1625         switch (act) {
1626         case XDP_REDIRECT:
1627                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1628                 if (err)
1629                         return err;
1630                 break;
1631         case XDP_TX:
1632                 err = tun_xdp_tx(tun->dev, xdp);
1633                 if (err < 0)
1634                         return err;
1635                 break;
1636         case XDP_PASS:
1637                 break;
1638         default:
1639                 bpf_warn_invalid_xdp_action(act);
1640                 /* fall through */
1641         case XDP_ABORTED:
1642                 trace_xdp_exception(tun->dev, xdp_prog, act);
1643                 /* fall through */
1644         case XDP_DROP:
1645                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1646                 break;
1647         }
1648
1649         return act;
1650 }
1651
1652 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1653                                      struct tun_file *tfile,
1654                                      struct iov_iter *from,
1655                                      struct virtio_net_hdr *hdr,
1656                                      int len, int *skb_xdp)
1657 {
1658         struct page_frag *alloc_frag = &current->task_frag;
1659         struct bpf_prog *xdp_prog;
1660         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1661         char *buf;
1662         size_t copied;
1663         int pad = TUN_RX_PAD;
1664         int err = 0;
1665
1666         rcu_read_lock();
1667         xdp_prog = rcu_dereference(tun->xdp_prog);
1668         if (xdp_prog)
1669                 pad += XDP_PACKET_HEADROOM;
1670         buflen += SKB_DATA_ALIGN(len + pad);
1671         rcu_read_unlock();
1672
1673         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1674         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1675                 return ERR_PTR(-ENOMEM);
1676
1677         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1678         copied = copy_page_from_iter(alloc_frag->page,
1679                                      alloc_frag->offset + pad,
1680                                      len, from);
1681         if (copied != len)
1682                 return ERR_PTR(-EFAULT);
1683
1684         /* There's a small window that XDP may be set after the check
1685          * of xdp_prog above, this should be rare and for simplicity
1686          * we do XDP on skb in case the headroom is not enough.
1687          */
1688         if (hdr->gso_type || !xdp_prog) {
1689                 *skb_xdp = 1;
1690                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1691         }
1692
1693         *skb_xdp = 0;
1694
1695         local_bh_disable();
1696         rcu_read_lock();
1697         xdp_prog = rcu_dereference(tun->xdp_prog);
1698         if (xdp_prog) {
1699                 struct xdp_buff xdp;
1700                 u32 act;
1701
1702                 xdp.data_hard_start = buf;
1703                 xdp.data = buf + pad;
1704                 xdp_set_data_meta_invalid(&xdp);
1705                 xdp.data_end = xdp.data + len;
1706                 xdp.rxq = &tfile->xdp_rxq;
1707
1708                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1709                 if (act == XDP_REDIRECT || act == XDP_TX) {
1710                         get_page(alloc_frag->page);
1711                         alloc_frag->offset += buflen;
1712                 }
1713                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1714                 if (err < 0)
1715                         goto err_xdp;
1716                 if (err == XDP_REDIRECT)
1717                         xdp_do_flush_map();
1718                 if (err != XDP_PASS)
1719                         goto out;
1720
1721                 pad = xdp.data - xdp.data_hard_start;
1722                 len = xdp.data_end - xdp.data;
1723         }
1724         rcu_read_unlock();
1725         local_bh_enable();
1726
1727         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1728
1729 err_xdp:
1730         put_page(alloc_frag->page);
1731 out:
1732         rcu_read_unlock();
1733         local_bh_enable();
1734         return NULL;
1735 }
1736
1737 /* Get packet from user space buffer */
1738 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1739                             void *msg_control, struct iov_iter *from,
1740                             int noblock, bool more)
1741 {
1742         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1743         struct sk_buff *skb;
1744         size_t total_len = iov_iter_count(from);
1745         size_t len = total_len, align = tun->align, linear;
1746         struct virtio_net_hdr gso = { 0 };
1747         struct tun_pcpu_stats *stats;
1748         int good_linear;
1749         int copylen;
1750         bool zerocopy = false;
1751         int err;
1752         u32 rxhash = 0;
1753         int skb_xdp = 1;
1754         bool frags = tun_napi_frags_enabled(tfile);
1755
1756         if (!(tun->flags & IFF_NO_PI)) {
1757                 if (len < sizeof(pi))
1758                         return -EINVAL;
1759                 len -= sizeof(pi);
1760
1761                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1762                         return -EFAULT;
1763         }
1764
1765         if (tun->flags & IFF_VNET_HDR) {
1766                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1767
1768                 if (len < vnet_hdr_sz)
1769                         return -EINVAL;
1770                 len -= vnet_hdr_sz;
1771
1772                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1773                         return -EFAULT;
1774
1775                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1776                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1777                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1778
1779                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1780                         return -EINVAL;
1781                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1782         }
1783
1784         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1785                 align += NET_IP_ALIGN;
1786                 if (unlikely(len < ETH_HLEN ||
1787                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1788                         return -EINVAL;
1789         }
1790
1791         good_linear = SKB_MAX_HEAD(align);
1792
1793         if (msg_control) {
1794                 struct iov_iter i = *from;
1795
1796                 /* There are 256 bytes to be copied in skb, so there is
1797                  * enough room for skb expand head in case it is used.
1798                  * The rest of the buffer is mapped from userspace.
1799                  */
1800                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1801                 if (copylen > good_linear)
1802                         copylen = good_linear;
1803                 linear = copylen;
1804                 iov_iter_advance(&i, copylen);
1805                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1806                         zerocopy = true;
1807         }
1808
1809         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1810                 /* For the packet that is not easy to be processed
1811                  * (e.g gso or jumbo packet), we will do it at after
1812                  * skb was created with generic XDP routine.
1813                  */
1814                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1815                 if (IS_ERR(skb)) {
1816                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1817                         return PTR_ERR(skb);
1818                 }
1819                 if (!skb)
1820                         return total_len;
1821         } else {
1822                 if (!zerocopy) {
1823                         copylen = len;
1824                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1825                                 linear = good_linear;
1826                         else
1827                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1828                 }
1829
1830                 if (frags) {
1831                         mutex_lock(&tfile->napi_mutex);
1832                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1833                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1834                          * If zerocopy is enabled, then this layout will be
1835                          * overwritten by zerocopy_sg_from_iter().
1836                          */
1837                         zerocopy = false;
1838                 } else {
1839                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1840                                             noblock);
1841                 }
1842
1843                 if (IS_ERR(skb)) {
1844                         if (PTR_ERR(skb) != -EAGAIN)
1845                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1846                         if (frags)
1847                                 mutex_unlock(&tfile->napi_mutex);
1848                         return PTR_ERR(skb);
1849                 }
1850
1851                 if (zerocopy)
1852                         err = zerocopy_sg_from_iter(skb, from);
1853                 else
1854                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1855
1856                 if (err) {
1857                         err = -EFAULT;
1858 drop:
1859                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1860                         kfree_skb(skb);
1861                         if (frags) {
1862                                 tfile->napi.skb = NULL;
1863                                 mutex_unlock(&tfile->napi_mutex);
1864                         }
1865
1866                         return err;
1867                 }
1868         }
1869
1870         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1871                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1872                 kfree_skb(skb);
1873                 if (frags) {
1874                         tfile->napi.skb = NULL;
1875                         mutex_unlock(&tfile->napi_mutex);
1876                 }
1877
1878                 return -EINVAL;
1879         }
1880
1881         switch (tun->flags & TUN_TYPE_MASK) {
1882         case IFF_TUN:
1883                 if (tun->flags & IFF_NO_PI) {
1884                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1885
1886                         switch (ip_version) {
1887                         case 4:
1888                                 pi.proto = htons(ETH_P_IP);
1889                                 break;
1890                         case 6:
1891                                 pi.proto = htons(ETH_P_IPV6);
1892                                 break;
1893                         default:
1894                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1895                                 kfree_skb(skb);
1896                                 return -EINVAL;
1897                         }
1898                 }
1899
1900                 skb_reset_mac_header(skb);
1901                 skb->protocol = pi.proto;
1902                 skb->dev = tun->dev;
1903                 break;
1904         case IFF_TAP:
1905                 if (!frags)
1906                         skb->protocol = eth_type_trans(skb, tun->dev);
1907                 break;
1908         }
1909
1910         /* copy skb_ubuf_info for callback when skb has no error */
1911         if (zerocopy) {
1912                 skb_shinfo(skb)->destructor_arg = msg_control;
1913                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1914                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1915         } else if (msg_control) {
1916                 struct ubuf_info *uarg = msg_control;
1917                 uarg->callback(uarg, false);
1918         }
1919
1920         skb_reset_network_header(skb);
1921         skb_probe_transport_header(skb);
1922
1923         if (skb_xdp) {
1924                 struct bpf_prog *xdp_prog;
1925                 int ret;
1926
1927                 local_bh_disable();
1928                 rcu_read_lock();
1929                 xdp_prog = rcu_dereference(tun->xdp_prog);
1930                 if (xdp_prog) {
1931                         ret = do_xdp_generic(xdp_prog, skb);
1932                         if (ret != XDP_PASS) {
1933                                 rcu_read_unlock();
1934                                 local_bh_enable();
1935                                 return total_len;
1936                         }
1937                 }
1938                 rcu_read_unlock();
1939                 local_bh_enable();
1940         }
1941
1942         /* Compute the costly rx hash only if needed for flow updates.
