drivers/net/tun.c

   1 /*
   2  *  TUN - Universal TUN/TAP device driver.
   3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   4  *
   5  *  This program is free software; you can redistribute it and/or modify
   6  *  it under the terms of the GNU General Public License as published by
   7  *  the Free Software Foundation; either version 2 of the License, or
   8  *  (at your option) any later version.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
  16  */
  17
  18 /*
  19  *  Changes:
  20  *
  21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  22  *    Add TUNSETLINK ioctl to set the link encapsulation
  23  *
  24  *  Mark Smith <markzzzsmith@yahoo.com.au>
  25  *    Use eth_random_addr() for tap MAC address.
  26  *
  27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  28  *    Fixes in packet dropping, queue length setting and queue wakeup.
  29  *    Increased default tx queue length.
  30  *    Added ethtool API.
  31  *    Minor cleanups
  32  *
  33  *  Daniel Podlejski <underley@underley.eu.org>
  34  *    Modifications for 2.3.99-pre5 kernel.
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #define DRV_NAME        "tun"
  40 #define DRV_VERSION     "1.6"
  41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
  42 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  43
  44 #include <linux/module.h>
  45 #include <linux/errno.h>
  46 #include <linux/kernel.h>
  47 #include <linux/sched/signal.h>
  48 #include <linux/major.h>
  49 #include <linux/slab.h>
  50 #include <linux/poll.h>
  51 #include <linux/fcntl.h>
  52 #include <linux/init.h>
  53 #include <linux/skbuff.h>
  54 #include <linux/netdevice.h>
  55 #include <linux/etherdevice.h>
  56 #include <linux/miscdevice.h>
  57 #include <linux/ethtool.h>
  58 #include <linux/rtnetlink.h>
  59 #include <linux/compat.h>
  60 #include <linux/if.h>
  61 #include <linux/if_arp.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/if_tun.h>
  64 #include <linux/if_vlan.h>
  65 #include <linux/crc32.h>
  66 #include <linux/nsproxy.h>
  67 #include <linux/virtio_net.h>
  68 #include <linux/rcupdate.h>
  69 #include <net/net_namespace.h>
  70 #include <net/netns/generic.h>
  71 #include <net/rtnetlink.h>
  72 #include <net/sock.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/uio.h>
  75 #include <linux/skb_array.h>
  76 #include <linux/bpf.h>
  77 #include <linux/bpf_trace.h>
  78 #include <linux/mutex.h>
  79
  80 #include <linux/uaccess.h>
  81 #include <linux/proc_fs.h>
  82
  83 /* Uncomment to enable debugging */
  84 /* #define TUN_DEBUG 1 */
  85
  86 #ifdef TUN_DEBUG
  87 static int debug;
  88
  89 #define tun_debug(level, tun, fmt, args...)                     \
  90 do {                                                            \
  91         if (tun->debug)                                         \
  92                 netdev_printk(level, tun->dev, fmt, ##args);    \
  93 } while (0)
  94 #define DBG1(level, fmt, args...)                               \
  95 do {                                                            \
  96         if (debug == 2)                                         \
  97                 printk(level fmt, ##args);                      \
  98 } while (0)
  99 #else
 100 #define tun_debug(level, tun, fmt, args...)                     \
 101 do {                                                            \
 102         if (0)                                                  \
 103                 netdev_printk(level, tun->dev, fmt, ##args);    \
 104 } while (0)
 105 #define DBG1(level, fmt, args...)                               \
 106 do {                                                            \
 107         if (0)                                                  \
 108                 printk(level fmt, ##args);                      \
 109 } while (0)
 110 #endif
 111
 112 #define TUN_HEADROOM 256
 113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 114
 115 /* TUN device flags */
 116
 117 /* IFF_ATTACH_QUEUE is never stored in device flags,
 118  * overload it to mean fasync when stored there.
 119  */
 120 #define TUN_FASYNC      IFF_ATTACH_QUEUE
 121 /* High bits in flags field are unused. */
 122 #define TUN_VNET_LE     0x80000000
 123 #define TUN_VNET_BE     0x40000000
 124
 125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 126                       IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 127
 128 #define GOODCOPY_LEN 128
 129
 130 #define FLT_EXACT_COUNT 8
 131 struct tap_filter {
 132         unsigned int    count;    /* Number of addrs. Zero means disabled */
 133         u32             mask[2];  /* Mask of the hashed addrs */
 134         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
 135 };
 136
 137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 138  * to max number of VCPUs in guest. */
 139 #define MAX_TAP_QUEUES 256
 140 #define MAX_TAP_FLOWS  4096
 141
 142 #define TUN_FLOW_EXPIRE (3 * HZ)
 143
 144 struct tun_pcpu_stats {
 145         u64 rx_packets;
 146         u64 rx_bytes;
 147         u64 tx_packets;
 148         u64 tx_bytes;
 149         struct u64_stats_sync syncp;
 150         u32 rx_dropped;
 151         u32 tx_dropped;
 152         u32 rx_frame_errors;
 153 };
 154
 155 /* A tun_file connects an open character device to a tuntap netdevice. It
 156  * also contains all socket related structures (except sock_fprog and tap_filter)
 157  * to serve as one transmit queue for tuntap device. The sock_fprog and
 158  * tap_filter were kept in tun_struct since they were used for filtering for the
 159  * netdevice not for a specific queue (at least I didn't see the requirement for
 160  * this).
 161  *
 162  * RCU usage:
 163  * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 164  * other can only be read while rcu_read_lock or rtnl_lock is held.
 165  */
 166 struct tun_file {
 167         struct sock sk;
 168         struct socket socket;
 169         struct socket_wq wq;
 170         struct tun_struct __rcu *tun;
 171         struct fasync_struct *fasync;
 172         /* only used for fasnyc */
 173         unsigned int flags;
 174         union {
 175                 u16 queue_index;
 176                 unsigned int ifindex;
 177         };
 178         struct napi_struct napi;
 179         bool napi_enabled;
 180         struct mutex napi_mutex;        /* Protects access to the above napi */
 181         struct list_head next;
 182         struct tun_struct *detached;
 183         struct ptr_ring tx_ring;
 184         struct xdp_rxq_info xdp_rxq;
 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;
 196 };
 197
 198 #define TUN_NUM_FLOW_ENTRIES 1024
 199
 200 struct tun_prog {
 201         struct rcu_head rcu;
 202         struct bpf_prog *prog;
 203 };
 204
 205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
 206  * device, socket filter, sndbuf and vnet header size were restore when the
 207  * file were attached to a persist device.
 208  */
 209 struct tun_struct {
 210         struct tun_file __rcu   *tfiles[MAX_TAP_QUEUES];
 211         unsigned int            numqueues;
 212         unsigned int            flags;
 213         kuid_t                  owner;
 214         kgid_t                  group;
 215
 216         struct net_device       *dev;
 217         netdev_features_t       set_features;
 218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 219                           NETIF_F_TSO6)
 220
 221         int                     align;
 222         int                     vnet_hdr_sz;
 223         int                     sndbuf;
 224         struct tap_filter       txflt;
 225         struct sock_fprog       fprog;
 226         /* protected by rtnl lock */
 227         bool                    filter_attached;
 228 #ifdef TUN_DEBUG
 229         int debug;
 230 #endif
 231         spinlock_t lock;
 232         struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 233         struct timer_list flow_gc_timer;
 234         unsigned long ageing_time;
 235         unsigned int numdisabled;
 236         struct list_head disabled;
 237         void *security;
 238         u32 flow_count;
 239         u32 rx_batched;
 240         struct tun_pcpu_stats __percpu *pcpu_stats;
 241         struct bpf_prog __rcu *xdp_prog;
 242         struct tun_prog __rcu *steering_prog;
 243         struct tun_prog __rcu *filter_prog;
 244 };
 245
 246 struct veth {
 247         __be16 h_vlan_proto;
 248         __be16 h_vlan_TCI;
 249 };
 250
 251 bool tun_is_xdp_buff(void *ptr)
 252 {
 253         return (unsigned long)ptr & TUN_XDP_FLAG;
 254 }
 255 EXPORT_SYMBOL(tun_is_xdp_buff);
 256
 257 void *tun_xdp_to_ptr(void *ptr)
 258 {
 259         return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
 260 }
 261 EXPORT_SYMBOL(tun_xdp_to_ptr);
 262
 263 void *tun_ptr_to_xdp(void *ptr)
 264 {
 265         return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
 266 }
 267 EXPORT_SYMBOL(tun_ptr_to_xdp);
 268
 269 static int tun_napi_receive(struct napi_struct *napi, int budget)
 270 {
 271         struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 272         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 273         struct sk_buff_head process_queue;
 274         struct sk_buff *skb;
 275         int received = 0;
 276
 277         __skb_queue_head_init(&process_queue);
 278
 279         spin_lock(&queue->lock);
 280         skb_queue_splice_tail_init(queue, &process_queue);
 281         spin_unlock(&queue->lock);
 282
 283         while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 284                 napi_gro_receive(napi, skb);
 285                 ++received;
 286         }
 287
 288         if (!skb_queue_empty(&process_queue)) {
 289                 spin_lock(&queue->lock);
 290                 skb_queue_splice(&process_queue, queue);
 291                 spin_unlock(&queue->lock);
 292         }
 293
 294         return received;
 295 }
 296
 297 static int tun_napi_poll(struct napi_struct *napi, int budget)
 298 {
 299         unsigned int received;
 300
 301         received = tun_napi_receive(napi, budget);
 302
 303         if (received < budget)
 304                 napi_complete_done(napi, received);
 305
 306         return received;
 307 }
 308
 309 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 310                           bool napi_en)
 311 {
 312         tfile->napi_enabled = napi_en;
 313         if (napi_en) {
 314                 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
 315                                NAPI_POLL_WEIGHT);
 316                 napi_enable(&tfile->napi);
 317                 mutex_init(&tfile->napi_mutex);
 318         }
 319 }
 320
 321 static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
 322 {
 323         if (tfile->napi_enabled)
 324                 napi_disable(&tfile->napi);
 325 }
 326
 327 static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
 328 {
 329         if (tfile->napi_enabled)
 330                 netif_napi_del(&tfile->napi);
 331 }
 332
 333 static bool tun_napi_frags_enabled(const struct tun_struct *tun)
 334 {
 335         return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
 336 }
 337
 338 #ifdef CONFIG_TUN_VNET_CROSS_LE
 339 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 340 {
 341         return tun->flags & TUN_VNET_BE ? false :
 342                 virtio_legacy_is_little_endian();
 343 }
 344
 345 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 346 {
 347         int be = !!(tun->flags & TUN_VNET_BE);
 348
 349         if (put_user(be, argp))
 350                 return -EFAULT;
 351
 352         return 0;
 353 }
 354
 355 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 356 {
