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         struct tun_struct *tun = netdev_priv(dev);
1018         int i;
1019
1020         netif_tx_start_all_queues(dev);
1021
1022         for (i = 0; i < tun->numqueues; i++) {
1023                 struct tun_file *tfile;
1024
1025                 tfile = rtnl_dereference(tun->tfiles[i]);
1026                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1027         }
1028
1029         return 0;
1030 }
1031
1032 /* Net device close. */
1033 static int tun_net_close(struct net_device *dev)
1034 {
1035         netif_tx_stop_all_queues(dev);
1036         return 0;
1037 }
1038
1039 /* Net device start xmit */
1040 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1041 {
1042 #ifdef CONFIG_RPS
1043         if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1044                 /* Select queue was not called for the skbuff, so we extract the
1045                  * RPS hash and save it into the flow_table here.
1046                  */
1047                 struct tun_flow_entry *e;
1048                 __u32 rxhash;
1049
1050                 rxhash = __skb_get_hash_symmetric(skb);
1051                 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1052                 if (e)
1053                         tun_flow_save_rps_rxhash(e, rxhash);
1054         }
1055 #endif
1056 }
1057
1058 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1059                                     struct sk_buff *skb,
1060                                     int len)
1061 {
1062         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1063
1064         if (prog)
1065                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1066
1067         return len;
1068 }
1069
1070 /* Net device start xmit */
1071 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1072 {
1073         struct tun_struct *tun = netdev_priv(dev);
1074         int txq = skb->queue_mapping;
1075         struct tun_file *tfile;
1076         int len = skb->len;
1077
1078         rcu_read_lock();
1079         tfile = rcu_dereference(tun->tfiles[txq]);
1080
1081         /* Drop packet if interface is not attached */
1082         if (!tfile)
1083                 goto drop;
1084
1085         if (!rcu_dereference(tun->steering_prog))
1086                 tun_automq_xmit(tun, skb);
1087
1088         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1089
1090         BUG_ON(!tfile);
1091
1092         /* Drop if the filter does not like it.
1093          * This is a noop if the filter is disabled.
1094          * Filter can be enabled only for the TAP devices. */
1095         if (!check_filter(&tun->txflt, skb))
1096                 goto drop;
1097
1098         if (tfile->socket.sk->sk_filter &&
1099             sk_filter(tfile->socket.sk, skb))
1100                 goto drop;
1101
1102         len = run_ebpf_filter(tun, skb, len);
1103         if (len == 0 || pskb_trim(skb, len))
1104                 goto drop;
1105
1106         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1107                 goto drop;
1108
1109         skb_tx_timestamp(skb);
1110
1111         /* Orphan the skb - required as we might hang on to it
1112          * for indefinite time.
1113          */
1114         skb_orphan(skb);
1115
1116         nf_reset(skb);
1117
1118         if (ptr_ring_produce(&tfile->tx_ring, skb))
1119                 goto drop;
1120
1121         /* Notify and wake up reader process */
1122         if (tfile->flags & TUN_FASYNC)
1123                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1124         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1125
1126         rcu_read_unlock();
1127         return NETDEV_TX_OK;
1128
1129 drop:
1130         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1131         skb_tx_error(skb);
1132         kfree_skb(skb);
1133         rcu_read_unlock();
1134         return NET_XMIT_DROP;
1135 }
1136
1137 static void tun_net_mclist(struct net_device *dev)
1138 {
1139         /*
1140          * This callback is supposed to deal with mc filter in
1141          * _rx_ path and has nothing to do with the _tx_ path.
1142          * In rx path we always accept everything userspace gives us.
1143          */
1144 }
1145
1146 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1147         netdev_features_t features)
1148 {
1149         struct tun_struct *tun = netdev_priv(dev);
1150
1151         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1152 }
1153
1154 static void tun_set_headroom(struct net_device *dev, int new_hr)
1155 {
1156         struct tun_struct *tun = netdev_priv(dev);
1157
1158         if (new_hr < NET_SKB_PAD)
1159                 new_hr = NET_SKB_PAD;
1160
1161         tun->align = new_hr;
1162 }
1163
1164 static void
1165 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1166 {
1167         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1168         struct tun_struct *tun = netdev_priv(dev);
1169         struct tun_pcpu_stats *p;
1170         int i;
1171
1172         for_each_possible_cpu(i) {
1173                 u64 rxpackets, rxbytes, txpackets, txbytes;
1174                 unsigned int start;
1175
1176                 p = per_cpu_ptr(tun->pcpu_stats, i);
1177                 do {
1178                         start = u64_stats_fetch_begin(&p->syncp);
1179                         rxpackets       = p->rx_packets;
1180                         rxbytes         = p->rx_bytes;
1181                         txpackets       = p->tx_packets;
1182                         txbytes         = p->tx_bytes;
1183                 } while (u64_stats_fetch_retry(&p->syncp, start));
1184
1185                 stats->rx_packets       += rxpackets;
1186                 stats->rx_bytes         += rxbytes;
1187                 stats->tx_packets       += txpackets;
1188                 stats->tx_bytes         += txbytes;
1189
1190                 /* u32 counters */
1191                 rx_dropped      += p->rx_dropped;
1192                 rx_frame_errors += p->rx_frame_errors;
1193                 tx_dropped      += p->tx_dropped;
1194         }
1195         stats->rx_dropped  = rx_dropped;
1196         stats->rx_frame_errors = rx_frame_errors;
1197         stats->tx_dropped = tx_dropped;
1198 }
1199
1200 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1201                        struct netlink_ext_ack *extack)
1202 {
1203         struct tun_struct *tun = netdev_priv(dev);
1204         struct tun_file *tfile;
1205         struct bpf_prog *old_prog;
1206         int i;
1207
1208         old_prog = rtnl_dereference(tun->xdp_prog);
1209         rcu_assign_pointer(tun->xdp_prog, prog);
1210         if (old_prog)
1211                 bpf_prog_put(old_prog);
1212
1213         for (i = 0; i < tun->numqueues; i++) {
1214                 tfile = rtnl_dereference(tun->tfiles[i]);
1215                 if (prog)
1216                         sock_set_flag(&tfile->sk, SOCK_XDP);
1217                 else
1218                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1219         }
1220         list_for_each_entry(tfile, &tun->disabled, next) {
1221                 if (prog)
1222                         sock_set_flag(&tfile->sk, SOCK_XDP);
1223                 else
1224                         sock_reset_flag(&tfile->sk, SOCK_XDP);
1225         }
1226
1227         return 0;
1228 }
1229
1230 static u32 tun_xdp_query(struct net_device *dev)
1231 {
1232         struct tun_struct *tun = netdev_priv(dev);
1233         const struct bpf_prog *xdp_prog;
1234
1235         xdp_prog = rtnl_dereference(tun->xdp_prog);
1236         if (xdp_prog)
1237                 return xdp_prog->aux->id;
1238
1239         return 0;
1240 }
1241
1242 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1243 {
1244         switch (xdp->command) {
1245         case XDP_SETUP_PROG:
1246                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1247         case XDP_QUERY_PROG:
1248                 xdp->prog_id = tun_xdp_query(dev);
1249                 return 0;
1250         default:
1251                 return -EINVAL;
1252         }
1253 }
1254
1255 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1256 {
1257         if (new_carrier) {
1258                 struct tun_struct *tun = netdev_priv(dev);
1259
1260                 if (!tun->numqueues)
1261                         return -EPERM;
1262
1263                 netif_carrier_on(dev);
1264         } else {
1265                 netif_carrier_off(dev);
1266         }
1267         return 0;
1268 }
1269
1270 static const struct net_device_ops tun_netdev_ops = {
1271         .ndo_uninit             = tun_net_uninit,
1272         .ndo_open               = tun_net_open,
1273         .ndo_stop               = tun_net_close,
1274         .ndo_start_xmit         = tun_net_xmit,
1275         .ndo_fix_features       = tun_net_fix_features,
1276         .ndo_select_queue       = tun_select_queue,
1277         .ndo_set_rx_headroom    = tun_set_headroom,
1278         .ndo_get_stats64        = tun_net_get_stats64,
1279         .ndo_change_carrier     = tun_net_change_carrier,
1280 };
1281
1282 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1283 {
1284         /* Notify and wake up reader process */
1285         if (tfile->flags & TUN_FASYNC)
1286                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1287         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1288 }
1289
1290 static int tun_xdp_xmit(struct net_device *dev, int n,
1291                         struct xdp_frame **frames, u32 flags)
1292 {
1293         struct tun_struct *tun = netdev_priv(dev);
1294         struct tun_file *tfile;
1295         u32 numqueues;
1296         int drops = 0;
1297         int cnt = n;
1298         int i;
1299
1300         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1301                 return -EINVAL;
1302
1303         rcu_read_lock();
1304
1305 resample:
1306         numqueues = READ_ONCE(tun->numqueues);
1307         if (!numqueues) {
1308                 rcu_read_unlock();
1309                 return -ENXIO; /* Caller will free/return all frames */
1310         }
1311
1312         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1313                                             numqueues]);
1314         if (unlikely(!tfile))
1315                 goto resample;
1316
1317         spin_lock(&tfile->tx_ring.producer_lock);
1318         for (i = 0; i < n; i++) {
1319                 struct xdp_frame *xdp = frames[i];
1320                 /* Encode the XDP flag into lowest bit for consumer to differ
1321                  * XDP buffer from sk_buff.