1943          * We may get a very small possibility of OOO during switching, not
1944          * worth to optimize.
1945          */
1946         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1947             !tfile->detached)
1948                 rxhash = __skb_get_hash_symmetric(skb);
1949
1950         rcu_read_lock();
1951         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1952                 err = -EIO;
1953                 rcu_read_unlock();
1954                 goto drop;
1955         }
1956
1957         if (frags) {
1958                 /* Exercise flow dissector code path. */
1959                 u32 headlen = eth_get_headlen(tun->dev, skb->data,
1960                                               skb_headlen(skb));
1961
1962                 if (unlikely(headlen > skb_headlen(skb))) {
1963                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1964                         napi_free_frags(&tfile->napi);
1965                         rcu_read_unlock();
1966                         mutex_unlock(&tfile->napi_mutex);
1967                         WARN_ON(1);
1968                         return -ENOMEM;
1969                 }
1970
1971                 local_bh_disable();
1972                 napi_gro_frags(&tfile->napi);
1973                 local_bh_enable();
1974                 mutex_unlock(&tfile->napi_mutex);
1975         } else if (tfile->napi_enabled) {
1976                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1977                 int queue_len;
1978
1979                 spin_lock_bh(&queue->lock);
1980                 __skb_queue_tail(queue, skb);
1981                 queue_len = skb_queue_len(queue);
1982                 spin_unlock(&queue->lock);
1983
1984                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1985                         napi_schedule(&tfile->napi);
1986
1987                 local_bh_enable();
1988         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1989                 tun_rx_batched(tun, tfile, skb, more);
1990         } else {
1991                 netif_rx_ni(skb);
1992         }
1993         rcu_read_unlock();
1994
1995         stats = get_cpu_ptr(tun->pcpu_stats);
1996         u64_stats_update_begin(&stats->syncp);
1997         stats->rx_packets++;
1998         stats->rx_bytes += len;
1999         u64_stats_update_end(&stats->syncp);
2000         put_cpu_ptr(stats);
2001
2002         if (rxhash)
2003                 tun_flow_update(tun, rxhash, tfile);
2004
2005         return total_len;
2006 }
2007
2008 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2009 {
2010         struct file *file = iocb->ki_filp;
2011         struct tun_file *tfile = file->private_data;
2012         struct tun_struct *tun = tun_get(tfile);
2013         ssize_t result;
2014
2015         if (!tun)
2016                 return -EBADFD;
2017
2018         result = tun_get_user(tun, tfile, NULL, from,
2019                               file->f_flags & O_NONBLOCK, false);
2020
2021         tun_put(tun);
2022         return result;
2023 }
2024
2025 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2026                                 struct tun_file *tfile,
2027                                 struct xdp_frame *xdp_frame,
2028                                 struct iov_iter *iter)
2029 {
2030         int vnet_hdr_sz = 0;
2031         size_t size = xdp_frame->len;
2032         struct tun_pcpu_stats *stats;
2033         size_t ret;
2034
2035         if (tun->flags & IFF_VNET_HDR) {
2036                 struct virtio_net_hdr gso = { 0 };
2037
2038                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2039                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2040                         return -EINVAL;
2041                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2042                              sizeof(gso)))
2043                         return -EFAULT;
2044                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2045         }
2046
2047         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2048
2049         stats = get_cpu_ptr(tun->pcpu_stats);
2050         u64_stats_update_begin(&stats->syncp);
2051         stats->tx_packets++;
2052         stats->tx_bytes += ret;
2053         u64_stats_update_end(&stats->syncp);
2054         put_cpu_ptr(tun->pcpu_stats);
2055
2056         return ret;
2057 }
2058
2059 /* Put packet to the user space buffer */
2060 static ssize_t tun_put_user(struct tun_struct *tun,
2061                             struct tun_file *tfile,
2062                             struct sk_buff *skb,
2063                             struct iov_iter *iter)
2064 {
2065         struct tun_pi pi = { 0, skb->protocol };
2066         struct tun_pcpu_stats *stats;
2067         ssize_t total;
2068         int vlan_offset = 0;
2069         int vlan_hlen = 0;
2070         int vnet_hdr_sz = 0;
2071
2072         if (skb_vlan_tag_present(skb))
2073                 vlan_hlen = VLAN_HLEN;
2074
2075         if (tun->flags & IFF_VNET_HDR)
2076                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2077
2078         total = skb->len + vlan_hlen + vnet_hdr_sz;
2079
2080         if (!(tun->flags & IFF_NO_PI)) {
2081                 if (iov_iter_count(iter) < sizeof(pi))
2082                         return -EINVAL;
2083
2084                 total += sizeof(pi);
2085                 if (iov_iter_count(iter) < total) {
2086                         /* Packet will be striped */
2087                         pi.flags |= TUN_PKT_STRIP;
2088                 }
2089
2090                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2091                         return -EFAULT;
2092         }
2093
2094         if (vnet_hdr_sz) {
2095                 struct virtio_net_hdr gso;
2096
2097                 if (iov_iter_count(iter) < vnet_hdr_sz)
2098                         return -EINVAL;
2099
2100                 if (virtio_net_hdr_from_skb(skb, &gso,
2101                                             tun_is_little_endian(tun), true,
2102                                             vlan_hlen)) {
2103                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2104                         pr_err("unexpected GSO type: "
2105                                "0x%x, gso_size %d, hdr_len %d\n",
2106                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2107                                tun16_to_cpu(tun, gso.hdr_len));
2108                         print_hex_dump(KERN_ERR, "tun: ",
2109                                        DUMP_PREFIX_NONE,
2110                                        16, 1, skb->head,
2111                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2112                         WARN_ON_ONCE(1);
2113                         return -EINVAL;
2114                 }
2115
2116                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2117                         return -EFAULT;
2118
2119                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2120         }
2121
2122         if (vlan_hlen) {
2123                 int ret;
2124                 struct veth veth;
2125
2126                 veth.h_vlan_proto = skb->vlan_proto;
2127                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2128
2129                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2130
2131                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2132                 if (ret || !iov_iter_count(iter))
2133                         goto done;
2134
2135                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2136                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2137                         goto done;
2138         }
2139
2140         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2141
2142 done:
2143         /* caller is in process context, */
2144         stats = get_cpu_ptr(tun->pcpu_stats);
2145         u64_stats_update_begin(&stats->syncp);
2146         stats->tx_packets++;
2147         stats->tx_bytes += skb->len + vlan_hlen;
2148         u64_stats_update_end(&stats->syncp);
2149         put_cpu_ptr(tun->pcpu_stats);
2150
2151         return total;
2152 }
2153
2154 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2155 {
2156         DECLARE_WAITQUEUE(wait, current);
2157         void *ptr = NULL;
2158         int error = 0;
2159
2160         ptr = ptr_ring_consume(&tfile->tx_ring);
2161         if (ptr)
2162                 goto out;
2163         if (noblock) {
2164                 error = -EAGAIN;
2165                 goto out;
2166         }
2167
2168         add_wait_queue(&tfile->wq.wait, &wait);
2169
2170         while (1) {
2171                 set_current_state(TASK_INTERRUPTIBLE);