 357         int be;
 358
 359         if (get_user(be, argp))
 360                 return -EFAULT;
 361
 362         if (be)
 363                 tun->flags |= TUN_VNET_BE;
 364         else
 365                 tun->flags &= ~TUN_VNET_BE;
 366
 367         return 0;
 368 }
 369 #else
 370 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
 371 {
 372         return virtio_legacy_is_little_endian();
 373 }
 374
 375 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
 376 {
 377         return -EINVAL;
 378 }
 379
 380 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
 381 {
 382         return -EINVAL;
 383 }
 384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
 385
 386 static inline bool tun_is_little_endian(struct tun_struct *tun)
 387 {
 388         return tun->flags & TUN_VNET_LE ||
 389                 tun_legacy_is_little_endian(tun);
 390 }
 391
 392 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
 393 {
 394         return __virtio16_to_cpu(tun_is_little_endian(tun), val);
 395 }
 396
 397 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
 398 {
 399         return __cpu_to_virtio16(tun_is_little_endian(tun), val);
 400 }
 401
 402 static inline u32 tun_hashfn(u32 rxhash)
 403 {
 404         return rxhash & 0x3ff;
 405 }
 406
 407 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 408 {
 409         struct tun_flow_entry *e;
 410
 411         hlist_for_each_entry_rcu(e, head, hash_link) {
 412                 if (e->rxhash == rxhash)
 413                         return e;
 414         }
 415         return NULL;
 416 }
 417
 418 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 419                                               struct hlist_head *head,
 420                                               u32 rxhash, u16 queue_index)
 421 {
 422         struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 423
 424         if (e) {
 425                 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
 426                           rxhash, queue_index);
 427                 e->updated = jiffies;
 428                 e->rxhash = rxhash;
 429                 e->rps_rxhash = 0;
 430                 e->queue_index = queue_index;
 431                 e->tun = tun;
 432                 hlist_add_head_rcu(&e->hash_link, head);
 433                 ++tun->flow_count;
 434         }
 435         return e;
 436 }
 437
 438 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 439 {
 440         tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
 441                   e->rxhash, e->queue_index);
 442         hlist_del_rcu(&e->hash_link);
 443         kfree_rcu(e, rcu);
 444         --tun->flow_count;
 445 }
 446
 447 static void tun_flow_flush(struct tun_struct *tun)
 448 {
 449         int i;
 450
 451         spin_lock_bh(&tun->lock);
 452         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 453                 struct tun_flow_entry *e;
 454                 struct hlist_node *n;
 455
 456                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 457                         tun_flow_delete(tun, e);
 458         }
 459         spin_unlock_bh(&tun->lock);
 460 }
 461
 462 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 463 {
 464         int i;
 465
 466         spin_lock_bh(&tun->lock);
 467         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 468                 struct tun_flow_entry *e;
 469                 struct hlist_node *n;
 470
 471                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 472                         if (e->queue_index == queue_index)
 473                                 tun_flow_delete(tun, e);
 474                 }
 475         }
 476         spin_unlock_bh(&tun->lock);
 477 }
 478
 479 static void tun_flow_cleanup(struct timer_list *t)
 480 {
 481         struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
 482         unsigned long delay = tun->ageing_time;
 483         unsigned long next_timer = jiffies + delay;
 484         unsigned long count = 0;
 485         int i;
 486
 487         tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
 488
 489         spin_lock(&tun->lock);
 490         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 491                 struct tun_flow_entry *e;
 492                 struct hlist_node *n;
 493
 494                 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 495                         unsigned long this_timer;
 496
 497                         this_timer = e->updated + delay;
 498                         if (time_before_eq(this_timer, jiffies)) {
 499                                 tun_flow_delete(tun, e);
 500                                 continue;
 501                         }
 502                         count++;
 503                         if (time_before(this_timer, next_timer))
 504                                 next_timer = this_timer;
 505                 }
 506         }
 507
 508         if (count)
 509                 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 510         spin_unlock(&tun->lock);
 511 }
 512
 513 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 514                             struct tun_file *tfile)
 515 {
 516         struct hlist_head *head;
 517         struct tun_flow_entry *e;
 518         unsigned long delay = tun->ageing_time;
 519         u16 queue_index = tfile->queue_index;
 520
 521         if (!rxhash)
 522                 return;
 523         else
 524                 head = &tun->flows[tun_hashfn(rxhash)];
 525
 526         rcu_read_lock();
 527
 528         /* We may get a very small possibility of OOO during switching, not
 529          * worth to optimize.*/
 530         if (tun->numqueues == 1 || tfile->detached)
 531                 goto unlock;
 532
 533         e = tun_flow_find(head, rxhash);
 534         if (likely(e)) {
 535                 /* TODO: keep queueing to old queue until it's empty? */
 536                 e->queue_index = queue_index;
 537                 e->updated = jiffies;
 538                 sock_rps_record_flow_hash(e->rps_rxhash);
 539         } else {
 540                 spin_lock_bh(&tun->lock);
 541                 if (!tun_flow_find(head, rxhash) &&
 542                     tun->flow_count < MAX_TAP_FLOWS)
 543                         tun_flow_create(tun, head, rxhash, queue_index);
 544
 545                 if (!timer_pending(&tun->flow_gc_timer))
 546                         mod_timer(&tun->flow_gc_timer,
 547                                   round_jiffies_up(jiffies + delay));
 548                 spin_unlock_bh(&tun->lock);
 549         }
 550
 551 unlock:
 552         rcu_read_unlock();
 553 }
 554
 555 /**
 556  * Save the hash received in the stack receive path and update the
 557  * flow_hash table accordingly.
 558  */
 559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 560 {
 561         if (unlikely(e->rps_rxhash != hash))
 562                 e->rps_rxhash = hash;
 563 }
 564
 565 /* We try to identify a flow through its rxhash first. The reason that
 566  * we do not check rxq no. is because some cards(e.g 82599), chooses
 567  * the rxq based on the txq where the last packet of the flow comes. As
 568  * the userspace application move between processors, we may get a
 569  * different rxq no. here. If we could not get rxhash, then we would
 570  * hope the rxq no. may help here.
 571  */
 572 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 573 {
 574         struct tun_flow_entry *e;
 575         u32 txq = 0;
 576         u32 numqueues = 0;
 577
 578         numqueues = READ_ONCE(tun->numqueues);
 579
 580         txq = __skb_get_hash_symmetric(skb);
 581         if (txq) {
 582                 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 583                 if (e) {
 584                         tun_flow_save_rps_rxhash(e, txq);
 585                         txq = e->queue_index;
 586                 } else
 587                         /* use multiply and shift instead of expensive divide */
 588                         txq = ((u64)txq * numqueues) >> 32;
 589         } else if (likely(skb_rx_queue_recorded(skb))) {
 590                 txq = skb_get_rx_queue(skb);
 591                 while (unlikely(txq >= numqueues))
 592                         txq -= numqueues;
 593         }
 594
 595         return txq;
 596 }
 597
 598 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 599 {
 600         struct tun_prog *prog;
 601         u16 ret = 0;
 602
 603         prog = rcu_dereference(tun->steering_prog);
 604         if (prog)
 605                 ret = bpf_prog_run_clear_cb(prog->prog, skb);
 606
 607         return ret % tun->numqueues;
 608 }
 609
 610 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 611                             void *accel_priv, select_queue_fallback_t fallback)
 612 {
 613         struct tun_struct *tun = netdev_priv(dev);
 614         u16 ret;
 615
 616         rcu_read_lock();
 617         if (rcu_dereference(tun->steering_prog))
 618                 ret = tun_ebpf_select_queue(tun, skb);
 619         else
 620                 ret = tun_automq_select_queue(tun, skb);
 621         rcu_read_unlock();
 622
 623         return ret;
 624 }
 625
 626 static inline bool tun_not_capable(struct tun_struct *tun)
 627 {
 628         const struct cred *cred = current_cred();
 629         struct net *net = dev_net(tun->dev);
 630
 631         return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 632                   (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 633                 !ns_capable(net->user_ns, CAP_NET_ADMIN);
 634 }
 635
 636 static void tun_set_real_num_queues(struct tun_struct *tun)
 637 {
 638         netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 639         netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 640 }
 641
 642 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 643 {
 644         tfile->detached = tun;
 645         list_add_tail(&tfile->next, &tun->disabled);
 646         ++tun->numdisabled;
 647 }
 648
 649 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 650 {
 651         struct tun_struct *tun = tfile->detached;
 652
 653         tfile->detached = NULL;
 654         list_del_init(&tfile->next);
 655         --tun->numdisabled;
 656         return tun;
 657 }
 658
 659 static void tun_ptr_free(void *ptr)
 660 {
 661         if (!ptr)
 662                 return;
 663         if (tun_is_xdp_buff(ptr)) {
 664                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
 665
 666                 put_page(virt_to_head_page(xdp->data));
 667         } else {
 668                 __skb_array_destroy_skb(ptr);
 669         }
 670 }
 671
 672 static void tun_queue_purge(struct tun_file *tfile)
 673 {
 674         void *ptr;
 675
 676         while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 677                 tun_ptr_free(ptr);
 678
 679         skb_queue_purge(&tfile->sk.sk_write_queue);
 680         skb_queue_purge(&tfile->sk.sk_error_queue);
 681 }
 682
 683 static void tun_cleanup_tx_ring(struct tun_file *tfile)
 684 {
 685         if (tfile->tx_ring.queue) {
 686                 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 687                 xdp_rxq_info_unreg(&tfile->xdp_rxq);
 688                 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
 689         }
 690 }
 691
 692 static void __tun_detach(struct tun_file *tfile, bool clean)
 693 {
 694         struct tun_file *ntfile;
 695         struct tun_struct *tun;
 696
 697         tun = rtnl_dereference(tfile->tun);
 698
 699         if (tun && clean) {
 700                 tun_napi_disable(tun, tfile);