1322                  */
1323                 void *frame = tun_xdp_to_ptr(xdp);
1324
1325                 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1326                         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1327                         xdp_return_frame_rx_napi(xdp);
1328                         drops++;
1329                 }
1330         }
1331         spin_unlock(&tfile->tx_ring.producer_lock);
1332
1333         if (flags & XDP_XMIT_FLUSH)
1334                 __tun_xdp_flush_tfile(tfile);
1335
1336         rcu_read_unlock();
1337         return cnt - drops;
1338 }
1339
1340 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1341 {
1342         struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1343
1344         if (unlikely(!frame))
1345                 return -EOVERFLOW;
1346
1347         return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1348 }
1349
1350 static const struct net_device_ops tap_netdev_ops = {
1351         .ndo_uninit             = tun_net_uninit,
1352         .ndo_open               = tun_net_open,
1353         .ndo_stop               = tun_net_close,
1354         .ndo_start_xmit         = tun_net_xmit,
1355         .ndo_fix_features       = tun_net_fix_features,
1356         .ndo_set_rx_mode        = tun_net_mclist,
1357         .ndo_set_mac_address    = eth_mac_addr,
1358         .ndo_validate_addr      = eth_validate_addr,
1359         .ndo_select_queue       = tun_select_queue,
1360         .ndo_features_check     = passthru_features_check,
1361         .ndo_set_rx_headroom    = tun_set_headroom,
1362         .ndo_get_stats64        = tun_net_get_stats64,
1363         .ndo_bpf                = tun_xdp,
1364         .ndo_xdp_xmit           = tun_xdp_xmit,
1365         .ndo_change_carrier     = tun_net_change_carrier,
1366 };
1367
1368 static void tun_flow_init(struct tun_struct *tun)
1369 {
1370         int i;
1371
1372         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1373                 INIT_HLIST_HEAD(&tun->flows[i]);
1374
1375         tun->ageing_time = TUN_FLOW_EXPIRE;
1376         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1377         mod_timer(&tun->flow_gc_timer,
1378                   round_jiffies_up(jiffies + tun->ageing_time));
1379 }
1380
1381 static void tun_flow_uninit(struct tun_struct *tun)
1382 {
1383         del_timer_sync(&tun->flow_gc_timer);
1384         tun_flow_flush(tun);
1385 }
1386
1387 #define MIN_MTU 68
1388 #define MAX_MTU 65535
1389
1390 /* Initialize net device. */
1391 static void tun_net_init(struct net_device *dev)
1392 {
1393         struct tun_struct *tun = netdev_priv(dev);
1394
1395         switch (tun->flags & TUN_TYPE_MASK) {
1396         case IFF_TUN:
1397                 dev->netdev_ops = &tun_netdev_ops;
1398
1399                 /* Point-to-Point TUN Device */
1400                 dev->hard_header_len = 0;
1401                 dev->addr_len = 0;
1402                 dev->mtu = 1500;
1403
1404                 /* Zero header length */
1405                 dev->type = ARPHRD_NONE;
1406                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1407                 break;
1408
1409         case IFF_TAP:
1410                 dev->netdev_ops = &tap_netdev_ops;
1411                 /* Ethernet TAP Device */
1412                 ether_setup(dev);
1413                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1414                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1415
1416                 eth_hw_addr_random(dev);
1417
1418                 break;
1419         }
1420
1421         dev->min_mtu = MIN_MTU;
1422         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1423 }
1424
1425 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1426 {
1427         struct sock *sk = tfile->socket.sk;
1428
1429         return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1430 }
1431
1432 /* Character device part */
1433
1434 /* Poll */
1435 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1436 {
1437         struct tun_file *tfile = file->private_data;
1438         struct tun_struct *tun = tun_get(tfile);
1439         struct sock *sk;
1440         __poll_t mask = 0;
1441
1442         if (!tun)
1443                 return EPOLLERR;
1444
1445         sk = tfile->socket.sk;
1446
1447         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1448
1449         poll_wait(file, sk_sleep(sk), wait);
1450
1451         if (!ptr_ring_empty(&tfile->tx_ring))
1452                 mask |= EPOLLIN | EPOLLRDNORM;
1453
1454         /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1455          * guarantee EPOLLOUT to be raised by either here or
1456          * tun_sock_write_space(). Then process could get notification
1457          * after it writes to a down device and meets -EIO.
1458          */
1459         if (tun_sock_writeable(tun, tfile) ||
1460             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1461              tun_sock_writeable(tun, tfile)))
1462                 mask |= EPOLLOUT | EPOLLWRNORM;
1463
1464         if (tun->dev->reg_state != NETREG_REGISTERED)
1465                 mask = EPOLLERR;
1466
1467         tun_put(tun);
1468         return mask;
1469 }
1470
1471 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1472                                             size_t len,
1473                                             const struct iov_iter *it)
1474 {
1475         struct sk_buff *skb;
1476         size_t linear;
1477         int err;
1478         int i;
1479
1480         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1481                 return ERR_PTR(-ENOMEM);
1482
1483         local_bh_disable();
1484         skb = napi_get_frags(&tfile->napi);
1485         local_bh_enable();
1486         if (!skb)
1487                 return ERR_PTR(-ENOMEM);
1488
1489         linear = iov_iter_single_seg_count(it);
1490         err = __skb_grow(skb, linear);
1491         if (err)
1492                 goto free;
1493
1494         skb->len = len;
1495         skb->data_len = len - linear;
1496         skb->truesize += skb->data_len;
1497
1498         for (i = 1; i < it->nr_segs; i++) {
1499                 size_t fragsz = it->iov[i].iov_len;
1500                 struct page *page;
1501                 void *frag;
1502
1503                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1504                         err = -EINVAL;
1505                         goto free;
1506                 }
1507                 frag = netdev_alloc_frag(fragsz);
1508                 if (!frag) {
1509                         err = -ENOMEM;
1510                         goto free;
1511                 }
1512                 page = virt_to_head_page(frag);
1513                 skb_fill_page_desc(skb, i - 1, page,
1514                                    frag - page_address(page), fragsz);
1515         }
1516
1517         return skb;
1518 free:
1519         /* frees skb and all frags allocated with napi_alloc_frag() */
1520         napi_free_frags(&tfile->napi);
1521         return ERR_PTR(err);
1522 }
1523
1524 /* prepad is the amount to reserve at front.  len is length after that.
1525  * linear is a hint as to how much to copy (usually headers). */
1526 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1527                                      size_t prepad, size_t len,
1528                                      size_t linear, int noblock)
1529 {
1530         struct sock *sk = tfile->socket.sk;
1531         struct sk_buff *skb;
1532         int err;
1533
1534         /* Under a page?  Don't bother with paged skb. */
1535         if (prepad + len < PAGE_SIZE || !linear)
1536                 linear = len;
1537
1538         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1539                                    &err, 0);
1540         if (!skb)
1541                 return ERR_PTR(err);
1542
1543         skb_reserve(skb, prepad);
1544         skb_put(skb, linear);
1545         skb->data_len = len - linear;
1546         skb->len += len - linear;
1547
1548         return skb;
1549 }
1550
1551 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1552                            struct sk_buff *skb, int more)
1553 {
1554         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1555         struct sk_buff_head process_queue;
1556         u32 rx_batched = tun->rx_batched;
1557         bool rcv = false;
1558
1559         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1560                 local_bh_disable();
1561                 skb_record_rx_queue(skb, tfile->queue_index);
1562                 netif_receive_skb(skb);
1563                 local_bh_enable();
1564                 return;
1565         }
1566
1567         spin_lock(&queue->lock);
1568         if (!more || skb_queue_len(queue) == rx_batched) {
1569                 __skb_queue_head_init(&process_queue);
1570                 skb_queue_splice_tail_init(queue, &process_queue);
1571                 rcv = true;
1572         } else {
1573                 __skb_queue_tail(queue, skb);
1574         }
1575         spin_unlock(&queue->lock);
1576
1577         if (rcv) {
1578                 struct sk_buff *nskb;
1579
1580                 local_bh_disable();
1581                 while ((nskb = __skb_dequeue(&process_queue))) {
1582                         skb_record_rx_queue(nskb, tfile->queue_index);
1583                         netif_receive_skb(nskb);
1584                 }
1585                 skb_record_rx_queue(skb, tfile->queue_index);
1586                 netif_receive_skb(skb);
1587                 local_bh_enable();
1588         }
1589 }
1590
1591 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1592                               int len, int noblock, bool zerocopy)
1593 {
1594         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1595                 return false;
1596
1597         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1598                 return false;
1599
1600         if (!noblock)
1601                 return false;
1602
1603         if (zerocopy)
1604                 return false;
1605
1606         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1607             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1608                 return false;
1609
1610         return true;
1611 }
1612
1613 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1614                                        int buflen, int len, int pad)
1615 {
1616         struct sk_buff *skb = build_skb(buf, buflen);
1617
1618         if (!skb)
1619                 return ERR_PTR(-ENOMEM);
1620
1621         skb_reserve(skb, pad);
1622         skb_put(skb, len);
1623
1624         get_page(alloc_frag->page);
1625         alloc_frag->offset += buflen;
1626
1627         return skb;
1628 }
1629
1630 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1631                        struct xdp_buff *xdp, u32 act)
1632 {
1633         int err;
1634
1635         switch (act) {
1636         case XDP_REDIRECT:
1637                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1638                 if (err)
1639                         return err;
1640                 break;
1641         case XDP_TX:
1642                 err = tun_xdp_tx(tun->dev, xdp);
1643                 if (err < 0)
1644                         return err;
1645                 break;
1646         case XDP_PASS:
1647                 break;
1648         default:
1649                 bpf_warn_invalid_xdp_action(act);
1650                 /* fall through */
1651         case XDP_ABORTED:
1652                 trace_xdp_exception(tun->dev, xdp_prog, act);
1653                 /* fall through */
1654         case XDP_DROP:
1655                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1656                 break;
1657         }
1658
1659         return act;
1660 }
1661
1662 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1663                                      struct tun_file *tfile,
1664                                      struct iov_iter *from,
1665                                      struct virtio_net_hdr *hdr,
1666                                      int len, int *skb_xdp)
1667 {
1668         struct page_frag *alloc_frag = &current->task_frag;
1669         struct bpf_prog *xdp_prog;
1670         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1671         char *buf;
1672         size_t copied;
1673         int pad = TUN_RX_PAD;
1674         int err = 0;
1675
1676         rcu_read_lock();
1677         xdp_prog = rcu_dereference(tun->xdp_prog);
1678         if (xdp_prog)
1679                 pad += XDP_PACKET_HEADROOM;
1680         buflen += SKB_DATA_ALIGN(len + pad);
1681         rcu_read_unlock();
1682
1683         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1684         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1685                 return ERR_PTR(-ENOMEM);
1686
1687         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1688         copied = copy_page_from_iter(alloc_frag->page,
1689                                      alloc_frag->offset + pad,
1690                                      len, from);
1691         if (copied != len)
1692                 return ERR_PTR(-EFAULT);
1693
1694         /* There's a small window that XDP may be set after the check
1695          * of xdp_prog above, this should be rare and for simplicity
1696          * we do XDP on skb in case the headroom is not enough.