2172                 ptr = ptr_ring_consume(&tfile->tx_ring);
2173                 if (ptr)
2174                         break;
2175                 if (signal_pending(current)) {
2176                         error = -ERESTARTSYS;
2177                         break;
2178                 }
2179                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2180                         error = -EFAULT;
2181                         break;
2182                 }
2183
2184                 schedule();
2185         }
2186
2187         __set_current_state(TASK_RUNNING);
2188         remove_wait_queue(&tfile->wq.wait, &wait);
2189
2190 out:
2191         *err = error;
2192         return ptr;
2193 }
2194
2195 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2196                            struct iov_iter *to,
2197                            int noblock, void *ptr)
2198 {
2199         ssize_t ret;
2200         int err;
2201
2202         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2203
2204         if (!iov_iter_count(to)) {
2205                 tun_ptr_free(ptr);
2206                 return 0;
2207         }
2208
2209         if (!ptr) {
2210                 /* Read frames from ring */
2211                 ptr = tun_ring_recv(tfile, noblock, &err);
2212                 if (!ptr)
2213                         return err;
2214         }
2215
2216         if (tun_is_xdp_frame(ptr)) {
2217                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2218
2219                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2220                 xdp_return_frame(xdpf);
2221         } else {
2222                 struct sk_buff *skb = ptr;
2223
2224                 ret = tun_put_user(tun, tfile, skb, to);
2225                 if (unlikely(ret < 0))
2226                         kfree_skb(skb);
2227                 else
2228                         consume_skb(skb);
2229         }
2230
2231         return ret;
2232 }
2233
2234 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2235 {
2236         struct file *file = iocb->ki_filp;
2237         struct tun_file *tfile = file->private_data;
2238         struct tun_struct *tun = tun_get(tfile);
2239         ssize_t len = iov_iter_count(to), ret;
2240
2241         if (!tun)
2242                 return -EBADFD;
2243         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2244         ret = min_t(ssize_t, ret, len);
2245         if (ret > 0)
2246                 iocb->ki_pos = ret;
2247         tun_put(tun);
2248         return ret;
2249 }
2250
2251 static void tun_prog_free(struct rcu_head *rcu)
2252 {
2253         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2254
2255         bpf_prog_destroy(prog->prog);
2256         kfree(prog);
2257 }
2258
2259 static int __tun_set_ebpf(struct tun_struct *tun,
2260                           struct tun_prog __rcu **prog_p,
2261                           struct bpf_prog *prog)
2262 {
2263         struct tun_prog *old, *new = NULL;
2264
2265         if (prog) {
2266                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2267                 if (!new)
2268                         return -ENOMEM;
2269                 new->prog = prog;
2270         }
2271
2272         spin_lock_bh(&tun->lock);
2273         old = rcu_dereference_protected(*prog_p,
2274                                         lockdep_is_held(&tun->lock));
2275         rcu_assign_pointer(*prog_p, new);
2276         spin_unlock_bh(&tun->lock);
2277
2278         if (old)
2279                 call_rcu(&old->rcu, tun_prog_free);
2280
2281         return 0;
2282 }
2283
2284 static void tun_free_netdev(struct net_device *dev)
2285 {
2286         struct tun_struct *tun = netdev_priv(dev);
2287
2288         BUG_ON(!(list_empty(&tun->disabled)));
2289         free_percpu(tun->pcpu_stats);
2290         tun_flow_uninit(tun);
2291         security_tun_dev_free_security(tun->security);
2292         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2293         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2294 }
2295
2296 static void tun_setup(struct net_device *dev)
2297 {
2298         struct tun_struct *tun = netdev_priv(dev);
2299
2300         tun->owner = INVALID_UID;
2301         tun->group = INVALID_GID;
2302         tun_default_link_ksettings(dev, &tun->link_ksettings);
2303
2304         dev->ethtool_ops = &tun_ethtool_ops;
2305         dev->needs_free_netdev = true;
2306         dev->priv_destructor = tun_free_netdev;
2307         /* We prefer our own queue length */
2308         dev->tx_queue_len = TUN_READQ_SIZE;
2309 }
2310
2311 /* Trivial set of netlink ops to allow deleting tun or tap
2312  * device with netlink.
2313  */
2314 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2315                         struct netlink_ext_ack *extack)
2316 {
2317         NL_SET_ERR_MSG(extack,
2318                        "tun/tap creation via rtnetlink is not supported.");
2319         return -EOPNOTSUPP;
2320 }
2321
2322 static size_t tun_get_size(const struct net_device *dev)
2323 {
2324         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2325         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2326
2327         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2328                nla_total_size(sizeof(gid_t)) + /* GROUP */
2329                nla_total_size(sizeof(u8)) + /* TYPE */
2330                nla_total_size(sizeof(u8)) + /* PI */
2331                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2332                nla_total_size(sizeof(u8)) + /* PERSIST */
2333                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2334                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2335                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2336                0;
2337 }
2338
2339 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2340 {
2341         struct tun_struct *tun = netdev_priv(dev);
2342
2343         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2344                 goto nla_put_failure;
2345         if (uid_valid(tun->owner) &&
2346             nla_put_u32(skb, IFLA_TUN_OWNER,
2347                         from_kuid_munged(current_user_ns(), tun->owner)))
2348                 goto nla_put_failure;
2349         if (gid_valid(tun->group) &&
2350             nla_put_u32(skb, IFLA_TUN_GROUP,
2351                         from_kgid_munged(current_user_ns(), tun->group)))
2352                 goto nla_put_failure;
2353         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2354                 goto nla_put_failure;
2355         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2356                 goto nla_put_failure;
2357         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2358                 goto nla_put_failure;
2359         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2360                        !!(tun->flags & IFF_MULTI_QUEUE)))
2361                 goto nla_put_failure;
2362         if (tun->flags & IFF_MULTI_QUEUE) {
2363                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2364                         goto nla_put_failure;
2365                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2366                                 tun->numdisabled))
2367                         goto nla_put_failure;
2368         }
2369
2370         return 0;
2371
2372 nla_put_failure:
2373         return -EMSGSIZE;
2374 }
2375
2376 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2377         .kind           = DRV_NAME,
2378         .priv_size      = sizeof(struct tun_struct),
2379         .setup          = tun_setup,
2380         .validate       = tun_validate,
2381         .get_size       = tun_get_size,
2382         .fill_info      = tun_fill_info,
2383 };
2384
2385 static void tun_sock_write_space(struct sock *sk)
2386 {
2387         struct tun_file *tfile;
2388         wait_queue_head_t *wqueue;
2389
2390         if (!sock_writeable(sk))
2391                 return;
2392
2393         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2394                 return;
2395
2396         wqueue = sk_sleep(sk);
2397         if (wqueue && waitqueue_active(wqueue))
2398                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2399                                                 EPOLLWRNORM | EPOLLWRBAND);
2400
2401         tfile = container_of(sk, struct tun_file, sk);
2402         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2403 }
2404
2405 static void tun_put_page(struct tun_page *tpage)
2406 {
2407         if (tpage->page)
2408                 __page_frag_cache_drain(tpage->page, tpage->count);
2409 }
2410
2411 static int tun_xdp_one(struct tun_struct *tun,
2412                        struct tun_file *tfile,