 701                 tun_napi_del(tun, tfile);
 702         }
 703
 704         if (tun && !tfile->detached) {
 705                 u16 index = tfile->queue_index;
 706                 BUG_ON(index >= tun->numqueues);
 707
 708                 rcu_assign_pointer(tun->tfiles[index],
 709                                    tun->tfiles[tun->numqueues - 1]);
 710                 ntfile = rtnl_dereference(tun->tfiles[index]);
 711                 ntfile->queue_index = index;
 712
 713                 --tun->numqueues;
 714                 if (clean) {
 715                         RCU_INIT_POINTER(tfile->tun, NULL);
 716                         sock_put(&tfile->sk);
 717                 } else
 718                         tun_disable_queue(tun, tfile);
 719
 720                 synchronize_net();
 721                 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 722                 /* Drop read queue */
 723                 tun_queue_purge(tfile);
 724                 tun_set_real_num_queues(tun);
 725         } else if (tfile->detached && clean) {
 726                 tun = tun_enable_queue(tfile);
 727                 sock_put(&tfile->sk);
 728         }
 729
 730         if (clean) {
 731                 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 732                         netif_carrier_off(tun->dev);
 733
 734                         if (!(tun->flags & IFF_PERSIST) &&
 735                             tun->dev->reg_state == NETREG_REGISTERED)
 736                                 unregister_netdevice(tun->dev);
 737                 }
 738                 tun_cleanup_tx_ring(tfile);
 739                 sock_put(&tfile->sk);
 740         }
 741 }
 742
 743 static void tun_detach(struct tun_file *tfile, bool clean)
 744 {
 745         rtnl_lock();
 746         __tun_detach(tfile, clean);
 747         rtnl_unlock();
 748 }
 749
 750 static void tun_detach_all(struct net_device *dev)
 751 {
 752         struct tun_struct *tun = netdev_priv(dev);
 753         struct tun_file *tfile, *tmp;
 754         int i, n = tun->numqueues;
 755
 756         for (i = 0; i < n; i++) {
 757                 tfile = rtnl_dereference(tun->tfiles[i]);
 758                 BUG_ON(!tfile);
 759                 tun_napi_disable(tun, tfile);
 760                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 761                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 762                 RCU_INIT_POINTER(tfile->tun, NULL);
 763                 --tun->numqueues;
 764         }
 765         list_for_each_entry(tfile, &tun->disabled, next) {
 766                 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 767                 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 768                 RCU_INIT_POINTER(tfile->tun, NULL);
 769         }
 770         BUG_ON(tun->numqueues != 0);
 771
 772         synchronize_net();
 773         for (i = 0; i < n; i++) {
 774                 tfile = rtnl_dereference(tun->tfiles[i]);
 775                 tun_napi_del(tun, tfile);
 776                 /* Drop read queue */
 777                 tun_queue_purge(tfile);
 778                 sock_put(&tfile->sk);
 779                 tun_cleanup_tx_ring(tfile);
 780         }
 781         list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 782                 tun_enable_queue(tfile);
 783                 tun_queue_purge(tfile);
 784                 sock_put(&tfile->sk);
 785                 tun_cleanup_tx_ring(tfile);
 786         }
 787         BUG_ON(tun->numdisabled != 0);
 788
 789         if (tun->flags & IFF_PERSIST)
 790                 module_put(THIS_MODULE);
 791 }
 792
 793 static int tun_attach(struct tun_struct *tun, struct file *file,
 794                       bool skip_filter, bool napi)
 795 {
 796         struct tun_file *tfile = file->private_data;
 797         struct net_device *dev = tun->dev;
 798         int err;
 799
 800         err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 801         if (err < 0)
 802                 goto out;
 803
 804         err = -EINVAL;
 805         if (rtnl_dereference(tfile->tun) && !tfile->detached)
 806                 goto out;
 807
 808         err = -EBUSY;
 809         if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 810                 goto out;
 811
 812         err = -E2BIG;
 813         if (!tfile->detached &&
 814             tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 815                 goto out;
 816
 817         err = 0;
 818
 819         /* Re-attach the filter to persist device */
 820         if (!skip_filter && (tun->filter_attached == true)) {
 821                 lock_sock(tfile->socket.sk);
 822                 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 823                 release_sock(tfile->socket.sk);
 824                 if (!err)
 825                         goto out;
 826         }
 827
 828         if (!tfile->detached &&
 829             ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
 830                 err = -ENOMEM;
 831                 goto out;
 832         }
 833
 834         tfile->queue_index = tun->numqueues;
 835         tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 836
 837         if (tfile->detached) {
 838                 /* Re-attach detached tfile, updating XDP queue_index */
 839                 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 840
 841                 if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 842                         tfile->xdp_rxq.queue_index = tfile->queue_index;
 843         } else {
 844                 /* Setup XDP RX-queue info, for new tfile getting attached */
 845                 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 846                                        tun->dev, tfile->queue_index);
 847                 if (err < 0)
 848                         goto out;
 849                 err = 0;
 850         }
 851
 852         rcu_assign_pointer(tfile->tun, tun);
 853         rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 854         tun->numqueues++;
 855
 856         if (tfile->detached) {
 857                 tun_enable_queue(tfile);
 858         } else {
 859                 sock_hold(&tfile->sk);
 860                 tun_napi_init(tun, tfile, napi);
 861         }
 862
 863         tun_set_real_num_queues(tun);
 864
 865         /* device is allowed to go away first, so no need to hold extra
 866          * refcnt.
 867          */
 868
 869 out:
 870         return err;
 871 }
 872
 873 static struct tun_struct *tun_get(struct tun_file *tfile)
 874 {
 875         struct tun_struct *tun;
 876
 877         rcu_read_lock();
 878         tun = rcu_dereference(tfile->tun);
 879         if (tun)
 880                 dev_hold(tun->dev);
 881         rcu_read_unlock();
 882
 883         return tun;
 884 }
 885
 886 static void tun_put(struct tun_struct *tun)
 887 {
 888         dev_put(tun->dev);
 889 }
 890
 891 /* TAP filtering */
 892 static void addr_hash_set(u32 *mask, const u8 *addr)
 893 {
 894         int n = ether_crc(ETH_ALEN, addr) >> 26;
 895         mask[n >> 5] |= (1 << (n & 31));
 896 }
 897
 898 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 899 {
 900         int n = ether_crc(ETH_ALEN, addr) >> 26;
 901         return mask[n >> 5] & (1 << (n & 31));
 902 }
 903
 904 static int update_filter(struct tap_filter *filter, void __user *arg)
 905 {
 906         struct { u8 u[ETH_ALEN]; } *addr;
 907         struct tun_filter uf;
 908         int err, alen, n, nexact;
 909
 910         if (copy_from_user(&uf, arg, sizeof(uf)))
 911                 return -EFAULT;
 912
 913         if (!uf.count) {
 914                 /* Disabled */
 915                 filter->count = 0;
 916                 return 0;
 917         }
 918
 919         alen = ETH_ALEN * uf.count;
 920         addr = memdup_user(arg + sizeof(uf), alen);
 921         if (IS_ERR(addr))
 922                 return PTR_ERR(addr);
 923
 924         /* The filter is updated without holding any locks. Which is
 925          * perfectly safe. We disable it first and in the worst
 926          * case we'll accept a few undesired packets. */
 927         filter->count = 0;
 928         wmb();
 929
 930         /* Use first set of addresses as an exact filter */
 931         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 932                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 933
 934         nexact = n;
 935
 936         /* Remaining multicast addresses are hashed,
 937          * unicast will leave the filter disabled. */
 938         memset(filter->mask, 0, sizeof(filter->mask));
 939         for (; n < uf.count; n++) {
 940                 if (!is_multicast_ether_addr(addr[n].u)) {
 941                         err = 0; /* no filter */
 942                         goto free_addr;
 943                 }
 944                 addr_hash_set(filter->mask, addr[n].u);
 945         }
 946
 947         /* For ALLMULTI just set the mask to all ones.
 948          * This overrides the mask populated above. */
 949         if ((uf.flags & TUN_FLT_ALLMULTI))
 950                 memset(filter->mask, ~0, sizeof(filter->mask));
 951
 952         /* Now enable the filter */
 953         wmb();
 954         filter->count = nexact;
 955
 956         /* Return the number of exact filters */
 957         err = nexact;
 958 free_addr:
 959         kfree(addr);
 960         return err;
 961 }
 962
 963 /* Returns: 0 - drop, !=0 - accept */
 964 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 965 {
 966         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 967          * at this point. */
 968         struct ethhdr *eh = (struct ethhdr *) skb->data;
 969         int i;
 970
 971         /* Exact match */
 972         for (i = 0; i < filter->count; i++)
 973                 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 974                         return 1;
 975
 976         /* Inexact match (multicast only) */
 977         if (is_multicast_ether_addr(eh->h_dest))
 978                 return addr_hash_test(filter->mask, eh->h_dest);
 979
 980         return 0;
 981 }
 982
 983 /*
 984  * Checks whether the packet is accepted or not.
 985  * Returns: 0 - drop, !=0 - accept
 986  */
 987 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 988 {
 989         if (!filter->count)
 990                 return 1;
 991
 992         return run_filter(filter, skb);
 993 }
 994
 995 /* Network device part of the driver */
 996
 997 static const struct ethtool_ops tun_ethtool_ops;
 998
 999 /* Net device detach from fd. */
1000 static void tun_net_uninit(struct net_device *dev)
1001 {
1002         tun_detach_all(dev);
1003 }
1004
1005 /* Net device open. */
1006 static int tun_net_open(struct net_device *dev)
1007 {
1008         struct tun_struct *tun = netdev_priv(dev);
1009         int i;
1010
1011         netif_tx_start_all_queues(dev);
1012
1013         for (i = 0; i < tun->numqueues; i++) {
1014                 struct tun_file *tfile;
1015
1016                 tfile = rtnl_dereference(tun->tfiles[i]);
1017                 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1018         }
1019
1020         return 0;
1021 }
1022
1023 /* Net device close. */
1024 static int tun_net_close(struct net_device *dev)
1025 {
1026         netif_tx_stop_all_queues(dev);
1027         return 0;
1028 }
1029
1030 /* Net device start xmit */
1031 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1032 {
1033 #ifdef CONFIG_RPS
1034         if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1035                 /* Select queue was not called for the skbuff, so we extract the
1036                  * RPS hash and save it into the flow_table here.