1697          */
1698         if (hdr->gso_type || !xdp_prog) {
1699                 *skb_xdp = 1;
1700                 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1701         }
1702
1703         *skb_xdp = 0;
1704
1705         local_bh_disable();
1706         rcu_read_lock();
1707         xdp_prog = rcu_dereference(tun->xdp_prog);
1708         if (xdp_prog) {
1709                 struct xdp_buff xdp;
1710                 u32 act;
1711
1712                 xdp.data_hard_start = buf;
1713                 xdp.data = buf + pad;
1714                 xdp_set_data_meta_invalid(&xdp);
1715                 xdp.data_end = xdp.data + len;
1716                 xdp.rxq = &tfile->xdp_rxq;
1717
1718                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1719                 if (act == XDP_REDIRECT || act == XDP_TX) {
1720                         get_page(alloc_frag->page);
1721                         alloc_frag->offset += buflen;
1722                 }
1723                 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1724                 if (err < 0)
1725                         goto err_xdp;
1726                 if (err == XDP_REDIRECT)
1727                         xdp_do_flush_map();
1728                 if (err != XDP_PASS)
1729                         goto out;
1730
1731                 pad = xdp.data - xdp.data_hard_start;
1732                 len = xdp.data_end - xdp.data;
1733         }
1734         rcu_read_unlock();
1735         local_bh_enable();
1736
1737         return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1738
1739 err_xdp:
1740         put_page(alloc_frag->page);
1741 out:
1742         rcu_read_unlock();
1743         local_bh_enable();
1744         return NULL;
1745 }
1746
1747 /* Get packet from user space buffer */
1748 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1749                             void *msg_control, struct iov_iter *from,
1750                             int noblock, bool more)
1751 {
1752         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1753         struct sk_buff *skb;
1754         size_t total_len = iov_iter_count(from);
1755         size_t len = total_len, align = tun->align, linear;
1756         struct virtio_net_hdr gso = { 0 };
1757         struct tun_pcpu_stats *stats;
1758         int good_linear;
1759         int copylen;
1760         bool zerocopy = false;
1761         int err;
1762         u32 rxhash = 0;
1763         int skb_xdp = 1;
1764         bool frags = tun_napi_frags_enabled(tfile);
1765
1766         if (!(tun->flags & IFF_NO_PI)) {
1767                 if (len < sizeof(pi))
1768                         return -EINVAL;
1769                 len -= sizeof(pi);
1770
1771                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1772                         return -EFAULT;
1773         }
1774
1775         if (tun->flags & IFF_VNET_HDR) {
1776                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1777
1778                 if (len < vnet_hdr_sz)
1779                         return -EINVAL;
1780                 len -= vnet_hdr_sz;
1781
1782                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1783                         return -EFAULT;
1784
1785                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1786                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1787                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1788
1789                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1790                         return -EINVAL;
1791                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1792         }
1793
1794         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1795                 align += NET_IP_ALIGN;
1796                 if (unlikely(len < ETH_HLEN ||
1797                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1798                         return -EINVAL;
1799         }
1800
1801         good_linear = SKB_MAX_HEAD(align);
1802
1803         if (msg_control) {
1804                 struct iov_iter i = *from;
1805
1806                 /* There are 256 bytes to be copied in skb, so there is
1807                  * enough room for skb expand head in case it is used.
1808                  * The rest of the buffer is mapped from userspace.
1809                  */
1810                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1811                 if (copylen > good_linear)
1812                         copylen = good_linear;
1813                 linear = copylen;
1814                 iov_iter_advance(&i, copylen);
1815                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1816                         zerocopy = true;
1817         }
1818
1819         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1820                 /* For the packet that is not easy to be processed
1821                  * (e.g gso or jumbo packet), we will do it at after
1822                  * skb was created with generic XDP routine.
1823                  */
1824                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1825                 if (IS_ERR(skb)) {
1826                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1827                         return PTR_ERR(skb);
1828                 }
1829                 if (!skb)
1830                         return total_len;
1831         } else {
1832                 if (!zerocopy) {
1833                         copylen = len;
1834                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1835                                 linear = good_linear;
1836                         else
1837                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1838                 }
1839
1840                 if (frags) {
1841                         mutex_lock(&tfile->napi_mutex);
1842                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1843                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1844                          * If zerocopy is enabled, then this layout will be
1845                          * overwritten by zerocopy_sg_from_iter().
1846                          */
1847                         zerocopy = false;
1848                 } else {
1849                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1850                                             noblock);
1851                 }
1852
1853                 if (IS_ERR(skb)) {
1854                         if (PTR_ERR(skb) != -EAGAIN)
1855                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1856                         if (frags)
1857                                 mutex_unlock(&tfile->napi_mutex);
1858                         return PTR_ERR(skb);
1859                 }
1860
1861                 if (zerocopy)
1862                         err = zerocopy_sg_from_iter(skb, from);
1863                 else
1864                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1865
1866                 if (err) {
1867                         err = -EFAULT;
1868 drop:
1869                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1870                         kfree_skb(skb);
1871                         if (frags) {
1872                                 tfile->napi.skb = NULL;
1873                                 mutex_unlock(&tfile->napi_mutex);
1874                         }
1875
1876                         return err;
1877                 }
1878         }
1879
1880         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1881                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1882                 kfree_skb(skb);
1883                 if (frags) {
1884                         tfile->napi.skb = NULL;
1885                         mutex_unlock(&tfile->napi_mutex);
1886                 }
1887
1888                 return -EINVAL;
1889         }
1890
1891         switch (tun->flags & TUN_TYPE_MASK) {
1892         case IFF_TUN:
1893                 if (tun->flags & IFF_NO_PI) {
1894                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1895
1896                         switch (ip_version) {
1897                         case 4:
1898                                 pi.proto = htons(ETH_P_IP);
1899                                 break;
1900                         case 6:
1901                                 pi.proto = htons(ETH_P_IPV6);
1902                                 break;
1903                         default:
1904                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1905                                 kfree_skb(skb);
1906                                 return -EINVAL;
1907                         }
1908                 }
1909
1910                 skb_reset_mac_header(skb);
1911                 skb->protocol = pi.proto;
1912                 skb->dev = tun->dev;
1913                 break;
1914         case IFF_TAP:
1915                 if (!frags)
1916                         skb->protocol = eth_type_trans(skb, tun->dev);
1917                 break;
1918         }
1919
1920         /* copy skb_ubuf_info for callback when skb has no error */
1921         if (zerocopy) {
1922                 skb_shinfo(skb)->destructor_arg = msg_control;
1923                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1924                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1925         } else if (msg_control) {
1926                 struct ubuf_info *uarg = msg_control;
1927                 uarg->callback(uarg, false);
1928         }
1929
1930         skb_reset_network_header(skb);
1931         skb_probe_transport_header(skb);
1932
1933         if (skb_xdp) {
1934                 struct bpf_prog *xdp_prog;
1935                 int ret;
1936
1937                 local_bh_disable();
1938                 rcu_read_lock();
1939                 xdp_prog = rcu_dereference(tun->xdp_prog);
1940                 if (xdp_prog) {
1941                         ret = do_xdp_generic(xdp_prog, skb);
1942                         if (ret != XDP_PASS) {
1943                                 rcu_read_unlock();
1944                                 local_bh_enable();
1945                                 return total_len;
1946                         }
1947                 }
1948                 rcu_read_unlock();
1949                 local_bh_enable();
1950         }
1951
1952         /* Compute the costly rx hash only if needed for flow updates.
1953          * We may get a very small possibility of OOO during switching, not
1954          * worth to optimize.