2413                        struct xdp_buff *xdp, int *flush,
2414                        struct tun_page *tpage)
2415 {
2416         unsigned int datasize = xdp->data_end - xdp->data;
2417         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2418         struct virtio_net_hdr *gso = &hdr->gso;
2419         struct tun_pcpu_stats *stats;
2420         struct bpf_prog *xdp_prog;
2421         struct sk_buff *skb = NULL;
2422         u32 rxhash = 0, act;
2423         int buflen = hdr->buflen;
2424         int err = 0;
2425         bool skb_xdp = false;
2426         struct page *page;
2427
2428         xdp_prog = rcu_dereference(tun->xdp_prog);
2429         if (xdp_prog) {
2430                 if (gso->gso_type) {
2431                         skb_xdp = true;
2432                         goto build;
2433                 }
2434                 xdp_set_data_meta_invalid(xdp);
2435                 xdp->rxq = &tfile->xdp_rxq;
2436
2437                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2438                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2439                 if (err < 0) {
2440                         put_page(virt_to_head_page(xdp->data));
2441                         return err;
2442                 }
2443
2444                 switch (err) {
2445                 case XDP_REDIRECT:
2446                         *flush = true;
2447                         /* fall through */
2448                 case XDP_TX:
2449                         return 0;
2450                 case XDP_PASS:
2451                         break;
2452                 default:
2453                         page = virt_to_head_page(xdp->data);
2454                         if (tpage->page == page) {
2455                                 ++tpage->count;
2456                         } else {
2457                                 tun_put_page(tpage);
2458                                 tpage->page = page;
2459                                 tpage->count = 1;
2460                         }
2461                         return 0;
2462                 }
2463         }
2464
2465 build:
2466         skb = build_skb(xdp->data_hard_start, buflen);
2467         if (!skb) {
2468                 err = -ENOMEM;
2469                 goto out;
2470         }
2471
2472         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2473         skb_put(skb, xdp->data_end - xdp->data);
2474
2475         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2476                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2477                 kfree_skb(skb);
2478                 err = -EINVAL;
2479                 goto out;
2480         }
2481
2482         skb->protocol = eth_type_trans(skb, tun->dev);
2483         skb_reset_network_header(skb);
2484         skb_probe_transport_header(skb);
2485
2486         if (skb_xdp) {
2487                 err = do_xdp_generic(xdp_prog, skb);
2488                 if (err != XDP_PASS)
2489                         goto out;
2490         }
2491
2492         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2493             !tfile->detached)
2494                 rxhash = __skb_get_hash_symmetric(skb);
2495
2496         skb_record_rx_queue(skb, tfile->queue_index);
2497         netif_receive_skb(skb);
2498
2499         /* No need for get_cpu_ptr() here since this function is
2500          * always called with bh disabled
2501          */
2502         stats = this_cpu_ptr(tun->pcpu_stats);
2503         u64_stats_update_begin(&stats->syncp);
2504         stats->rx_packets++;
2505         stats->rx_bytes += datasize;
2506         u64_stats_update_end(&stats->syncp);
2507
2508         if (rxhash)
2509                 tun_flow_update(tun, rxhash, tfile);
2510
2511 out:
2512         return err;
2513 }
2514
2515 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2516 {
2517         int ret, i;
2518         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2519         struct tun_struct *tun = tun_get(tfile);
2520         struct tun_msg_ctl *ctl = m->msg_control;
2521         struct xdp_buff *xdp;
2522
2523         if (!tun)
2524                 return -EBADFD;
2525
2526         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2527                 struct tun_page tpage;
2528                 int n = ctl->num;
2529                 int flush = 0;
2530
2531                 memset(&tpage, 0, sizeof(tpage));
2532
2533                 local_bh_disable();
2534                 rcu_read_lock();
2535
2536                 for (i = 0; i < n; i++) {
2537                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2538                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2539                 }
2540
2541                 if (flush)
2542                         xdp_do_flush_map();
2543
2544                 rcu_read_unlock();
2545                 local_bh_enable();
2546
2547                 tun_put_page(&tpage);
2548
2549                 ret = total_len;
2550                 goto out;
2551         }
2552
2553         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2554                            m->msg_flags & MSG_DONTWAIT,
2555                            m->msg_flags & MSG_MORE);
2556 out:
2557         tun_put(tun);
2558         return ret;
2559 }
2560
2561 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2562                        int flags)
2563 {
2564         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2565         struct tun_struct *tun = tun_get(tfile);
2566         void *ptr = m->msg_control;
2567         int ret;
2568
2569         if (!tun) {
2570                 ret = -EBADFD;
2571                 goto out_free;
2572         }
2573
2574         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2575                 ret = -EINVAL;
2576                 goto out_put_tun;
2577         }
2578         if (flags & MSG_ERRQUEUE) {
2579                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2580                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2581                 goto out;
2582         }
2583         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2584         if (ret > (ssize_t)total_len) {
2585                 m->msg_flags |= MSG_TRUNC;
2586                 ret = flags & MSG_TRUNC ? ret : total_len;
2587         }
2588 out:
2589         tun_put(tun);
2590         return ret;
2591
2592 out_put_tun:
2593         tun_put(tun);
2594 out_free:
2595         tun_ptr_free(ptr);
2596         return ret;
2597 }
2598
2599 static int tun_ptr_peek_len(void *ptr)
2600 {
2601         if (likely(ptr)) {
2602                 if (tun_is_xdp_frame(ptr)) {
2603                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2604
2605                         return xdpf->len;
2606                 }
2607                 return __skb_array_len_with_tag(ptr);
2608         } else {
2609                 return 0;
2610         }
2611 }
2612
2613 static int tun_peek_len(struct socket *sock)
2614 {
2615         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2616         struct tun_struct *tun;
2617         int ret = 0;
2618
2619         tun = tun_get(tfile);
2620         if (!tun)
2621                 return 0;
2622
2623         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2624         tun_put(tun);
2625
2626         return ret;
2627 }
2628
2629 /* Ops structure to mimic raw sockets with tun */
2630 static const struct proto_ops tun_socket_ops = {
2631         .peek_len = tun_peek_len,
2632         .sendmsg = tun_sendmsg,
2633         .recvmsg = tun_recvmsg,
2634 };
2635
2636 static struct proto tun_proto = {
2637         .name           = "tun",
2638         .owner          = THIS_MODULE,
2639         .obj_size       = sizeof(struct tun_file),
2640 };
2641
2642 static int tun_flags(struct tun_struct *tun)
2643 {
2644         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2645 }
2646
2647 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2648                               char *buf)
2649 {
2650         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2651         return sprintf(buf, "0x%x\n", tun_flags(tun));
2652 }
2653
2654 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2655                               char *buf)
2656 {
2657         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2658         return uid_valid(tun->owner)?
2659                 sprintf(buf, "%u\n",
2660                         from_kuid_munged(current_user_ns(), tun->owner)):
2661                 sprintf(buf, "-1\n");
2662 }
2663
2664 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2665                               char *buf)
2666 {
2667         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2668         return gid_valid(tun->group) ?