1037                  */
1038                 __u32 rxhash;
1039
1040                 rxhash = __skb_get_hash_symmetric(skb);
1041                 if (rxhash) {
1042                         struct tun_flow_entry *e;
1043                         e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1044                                         rxhash);
1045                         if (e)
1046                                 tun_flow_save_rps_rxhash(e, rxhash);
1047                 }
1048         }
1049 #endif
1050 }
1051
1052 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1053                                     struct sk_buff *skb,
1054                                     int len)
1055 {
1056         struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1057
1058         if (prog)
1059                 len = bpf_prog_run_clear_cb(prog->prog, skb);
1060
1061         return len;
1062 }
1063
1064 /* Net device start xmit */
1065 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1066 {
1067         struct tun_struct *tun = netdev_priv(dev);
1068         int txq = skb->queue_mapping;
1069         struct tun_file *tfile;
1070         int len = skb->len;
1071
1072         rcu_read_lock();
1073         tfile = rcu_dereference(tun->tfiles[txq]);
1074
1075         /* Drop packet if interface is not attached */
1076         if (txq >= tun->numqueues)
1077                 goto drop;
1078
1079         if (!rcu_dereference(tun->steering_prog))
1080                 tun_automq_xmit(tun, skb);
1081
1082         tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1083
1084         BUG_ON(!tfile);
1085
1086         /* Drop if the filter does not like it.
1087          * This is a noop if the filter is disabled.
1088          * Filter can be enabled only for the TAP devices. */
1089         if (!check_filter(&tun->txflt, skb))
1090                 goto drop;
1091
1092         if (tfile->socket.sk->sk_filter &&
1093             sk_filter(tfile->socket.sk, skb))
1094                 goto drop;
1095
1096         len = run_ebpf_filter(tun, skb, len);
1097
1098         /* Trim extra bytes since we may insert vlan proto & TCI
1099          * in tun_put_user().
1100          */
1101         len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1102         if (len <= 0 || pskb_trim(skb, len))
1103                 goto drop;
1104
1105         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1106                 goto drop;
1107
1108         skb_tx_timestamp(skb);
1109
1110         /* Orphan the skb - required as we might hang on to it
1111          * for indefinite time.
1112          */
1113         skb_orphan(skb);
1114
1115         nf_reset(skb);
1116
1117         if (ptr_ring_produce(&tfile->tx_ring, skb))
1118                 goto drop;
1119
1120         /* Notify and wake up reader process */
1121         if (tfile->flags & TUN_FASYNC)
1122                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1123         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1124
1125         rcu_read_unlock();
1126         return NETDEV_TX_OK;
1127
1128 drop:
1129         this_cpu_inc(tun->pcpu_stats->tx_dropped);
1130         skb_tx_error(skb);
1131         kfree_skb(skb);
1132         rcu_read_unlock();
1133         return NET_XMIT_DROP;
1134 }
1135
1136 static void tun_net_mclist(struct net_device *dev)
1137 {
1138         /*
1139          * This callback is supposed to deal with mc filter in
1140          * _rx_ path and has nothing to do with the _tx_ path.
1141          * In rx path we always accept everything userspace gives us.
1142          */
1143 }
1144
1145 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1146         netdev_features_t features)
1147 {
1148         struct tun_struct *tun = netdev_priv(dev);
1149
1150         return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1151 }
1152 #ifdef CONFIG_NET_POLL_CONTROLLER
1153 static void tun_poll_controller(struct net_device *dev)
1154 {
1155         /*
1156          * Tun only receives frames when:
1157          * 1) the char device endpoint gets data from user space
1158          * 2) the tun socket gets a sendmsg call from user space
1159          * If NAPI is not enabled, since both of those are synchronous
1160          * operations, we are guaranteed never to have pending data when we poll
1161          * for it so there is nothing to do here but return.
1162          * We need this though so netpoll recognizes us as an interface that
1163          * supports polling, which enables bridge devices in virt setups to
1164          * still use netconsole
1165          * If NAPI is enabled, however, we need to schedule polling for all
1166          * queues unless we are using napi_gro_frags(), which we call in
1167          * process context and not in NAPI context.
1168          */
1169         struct tun_struct *tun = netdev_priv(dev);
1170
1171         if (tun->flags & IFF_NAPI) {
1172                 struct tun_file *tfile;
1173                 int i;
1174
1175                 if (tun_napi_frags_enabled(tun))
1176                         return;
1177
1178                 rcu_read_lock();
1179                 for (i = 0; i < tun->numqueues; i++) {
1180                         tfile = rcu_dereference(tun->tfiles[i]);
1181                         if (tfile->napi_enabled)
1182                                 napi_schedule(&tfile->napi);
1183                 }
1184                 rcu_read_unlock();
1185         }
1186         return;
1187 }
1188 #endif
1189
1190 static void tun_set_headroom(struct net_device *dev, int new_hr)
1191 {
1192         struct tun_struct *tun = netdev_priv(dev);
1193
1194         if (new_hr < NET_SKB_PAD)
1195                 new_hr = NET_SKB_PAD;
1196
1197         tun->align = new_hr;
1198 }
1199
1200 static void
1201 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1202 {
1203         u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1204         struct tun_struct *tun = netdev_priv(dev);
1205         struct tun_pcpu_stats *p;
1206         int i;
1207
1208         for_each_possible_cpu(i) {
1209                 u64 rxpackets, rxbytes, txpackets, txbytes;
1210                 unsigned int start;
1211
1212                 p = per_cpu_ptr(tun->pcpu_stats, i);
1213                 do {
1214                         start = u64_stats_fetch_begin(&p->syncp);
1215                         rxpackets       = p->rx_packets;
1216                         rxbytes         = p->rx_bytes;
1217                         txpackets       = p->tx_packets;
1218                         txbytes         = p->tx_bytes;
1219                 } while (u64_stats_fetch_retry(&p->syncp, start));
1220
1221                 stats->rx_packets       += rxpackets;
1222                 stats->rx_bytes         += rxbytes;
1223                 stats->tx_packets       += txpackets;
1224                 stats->tx_bytes         += txbytes;
1225
1226                 /* u32 counters */
1227                 rx_dropped      += p->rx_dropped;
1228                 rx_frame_errors += p->rx_frame_errors;
1229                 tx_dropped      += p->tx_dropped;
1230         }
1231         stats->rx_dropped  = rx_dropped;
1232         stats->rx_frame_errors = rx_frame_errors;
1233         stats->tx_dropped = tx_dropped;
1234 }
1235
1236 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1237                        struct netlink_ext_ack *extack)
1238 {
1239         struct tun_struct *tun = netdev_priv(dev);
1240         struct bpf_prog *old_prog;
1241
1242         old_prog = rtnl_dereference(tun->xdp_prog);
1243         rcu_assign_pointer(tun->xdp_prog, prog);
1244         if (old_prog)
1245                 bpf_prog_put(old_prog);
1246
1247         return 0;
1248 }
1249
1250 static u32 tun_xdp_query(struct net_device *dev)
1251 {
1252         struct tun_struct *tun = netdev_priv(dev);
1253         const struct bpf_prog *xdp_prog;
1254
1255         xdp_prog = rtnl_dereference(tun->xdp_prog);
1256         if (xdp_prog)
1257                 return xdp_prog->aux->id;
1258
1259         return 0;
1260 }
1261
1262 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1263 {
1264         switch (xdp->command) {
1265         case XDP_SETUP_PROG:
1266                 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1267         case XDP_QUERY_PROG:
1268                 xdp->prog_id = tun_xdp_query(dev);
1269                 xdp->prog_attached = !!xdp->prog_id;
1270                 return 0;
1271         default:
1272                 return -EINVAL;
1273         }
1274 }
1275
1276 static const struct net_device_ops tun_netdev_ops = {
1277         .ndo_uninit             = tun_net_uninit,
1278         .ndo_open               = tun_net_open,
1279         .ndo_stop               = tun_net_close,
1280         .ndo_start_xmit         = tun_net_xmit,
1281         .ndo_fix_features       = tun_net_fix_features,
1282         .ndo_select_queue       = tun_select_queue,
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284         .ndo_poll_controller    = tun_poll_controller,
1285 #endif
1286         .ndo_set_rx_headroom    = tun_set_headroom,
1287         .ndo_get_stats64        = tun_net_get_stats64,
1288 };
1289
1290 static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1291 {
1292         struct tun_struct *tun = netdev_priv(dev);
1293         struct xdp_buff *buff = xdp->data_hard_start;
1294         int headroom = xdp->data - xdp->data_hard_start;
1295         struct tun_file *tfile;
1296         u32 numqueues;
1297         int ret = 0;
1298
1299         /* Assure headroom is available and buff is properly aligned */
1300         if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1301                 return -ENOSPC;
1302
1303         *buff = *xdp;
1304
1305         rcu_read_lock();
1306
1307         numqueues = READ_ONCE(tun->numqueues);
1308         if (!numqueues) {
1309                 ret = -ENOSPC;
1310                 goto out;
1311         }
1312
1313         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1314                                             numqueues]);
1315         /* Encode the XDP flag into lowest bit for consumer to differ
1316          * XDP buffer from sk_buff.