1955          */
1956         if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1957             !tfile->detached)
1958                 rxhash = __skb_get_hash_symmetric(skb);
1959
1960         rcu_read_lock();
1961         if (unlikely(!(tun->dev->flags & IFF_UP))) {
1962                 err = -EIO;
1963                 rcu_read_unlock();
1964                 goto drop;
1965         }
1966
1967         if (frags) {
1968                 /* Exercise flow dissector code path. */
1969                 u32 headlen = eth_get_headlen(tun->dev, skb->data,
1970                                               skb_headlen(skb));
1971
1972                 if (unlikely(headlen > skb_headlen(skb))) {
1973                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1974                         napi_free_frags(&tfile->napi);
1975                         rcu_read_unlock();
1976                         mutex_unlock(&tfile->napi_mutex);
1977                         WARN_ON(1);
1978                         return -ENOMEM;
1979                 }
1980
1981                 local_bh_disable();
1982                 napi_gro_frags(&tfile->napi);
1983                 local_bh_enable();
1984                 mutex_unlock(&tfile->napi_mutex);
1985         } else if (tfile->napi_enabled) {
1986                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1987                 int queue_len;
1988
1989                 spin_lock_bh(&queue->lock);
1990                 __skb_queue_tail(queue, skb);
1991                 queue_len = skb_queue_len(queue);
1992                 spin_unlock(&queue->lock);
1993
1994                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1995                         napi_schedule(&tfile->napi);
1996
1997                 local_bh_enable();
1998         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1999                 tun_rx_batched(tun, tfile, skb, more);
2000         } else {
2001                 netif_rx_ni(skb);
2002         }
2003         rcu_read_unlock();
2004
2005         stats = get_cpu_ptr(tun->pcpu_stats);
2006         u64_stats_update_begin(&stats->syncp);
2007         stats->rx_packets++;
2008         stats->rx_bytes += len;
2009         u64_stats_update_end(&stats->syncp);
2010         put_cpu_ptr(stats);
2011
2012         if (rxhash)
2013                 tun_flow_update(tun, rxhash, tfile);
2014
2015         return total_len;
2016 }
2017
2018 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2019 {
2020         struct file *file = iocb->ki_filp;
2021         struct tun_file *tfile = file->private_data;
2022         struct tun_struct *tun = tun_get(tfile);
2023         ssize_t result;
2024
2025         if (!tun)
2026                 return -EBADFD;
2027
2028         result = tun_get_user(tun, tfile, NULL, from,
2029                               file->f_flags & O_NONBLOCK, false);
2030
2031         tun_put(tun);
2032         return result;
2033 }
2034
2035 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2036                                 struct tun_file *tfile,
2037                                 struct xdp_frame *xdp_frame,
2038                                 struct iov_iter *iter)
2039 {
2040         int vnet_hdr_sz = 0;
2041         size_t size = xdp_frame->len;
2042         struct tun_pcpu_stats *stats;
2043         size_t ret;
2044
2045         if (tun->flags & IFF_VNET_HDR) {
2046                 struct virtio_net_hdr gso = { 0 };
2047
2048                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2049                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2050                         return -EINVAL;
2051                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2052                              sizeof(gso)))
2053                         return -EFAULT;
2054                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2055         }
2056
2057         ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2058
2059         stats = get_cpu_ptr(tun->pcpu_stats);
2060         u64_stats_update_begin(&stats->syncp);
2061         stats->tx_packets++;
2062         stats->tx_bytes += ret;
2063         u64_stats_update_end(&stats->syncp);
2064         put_cpu_ptr(tun->pcpu_stats);
2065
2066         return ret;
2067 }
2068
2069 /* Put packet to the user space buffer */
2070 static ssize_t tun_put_user(struct tun_struct *tun,
2071                             struct tun_file *tfile,
2072                             struct sk_buff *skb,
2073                             struct iov_iter *iter)
2074 {
2075         struct tun_pi pi = { 0, skb->protocol };
2076         struct tun_pcpu_stats *stats;
2077         ssize_t total;
2078         int vlan_offset = 0;
2079         int vlan_hlen = 0;
2080         int vnet_hdr_sz = 0;
2081
2082         if (skb_vlan_tag_present(skb))
2083                 vlan_hlen = VLAN_HLEN;
2084
2085         if (tun->flags & IFF_VNET_HDR)
2086                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2087
2088         total = skb->len + vlan_hlen + vnet_hdr_sz;
2089
2090         if (!(tun->flags & IFF_NO_PI)) {
2091                 if (iov_iter_count(iter) < sizeof(pi))
2092                         return -EINVAL;
2093
2094                 total += sizeof(pi);
2095                 if (iov_iter_count(iter) < total) {
2096                         /* Packet will be striped */
2097                         pi.flags |= TUN_PKT_STRIP;
2098                 }
2099
2100                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2101                         return -EFAULT;
2102         }
2103
2104         if (vnet_hdr_sz) {
2105                 struct virtio_net_hdr gso;
2106
2107                 if (iov_iter_count(iter) < vnet_hdr_sz)
2108                         return -EINVAL;
2109
2110                 if (virtio_net_hdr_from_skb(skb, &gso,
2111                                             tun_is_little_endian(tun), true,
2112                                             vlan_hlen)) {
2113                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2114                         pr_err("unexpected GSO type: "
2115                                "0x%x, gso_size %d, hdr_len %d\n",
2116                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2117                                tun16_to_cpu(tun, gso.hdr_len));
2118                         print_hex_dump(KERN_ERR, "tun: ",
2119                                        DUMP_PREFIX_NONE,
2120                                        16, 1, skb->head,
2121                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2122                         WARN_ON_ONCE(1);
2123                         return -EINVAL;
2124                 }
2125
2126                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2127                         return -EFAULT;
2128
2129                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2130         }
2131
2132         if (vlan_hlen) {
2133                 int ret;
2134                 struct veth veth;
2135
2136                 veth.h_vlan_proto = skb->vlan_proto;
2137                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2138
2139                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2140
2141                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2142                 if (ret || !iov_iter_count(iter))
2143                         goto done;
2144
2145                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2146                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2147                         goto done;
2148         }
2149
2150         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2151
2152 done:
2153         /* caller is in process context, */
2154         stats = get_cpu_ptr(tun->pcpu_stats);
2155         u64_stats_update_begin(&stats->syncp);
2156         stats->tx_packets++;
2157         stats->tx_bytes += skb->len + vlan_hlen;
2158         u64_stats_update_end(&stats->syncp);
2159         put_cpu_ptr(tun->pcpu_stats);
2160
2161         return total;
2162 }
2163
2164 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2165 {
2166         DECLARE_WAITQUEUE(wait, current);
2167         void *ptr = NULL;
2168         int error = 0;
2169
2170         ptr = ptr_ring_consume(&tfile->tx_ring);
2171         if (ptr)
2172                 goto out;
2173         if (noblock) {
2174                 error = -EAGAIN;
2175                 goto out;
2176         }
2177
2178         add_wait_queue(&tfile->wq.wait, &wait);
2179
2180         while (1) {
2181                 set_current_state(TASK_INTERRUPTIBLE);
2182                 ptr = ptr_ring_consume(&tfile->tx_ring);
2183                 if (ptr)
2184                         break;
2185                 if (signal_pending(current)) {
2186                         error = -ERESTARTSYS;
2187                         break;
2188                 }
2189                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2190                         error = -EFAULT;
2191                         break;
2192                 }
2193
2194                 schedule();
2195         }
2196
2197         __set_current_state(TASK_RUNNING);
2198         remove_wait_queue(&tfile->wq.wait, &wait);
2199
2200 out:
2201         *err = error;
2202         return ptr;
2203 }
2204
2205 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2206                            struct iov_iter *to,
2207                            int noblock, void *ptr)
2208 {
2209         ssize_t ret;
2210         int err;
2211
2212         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2213
2214         if (!iov_iter_count(to)) {
2215                 tun_ptr_free(ptr);
2216                 return 0;
2217         }
2218
2219         if (!ptr) {
2220                 /* Read frames from ring */
2221                 ptr = tun_ring_recv(tfile, noblock, &err);
2222                 if (!ptr)
2223                         return err;
2224         }
2225
2226         if (tun_is_xdp_frame(ptr)) {
2227                 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2228
2229                 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2230                 xdp_return_frame(xdpf);
2231         } else {
2232                 struct sk_buff *skb = ptr;
2233
2234                 ret = tun_put_user(tun, tfile, skb, to);
2235                 if (unlikely(ret < 0))
2236                         kfree_skb(skb);
2237                 else
2238                         consume_skb(skb);
2239         }
2240
2241         return ret;
2242 }
2243
2244 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2245 {
2246         struct file *file = iocb->ki_filp;
2247         struct tun_file *tfile = file->private_data;
2248         struct tun_struct *tun = tun_get(tfile);
2249         ssize_t len = iov_iter_count(to), ret;
2250
2251         if (!tun)
2252                 return -EBADFD;
2253         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2254         ret = min_t(ssize_t, ret, len);
2255         if (ret > 0)
2256                 iocb->ki_pos = ret;
2257         tun_put(tun);
2258         return ret;
2259 }
2260
2261 static void tun_prog_free(struct rcu_head *rcu)
2262 {
2263         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2264
2265         bpf_prog_destroy(prog->prog);
2266         kfree(prog);
2267 }
2268
2269 static int __tun_set_ebpf(struct tun_struct *tun,
2270                           struct tun_prog __rcu **prog_p,
2271                           struct bpf_prog *prog)
2272 {
2273         struct tun_prog *old, *new = NULL;
2274
2275         if (prog) {
2276                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2277                 if (!new)
2278                         return -ENOMEM;
2279                 new->prog = prog;
2280         }
2281
2282         spin_lock_bh(&tun->lock);
2283         old = rcu_dereference_protected(*prog_p,
2284                                         lockdep_is_held(&tun->lock));
2285         rcu_assign_pointer(*prog_p, new);
2286         spin_unlock_bh(&tun->lock);
2287
2288         if (old)
2289                 call_rcu(&old->rcu, tun_prog_free);
2290
2291         return 0;
2292 }
2293
2294 static void tun_free_netdev(struct net_device *dev)
2295 {
2296         struct tun_struct *tun = netdev_priv(dev);
2297
2298         BUG_ON(!(list_empty(&tun->disabled)));
2299         free_percpu(tun->pcpu_stats);
2300         tun_flow_uninit(tun);
2301         security_tun_dev_free_security(tun->security);
2302         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2303         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2304 }
2305
2306 static void tun_setup(struct net_device *dev)
2307 {
2308         struct tun_struct *tun = netdev_priv(dev);
2309
2310         tun->owner = INVALID_UID;
2311         tun->group = INVALID_GID;
2312         tun_default_link_ksettings(dev, &tun->link_ksettings);
2313
2314         dev->ethtool_ops = &tun_ethtool_ops;
2315         dev->needs_free_netdev = true;
2316         dev->priv_destructor = tun_free_netdev;
2317         /* We prefer our own queue length */
2318         dev->tx_queue_len = TUN_READQ_SIZE;
2319 }
2320
2321 /* Trivial set of netlink ops to allow deleting tun or tap
2322  * device with netlink.