2669                 sprintf(buf, "%u\n",
2670                         from_kgid_munged(current_user_ns(), tun->group)):
2671                 sprintf(buf, "-1\n");
2672 }
2673
2674 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2675 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2676 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2677
2678 static struct attribute *tun_dev_attrs[] = {
2679         &dev_attr_tun_flags.attr,
2680         &dev_attr_owner.attr,
2681         &dev_attr_group.attr,
2682         NULL
2683 };
2684
2685 static const struct attribute_group tun_attr_group = {
2686         .attrs = tun_dev_attrs
2687 };
2688
2689 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2690 {
2691         struct tun_struct *tun;
2692         struct tun_file *tfile = file->private_data;
2693         struct net_device *dev;
2694         int err;
2695
2696         if (tfile->detached)
2697                 return -EINVAL;
2698
2699         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2700                 if (!capable(CAP_NET_ADMIN))
2701                         return -EPERM;
2702
2703                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2704                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2705                         return -EINVAL;
2706         }
2707
2708         dev = __dev_get_by_name(net, ifr->ifr_name);
2709         if (dev) {
2710                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2711                         return -EBUSY;
2712                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2713                         tun = netdev_priv(dev);
2714                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2715                         tun = netdev_priv(dev);
2716                 else
2717                         return -EINVAL;
2718
2719                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2720                     !!(tun->flags & IFF_MULTI_QUEUE))
2721                         return -EINVAL;
2722
2723                 if (tun_not_capable(tun))
2724                         return -EPERM;
2725                 err = security_tun_dev_open(tun->security);
2726                 if (err < 0)
2727                         return err;
2728
2729                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2730                                  ifr->ifr_flags & IFF_NAPI,
2731                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2732                 if (err < 0)
2733                         return err;
2734
2735                 if (tun->flags & IFF_MULTI_QUEUE &&
2736                     (tun->numqueues + tun->numdisabled > 1)) {
2737                         /* One or more queue has already been attached, no need
2738                          * to initialize the device again.
2739                          */
2740                         netdev_state_change(dev);
2741                         return 0;
2742                 }
2743
2744                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2745                               (ifr->ifr_flags & TUN_FEATURES);
2746
2747                 netdev_state_change(dev);
2748         } else {
2749                 char *name;
2750                 unsigned long flags = 0;
2751                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2752                              MAX_TAP_QUEUES : 1;
2753
2754                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2755                         return -EPERM;
2756                 err = security_tun_dev_create();
2757                 if (err < 0)
2758                         return err;
2759
2760                 /* Set dev type */
2761                 if (ifr->ifr_flags & IFF_TUN) {
2762                         /* TUN device */
2763                         flags |= IFF_TUN;
2764                         name = "tun%d";
2765                 } else if (ifr->ifr_flags & IFF_TAP) {
2766                         /* TAP device */
2767                         flags |= IFF_TAP;
2768                         name = "tap%d";
2769                 } else
2770                         return -EINVAL;
2771
2772                 if (*ifr->ifr_name)
2773                         name = ifr->ifr_name;
2774
2775                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2776                                        NET_NAME_UNKNOWN, tun_setup, queues,
2777                                        queues);
2778
2779                 if (!dev)
2780                         return -ENOMEM;
2781                 err = dev_get_valid_name(net, dev, name);
2782                 if (err < 0)
2783                         goto err_free_dev;
2784
2785                 dev_net_set(dev, net);
2786                 dev->rtnl_link_ops = &tun_link_ops;
2787                 dev->ifindex = tfile->ifindex;
2788                 dev->sysfs_groups[0] = &tun_attr_group;
2789
2790                 tun = netdev_priv(dev);
2791                 tun->dev = dev;
2792                 tun->flags = flags;
2793                 tun->txflt.count = 0;
2794                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2795
2796                 tun->align = NET_SKB_PAD;
2797                 tun->filter_attached = false;
2798                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2799                 tun->rx_batched = 0;
2800                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2801
2802                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2803                 if (!tun->pcpu_stats) {
2804                         err = -ENOMEM;
2805                         goto err_free_dev;
2806                 }
2807
2808                 spin_lock_init(&tun->lock);
2809
2810                 err = security_tun_dev_alloc_security(&tun->security);
2811                 if (err < 0)
2812                         goto err_free_stat;
2813
2814                 tun_net_init(dev);
2815                 tun_flow_init(tun);
2816
2817                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2818                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2819                                    NETIF_F_HW_VLAN_STAG_TX;
2820                 dev->features = dev->hw_features | NETIF_F_LLTX;
2821                 dev->vlan_features = dev->features &
2822                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2823                                        NETIF_F_HW_VLAN_STAG_TX);
2824
2825                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2826                               (ifr->ifr_flags & TUN_FEATURES);
2827
2828                 INIT_LIST_HEAD(&tun->disabled);
2829                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2830                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2831                 if (err < 0)
2832                         goto err_free_flow;
2833
2834                 err = register_netdevice(tun->dev);
2835                 if (err < 0)
2836                         goto err_detach;
2837         }
2838
2839         netif_carrier_on(tun->dev);
2840
2841         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2842
2843         /* Make sure persistent devices do not get stuck in
2844          * xoff state.
2845          */
2846         if (netif_running(tun->dev))
2847                 netif_tx_wake_all_queues(tun->dev);
2848
2849         strcpy(ifr->ifr_name, tun->dev->name);
2850         return 0;
2851
2852 err_detach:
2853         tun_detach_all(dev);
2854         /* register_netdevice() already called tun_free_netdev() */
2855         goto err_free_dev;
2856
2857 err_free_flow:
2858         tun_flow_uninit(tun);
2859         security_tun_dev_free_security(tun->security);
2860 err_free_stat:
2861         free_percpu(tun->pcpu_stats);
2862 err_free_dev:
2863         free_netdev(dev);
2864         return err;
2865 }
2866
2867 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2868 {
2869         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2870
2871         strcpy(ifr->ifr_name, tun->dev->name);
2872
2873         ifr->ifr_flags = tun_flags(tun);
2874
2875 }
2876
2877 /* This is like a cut-down ethtool ops, except done via tun fd so no
2878  * privs required. */
2879 static int set_offload(struct tun_struct *tun, unsigned long arg)
2880 {
2881         netdev_features_t features = 0;
2882
2883         if (arg & TUN_F_CSUM) {
2884                 features |= NETIF_F_HW_CSUM;
2885                 arg &= ~TUN_F_CSUM;
2886
2887                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2888                         if (arg & TUN_F_TSO_ECN) {
2889                                 features |= NETIF_F_TSO_ECN;
2890                                 arg &= ~TUN_F_TSO_ECN;
2891                         }
2892                         if (arg & TUN_F_TSO4)
2893                                 features |= NETIF_F_TSO;
2894                         if (arg & TUN_F_TSO6)
2895                                 features |= NETIF_F_TSO6;
2896                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2897                 }
2898
2899                 arg &= ~TUN_F_UFO;
2900         }
2901
2902         /* This gives the user a way to test for new features in future by
2903          * trying to set them. */
2904         if (arg)
2905                 return -EINVAL;
2906
2907         tun->set_features = features;
2908         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2909         tun->dev->wanted_features |= features;