1317          */
1318         if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1319                 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1320                 ret = -ENOSPC;
1321         }
1322
1323 out:
1324         rcu_read_unlock();
1325         return ret;
1326 }
1327
1328 static void tun_xdp_flush(struct net_device *dev)
1329 {
1330         struct tun_struct *tun = netdev_priv(dev);
1331         struct tun_file *tfile;
1332         u32 numqueues;
1333
1334         rcu_read_lock();
1335
1336         numqueues = READ_ONCE(tun->numqueues);
1337         if (!numqueues)
1338                 goto out;
1339
1340         tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1341                                             numqueues]);
1342         /* Notify and wake up reader process */
1343         if (tfile->flags & TUN_FASYNC)
1344                 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1345         tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1346
1347 out:
1348         rcu_read_unlock();
1349 }
1350
1351 static const struct net_device_ops tap_netdev_ops = {
1352         .ndo_uninit             = tun_net_uninit,
1353         .ndo_open               = tun_net_open,
1354         .ndo_stop               = tun_net_close,
1355         .ndo_start_xmit         = tun_net_xmit,
1356         .ndo_fix_features       = tun_net_fix_features,
1357         .ndo_set_rx_mode        = tun_net_mclist,
1358         .ndo_set_mac_address    = eth_mac_addr,
1359         .ndo_validate_addr      = eth_validate_addr,
1360         .ndo_select_queue       = tun_select_queue,
1361 #ifdef CONFIG_NET_POLL_CONTROLLER
1362         .ndo_poll_controller    = tun_poll_controller,
1363 #endif
1364         .ndo_features_check     = passthru_features_check,
1365         .ndo_set_rx_headroom    = tun_set_headroom,
1366         .ndo_get_stats64        = tun_net_get_stats64,
1367         .ndo_bpf                = tun_xdp,
1368         .ndo_xdp_xmit           = tun_xdp_xmit,
1369         .ndo_xdp_flush          = tun_xdp_flush,
1370 };
1371
1372 static void tun_flow_init(struct tun_struct *tun)
1373 {
1374         int i;
1375
1376         for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1377                 INIT_HLIST_HEAD(&tun->flows[i]);
1378
1379         tun->ageing_time = TUN_FLOW_EXPIRE;
1380         timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1381         mod_timer(&tun->flow_gc_timer,
1382                   round_jiffies_up(jiffies + tun->ageing_time));
1383 }
1384
1385 static void tun_flow_uninit(struct tun_struct *tun)
1386 {
1387         del_timer_sync(&tun->flow_gc_timer);
1388         tun_flow_flush(tun);
1389 }
1390
1391 #define MIN_MTU 68
1392 #define MAX_MTU 65535
1393
1394 /* Initialize net device. */
1395 static void tun_net_init(struct net_device *dev)
1396 {
1397         struct tun_struct *tun = netdev_priv(dev);
1398
1399         switch (tun->flags & TUN_TYPE_MASK) {
1400         case IFF_TUN:
1401                 dev->netdev_ops = &tun_netdev_ops;
1402
1403                 /* Point-to-Point TUN Device */
1404                 dev->hard_header_len = 0;
1405                 dev->addr_len = 0;
1406                 dev->mtu = 1500;
1407
1408                 /* Zero header length */
1409                 dev->type = ARPHRD_NONE;
1410                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1411                 break;
1412
1413         case IFF_TAP:
1414                 dev->netdev_ops = &tap_netdev_ops;
1415                 /* Ethernet TAP Device */
1416                 ether_setup(dev);
1417                 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1418                 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1419
1420                 eth_hw_addr_random(dev);
1421
1422                 break;
1423         }
1424
1425         dev->min_mtu = MIN_MTU;
1426         dev->max_mtu = MAX_MTU - dev->hard_header_len;
1427 }
1428
1429 /* Character device part */
1430
1431 /* Poll */
1432 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1433 {
1434         struct tun_file *tfile = file->private_data;
1435         struct tun_struct *tun = tun_get(tfile);
1436         struct sock *sk;
1437         __poll_t mask = 0;
1438
1439         if (!tun)
1440                 return EPOLLERR;
1441
1442         sk = tfile->socket.sk;
1443
1444         tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1445
1446         poll_wait(file, sk_sleep(sk), wait);
1447
1448         if (!ptr_ring_empty(&tfile->tx_ring))
1449                 mask |= EPOLLIN | EPOLLRDNORM;
1450
1451         if (tun->dev->flags & IFF_UP &&
1452             (sock_writeable(sk) ||
1453              (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1454               sock_writeable(sk))))
1455                 mask |= EPOLLOUT | EPOLLWRNORM;
1456
1457         if (tun->dev->reg_state != NETREG_REGISTERED)
1458                 mask = EPOLLERR;
1459
1460         tun_put(tun);
1461         return mask;
1462 }
1463
1464 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1465                                             size_t len,
1466                                             const struct iov_iter *it)
1467 {
1468         struct sk_buff *skb;
1469         size_t linear;
1470         int err;
1471         int i;
1472
1473         if (it->nr_segs > MAX_SKB_FRAGS + 1)
1474                 return ERR_PTR(-ENOMEM);
1475
1476         local_bh_disable();
1477         skb = napi_get_frags(&tfile->napi);
1478         local_bh_enable();
1479         if (!skb)
1480                 return ERR_PTR(-ENOMEM);
1481
1482         linear = iov_iter_single_seg_count(it);
1483         err = __skb_grow(skb, linear);
1484         if (err)
1485                 goto free;
1486
1487         skb->len = len;
1488         skb->data_len = len - linear;
1489         skb->truesize += skb->data_len;
1490
1491         for (i = 1; i < it->nr_segs; i++) {
1492                 struct page_frag *pfrag = &current->task_frag;
1493                 size_t fragsz = it->iov[i].iov_len;
1494
1495                 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1496                         err = -EINVAL;
1497                         goto free;
1498                 }
1499
1500                 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
1501                         err = -ENOMEM;
1502                         goto free;
1503                 }
1504
1505                 skb_fill_page_desc(skb, i - 1, pfrag->page,
1506                                    pfrag->offset, fragsz);
1507                 page_ref_inc(pfrag->page);
1508                 pfrag->offset += fragsz;
1509         }
1510
1511         return skb;
1512 free:
1513         /* frees skb and all frags allocated with napi_alloc_frag() */
1514         napi_free_frags(&tfile->napi);
1515         return ERR_PTR(err);
1516 }
1517
1518 /* prepad is the amount to reserve at front.  len is length after that.
1519  * linear is a hint as to how much to copy (usually headers). */
1520 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1521                                      size_t prepad, size_t len,
1522                                      size_t linear, int noblock)
1523 {
1524         struct sock *sk = tfile->socket.sk;
1525         struct sk_buff *skb;
1526         int err;
1527
1528         /* Under a page?  Don't bother with paged skb. */
1529         if (prepad + len < PAGE_SIZE || !linear)
1530                 linear = len;
1531
1532         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1533                                    &err, 0);
1534         if (!skb)
1535                 return ERR_PTR(err);
1536
1537         skb_reserve(skb, prepad);
1538         skb_put(skb, linear);
1539         skb->data_len = len - linear;
1540         skb->len += len - linear;
1541
1542         return skb;
1543 }
1544
1545 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1546                            struct sk_buff *skb, int more)
1547 {
1548         struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1549         struct sk_buff_head process_queue;
1550         u32 rx_batched = tun->rx_batched;
1551         bool rcv = false;
1552
1553         if (!rx_batched || (!more && skb_queue_empty(queue))) {
1554                 local_bh_disable();
1555                 netif_receive_skb(skb);
1556                 local_bh_enable();
1557                 return;
1558         }
1559
1560         spin_lock(&queue->lock);
1561         if (!more || skb_queue_len(queue) == rx_batched) {
1562                 __skb_queue_head_init(&process_queue);
1563                 skb_queue_splice_tail_init(queue, &process_queue);
1564                 rcv = true;
1565         } else {
1566                 __skb_queue_tail(queue, skb);
1567         }
1568         spin_unlock(&queue->lock);
1569
1570         if (rcv) {
1571                 struct sk_buff *nskb;
1572
1573                 local_bh_disable();
1574                 while ((nskb = __skb_dequeue(&process_queue)))
1575                         netif_receive_skb(nskb);
1576                 netif_receive_skb(skb);
1577                 local_bh_enable();
1578         }
1579 }
1580
1581 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1582                               int len, int noblock, bool zerocopy)
1583 {
1584         if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1585                 return false;
1586
1587         if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1588                 return false;
1589
1590         if (!noblock)
1591                 return false;
1592
1593         if (zerocopy)
1594                 return false;
1595
1596         if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1597             SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1598                 return false;
1599
1600         return true;
1601 }
1602
1603 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1604                                      struct tun_file *tfile,
1605                                      struct iov_iter *from,
1606                                      struct virtio_net_hdr *hdr,
1607                                      int len, int *skb_xdp)
1608 {
1609         struct page_frag *alloc_frag = &current->task_frag;
1610         struct sk_buff *skb;
1611         struct bpf_prog *xdp_prog;
1612         int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1613         unsigned int delta = 0;
1614         char *buf;
1615         size_t copied;
1616         bool xdp_xmit = false;
1617         int err, pad = TUN_RX_PAD;
1618
1619         rcu_read_lock();
1620         xdp_prog = rcu_dereference(tun->xdp_prog);
1621         if (xdp_prog)
1622                 pad += TUN_HEADROOM;
1623         buflen += SKB_DATA_ALIGN(len + pad);
1624         rcu_read_unlock();
1625
1626         alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1627         if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1628                 return ERR_PTR(-ENOMEM);
1629
1630         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1631         copied = copy_page_from_iter(alloc_frag->page,
1632                                      alloc_frag->offset + pad,
1633                                      len, from);
1634         if (copied != len)
1635                 return ERR_PTR(-EFAULT);
1636
1637         /* There's a small window that XDP may be set after the check
1638          * of xdp_prog above, this should be rare and for simplicity
1639          * we do XDP on skb in case the headroom is not enough.