2323  */
2324 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2325                         struct netlink_ext_ack *extack)
2326 {
2327         NL_SET_ERR_MSG(extack,
2328                        "tun/tap creation via rtnetlink is not supported.");
2329         return -EOPNOTSUPP;
2330 }
2331
2332 static size_t tun_get_size(const struct net_device *dev)
2333 {
2334         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2335         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2336
2337         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2338                nla_total_size(sizeof(gid_t)) + /* GROUP */
2339                nla_total_size(sizeof(u8)) + /* TYPE */
2340                nla_total_size(sizeof(u8)) + /* PI */
2341                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2342                nla_total_size(sizeof(u8)) + /* PERSIST */
2343                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2344                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2345                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2346                0;
2347 }
2348
2349 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2350 {
2351         struct tun_struct *tun = netdev_priv(dev);
2352
2353         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2354                 goto nla_put_failure;
2355         if (uid_valid(tun->owner) &&
2356             nla_put_u32(skb, IFLA_TUN_OWNER,
2357                         from_kuid_munged(current_user_ns(), tun->owner)))
2358                 goto nla_put_failure;
2359         if (gid_valid(tun->group) &&
2360             nla_put_u32(skb, IFLA_TUN_GROUP,
2361                         from_kgid_munged(current_user_ns(), tun->group)))
2362                 goto nla_put_failure;
2363         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2364                 goto nla_put_failure;
2365         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2366                 goto nla_put_failure;
2367         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2368                 goto nla_put_failure;
2369         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2370                        !!(tun->flags & IFF_MULTI_QUEUE)))
2371                 goto nla_put_failure;
2372         if (tun->flags & IFF_MULTI_QUEUE) {
2373                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2374                         goto nla_put_failure;
2375                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2376                                 tun->numdisabled))
2377                         goto nla_put_failure;
2378         }
2379
2380         return 0;
2381
2382 nla_put_failure:
2383         return -EMSGSIZE;
2384 }
2385
2386 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2387         .kind           = DRV_NAME,
2388         .priv_size      = sizeof(struct tun_struct),
2389         .setup          = tun_setup,
2390         .validate       = tun_validate,
2391         .get_size       = tun_get_size,
2392         .fill_info      = tun_fill_info,
2393 };
2394
2395 static void tun_sock_write_space(struct sock *sk)
2396 {
2397         struct tun_file *tfile;
2398         wait_queue_head_t *wqueue;
2399
2400         if (!sock_writeable(sk))
2401                 return;
2402
2403         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2404                 return;
2405
2406         wqueue = sk_sleep(sk);
2407         if (wqueue && waitqueue_active(wqueue))
2408                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2409                                                 EPOLLWRNORM | EPOLLWRBAND);
2410
2411         tfile = container_of(sk, struct tun_file, sk);
2412         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2413 }
2414
2415 static void tun_put_page(struct tun_page *tpage)
2416 {
2417         if (tpage->page)
2418                 __page_frag_cache_drain(tpage->page, tpage->count);
2419 }
2420
2421 static int tun_xdp_one(struct tun_struct *tun,
2422                        struct tun_file *tfile,
2423                        struct xdp_buff *xdp, int *flush,
2424                        struct tun_page *tpage)
2425 {
2426         unsigned int datasize = xdp->data_end - xdp->data;
2427         struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2428         struct virtio_net_hdr *gso = &hdr->gso;
2429         struct tun_pcpu_stats *stats;
2430         struct bpf_prog *xdp_prog;
2431         struct sk_buff *skb = NULL;
2432         u32 rxhash = 0, act;
2433         int buflen = hdr->buflen;
2434         int err = 0;
2435         bool skb_xdp = false;
2436         struct page *page;
2437
2438         xdp_prog = rcu_dereference(tun->xdp_prog);
2439         if (xdp_prog) {
2440                 if (gso->gso_type) {
2441                         skb_xdp = true;
2442                         goto build;
2443                 }
2444                 xdp_set_data_meta_invalid(xdp);
2445                 xdp->rxq = &tfile->xdp_rxq;
2446
2447                 act = bpf_prog_run_xdp(xdp_prog, xdp);
2448                 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2449                 if (err < 0) {
2450                         put_page(virt_to_head_page(xdp->data));
2451                         return err;
2452                 }
2453
2454                 switch (err) {
2455                 case XDP_REDIRECT:
2456                         *flush = true;
2457                         /* fall through */
2458                 case XDP_TX:
2459                         return 0;
2460                 case XDP_PASS:
2461                         break;
2462                 default:
2463                         page = virt_to_head_page(xdp->data);
2464                         if (tpage->page == page) {
2465                                 ++tpage->count;
2466                         } else {
2467                                 tun_put_page(tpage);
2468                                 tpage->page = page;
2469                                 tpage->count = 1;
2470                         }
2471                         return 0;
2472                 }
2473         }
2474
2475 build:
2476         skb = build_skb(xdp->data_hard_start, buflen);
2477         if (!skb) {
2478                 err = -ENOMEM;
2479                 goto out;
2480         }
2481
2482         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2483         skb_put(skb, xdp->data_end - xdp->data);
2484
2485         if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2486                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2487                 kfree_skb(skb);
2488                 err = -EINVAL;
2489                 goto out;
2490         }
2491
2492         skb->protocol = eth_type_trans(skb, tun->dev);
2493         skb_reset_network_header(skb);
2494         skb_probe_transport_header(skb);
2495
2496         if (skb_xdp) {
2497                 err = do_xdp_generic(xdp_prog, skb);
2498                 if (err != XDP_PASS)
2499                         goto out;
2500         }
2501
2502         if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2503             !tfile->detached)
2504                 rxhash = __skb_get_hash_symmetric(skb);
2505
2506         skb_record_rx_queue(skb, tfile->queue_index);
2507         netif_receive_skb(skb);
2508
2509         /* No need for get_cpu_ptr() here since this function is
2510          * always called with bh disabled
2511          */
2512         stats = this_cpu_ptr(tun->pcpu_stats);
2513         u64_stats_update_begin(&stats->syncp);
2514         stats->rx_packets++;
2515         stats->rx_bytes += datasize;
2516         u64_stats_update_end(&stats->syncp);
2517
2518         if (rxhash)
2519                 tun_flow_update(tun, rxhash, tfile);
2520
2521 out:
2522         return err;
2523 }
2524
2525 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2526 {
2527         int ret, i;
2528         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2529         struct tun_struct *tun = tun_get(tfile);
2530         struct tun_msg_ctl *ctl = m->msg_control;
2531         struct xdp_buff *xdp;
2532
2533         if (!tun)
2534                 return -EBADFD;
2535
2536         if (ctl && (ctl->type == TUN_MSG_PTR)) {
2537                 struct tun_page tpage;
2538                 int n = ctl->num;
2539                 int flush = 0;
2540
2541                 memset(&tpage, 0, sizeof(tpage));
2542
2543                 local_bh_disable();
2544                 rcu_read_lock();
2545
2546                 for (i = 0; i < n; i++) {
2547                         xdp = &((struct xdp_buff *)ctl->ptr)[i];
2548                         tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2549                 }
2550
2551                 if (flush)
2552                         xdp_do_flush_map();
2553
2554                 rcu_read_unlock();
2555                 local_bh_enable();
2556
2557                 tun_put_page(&tpage);
2558
2559                 ret = total_len;
2560                 goto out;
2561         }
2562
2563         ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2564                            m->msg_flags & MSG_DONTWAIT,
2565                            m->msg_flags & MSG_MORE);
2566 out:
2567         tun_put(tun);
2568         return ret;
2569 }
2570
2571 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2572                        int flags)
2573 {
2574         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2575         struct tun_struct *tun = tun_get(tfile);
2576         void *ptr = m->msg_control;
2577         int ret;
2578
2579         if (!tun) {
2580                 ret = -EBADFD;
2581                 goto out_free;
2582         }
2583
2584         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2585                 ret = -EINVAL;
2586                 goto out_put_tun;
2587         }
2588         if (flags & MSG_ERRQUEUE) {
2589                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2590                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2591                 goto out;
2592         }
2593         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2594         if (ret > (ssize_t)total_len) {
2595                 m->msg_flags |= MSG_TRUNC;
2596                 ret = flags & MSG_TRUNC ? ret : total_len;
2597         }
2598 out:
2599         tun_put(tun);
2600         return ret;
2601
2602 out_put_tun:
2603         tun_put(tun);
2604 out_free:
2605         tun_ptr_free(ptr);
2606         return ret;
2607 }
2608
2609 static int tun_ptr_peek_len(void *ptr)
2610 {
2611         if (likely(ptr)) {
2612                 if (tun_is_xdp_frame(ptr)) {
2613                         struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2614
2615                         return xdpf->len;
2616                 }
2617                 return __skb_array_len_with_tag(ptr);
2618         } else {
2619                 return 0;
2620         }
2621 }
2622
2623 static int tun_peek_len(struct socket *sock)
2624 {
2625         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2626         struct tun_struct *tun;
2627         int ret = 0;
2628
2629         tun = tun_get(tfile);
2630         if (!tun)
2631                 return 0;
2632
2633         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2634         tun_put(tun);
2635
2636         return ret;
2637 }
2638
2639 /* Ops structure to mimic raw sockets with tun */
2640 static const struct proto_ops tun_socket_ops = {
2641         .peek_len = tun_peek_len,
2642         .sendmsg = tun_sendmsg,
2643         .recvmsg = tun_recvmsg,
2644 };
2645
2646 static struct proto tun_proto = {
2647         .name           = "tun",
2648         .owner          = THIS_MODULE,
2649         .obj_size       = sizeof(struct tun_file),
2650 };
2651
2652 static int tun_flags(struct tun_struct *tun)
2653 {
2654         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2655 }
2656
2657 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2658                               char *buf)
2659 {
2660         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2661         return sprintf(buf, "0x%x\n", tun_flags(tun));
2662 }
2663
2664 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2665                               char *buf)
2666 {
2667         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2668         return uid_valid(tun->owner)?
2669                 sprintf(buf, "%u\n",
2670                         from_kuid_munged(current_user_ns(), tun->owner)):
2671                 sprintf(buf, "-1\n");
2672 }
2673
2674 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2675                               char *buf)
2676 {
2677         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2678         return gid_valid(tun->group) ?
2679                 sprintf(buf, "%u\n",
2680                         from_kgid_munged(current_user_ns(), tun->group)):
2681                 sprintf(buf, "-1\n");
2682 }
2683
2684 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2685 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2686 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2687
2688 static struct attribute *tun_dev_attrs[] = {
2689         &dev_attr_tun_flags.attr,
2690         &dev_attr_owner.attr,
2691         &dev_attr_group.attr,
2692         NULL
2693 };
2694
2695 static const struct attribute_group tun_attr_group = {
2696         .attrs = tun_dev_attrs
2697 };
2698
2699 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2700 {
2701         struct tun_struct *tun;
2702         struct tun_file *tfile = file->private_data;
2703         struct net_device *dev;
2704         int err;
2705
2706         if (tfile->detached)
2707                 return -EINVAL;
2708
2709         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2710                 if (!capable(CAP_NET_ADMIN))
2711                         return -EPERM;
2712
2713                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2714                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2715                         return -EINVAL;
2716         }
2717
2718         dev = __dev_get_by_name(net, ifr->ifr_name);
2719         if (dev) {
2720                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2721                         return -EBUSY;
2722                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2723                         tun = netdev_priv(dev);
2724                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2725                         tun = netdev_priv(dev);
2726                 else
2727                         return -EINVAL;
2728
2729                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2730                     !!(tun->flags & IFF_MULTI_QUEUE))
2731                         return -EINVAL;
2732
2733                 if (tun_not_capable(tun))
2734                         return -EPERM;
2735                 err = security_tun_dev_open(tun->security);
2736                 if (err < 0)
2737                         return err;
2738
2739                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2740                                  ifr->ifr_flags & IFF_NAPI,
2741                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2742                 if (err < 0)
2743                         return err;
2744
2745                 if (tun->flags & IFF_MULTI_QUEUE &&
2746                     (tun->numqueues + tun->numdisabled > 1)) {
2747                         /* One or more queue has already been attached, no need
2748                          * to initialize the device again.