2910         netdev_update_features(tun->dev);
2911
2912         return 0;
2913 }
2914
2915 static void tun_detach_filter(struct tun_struct *tun, int n)
2916 {
2917         int i;
2918         struct tun_file *tfile;
2919
2920         for (i = 0; i < n; i++) {
2921                 tfile = rtnl_dereference(tun->tfiles[i]);
2922                 lock_sock(tfile->socket.sk);
2923                 sk_detach_filter(tfile->socket.sk);
2924                 release_sock(tfile->socket.sk);
2925         }
2926
2927         tun->filter_attached = false;
2928 }
2929
2930 static int tun_attach_filter(struct tun_struct *tun)
2931 {
2932         int i, ret = 0;
2933         struct tun_file *tfile;
2934
2935         for (i = 0; i < tun->numqueues; i++) {
2936                 tfile = rtnl_dereference(tun->tfiles[i]);
2937                 lock_sock(tfile->socket.sk);
2938                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2939                 release_sock(tfile->socket.sk);
2940                 if (ret) {
2941                         tun_detach_filter(tun, i);
2942                         return ret;
2943                 }
2944         }
2945
2946         tun->filter_attached = true;
2947         return ret;
2948 }
2949
2950 static void tun_set_sndbuf(struct tun_struct *tun)
2951 {
2952         struct tun_file *tfile;
2953         int i;
2954
2955         for (i = 0; i < tun->numqueues; i++) {
2956                 tfile = rtnl_dereference(tun->tfiles[i]);
2957                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2958         }
2959 }
2960
2961 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2962 {
2963         struct tun_file *tfile = file->private_data;
2964         struct tun_struct *tun;
2965         int ret = 0;
2966
2967         rtnl_lock();
2968
2969         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2970                 tun = tfile->detached;
2971                 if (!tun) {
2972                         ret = -EINVAL;
2973                         goto unlock;
2974                 }
2975                 ret = security_tun_dev_attach_queue(tun->security);
2976                 if (ret < 0)
2977                         goto unlock;
2978                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2979                                  tun->flags & IFF_NAPI_FRAGS);
2980         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2981                 tun = rtnl_dereference(tfile->tun);
2982                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2983                         ret = -EINVAL;
2984                 else
2985                         __tun_detach(tfile, false);
2986         } else
2987                 ret = -EINVAL;
2988
2989         if (ret >= 0)
2990                 netdev_state_change(tun->dev);
2991
2992 unlock:
2993         rtnl_unlock();
2994         return ret;
2995 }
2996
2997 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2998                         void __user *data)
2999 {
3000         struct bpf_prog *prog;
3001         int fd;
3002
3003         if (copy_from_user(&fd, data, sizeof(fd)))
3004                 return -EFAULT;
3005
3006         if (fd == -1) {
3007                 prog = NULL;
3008         } else {
3009                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3010                 if (IS_ERR(prog))
3011                         return PTR_ERR(prog);
3012         }
3013
3014         return __tun_set_ebpf(tun, prog_p, prog);
3015 }
3016
3017 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3018                             unsigned long arg, int ifreq_len)
3019 {
3020         struct tun_file *tfile = file->private_data;
3021         struct net *net = sock_net(&tfile->sk);
3022         struct tun_struct *tun;
3023         void __user* argp = (void __user*)arg;
3024         unsigned int ifindex, carrier;
3025         struct ifreq ifr;
3026         kuid_t owner;
3027         kgid_t group;
3028         int sndbuf;
3029         int vnet_hdr_sz;
3030         int le;
3031         int ret;
3032         bool do_notify = false;
3033
3034         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3035             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3036                 if (copy_from_user(&ifr, argp, ifreq_len))
3037                         return -EFAULT;
3038         } else {
3039                 memset(&ifr, 0, sizeof(ifr));
3040         }
3041         if (cmd == TUNGETFEATURES) {
3042                 /* Currently this just means: "what IFF flags are valid?".
3043                  * This is needed because we never checked for invalid flags on
3044                  * TUNSETIFF.
3045                  */
3046                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3047                                 (unsigned int __user*)argp);
3048         } else if (cmd == TUNSETQUEUE) {
3049                 return tun_set_queue(file, &ifr);
3050         } else if (cmd == SIOCGSKNS) {
3051                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3052                         return -EPERM;
3053                 return open_related_ns(&net->ns, get_net_ns);
3054         }
3055
3056         ret = 0;
3057         rtnl_lock();
3058
3059         tun = tun_get(tfile);
3060         if (cmd == TUNSETIFF) {
3061                 ret = -EEXIST;
3062                 if (tun)
3063                         goto unlock;
3064
3065                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3066
3067                 ret = tun_set_iff(net, file, &ifr);
3068
3069                 if (ret)
3070                         goto unlock;
3071
3072                 if (copy_to_user(argp, &ifr, ifreq_len))
3073                         ret = -EFAULT;
3074                 goto unlock;
3075         }
3076         if (cmd == TUNSETIFINDEX) {
3077                 ret = -EPERM;
3078                 if (tun)
3079                         goto unlock;
3080
3081                 ret = -EFAULT;
3082                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3083                         goto unlock;
3084
3085                 ret = 0;
3086                 tfile->ifindex = ifindex;
3087                 goto unlock;
3088         }
3089
3090         ret = -EBADFD;
3091         if (!tun)
3092                 goto unlock;
3093
3094         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3095
3096         net = dev_net(tun->dev);
3097         ret = 0;
3098         switch (cmd) {
3099         case TUNGETIFF:
3100                 tun_get_iff(tun, &ifr);
3101
3102                 if (tfile->detached)
3103                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3104                 if (!tfile->socket.sk->sk_filter)
3105                         ifr.ifr_flags |= IFF_NOFILTER;
3106
3107                 if (copy_to_user(argp, &ifr, ifreq_len))
3108                         ret = -EFAULT;
3109                 break;
3110
3111         case TUNSETNOCSUM:
3112                 /* Disable/Enable checksum */
3113
3114                 /* [unimplemented] */
3115                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3116                           arg ? "disabled" : "enabled");
3117                 break;
3118
3119         case TUNSETPERSIST:
3120                 /* Disable/Enable persist mode. Keep an extra reference to the
3121                  * module to prevent the module being unprobed.
3122                  */
3123                 if (arg && !(tun->flags & IFF_PERSIST)) {
3124                         tun->flags |= IFF_PERSIST;
3125                         __module_get(THIS_MODULE);
3126                         do_notify = true;
3127                 }
3128                 if (!arg && (tun->flags & IFF_PERSIST)) {
3129                         tun->flags &= ~IFF_PERSIST;
3130                         module_put(THIS_MODULE);
3131                         do_notify = true;
3132                 }
3133
3134                 tun_debug(KERN_INFO, tun, "persist %s\n",
3135                           arg ? "enabled" : "disabled");
3136                 break;
3137
3138         case TUNSETOWNER:
3139                 /* Set owner of the device */
3140                 owner = make_kuid(current_user_ns(), arg);
3141                 if (!uid_valid(owner)) {
3142                         ret = -EINVAL;
3143                         break;
3144                 }
3145                 tun->owner = owner;
3146                 do_notify = true;
3147                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3148                           from_kuid(&init_user_ns, tun->owner));
3149                 break;
3150
3151         case TUNSETGROUP:
3152                 /* Set group of the device */
3153                 group = make_kgid(current_user_ns(), arg);
3154                 if (!gid_valid(group)) {
3155                         ret = -EINVAL;
3156                         break;
3157                 }
3158                 tun->group = group;
3159                 do_notify = true;
3160                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3161                           from_kgid(&init_user_ns, tun->group));
3162                 break;
3163
3164         case TUNSETLINK:
3165                 /* Only allow setting the type when the interface is down */
3166                 if (tun->dev->flags & IFF_UP) {
3167                         tun_debug(KERN_INFO, tun,
3168                                   "Linktype set failed because interface is up\n");
3169                         ret = -EBUSY;
3170                 } else {