1640          */
1641         if (hdr->gso_type || !xdp_prog)
1642                 *skb_xdp = 1;
1643         else
1644                 *skb_xdp = 0;
1645
1646         preempt_disable();
1647         rcu_read_lock();
1648         xdp_prog = rcu_dereference(tun->xdp_prog);
1649         if (xdp_prog && !*skb_xdp) {
1650                 struct xdp_buff xdp;
1651                 void *orig_data;
1652                 u32 act;
1653
1654                 xdp.data_hard_start = buf;
1655                 xdp.data = buf + pad;
1656                 xdp_set_data_meta_invalid(&xdp);
1657                 xdp.data_end = xdp.data + len;
1658                 xdp.rxq = &tfile->xdp_rxq;
1659                 orig_data = xdp.data;
1660                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1661
1662                 switch (act) {
1663                 case XDP_REDIRECT:
1664                         get_page(alloc_frag->page);
1665                         alloc_frag->offset += buflen;
1666                         err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1667                         xdp_do_flush_map();
1668                         if (err)
1669                                 goto err_redirect;
1670                         rcu_read_unlock();
1671                         preempt_enable();
1672                         return NULL;
1673                 case XDP_TX:
1674                         xdp_xmit = true;
1675                         /* fall through */
1676                 case XDP_PASS:
1677                         delta = orig_data - xdp.data;
1678                         break;
1679                 default:
1680                         bpf_warn_invalid_xdp_action(act);
1681                         /* fall through */
1682                 case XDP_ABORTED:
1683                         trace_xdp_exception(tun->dev, xdp_prog, act);
1684                         /* fall through */
1685                 case XDP_DROP:
1686                         goto err_xdp;
1687                 }
1688         }
1689
1690         skb = build_skb(buf, buflen);
1691         if (!skb) {
1692                 rcu_read_unlock();
1693                 preempt_enable();
1694                 return ERR_PTR(-ENOMEM);
1695         }
1696
1697         skb_reserve(skb, pad - delta);
1698         skb_put(skb, len + delta);
1699         get_page(alloc_frag->page);
1700         alloc_frag->offset += buflen;
1701
1702         if (xdp_xmit) {
1703                 skb->dev = tun->dev;
1704                 generic_xdp_tx(skb, xdp_prog);
1705                 rcu_read_unlock();
1706                 preempt_enable();
1707                 return NULL;
1708         }
1709
1710         rcu_read_unlock();
1711         preempt_enable();
1712
1713         return skb;
1714
1715 err_redirect:
1716         put_page(alloc_frag->page);
1717 err_xdp:
1718         rcu_read_unlock();
1719         preempt_enable();
1720         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1721         return NULL;
1722 }
1723
1724 /* Get packet from user space buffer */
1725 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1726                             void *msg_control, struct iov_iter *from,
1727                             int noblock, bool more)
1728 {
1729         struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1730         struct sk_buff *skb;
1731         size_t total_len = iov_iter_count(from);
1732         size_t len = total_len, align = tun->align, linear;
1733         struct virtio_net_hdr gso = { 0 };
1734         struct tun_pcpu_stats *stats;
1735         int good_linear;
1736         int copylen;
1737         bool zerocopy = false;
1738         int err;
1739         u32 rxhash = 0;
1740         int skb_xdp = 1;
1741         bool frags = tun_napi_frags_enabled(tun);
1742
1743         if (!(tun->dev->flags & IFF_UP))
1744                 return -EIO;
1745
1746         if (!(tun->flags & IFF_NO_PI)) {
1747                 if (len < sizeof(pi))
1748                         return -EINVAL;
1749                 len -= sizeof(pi);
1750
1751                 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1752                         return -EFAULT;
1753         }
1754
1755         if (tun->flags & IFF_VNET_HDR) {
1756                 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1757
1758                 if (len < vnet_hdr_sz)
1759                         return -EINVAL;
1760                 len -= vnet_hdr_sz;
1761
1762                 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1763                         return -EFAULT;
1764
1765                 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1766                     tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1767                         gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1768
1769                 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1770                         return -EINVAL;
1771                 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1772         }
1773
1774         if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1775                 align += NET_IP_ALIGN;
1776                 if (unlikely(len < ETH_HLEN ||
1777                              (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1778                         return -EINVAL;
1779         }
1780
1781         good_linear = SKB_MAX_HEAD(align);
1782
1783         if (msg_control) {
1784                 struct iov_iter i = *from;
1785
1786                 /* There are 256 bytes to be copied in skb, so there is
1787                  * enough room for skb expand head in case it is used.
1788                  * The rest of the buffer is mapped from userspace.
1789                  */
1790                 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1791                 if (copylen > good_linear)
1792                         copylen = good_linear;
1793                 linear = copylen;
1794                 iov_iter_advance(&i, copylen);
1795                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1796                         zerocopy = true;
1797         }
1798
1799         if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1800                 /* For the packet that is not easy to be processed
1801                  * (e.g gso or jumbo packet), we will do it at after
1802                  * skb was created with generic XDP routine.
1803                  */
1804                 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1805                 if (IS_ERR(skb)) {
1806                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1807                         return PTR_ERR(skb);
1808                 }
1809                 if (!skb)
1810                         return total_len;
1811         } else {
1812                 if (!zerocopy) {
1813                         copylen = len;
1814                         if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1815                                 linear = good_linear;
1816                         else
1817                                 linear = tun16_to_cpu(tun, gso.hdr_len);
1818                 }
1819
1820                 if (frags) {
1821                         mutex_lock(&tfile->napi_mutex);
1822                         skb = tun_napi_alloc_frags(tfile, copylen, from);
1823                         /* tun_napi_alloc_frags() enforces a layout for the skb.
1824                          * If zerocopy is enabled, then this layout will be
1825                          * overwritten by zerocopy_sg_from_iter().
1826                          */
1827                         zerocopy = false;
1828                 } else {
1829                         skb = tun_alloc_skb(tfile, align, copylen, linear,
1830                                             noblock);
1831                 }
1832
1833                 if (IS_ERR(skb)) {
1834                         if (PTR_ERR(skb) != -EAGAIN)
1835                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1836                         if (frags)
1837                                 mutex_unlock(&tfile->napi_mutex);
1838                         return PTR_ERR(skb);
1839                 }
1840
1841                 if (zerocopy)
1842                         err = zerocopy_sg_from_iter(skb, from);
1843                 else
1844                         err = skb_copy_datagram_from_iter(skb, 0, from, len);
1845
1846                 if (err) {
1847                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1848                         kfree_skb(skb);
1849                         if (frags) {
1850                                 tfile->napi.skb = NULL;
1851                                 mutex_unlock(&tfile->napi_mutex);
1852                         }
1853
1854                         return -EFAULT;
1855                 }
1856         }
1857
1858         if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1859                 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1860                 kfree_skb(skb);
1861                 if (frags) {
1862                         tfile->napi.skb = NULL;
1863                         mutex_unlock(&tfile->napi_mutex);
1864                 }
1865
1866                 return -EINVAL;
1867         }
1868
1869         switch (tun->flags & TUN_TYPE_MASK) {
1870         case IFF_TUN:
1871                 if (tun->flags & IFF_NO_PI) {
1872                         u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1873
1874                         switch (ip_version) {
1875                         case 4:
1876                                 pi.proto = htons(ETH_P_IP);
1877                                 break;
1878                         case 6:
1879                                 pi.proto = htons(ETH_P_IPV6);
1880                                 break;
1881                         default:
1882                                 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1883                                 kfree_skb(skb);
1884                                 return -EINVAL;
1885                         }
1886                 }
1887
1888                 skb_reset_mac_header(skb);
1889                 skb->protocol = pi.proto;
1890                 skb->dev = tun->dev;
1891                 break;
1892         case IFF_TAP:
1893                 if (!frags)
1894                         skb->protocol = eth_type_trans(skb, tun->dev);
1895                 break;
1896         }
1897
1898         /* copy skb_ubuf_info for callback when skb has no error */
1899         if (zerocopy) {
1900                 skb_shinfo(skb)->destructor_arg = msg_control;
1901                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1902                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1903         } else if (msg_control) {
1904                 struct ubuf_info *uarg = msg_control;
1905                 uarg->callback(uarg, false);
1906         }
1907
1908         skb_reset_network_header(skb);
1909         skb_probe_transport_header(skb, 0);
1910
1911         if (skb_xdp) {
1912                 struct bpf_prog *xdp_prog;
1913                 int ret;
1914
1915                 rcu_read_lock();
1916                 xdp_prog = rcu_dereference(tun->xdp_prog);
1917                 if (xdp_prog) {
1918                         ret = do_xdp_generic(xdp_prog, skb);
1919                         if (ret != XDP_PASS) {
1920                                 rcu_read_unlock();
1921                                 return total_len;
1922                         }
1923                 }
1924                 rcu_read_unlock();
1925         }
1926
1927         rcu_read_lock();
1928         if (!rcu_dereference(tun->steering_prog))
1929                 rxhash = __skb_get_hash_symmetric(skb);
1930         rcu_read_unlock();
1931
1932         if (frags) {
1933                 /* Exercise flow dissector code path. */
1934                 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1935
1936                 if (unlikely(headlen > skb_headlen(skb))) {
1937                         this_cpu_inc(tun->pcpu_stats->rx_dropped);
1938                         napi_free_frags(&tfile->napi);
1939                         mutex_unlock(&tfile->napi_mutex);