2749                          */
2750                         netdev_state_change(dev);
2751                         return 0;
2752                 }
2753
2754                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2755                               (ifr->ifr_flags & TUN_FEATURES);
2756
2757                 netdev_state_change(dev);
2758         } else {
2759                 char *name;
2760                 unsigned long flags = 0;
2761                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2762                              MAX_TAP_QUEUES : 1;
2763
2764                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2765                         return -EPERM;
2766                 err = security_tun_dev_create();
2767                 if (err < 0)
2768                         return err;
2769
2770                 /* Set dev type */
2771                 if (ifr->ifr_flags & IFF_TUN) {
2772                         /* TUN device */
2773                         flags |= IFF_TUN;
2774                         name = "tun%d";
2775                 } else if (ifr->ifr_flags & IFF_TAP) {
2776                         /* TAP device */
2777                         flags |= IFF_TAP;
2778                         name = "tap%d";
2779                 } else
2780                         return -EINVAL;
2781
2782                 if (*ifr->ifr_name)
2783                         name = ifr->ifr_name;
2784
2785                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2786                                        NET_NAME_UNKNOWN, tun_setup, queues,
2787                                        queues);
2788
2789                 if (!dev)
2790                         return -ENOMEM;
2791                 err = dev_get_valid_name(net, dev, name);
2792                 if (err < 0)
2793                         goto err_free_dev;
2794
2795                 dev_net_set(dev, net);
2796                 dev->rtnl_link_ops = &tun_link_ops;
2797                 dev->ifindex = tfile->ifindex;
2798                 dev->sysfs_groups[0] = &tun_attr_group;
2799
2800                 tun = netdev_priv(dev);
2801                 tun->dev = dev;
2802                 tun->flags = flags;
2803                 tun->txflt.count = 0;
2804                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2805
2806                 tun->align = NET_SKB_PAD;
2807                 tun->filter_attached = false;
2808                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2809                 tun->rx_batched = 0;
2810                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2811
2812                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2813                 if (!tun->pcpu_stats) {
2814                         err = -ENOMEM;
2815                         goto err_free_dev;
2816                 }
2817
2818                 spin_lock_init(&tun->lock);
2819
2820                 err = security_tun_dev_alloc_security(&tun->security);
2821                 if (err < 0)
2822                         goto err_free_stat;
2823
2824                 tun_net_init(dev);
2825                 tun_flow_init(tun);
2826
2827                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2828                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2829                                    NETIF_F_HW_VLAN_STAG_TX;
2830                 dev->features = dev->hw_features | NETIF_F_LLTX;
2831                 dev->vlan_features = dev->features &
2832                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2833                                        NETIF_F_HW_VLAN_STAG_TX);
2834
2835                 tun->flags = (tun->flags & ~TUN_FEATURES) |
2836                               (ifr->ifr_flags & TUN_FEATURES);
2837
2838                 INIT_LIST_HEAD(&tun->disabled);
2839                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2840                                  ifr->ifr_flags & IFF_NAPI_FRAGS);
2841                 if (err < 0)
2842                         goto err_free_flow;
2843
2844                 err = register_netdevice(tun->dev);
2845                 if (err < 0)
2846                         goto err_detach;
2847         }
2848
2849         netif_carrier_on(tun->dev);
2850
2851         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2852
2853         /* Make sure persistent devices do not get stuck in
2854          * xoff state.
2855          */
2856         if (netif_running(tun->dev))
2857                 netif_tx_wake_all_queues(tun->dev);
2858
2859         strcpy(ifr->ifr_name, tun->dev->name);
2860         return 0;
2861
2862 err_detach:
2863         tun_detach_all(dev);
2864         /* register_netdevice() already called tun_free_netdev() */
2865         goto err_free_dev;
2866
2867 err_free_flow:
2868         tun_flow_uninit(tun);
2869         security_tun_dev_free_security(tun->security);
2870 err_free_stat:
2871         free_percpu(tun->pcpu_stats);
2872 err_free_dev:
2873         free_netdev(dev);
2874         return err;
2875 }
2876
2877 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2878 {
2879         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2880
2881         strcpy(ifr->ifr_name, tun->dev->name);
2882
2883         ifr->ifr_flags = tun_flags(tun);
2884
2885 }
2886
2887 /* This is like a cut-down ethtool ops, except done via tun fd so no
2888  * privs required. */
2889 static int set_offload(struct tun_struct *tun, unsigned long arg)
2890 {
2891         netdev_features_t features = 0;
2892
2893         if (arg & TUN_F_CSUM) {
2894                 features |= NETIF_F_HW_CSUM;
2895                 arg &= ~TUN_F_CSUM;
2896
2897                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2898                         if (arg & TUN_F_TSO_ECN) {
2899                                 features |= NETIF_F_TSO_ECN;
2900                                 arg &= ~TUN_F_TSO_ECN;
2901                         }
2902                         if (arg & TUN_F_TSO4)
2903                                 features |= NETIF_F_TSO;
2904                         if (arg & TUN_F_TSO6)
2905                                 features |= NETIF_F_TSO6;
2906                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2907                 }
2908
2909                 arg &= ~TUN_F_UFO;
2910         }
2911
2912         /* This gives the user a way to test for new features in future by
2913          * trying to set them. */
2914         if (arg)
2915                 return -EINVAL;
2916
2917         tun->set_features = features;
2918         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2919         tun->dev->wanted_features |= features;
2920         netdev_update_features(tun->dev);
2921
2922         return 0;
2923 }
2924
2925 static void tun_detach_filter(struct tun_struct *tun, int n)
2926 {
2927         int i;
2928         struct tun_file *tfile;
2929
2930         for (i = 0; i < n; i++) {
2931                 tfile = rtnl_dereference(tun->tfiles[i]);
2932                 lock_sock(tfile->socket.sk);
2933                 sk_detach_filter(tfile->socket.sk);
2934                 release_sock(tfile->socket.sk);
2935         }
2936
2937         tun->filter_attached = false;
2938 }
2939
2940 static int tun_attach_filter(struct tun_struct *tun)
2941 {
2942         int i, ret = 0;
2943         struct tun_file *tfile;
2944
2945         for (i = 0; i < tun->numqueues; i++) {
2946                 tfile = rtnl_dereference(tun->tfiles[i]);
2947                 lock_sock(tfile->socket.sk);
2948                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2949                 release_sock(tfile->socket.sk);
2950                 if (ret) {
2951                         tun_detach_filter(tun, i);
2952                         return ret;
2953                 }
2954         }
2955
2956         tun->filter_attached = true;
2957         return ret;
2958 }
2959
2960 static void tun_set_sndbuf(struct tun_struct *tun)
2961 {
2962         struct tun_file *tfile;
2963         int i;
2964
2965         for (i = 0; i < tun->numqueues; i++) {
2966                 tfile = rtnl_dereference(tun->tfiles[i]);
2967                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2968         }
2969 }
2970
2971 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2972 {
2973         struct tun_file *tfile = file->private_data;
2974         struct tun_struct *tun;
2975         int ret = 0;
2976
2977         rtnl_lock();
2978
2979         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2980                 tun = tfile->detached;
2981                 if (!tun) {
2982                         ret = -EINVAL;
2983                         goto unlock;
2984                 }
2985                 ret = security_tun_dev_attach_queue(tun->security);
2986                 if (ret < 0)
2987                         goto unlock;
2988                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2989                                  tun->flags & IFF_NAPI_FRAGS);
2990         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2991                 tun = rtnl_dereference(tfile->tun);
2992                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2993                         ret = -EINVAL;
2994                 else
2995                         __tun_detach(tfile, false);
2996         } else
2997                 ret = -EINVAL;
2998
2999         if (ret >= 0)
3000                 netdev_state_change(tun->dev);
3001
3002 unlock:
3003         rtnl_unlock();
3004         return ret;
3005 }
3006
3007 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3008                         void __user *data)
3009 {
3010         struct bpf_prog *prog;
3011         int fd;
3012
3013         if (copy_from_user(&fd, data, sizeof(fd)))
3014                 return -EFAULT;
3015
3016         if (fd == -1) {
3017                 prog = NULL;
3018         } else {
3019                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3020                 if (IS_ERR(prog))
3021                         return PTR_ERR(prog);
3022         }
3023
3024         return __tun_set_ebpf(tun, prog_p, prog);
3025 }
3026
3027 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3028                             unsigned long arg, int ifreq_len)
3029 {
3030         struct tun_file *tfile = file->private_data;
3031         struct net *net = sock_net(&tfile->sk);
3032         struct tun_struct *tun;
3033         void __user* argp = (void __user*)arg;
3034         unsigned int ifindex, carrier;
3035         struct ifreq ifr;
3036         kuid_t owner;
3037         kgid_t group;
3038         int sndbuf;
3039         int vnet_hdr_sz;
3040         int le;
3041         int ret;
3042         bool do_notify = false;
3043
3044         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3045             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3046                 if (copy_from_user(&ifr, argp, ifreq_len))
3047                         return -EFAULT;
3048         } else {
3049                 memset(&ifr, 0, sizeof(ifr));
3050         }
3051         if (cmd == TUNGETFEATURES) {
3052                 /* Currently this just means: "what IFF flags are valid?".
3053                  * This is needed because we never checked for invalid flags on
3054                  * TUNSETIFF.