3171                         tun->dev->type = (int) arg;
3172                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3173                                   tun->dev->type);
3174                         ret = 0;
3175                 }
3176                 break;
3177
3178 #ifdef TUN_DEBUG
3179         case TUNSETDEBUG:
3180                 tun->debug = arg;
3181                 break;
3182 #endif
3183         case TUNSETOFFLOAD:
3184                 ret = set_offload(tun, arg);
3185                 break;
3186
3187         case TUNSETTXFILTER:
3188                 /* Can be set only for TAPs */
3189                 ret = -EINVAL;
3190                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3191                         break;
3192                 ret = update_filter(&tun->txflt, (void __user *)arg);
3193                 break;
3194
3195         case SIOCGIFHWADDR:
3196                 /* Get hw address */
3197                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3198                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3199                 if (copy_to_user(argp, &ifr, ifreq_len))
3200                         ret = -EFAULT;
3201                 break;
3202
3203         case SIOCSIFHWADDR:
3204                 /* Set hw address */
3205                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3206                           ifr.ifr_hwaddr.sa_data);
3207
3208                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3209                 break;
3210
3211         case TUNGETSNDBUF:
3212                 sndbuf = tfile->socket.sk->sk_sndbuf;
3213                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3214                         ret = -EFAULT;
3215                 break;
3216
3217         case TUNSETSNDBUF:
3218                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3219                         ret = -EFAULT;
3220                         break;
3221                 }
3222                 if (sndbuf <= 0) {
3223                         ret = -EINVAL;
3224                         break;
3225                 }
3226
3227                 tun->sndbuf = sndbuf;
3228                 tun_set_sndbuf(tun);
3229                 break;
3230
3231         case TUNGETVNETHDRSZ:
3232                 vnet_hdr_sz = tun->vnet_hdr_sz;
3233                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3234                         ret = -EFAULT;
3235                 break;
3236
3237         case TUNSETVNETHDRSZ:
3238                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3239                         ret = -EFAULT;
3240                         break;
3241                 }
3242                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3243                         ret = -EINVAL;
3244                         break;
3245                 }
3246
3247                 tun->vnet_hdr_sz = vnet_hdr_sz;
3248                 break;
3249
3250         case TUNGETVNETLE:
3251                 le = !!(tun->flags & TUN_VNET_LE);
3252                 if (put_user(le, (int __user *)argp))
3253                         ret = -EFAULT;
3254                 break;
3255
3256         case TUNSETVNETLE:
3257                 if (get_user(le, (int __user *)argp)) {
3258                         ret = -EFAULT;
3259                         break;
3260                 }
3261                 if (le)
3262                         tun->flags |= TUN_VNET_LE;
3263                 else
3264                         tun->flags &= ~TUN_VNET_LE;
3265                 break;
3266
3267         case TUNGETVNETBE:
3268                 ret = tun_get_vnet_be(tun, argp);
3269                 break;
3270
3271         case TUNSETVNETBE:
3272                 ret = tun_set_vnet_be(tun, argp);
3273                 break;
3274
3275         case TUNATTACHFILTER:
3276                 /* Can be set only for TAPs */
3277                 ret = -EINVAL;
3278                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3279                         break;
3280                 ret = -EFAULT;
3281                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3282                         break;
3283
3284                 ret = tun_attach_filter(tun);
3285                 break;
3286
3287         case TUNDETACHFILTER:
3288                 /* Can be set only for TAPs */
3289                 ret = -EINVAL;
3290                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3291                         break;
3292                 ret = 0;
3293                 tun_detach_filter(tun, tun->numqueues);
3294                 break;
3295
3296         case TUNGETFILTER:
3297                 ret = -EINVAL;
3298                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3299                         break;
3300                 ret = -EFAULT;
3301                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3302                         break;
3303                 ret = 0;
3304                 break;
3305
3306         case TUNSETSTEERINGEBPF:
3307                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3308                 break;
3309
3310         case TUNSETFILTEREBPF:
3311                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3312                 break;
3313
3314         case TUNSETCARRIER:
3315                 ret = -EFAULT;
3316                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3317                         goto unlock;
3318
3319                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3320                 break;
3321
3322         case TUNGETDEVNETNS:
3323                 ret = -EPERM;
3324                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3325                         goto unlock;
3326                 ret = open_related_ns(&net->ns, get_net_ns);
3327                 break;
3328
3329         default:
3330                 ret = -EINVAL;
3331                 break;
3332         }
3333
3334         if (do_notify)
3335                 netdev_state_change(tun->dev);
3336
3337 unlock:
3338         rtnl_unlock();
3339         if (tun)
3340                 tun_put(tun);
3341         return ret;
3342 }
3343
3344 static long tun_chr_ioctl(struct file *file,
3345                           unsigned int cmd, unsigned long arg)
3346 {
3347         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3348 }
3349
3350 #ifdef CONFIG_COMPAT
3351 static long tun_chr_compat_ioctl(struct file *file,
3352                          unsigned int cmd, unsigned long arg)
3353 {
3354         switch (cmd) {
3355         case TUNSETIFF:
3356         case TUNGETIFF:
3357         case TUNSETTXFILTER:
3358         case TUNGETSNDBUF:
3359         case TUNSETSNDBUF:
3360         case SIOCGIFHWADDR:
3361         case SIOCSIFHWADDR:
3362                 arg = (unsigned long)compat_ptr(arg);
3363                 break;
3364         default:
3365                 arg = (compat_ulong_t)arg;
3366                 break;
3367         }
3368
3369         /*
3370          * compat_ifreq is shorter than ifreq, so we must not access beyond
3371          * the end of that structure. All fields that are used in this
3372          * driver are compatible though, we don't need to convert the
3373          * contents.
3374          */
3375         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3376 }
3377 #endif /* CONFIG_COMPAT */
3378
3379 static int tun_chr_fasync(int fd, struct file *file, int on)
3380 {
3381         struct tun_file *tfile = file->private_data;
3382         int ret;
3383
3384         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3385                 goto out;
3386
3387         if (on) {
3388                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3389                 tfile->flags |= TUN_FASYNC;
3390         } else
3391                 tfile->flags &= ~TUN_FASYNC;
3392         ret = 0;
3393 out:
3394         return ret;
3395 }
3396
3397 static int tun_chr_open(struct inode *inode, struct file * file)
3398 {
3399         struct net *net = current->nsproxy->net_ns;
3400         struct tun_file *tfile;
3401
3402         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3403
3404         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3405                                             &tun_proto, 0);
3406         if (!tfile)
3407                 return -ENOMEM;
3408         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3409                 sk_free(&tfile->sk);
3410                 return -ENOMEM;
3411         }
3412
3413         mutex_init(&tfile->napi_mutex);
3414         RCU_INIT_POINTER(tfile->tun, NULL);
3415         tfile->flags = 0;
3416         tfile->ifindex = 0;
3417
3418         init_waitqueue_head(&tfile->wq.wait);
3419         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3420
3421         tfile->socket.file = file;
3422         tfile->socket.ops = &tun_socket_ops;
3423
3424         sock_init_data(&tfile->socket, &tfile->sk);
3425
3426         tfile->sk.sk_write_space = tun_sock_write_space;
3427         tfile->sk.sk_sndbuf = INT_MAX;
3428
3429         file->private_data = tfile;
3430         INIT_LIST_HEAD(&tfile->next);
3431
3432         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3433
3434         return 0;
3435 }
3436
3437 static int tun_chr_close(struct inode *inode, struct file *file)
3438 {
3439         struct tun_file *tfile = file->private_data;
3440
3441         tun_detach(tfile, true);
3442
3443         return 0;
3444 }
3445
3446 #ifdef CONFIG_PROC_FS