1940                         WARN_ON(1);
1941                         return -ENOMEM;
1942                 }
1943
1944                 local_bh_disable();
1945                 napi_gro_frags(&tfile->napi);
1946                 local_bh_enable();
1947                 mutex_unlock(&tfile->napi_mutex);
1948         } else if (tfile->napi_enabled) {
1949                 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1950                 int queue_len;
1951
1952                 spin_lock_bh(&queue->lock);
1953                 __skb_queue_tail(queue, skb);
1954                 queue_len = skb_queue_len(queue);
1955                 spin_unlock(&queue->lock);
1956
1957                 if (!more || queue_len > NAPI_POLL_WEIGHT)
1958                         napi_schedule(&tfile->napi);
1959
1960                 local_bh_enable();
1961         } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1962                 tun_rx_batched(tun, tfile, skb, more);
1963         } else {
1964                 netif_rx_ni(skb);
1965         }
1966
1967         stats = get_cpu_ptr(tun->pcpu_stats);
1968         u64_stats_update_begin(&stats->syncp);
1969         stats->rx_packets++;
1970         stats->rx_bytes += len;
1971         u64_stats_update_end(&stats->syncp);
1972         put_cpu_ptr(stats);
1973
1974         if (rxhash)
1975                 tun_flow_update(tun, rxhash, tfile);
1976
1977         return total_len;
1978 }
1979
1980 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1981 {
1982         struct file *file = iocb->ki_filp;
1983         struct tun_file *tfile = file->private_data;
1984         struct tun_struct *tun = tun_get(tfile);
1985         ssize_t result;
1986
1987         if (!tun)
1988                 return -EBADFD;
1989
1990         result = tun_get_user(tun, tfile, NULL, from,
1991                               file->f_flags & O_NONBLOCK, false);
1992
1993         tun_put(tun);
1994         return result;
1995 }
1996
1997 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1998                                 struct tun_file *tfile,
1999                                 struct xdp_buff *xdp,
2000                                 struct iov_iter *iter)
2001 {
2002         int vnet_hdr_sz = 0;
2003         size_t size = xdp->data_end - xdp->data;
2004         struct tun_pcpu_stats *stats;
2005         size_t ret;
2006
2007         if (tun->flags & IFF_VNET_HDR) {
2008                 struct virtio_net_hdr gso = { 0 };
2009
2010                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2011                 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2012                         return -EINVAL;
2013                 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2014                              sizeof(gso)))
2015                         return -EFAULT;
2016                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2017         }
2018
2019         ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2020
2021         stats = get_cpu_ptr(tun->pcpu_stats);
2022         u64_stats_update_begin(&stats->syncp);
2023         stats->tx_packets++;
2024         stats->tx_bytes += ret;
2025         u64_stats_update_end(&stats->syncp);
2026         put_cpu_ptr(tun->pcpu_stats);
2027
2028         return ret;
2029 }
2030
2031 /* Put packet to the user space buffer */
2032 static ssize_t tun_put_user(struct tun_struct *tun,
2033                             struct tun_file *tfile,
2034                             struct sk_buff *skb,
2035                             struct iov_iter *iter)
2036 {
2037         struct tun_pi pi = { 0, skb->protocol };
2038         struct tun_pcpu_stats *stats;
2039         ssize_t total;
2040         int vlan_offset = 0;
2041         int vlan_hlen = 0;
2042         int vnet_hdr_sz = 0;
2043
2044         if (skb_vlan_tag_present(skb))
2045                 vlan_hlen = VLAN_HLEN;
2046
2047         if (tun->flags & IFF_VNET_HDR)
2048                 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2049
2050         total = skb->len + vlan_hlen + vnet_hdr_sz;
2051
2052         if (!(tun->flags & IFF_NO_PI)) {
2053                 if (iov_iter_count(iter) < sizeof(pi))
2054                         return -EINVAL;
2055
2056                 total += sizeof(pi);
2057                 if (iov_iter_count(iter) < total) {
2058                         /* Packet will be striped */
2059                         pi.flags |= TUN_PKT_STRIP;
2060                 }
2061
2062                 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2063                         return -EFAULT;
2064         }
2065
2066         if (vnet_hdr_sz) {
2067                 struct virtio_net_hdr gso;
2068
2069                 if (iov_iter_count(iter) < vnet_hdr_sz)
2070                         return -EINVAL;
2071
2072                 if (virtio_net_hdr_from_skb(skb, &gso,
2073                                             tun_is_little_endian(tun), true)) {
2074                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2075                         pr_err("unexpected GSO type: "
2076                                "0x%x, gso_size %d, hdr_len %d\n",
2077                                sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2078                                tun16_to_cpu(tun, gso.hdr_len));
2079                         print_hex_dump(KERN_ERR, "tun: ",
2080                                        DUMP_PREFIX_NONE,
2081                                        16, 1, skb->head,
2082                                        min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2083                         WARN_ON_ONCE(1);
2084                         return -EINVAL;
2085                 }
2086
2087                 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2088                         return -EFAULT;
2089
2090                 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2091         }
2092
2093         if (vlan_hlen) {
2094                 int ret;
2095                 struct veth veth;
2096
2097                 veth.h_vlan_proto = skb->vlan_proto;
2098                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2099
2100                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2101
2102                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2103                 if (ret || !iov_iter_count(iter))
2104                         goto done;
2105
2106                 ret = copy_to_iter(&veth, sizeof(veth), iter);
2107                 if (ret != sizeof(veth) || !iov_iter_count(iter))
2108                         goto done;
2109         }
2110
2111         skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2112
2113 done:
2114         /* caller is in process context, */
2115         stats = get_cpu_ptr(tun->pcpu_stats);
2116         u64_stats_update_begin(&stats->syncp);
2117         stats->tx_packets++;
2118         stats->tx_bytes += skb->len + vlan_hlen;
2119         u64_stats_update_end(&stats->syncp);
2120         put_cpu_ptr(tun->pcpu_stats);
2121
2122         return total;
2123 }
2124
2125 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2126 {
2127         DECLARE_WAITQUEUE(wait, current);
2128         void *ptr = NULL;
2129         int error = 0;
2130
2131         ptr = ptr_ring_consume(&tfile->tx_ring);
2132         if (ptr)
2133                 goto out;
2134         if (noblock) {
2135                 error = -EAGAIN;
2136                 goto out;
2137         }
2138
2139         add_wait_queue(&tfile->wq.wait, &wait);
2140         current->state = TASK_INTERRUPTIBLE;
2141
2142         while (1) {
2143                 ptr = ptr_ring_consume(&tfile->tx_ring);
2144                 if (ptr)
2145                         break;
2146                 if (signal_pending(current)) {
2147                         error = -ERESTARTSYS;
2148                         break;
2149                 }
2150                 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2151                         error = -EFAULT;
2152                         break;
2153                 }
2154
2155                 schedule();
2156         }
2157
2158         current->state = TASK_RUNNING;
2159         remove_wait_queue(&tfile->wq.wait, &wait);
2160
2161 out:
2162         *err = error;
2163         return ptr;
2164 }
2165
2166 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2167                            struct iov_iter *to,
2168                            int noblock, void *ptr)
2169 {
2170         ssize_t ret;
2171         int err;
2172
2173         tun_debug(KERN_INFO, tun, "tun_do_read\n");
2174
2175         if (!iov_iter_count(to)) {
2176                 tun_ptr_free(ptr);
2177                 return 0;
2178         }
2179
2180         if (!ptr) {
2181                 /* Read frames from ring */
2182                 ptr = tun_ring_recv(tfile, noblock, &err);
2183                 if (!ptr)
2184                         return err;
2185         }
2186
2187         if (tun_is_xdp_buff(ptr)) {
2188                 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2189
2190                 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2191                 put_page(virt_to_head_page(xdp->data));
2192         } else {
2193                 struct sk_buff *skb = ptr;
2194
2195                 ret = tun_put_user(tun, tfile, skb, to);
2196                 if (unlikely(ret < 0))
2197                         kfree_skb(skb);
2198                 else
2199                         consume_skb(skb);
2200         }
2201
2202         return ret;
2203 }
2204
2205 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2206 {
2207         struct file *file = iocb->ki_filp;
2208         struct tun_file *tfile = file->private_data;
2209         struct tun_struct *tun = tun_get(tfile);
2210         ssize_t len = iov_iter_count(to), ret;
2211
2212         if (!tun)
2213                 return -EBADFD;
2214         ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2215         ret = min_t(ssize_t, ret, len);
2216         if (ret > 0)
2217                 iocb->ki_pos = ret;
2218         tun_put(tun);
2219         return ret;
2220 }
2221
2222 static void tun_prog_free(struct rcu_head *rcu)
2223 {
2224         struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2225
2226         bpf_prog_destroy(prog->prog);
2227         kfree(prog);
2228 }
2229
2230 static int __tun_set_ebpf(struct tun_struct *tun,
2231                           struct tun_prog __rcu **prog_p,
2232                           struct bpf_prog *prog)
2233 {
2234         struct tun_prog *old, *new = NULL;
2235
2236         if (prog) {
2237                 new = kmalloc(sizeof(*new), GFP_KERNEL);
2238                 if (!new)
2239                         return -ENOMEM;
2240                 new->prog = prog;
2241         }
2242
2243         spin_lock_bh(&tun->lock);
2244         old = rcu_dereference_protected(*prog_p,
2245                                         lockdep_is_held(&tun->lock));
2246         rcu_assign_pointer(*prog_p, new);
2247         spin_unlock_bh(&tun->lock);
2248
2249         if (old)
2250                 call_rcu(&old->rcu, tun_prog_free);
2251
2252         return 0;
2253 }
2254
2255 static void tun_free_netdev(struct net_device *dev)
2256 {
2257         struct tun_struct *tun = netdev_priv(dev);
2258
2259         BUG_ON(!(list_empty(&tun->disabled)));
2260         free_percpu(tun->pcpu_stats);
2261         tun_flow_uninit(tun);
2262         security_tun_dev_free_security(tun->security);
2263         __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2264         __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2265 }
2266
2267 static void tun_setup(struct net_device *dev)
2268 {
2269         struct tun_struct *tun = netdev_priv(dev);
2270
2271         tun->owner = INVALID_UID;
2272         tun->group = INVALID_GID;
2273
2274         dev->ethtool_ops = &tun_ethtool_ops;
2275         dev->needs_free_netdev = true;
2276         dev->priv_destructor = tun_free_netdev;
2277         /* We prefer our own queue length */
2278         dev->tx_queue_len = TUN_READQ_SIZE;
2279 }
2280
2281 /* Trivial set of netlink ops to allow deleting tun or tap
2282  * device with netlink.