3055                  */
3056                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3057                                 (unsigned int __user*)argp);
3058         } else if (cmd == TUNSETQUEUE) {
3059                 return tun_set_queue(file, &ifr);
3060         } else if (cmd == SIOCGSKNS) {
3061                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3062                         return -EPERM;
3063                 return open_related_ns(&net->ns, get_net_ns);
3064         }
3065
3066         ret = 0;
3067         rtnl_lock();
3068
3069         tun = tun_get(tfile);
3070         if (cmd == TUNSETIFF) {
3071                 ret = -EEXIST;
3072                 if (tun)
3073                         goto unlock;
3074
3075                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3076
3077                 ret = tun_set_iff(net, file, &ifr);
3078
3079                 if (ret)
3080                         goto unlock;
3081
3082                 if (copy_to_user(argp, &ifr, ifreq_len))
3083                         ret = -EFAULT;
3084                 goto unlock;
3085         }
3086         if (cmd == TUNSETIFINDEX) {
3087                 ret = -EPERM;
3088                 if (tun)
3089                         goto unlock;
3090
3091                 ret = -EFAULT;
3092                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3093                         goto unlock;
3094
3095                 ret = 0;
3096                 tfile->ifindex = ifindex;
3097                 goto unlock;
3098         }
3099
3100         ret = -EBADFD;
3101         if (!tun)
3102                 goto unlock;
3103
3104         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3105
3106         net = dev_net(tun->dev);
3107         ret = 0;
3108         switch (cmd) {
3109         case TUNGETIFF:
3110                 tun_get_iff(tun, &ifr);
3111
3112                 if (tfile->detached)
3113                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
3114                 if (!tfile->socket.sk->sk_filter)
3115                         ifr.ifr_flags |= IFF_NOFILTER;
3116
3117                 if (copy_to_user(argp, &ifr, ifreq_len))
3118                         ret = -EFAULT;
3119                 break;
3120
3121         case TUNSETNOCSUM:
3122                 /* Disable/Enable checksum */
3123
3124                 /* [unimplemented] */
3125                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3126                           arg ? "disabled" : "enabled");
3127                 break;
3128
3129         case TUNSETPERSIST:
3130                 /* Disable/Enable persist mode. Keep an extra reference to the
3131                  * module to prevent the module being unprobed.
3132                  */
3133                 if (arg && !(tun->flags & IFF_PERSIST)) {
3134                         tun->flags |= IFF_PERSIST;
3135                         __module_get(THIS_MODULE);
3136                         do_notify = true;
3137                 }
3138                 if (!arg && (tun->flags & IFF_PERSIST)) {
3139                         tun->flags &= ~IFF_PERSIST;
3140                         module_put(THIS_MODULE);
3141                         do_notify = true;
3142                 }
3143
3144                 tun_debug(KERN_INFO, tun, "persist %s\n",
3145                           arg ? "enabled" : "disabled");
3146                 break;
3147
3148         case TUNSETOWNER:
3149                 /* Set owner of the device */
3150                 owner = make_kuid(current_user_ns(), arg);
3151                 if (!uid_valid(owner)) {
3152                         ret = -EINVAL;
3153                         break;
3154                 }
3155                 tun->owner = owner;
3156                 do_notify = true;
3157                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3158                           from_kuid(&init_user_ns, tun->owner));
3159                 break;
3160
3161         case TUNSETGROUP:
3162                 /* Set group of the device */
3163                 group = make_kgid(current_user_ns(), arg);
3164                 if (!gid_valid(group)) {
3165                         ret = -EINVAL;
3166                         break;
3167                 }
3168                 tun->group = group;
3169                 do_notify = true;
3170                 tun_debug(KERN_INFO, tun, "group set to %u\n",
3171                           from_kgid(&init_user_ns, tun->group));
3172                 break;
3173
3174         case TUNSETLINK:
3175                 /* Only allow setting the type when the interface is down */
3176                 if (tun->dev->flags & IFF_UP) {
3177                         tun_debug(KERN_INFO, tun,
3178                                   "Linktype set failed because interface is up\n");
3179                         ret = -EBUSY;
3180                 } else {
3181                         tun->dev->type = (int) arg;
3182                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3183                                   tun->dev->type);
3184                         ret = 0;
3185                 }
3186                 break;
3187
3188 #ifdef TUN_DEBUG
3189         case TUNSETDEBUG:
3190                 tun->debug = arg;
3191                 break;
3192 #endif
3193         case TUNSETOFFLOAD:
3194                 ret = set_offload(tun, arg);
3195                 break;
3196
3197         case TUNSETTXFILTER:
3198                 /* Can be set only for TAPs */
3199                 ret = -EINVAL;
3200                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3201                         break;
3202                 ret = update_filter(&tun->txflt, (void __user *)arg);
3203                 break;
3204
3205         case SIOCGIFHWADDR:
3206                 /* Get hw address */
3207                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3208                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3209                 if (copy_to_user(argp, &ifr, ifreq_len))
3210                         ret = -EFAULT;
3211                 break;
3212
3213         case SIOCSIFHWADDR:
3214                 /* Set hw address */
3215                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3216                           ifr.ifr_hwaddr.sa_data);
3217
3218                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3219                 break;
3220
3221         case TUNGETSNDBUF:
3222                 sndbuf = tfile->socket.sk->sk_sndbuf;
3223                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3224                         ret = -EFAULT;
3225                 break;
3226
3227         case TUNSETSNDBUF:
3228                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3229                         ret = -EFAULT;
3230                         break;
3231                 }
3232                 if (sndbuf <= 0) {
3233                         ret = -EINVAL;
3234                         break;
3235                 }
3236
3237                 tun->sndbuf = sndbuf;
3238                 tun_set_sndbuf(tun);
3239                 break;
3240
3241         case TUNGETVNETHDRSZ:
3242                 vnet_hdr_sz = tun->vnet_hdr_sz;
3243                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3244                         ret = -EFAULT;
3245                 break;
3246
3247         case TUNSETVNETHDRSZ:
3248                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3249                         ret = -EFAULT;
3250                         break;
3251                 }
3252                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3253                         ret = -EINVAL;
3254                         break;
3255                 }
3256
3257                 tun->vnet_hdr_sz = vnet_hdr_sz;
3258                 break;
3259
3260         case TUNGETVNETLE:
3261                 le = !!(tun->flags & TUN_VNET_LE);
3262                 if (put_user(le, (int __user *)argp))
3263                         ret = -EFAULT;
3264                 break;
3265
3266         case TUNSETVNETLE:
3267                 if (get_user(le, (int __user *)argp)) {
3268                         ret = -EFAULT;
3269                         break;
3270                 }
3271                 if (le)
3272                         tun->flags |= TUN_VNET_LE;
3273                 else
3274                         tun->flags &= ~TUN_VNET_LE;
3275                 break;
3276
3277         case TUNGETVNETBE:
3278                 ret = tun_get_vnet_be(tun, argp);
3279                 break;
3280
3281         case TUNSETVNETBE:
3282                 ret = tun_set_vnet_be(tun, argp);
3283                 break;
3284
3285         case TUNATTACHFILTER:
3286                 /* Can be set only for TAPs */
3287                 ret = -EINVAL;
3288                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3289                         break;
3290                 ret = -EFAULT;
3291                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3292                         break;
3293
3294                 ret = tun_attach_filter(tun);
3295                 break;
3296
3297         case TUNDETACHFILTER:
3298                 /* Can be set only for TAPs */
3299                 ret = -EINVAL;
3300                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3301                         break;
3302                 ret = 0;
3303                 tun_detach_filter(tun, tun->numqueues);
3304                 break;
3305
3306         case TUNGETFILTER:
3307                 ret = -EINVAL;
3308                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3309                         break;
3310                 ret = -EFAULT;
3311                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3312                         break;
3313                 ret = 0;
3314                 break;
3315
3316         case TUNSETSTEERINGEBPF:
3317                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3318                 break;
3319
3320         case TUNSETFILTEREBPF:
3321                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3322                 break;
3323
3324         case TUNSETCARRIER:
3325                 ret = -EFAULT;
3326                 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3327                         goto unlock;
3328
3329                 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3330                 break;
3331
3332         case TUNGETDEVNETNS:
3333                 ret = -EPERM;
3334                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3335                         goto unlock;
3336                 ret = open_related_ns(&net->ns, get_net_ns);
3337                 break;
3338
3339         default:
3340                 ret = -EINVAL;
3341                 break;
3342         }
3343
3344         if (do_notify)
3345                 netdev_state_change(tun->dev);
3346
3347 unlock:
3348         rtnl_unlock();
3349         if (tun)
3350                 tun_put(tun);
3351         return ret;
3352 }
3353
3354 static long tun_chr_ioctl(struct file *file,
3355                           unsigned int cmd, unsigned long arg)
3356 {
3357         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3358 }
3359
3360 #ifdef CONFIG_COMPAT
3361 static long tun_chr_compat_ioctl(struct file *file,
3362                          unsigned int cmd, unsigned long arg)
3363 {
3364         switch (cmd) {
3365         case TUNSETIFF:
3366         case TUNGETIFF:
3367         case TUNSETTXFILTER:
3368         case TUNGETSNDBUF:
3369         case TUNSETSNDBUF:
3370         case SIOCGIFHWADDR:
3371         case SIOCSIFHWADDR:
3372                 arg = (unsigned long)compat_ptr(arg);
3373                 break;
3374         default:
3375                 arg = (compat_ulong_t)arg;
3376                 break;
3377         }
3378
3379         /*
3380          * compat_ifreq is shorter than ifreq, so we must not access beyond
3381          * the end of that structure. All fields that are used in this
3382          * driver are compatible though, we don't need to convert the
3383          * contents.