3447 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3448 {
3449         struct tun_file *tfile = file->private_data;
3450         struct tun_struct *tun;
3451         struct ifreq ifr;
3452
3453         memset(&ifr, 0, sizeof(ifr));
3454
3455         rtnl_lock();
3456         tun = tun_get(tfile);
3457         if (tun)
3458                 tun_get_iff(tun, &ifr);
3459         rtnl_unlock();
3460
3461         if (tun)
3462                 tun_put(tun);
3463
3464         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3465 }
3466 #endif
3467
3468 static const struct file_operations tun_fops = {
3469         .owner  = THIS_MODULE,
3470         .llseek = no_llseek,
3471         .read_iter  = tun_chr_read_iter,
3472         .write_iter = tun_chr_write_iter,
3473         .poll   = tun_chr_poll,
3474         .unlocked_ioctl = tun_chr_ioctl,
3475 #ifdef CONFIG_COMPAT
3476         .compat_ioctl = tun_chr_compat_ioctl,
3477 #endif
3478         .open   = tun_chr_open,
3479         .release = tun_chr_close,
3480         .fasync = tun_chr_fasync,
3481 #ifdef CONFIG_PROC_FS
3482         .show_fdinfo = tun_chr_show_fdinfo,
3483 #endif
3484 };
3485
3486 static struct miscdevice tun_miscdev = {
3487         .minor = TUN_MINOR,
3488         .name = "tun",
3489         .nodename = "net/tun",
3490         .fops = &tun_fops,
3491 };
3492
3493 /* ethtool interface */
3494
3495 static void tun_default_link_ksettings(struct net_device *dev,
3496                                        struct ethtool_link_ksettings *cmd)
3497 {
3498         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3499         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3500         cmd->base.speed         = SPEED_10;
3501         cmd->base.duplex        = DUPLEX_FULL;
3502         cmd->base.port          = PORT_TP;
3503         cmd->base.phy_address   = 0;
3504         cmd->base.autoneg       = AUTONEG_DISABLE;
3505 }
3506
3507 static int tun_get_link_ksettings(struct net_device *dev,
3508                                   struct ethtool_link_ksettings *cmd)
3509 {
3510         struct tun_struct *tun = netdev_priv(dev);
3511
3512         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3513         return 0;
3514 }
3515
3516 static int tun_set_link_ksettings(struct net_device *dev,
3517                                   const struct ethtool_link_ksettings *cmd)
3518 {
3519         struct tun_struct *tun = netdev_priv(dev);
3520
3521         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3522         return 0;
3523 }
3524
3525 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3526 {
3527         struct tun_struct *tun = netdev_priv(dev);
3528
3529         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3530         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3531
3532         switch (tun->flags & TUN_TYPE_MASK) {
3533         case IFF_TUN:
3534                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3535                 break;
3536         case IFF_TAP:
3537                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3538                 break;
3539         }
3540 }
3541
3542 static u32 tun_get_msglevel(struct net_device *dev)
3543 {
3544 #ifdef TUN_DEBUG
3545         struct tun_struct *tun = netdev_priv(dev);
3546         return tun->debug;
3547 #else
3548         return -EOPNOTSUPP;
3549 #endif
3550 }
3551
3552 static void tun_set_msglevel(struct net_device *dev, u32 value)
3553 {
3554 #ifdef TUN_DEBUG
3555         struct tun_struct *tun = netdev_priv(dev);
3556         tun->debug = value;
3557 #endif
3558 }
3559
3560 static int tun_get_coalesce(struct net_device *dev,
3561                             struct ethtool_coalesce *ec)
3562 {
3563         struct tun_struct *tun = netdev_priv(dev);
3564
3565         ec->rx_max_coalesced_frames = tun->rx_batched;
3566
3567         return 0;
3568 }
3569
3570 static int tun_set_coalesce(struct net_device *dev,
3571                             struct ethtool_coalesce *ec)
3572 {
3573         struct tun_struct *tun = netdev_priv(dev);
3574
3575         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3576                 tun->rx_batched = NAPI_POLL_WEIGHT;
3577         else
3578                 tun->rx_batched = ec->rx_max_coalesced_frames;
3579
3580         return 0;
3581 }
3582
3583 static const struct ethtool_ops tun_ethtool_ops = {
3584         .get_drvinfo    = tun_get_drvinfo,
3585         .get_msglevel   = tun_get_msglevel,
3586         .set_msglevel   = tun_set_msglevel,
3587         .get_link       = ethtool_op_get_link,
3588         .get_ts_info    = ethtool_op_get_ts_info,
3589         .get_coalesce   = tun_get_coalesce,
3590         .set_coalesce   = tun_set_coalesce,
3591         .get_link_ksettings = tun_get_link_ksettings,
3592         .set_link_ksettings = tun_set_link_ksettings,
3593 };
3594
3595 static int tun_queue_resize(struct tun_struct *tun)
3596 {
3597         struct net_device *dev = tun->dev;
3598         struct tun_file *tfile;
3599         struct ptr_ring **rings;
3600         int n = tun->numqueues + tun->numdisabled;
3601         int ret, i;
3602
3603         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3604         if (!rings)
3605                 return -ENOMEM;
3606
3607         for (i = 0; i < tun->numqueues; i++) {
3608                 tfile = rtnl_dereference(tun->tfiles[i]);
3609                 rings[i] = &tfile->tx_ring;
3610         }
3611         list_for_each_entry(tfile, &tun->disabled, next)
3612                 rings[i++] = &tfile->tx_ring;
3613
3614         ret = ptr_ring_resize_multiple(rings, n,
3615                                        dev->tx_queue_len, GFP_KERNEL,
3616                                        tun_ptr_free);
3617
3618         kfree(rings);
3619         return ret;
3620 }
3621
3622 static int tun_device_event(struct notifier_block *unused,
3623                             unsigned long event, void *ptr)
3624 {
3625         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3626         struct tun_struct *tun = netdev_priv(dev);
3627         int i;
3628
3629         if (dev->rtnl_link_ops != &tun_link_ops)
3630                 return NOTIFY_DONE;
3631
3632         switch (event) {
3633         case NETDEV_CHANGE_TX_QUEUE_LEN:
3634                 if (tun_queue_resize(tun))
3635                         return NOTIFY_BAD;
3636                 break;
3637         case NETDEV_UP:
3638                 for (i = 0; i < tun->numqueues; i++) {
3639                         struct tun_file *tfile;
3640
3641                         tfile = rtnl_dereference(tun->tfiles[i]);
3642                         tfile->socket.sk->sk_write_space(tfile->socket.sk);
3643                 }
3644                 break;
3645         default:
3646                 break;
3647         }
3648
3649         return NOTIFY_DONE;
3650 }
3651
3652 static struct notifier_block tun_notifier_block __read_mostly = {
3653         .notifier_call  = tun_device_event,
3654 };
3655
3656 static int __init tun_init(void)
3657 {
3658         int ret = 0;
3659
3660         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3661
3662         ret = rtnl_link_register(&tun_link_ops);
3663         if (ret) {
3664                 pr_err("Can't register link_ops\n");
3665                 goto err_linkops;
3666         }
3667
3668         ret = misc_register(&tun_miscdev);
3669         if (ret) {
3670                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3671                 goto err_misc;
3672         }
3673
3674         ret = register_netdevice_notifier(&tun_notifier_block);
3675         if (ret) {
3676                 pr_err("Can't register netdevice notifier\n");
3677                 goto err_notifier;
3678         }
3679
3680         return  0;
3681
3682 err_notifier:
3683         misc_deregister(&tun_miscdev);
3684 err_misc:
3685         rtnl_link_unregister(&tun_link_ops);
3686 err_linkops:
3687         return ret;
3688 }
3689
3690 static void tun_cleanup(void)
3691 {
3692         misc_deregister(&tun_miscdev);
3693         rtnl_link_unregister(&tun_link_ops);
3694         unregister_netdevice_notifier(&tun_notifier_block);
3695 }
3696
3697 /* Get an underlying socket object from tun file.  Returns error unless file is
3698  * attached to a device.  The returned object works like a packet socket, it
3699  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3700  * holding a reference to the file for as long as the socket is in use. */
3701 struct socket *tun_get_socket(struct file *file)
3702 {
3703         struct tun_file *tfile;
3704         if (file->f_op != &tun_fops)
3705                 return ERR_PTR(-EINVAL);
3706         tfile = file->private_data;
3707         if (!tfile)
3708                 return ERR_PTR(-EBADFD);
3709         return &tfile->socket;
3710 }
3711 EXPORT_SYMBOL_GPL(tun_get_socket);
3712
3713 struct ptr_ring *tun_get_tx_ring(struct file *file)
3714 {
3715         struct tun_file *tfile;
3716
3717         if (file->f_op != &tun_fops)
3718                 return ERR_PTR(-EINVAL);
3719         tfile = file->private_data;
3720         if (!tfile)
3721                 return ERR_PTR(-EBADFD);
3722         return &tfile->tx_ring;
3723 }
3724 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3725
3726 module_init(tun_init);
3727 module_exit(tun_cleanup);
3728 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3729 MODULE_AUTHOR(DRV_COPYRIGHT);
3730 MODULE_LICENSE("GPL");
3731 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3732 MODULE_ALIAS("devname:net/tun");