3384          */
3385         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3386 }
3387 #endif /* CONFIG_COMPAT */
3388
3389 static int tun_chr_fasync(int fd, struct file *file, int on)
3390 {
3391         struct tun_file *tfile = file->private_data;
3392         int ret;
3393
3394         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3395                 goto out;
3396
3397         if (on) {
3398                 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3399                 tfile->flags |= TUN_FASYNC;
3400         } else
3401                 tfile->flags &= ~TUN_FASYNC;
3402         ret = 0;
3403 out:
3404         return ret;
3405 }
3406
3407 static int tun_chr_open(struct inode *inode, struct file * file)
3408 {
3409         struct net *net = current->nsproxy->net_ns;
3410         struct tun_file *tfile;
3411
3412         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3413
3414         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3415                                             &tun_proto, 0);
3416         if (!tfile)
3417                 return -ENOMEM;
3418         if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3419                 sk_free(&tfile->sk);
3420                 return -ENOMEM;
3421         }
3422
3423         mutex_init(&tfile->napi_mutex);
3424         RCU_INIT_POINTER(tfile->tun, NULL);
3425         tfile->flags = 0;
3426         tfile->ifindex = 0;
3427
3428         init_waitqueue_head(&tfile->wq.wait);
3429         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3430
3431         tfile->socket.file = file;
3432         tfile->socket.ops = &tun_socket_ops;
3433
3434         sock_init_data(&tfile->socket, &tfile->sk);
3435
3436         tfile->sk.sk_write_space = tun_sock_write_space;
3437         tfile->sk.sk_sndbuf = INT_MAX;
3438
3439         file->private_data = tfile;
3440         INIT_LIST_HEAD(&tfile->next);
3441
3442         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3443
3444         return 0;
3445 }
3446
3447 static int tun_chr_close(struct inode *inode, struct file *file)
3448 {
3449         struct tun_file *tfile = file->private_data;
3450
3451         tun_detach(tfile, true);
3452
3453         return 0;
3454 }
3455
3456 #ifdef CONFIG_PROC_FS
3457 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3458 {
3459         struct tun_file *tfile = file->private_data;
3460         struct tun_struct *tun;
3461         struct ifreq ifr;
3462
3463         memset(&ifr, 0, sizeof(ifr));
3464
3465         rtnl_lock();
3466         tun = tun_get(tfile);
3467         if (tun)
3468                 tun_get_iff(tun, &ifr);
3469         rtnl_unlock();
3470
3471         if (tun)
3472                 tun_put(tun);
3473
3474         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3475 }
3476 #endif
3477
3478 static const struct file_operations tun_fops = {
3479         .owner  = THIS_MODULE,
3480         .llseek = no_llseek,
3481         .read_iter  = tun_chr_read_iter,
3482         .write_iter = tun_chr_write_iter,
3483         .poll   = tun_chr_poll,
3484         .unlocked_ioctl = tun_chr_ioctl,
3485 #ifdef CONFIG_COMPAT
3486         .compat_ioctl = tun_chr_compat_ioctl,
3487 #endif
3488         .open   = tun_chr_open,
3489         .release = tun_chr_close,
3490         .fasync = tun_chr_fasync,
3491 #ifdef CONFIG_PROC_FS
3492         .show_fdinfo = tun_chr_show_fdinfo,
3493 #endif
3494 };
3495
3496 static struct miscdevice tun_miscdev = {
3497         .minor = TUN_MINOR,
3498         .name = "tun",
3499         .nodename = "net/tun",
3500         .fops = &tun_fops,
3501 };
3502
3503 /* ethtool interface */
3504
3505 static void tun_default_link_ksettings(struct net_device *dev,
3506                                        struct ethtool_link_ksettings *cmd)
3507 {
3508         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3509         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3510         cmd->base.speed         = SPEED_10;
3511         cmd->base.duplex        = DUPLEX_FULL;
3512         cmd->base.port          = PORT_TP;
3513         cmd->base.phy_address   = 0;
3514         cmd->base.autoneg       = AUTONEG_DISABLE;
3515 }
3516
3517 static int tun_get_link_ksettings(struct net_device *dev,
3518                                   struct ethtool_link_ksettings *cmd)
3519 {
3520         struct tun_struct *tun = netdev_priv(dev);
3521
3522         memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3523         return 0;
3524 }
3525
3526 static int tun_set_link_ksettings(struct net_device *dev,
3527                                   const struct ethtool_link_ksettings *cmd)
3528 {
3529         struct tun_struct *tun = netdev_priv(dev);
3530
3531         memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3532         return 0;
3533 }
3534
3535 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3536 {
3537         struct tun_struct *tun = netdev_priv(dev);
3538
3539         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3540         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3541
3542         switch (tun->flags & TUN_TYPE_MASK) {
3543         case IFF_TUN:
3544                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3545                 break;
3546         case IFF_TAP:
3547                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3548                 break;
3549         }
3550 }
3551
3552 static u32 tun_get_msglevel(struct net_device *dev)
3553 {
3554 #ifdef TUN_DEBUG
3555         struct tun_struct *tun = netdev_priv(dev);
3556         return tun->debug;
3557 #else
3558         return -EOPNOTSUPP;
3559 #endif
3560 }
3561
3562 static void tun_set_msglevel(struct net_device *dev, u32 value)
3563 {
3564 #ifdef TUN_DEBUG
3565         struct tun_struct *tun = netdev_priv(dev);
3566         tun->debug = value;
3567 #endif
3568 }
3569
3570 static int tun_get_coalesce(struct net_device *dev,
3571                             struct ethtool_coalesce *ec)
3572 {
3573         struct tun_struct *tun = netdev_priv(dev);
3574
3575         ec->rx_max_coalesced_frames = tun->rx_batched;
3576
3577         return 0;
3578 }
3579
3580 static int tun_set_coalesce(struct net_device *dev,
3581                             struct ethtool_coalesce *ec)
3582 {
3583         struct tun_struct *tun = netdev_priv(dev);
3584
3585         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3586                 tun->rx_batched = NAPI_POLL_WEIGHT;
3587         else
3588                 tun->rx_batched = ec->rx_max_coalesced_frames;
3589
3590         return 0;
3591 }
3592
3593 static const struct ethtool_ops tun_ethtool_ops = {
3594         .get_drvinfo    = tun_get_drvinfo,
3595         .get_msglevel   = tun_get_msglevel,
3596         .set_msglevel   = tun_set_msglevel,
3597         .get_link       = ethtool_op_get_link,
3598         .get_ts_info    = ethtool_op_get_ts_info,
3599         .get_coalesce   = tun_get_coalesce,
3600         .set_coalesce   = tun_set_coalesce,
3601         .get_link_ksettings = tun_get_link_ksettings,
3602         .set_link_ksettings = tun_set_link_ksettings,
3603 };
3604
3605 static int tun_queue_resize(struct tun_struct *tun)
3606 {
3607         struct net_device *dev = tun->dev;
3608         struct tun_file *tfile;
3609         struct ptr_ring **rings;
3610         int n = tun->numqueues + tun->numdisabled;
3611         int ret, i;
3612
3613         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3614         if (!rings)
3615                 return -ENOMEM;
3616
3617         for (i = 0; i < tun->numqueues; i++) {
3618                 tfile = rtnl_dereference(tun->tfiles[i]);
3619                 rings[i] = &tfile->tx_ring;
3620         }
3621         list_for_each_entry(tfile, &tun->disabled, next)
3622                 rings[i++] = &tfile->tx_ring;
3623
3624         ret = ptr_ring_resize_multiple(rings, n,
3625                                        dev->tx_queue_len, GFP_KERNEL,
3626                                        tun_ptr_free);
3627
3628         kfree(rings);
3629         return ret;
3630 }
3631
3632 static int tun_device_event(struct notifier_block *unused,
3633                             unsigned long event, void *ptr)
3634 {
3635         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3636         struct tun_struct *tun = netdev_priv(dev);
3637
3638         if (dev->rtnl_link_ops != &tun_link_ops)
3639                 return NOTIFY_DONE;
3640
3641         switch (event) {
3642         case NETDEV_CHANGE_TX_QUEUE_LEN:
3643                 if (tun_queue_resize(tun))
3644                         return NOTIFY_BAD;
3645                 break;
3646         default:
3647                 break;
3648         }
3649
3650         return NOTIFY_DONE;
3651 }
3652
3653 static struct notifier_block tun_notifier_block __read_mostly = {
3654         .notifier_call  = tun_device_event,
3655 };
3656
3657 static int __init tun_init(void)
3658 {
3659         int ret = 0;
3660
3661         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3662
3663         ret = rtnl_link_register(&tun_link_ops);
3664         if (ret) {
3665                 pr_err("Can't register link_ops\n");
3666                 goto err_linkops;
3667         }
3668
3669         ret = misc_register(&tun_miscdev);
3670         if (ret) {
3671                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3672                 goto err_misc;
3673         }
3674
3675         ret = register_netdevice_notifier(&tun_notifier_block);
3676         if (ret) {
3677                 pr_err("Can't register netdevice notifier\n");
3678                 goto err_notifier;
3679         }
3680
3681         return  0;
3682
3683 err_notifier:
3684         misc_deregister(&tun_miscdev);
3685 err_misc:
3686         rtnl_link_unregister(&tun_link_ops);
3687 err_linkops:
3688         return ret;
3689 }
3690
3691 static void tun_cleanup(void)
3692 {
3693         misc_deregister(&tun_miscdev);
3694         rtnl_link_unregister(&tun_link_ops);
3695         unregister_netdevice_notifier(&tun_notifier_block);
3696 }
3697
3698 /* Get an underlying socket object from tun file.  Returns error unless file is
3699  * attached to a device.  The returned object works like a packet socket, it
3700  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3701  * holding a reference to the file for as long as the socket is in use. */
3702 struct socket *tun_get_socket(struct file *file)
3703 {
3704         struct tun_file *tfile;
3705         if (file->f_op != &tun_fops)
3706                 return ERR_PTR(-EINVAL);
3707         tfile = file->private_data;
3708         if (!tfile)
3709                 return ERR_PTR(-EBADFD);
3710         return &tfile->socket;
3711 }
3712 EXPORT_SYMBOL_GPL(tun_get_socket);
3713
3714 struct ptr_ring *tun_get_tx_ring(struct file *file)
3715 {
3716         struct tun_file *tfile;
3717
3718         if (file->f_op != &tun_fops)
3719                 return ERR_PTR(-EINVAL);
3720         tfile = file->private_data;
3721         if (!tfile)
3722                 return ERR_PTR(-EBADFD);
3723         return &tfile->tx_ring;
3724 }
3725 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3726
3727 module_init(tun_init);
3728 module_exit(tun_cleanup);
3729 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3730 MODULE_AUTHOR(DRV_COPYRIGHT);
3731 MODULE_LICENSE("GPL");
3732 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3733 MODULE_ALIAS("devname:net/tun");