2283  */
2284 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2285                         struct netlink_ext_ack *extack)
2286 {
2287         return -EINVAL;
2288 }
2289
2290 static size_t tun_get_size(const struct net_device *dev)
2291 {
2292         BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2293         BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2294
2295         return nla_total_size(sizeof(uid_t)) + /* OWNER */
2296                nla_total_size(sizeof(gid_t)) + /* GROUP */
2297                nla_total_size(sizeof(u8)) + /* TYPE */
2298                nla_total_size(sizeof(u8)) + /* PI */
2299                nla_total_size(sizeof(u8)) + /* VNET_HDR */
2300                nla_total_size(sizeof(u8)) + /* PERSIST */
2301                nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2302                nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2303                nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2304                0;
2305 }
2306
2307 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2308 {
2309         struct tun_struct *tun = netdev_priv(dev);
2310
2311         if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2312                 goto nla_put_failure;
2313         if (uid_valid(tun->owner) &&
2314             nla_put_u32(skb, IFLA_TUN_OWNER,
2315                         from_kuid_munged(current_user_ns(), tun->owner)))
2316                 goto nla_put_failure;
2317         if (gid_valid(tun->group) &&
2318             nla_put_u32(skb, IFLA_TUN_GROUP,
2319                         from_kgid_munged(current_user_ns(), tun->group)))
2320                 goto nla_put_failure;
2321         if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2322                 goto nla_put_failure;
2323         if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2324                 goto nla_put_failure;
2325         if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2326                 goto nla_put_failure;
2327         if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2328                        !!(tun->flags & IFF_MULTI_QUEUE)))
2329                 goto nla_put_failure;
2330         if (tun->flags & IFF_MULTI_QUEUE) {
2331                 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2332                         goto nla_put_failure;
2333                 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2334                                 tun->numdisabled))
2335                         goto nla_put_failure;
2336         }
2337
2338         return 0;
2339
2340 nla_put_failure:
2341         return -EMSGSIZE;
2342 }
2343
2344 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2345         .kind           = DRV_NAME,
2346         .priv_size      = sizeof(struct tun_struct),
2347         .setup          = tun_setup,
2348         .validate       = tun_validate,
2349         .get_size       = tun_get_size,
2350         .fill_info      = tun_fill_info,
2351 };
2352
2353 static void tun_sock_write_space(struct sock *sk)
2354 {
2355         struct tun_file *tfile;
2356         wait_queue_head_t *wqueue;
2357
2358         if (!sock_writeable(sk))
2359                 return;
2360
2361         if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2362                 return;
2363
2364         wqueue = sk_sleep(sk);
2365         if (wqueue && waitqueue_active(wqueue))
2366                 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2367                                                 EPOLLWRNORM | EPOLLWRBAND);
2368
2369         tfile = container_of(sk, struct tun_file, sk);
2370         kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2371 }
2372
2373 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2374 {
2375         int ret;
2376         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2377         struct tun_struct *tun = tun_get(tfile);
2378
2379         if (!tun)
2380                 return -EBADFD;
2381
2382         ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
2383                            m->msg_flags & MSG_DONTWAIT,
2384                            m->msg_flags & MSG_MORE);
2385         tun_put(tun);
2386         return ret;
2387 }
2388
2389 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2390                        int flags)
2391 {
2392         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2393         struct tun_struct *tun = tun_get(tfile);
2394         void *ptr = m->msg_control;
2395         int ret;
2396
2397         if (!tun) {
2398                 ret = -EBADFD;
2399                 goto out_free;
2400         }
2401
2402         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2403                 ret = -EINVAL;
2404                 goto out_put_tun;
2405         }
2406         if (flags & MSG_ERRQUEUE) {
2407                 ret = sock_recv_errqueue(sock->sk, m, total_len,
2408                                          SOL_PACKET, TUN_TX_TIMESTAMP);
2409                 goto out;
2410         }
2411         ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2412         if (ret > (ssize_t)total_len) {
2413                 m->msg_flags |= MSG_TRUNC;
2414                 ret = flags & MSG_TRUNC ? ret : total_len;
2415         }
2416 out:
2417         tun_put(tun);
2418         return ret;
2419
2420 out_put_tun:
2421         tun_put(tun);
2422 out_free:
2423         tun_ptr_free(ptr);
2424         return ret;
2425 }
2426
2427 static int tun_ptr_peek_len(void *ptr)
2428 {
2429         if (likely(ptr)) {
2430                 if (tun_is_xdp_buff(ptr)) {
2431                         struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2432
2433                         return xdp->data_end - xdp->data;
2434                 }
2435                 return __skb_array_len_with_tag(ptr);
2436         } else {
2437                 return 0;
2438         }
2439 }
2440
2441 static int tun_peek_len(struct socket *sock)
2442 {
2443         struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2444         struct tun_struct *tun;
2445         int ret = 0;
2446
2447         tun = tun_get(tfile);
2448         if (!tun)
2449                 return 0;
2450
2451         ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2452         tun_put(tun);
2453
2454         return ret;
2455 }
2456
2457 /* Ops structure to mimic raw sockets with tun */
2458 static const struct proto_ops tun_socket_ops = {
2459         .peek_len = tun_peek_len,
2460         .sendmsg = tun_sendmsg,
2461         .recvmsg = tun_recvmsg,
2462 };
2463
2464 static struct proto tun_proto = {
2465         .name           = "tun",
2466         .owner          = THIS_MODULE,
2467         .obj_size       = sizeof(struct tun_file),
2468 };
2469
2470 static int tun_flags(struct tun_struct *tun)
2471 {
2472         return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2473 }
2474
2475 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2476                               char *buf)
2477 {
2478         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2479         return sprintf(buf, "0x%x\n", tun_flags(tun));
2480 }
2481
2482 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2483                               char *buf)
2484 {
2485         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2486         return uid_valid(tun->owner)?
2487                 sprintf(buf, "%u\n",
2488                         from_kuid_munged(current_user_ns(), tun->owner)):
2489                 sprintf(buf, "-1\n");
2490 }
2491
2492 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2493                               char *buf)
2494 {
2495         struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2496         return gid_valid(tun->group) ?
2497                 sprintf(buf, "%u\n",
2498                         from_kgid_munged(current_user_ns(), tun->group)):
2499                 sprintf(buf, "-1\n");
2500 }
2501
2502 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2503 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2504 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2505
2506 static struct attribute *tun_dev_attrs[] = {
2507         &dev_attr_tun_flags.attr,
2508         &dev_attr_owner.attr,
2509         &dev_attr_group.attr,
2510         NULL
2511 };
2512
2513 static const struct attribute_group tun_attr_group = {
2514         .attrs = tun_dev_attrs
2515 };
2516
2517 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2518 {
2519         struct tun_struct *tun;
2520         struct tun_file *tfile = file->private_data;
2521         struct net_device *dev;
2522         int err;
2523
2524         if (tfile->detached)
2525                 return -EINVAL;
2526
2527         if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2528                 if (!capable(CAP_NET_ADMIN))
2529                         return -EPERM;
2530
2531                 if (!(ifr->ifr_flags & IFF_NAPI) ||
2532                     (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2533                         return -EINVAL;
2534         }
2535
2536         dev = __dev_get_by_name(net, ifr->ifr_name);
2537         if (dev) {
2538                 if (ifr->ifr_flags & IFF_TUN_EXCL)
2539                         return -EBUSY;
2540                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2541                         tun = netdev_priv(dev);
2542                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2543                         tun = netdev_priv(dev);
2544                 else
2545                         return -EINVAL;
2546
2547                 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2548                     !!(tun->flags & IFF_MULTI_QUEUE))
2549                         return -EINVAL;
2550
2551                 if (tun_not_capable(tun))
2552                         return -EPERM;
2553                 err = security_tun_dev_open(tun->security);
2554                 if (err < 0)
2555                         return err;
2556
2557                 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2558                                  ifr->ifr_flags & IFF_NAPI);
2559                 if (err < 0)
2560                         return err;
2561
2562                 if (tun->flags & IFF_MULTI_QUEUE &&
2563                     (tun->numqueues + tun->numdisabled > 1)) {
2564                         /* One or more queue has already been attached, no need
2565                          * to initialize the device again.
2566                          */
2567                         return 0;
2568                 }
2569         }
2570         else {
2571                 char *name;
2572                 unsigned long flags = 0;
2573                 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2574                              MAX_TAP_QUEUES : 1;
2575
2576                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2577                         return -EPERM;
2578                 err = security_tun_dev_create();
2579                 if (err < 0)
2580                         return err;
2581
2582                 /* Set dev type */
2583                 if (ifr->ifr_flags & IFF_TUN) {
2584                         /* TUN device */
2585                         flags |= IFF_TUN;
2586                         name = "tun%d";
2587                 } else if (ifr->ifr_flags & IFF_TAP) {
2588                         /* TAP device */
2589                         flags |= IFF_TAP;
2590                         name = "tap%d";
2591                 } else
2592                         return -EINVAL;
2593
2594                 if (*ifr->ifr_name)
2595                         name = ifr->ifr_name;
2596
2597                 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2598                                        NET_NAME_UNKNOWN, tun_setup, queues,
2599                                        queues);
2600
2601                 if (!dev)
2602                         return -ENOMEM;
2603                 err = dev_get_valid_name(net, dev, name);
2604                 if (err < 0)
2605                         goto err_free_dev;
2606
2607                 dev_net_set(dev, net);
2608                 dev->rtnl_link_ops = &tun_link_ops;
2609                 dev->ifindex = tfile->ifindex;
2610                 dev->sysfs_groups[0] = &tun_attr_group;
2611
2612                 tun = netdev_priv(dev);
2613                 tun->dev = dev;
2614                 tun->flags = flags;
2615                 tun->txflt.count = 0;
2616                 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2617
2618                 tun->align = NET_SKB_PAD;
2619                 tun->filter_attached = false;
2620                 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2621                 tun->rx_batched = 0;
2622                 RCU_INIT_POINTER(tun->steering_prog, NULL);
2623
2624                 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2625                 if (!tun->pcpu_stats) {
2626                         err = -ENOMEM;
2627                         goto err_free_dev;
2628                 }
2629
2630                 spin_lock_init(&tun->lock);
2631
2632                 err = security_tun_dev_alloc_security(&tun->security);
2633                 if (err < 0)
2634                         goto err_free_stat;
2635
2636                 tun_net_init(dev);
2637                 tun_flow_init(tun);
2638
2639                 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2640                                    TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2641                                    NETIF_F_HW_VLAN_STAG_TX;
2642                 dev->features = dev->hw_features | NETIF_F_LLTX;
2643                 dev->vlan_features = dev->features &
2644                                      ~(NETIF_F_HW_VLAN_CTAG_TX |
2645                                        NETIF_F_HW_VLAN_STAG_TX);
2646
2647                 INIT_LIST_HEAD(&tun->disabled);
2648                 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
2649                 if (err < 0)
2650                         goto err_free_flow;
2651
2652                 err = register_netdevice(tun->dev);
2653                 if (err < 0)
2654                         goto err_detach;
2655         }
2656
2657         netif_carrier_on(tun->dev);
2658
2659         tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2660
2661         tun->flags = (tun->flags & ~TUN_FEATURES) |
2662                 (ifr->ifr_flags & TUN_FEATURES);
2663
2664         /* Make sure persistent devices do not get stuck in
2665          * xoff state.
2666          */
2667         if (netif_running(tun->dev))
2668                 netif_tx_wake_all_queues(tun->dev);
2669
2670         strcpy(ifr->ifr_name, tun->dev->name);
2671         return 0;
2672
2673 err_detach:
2674         tun_detach_all(dev);
2675         /* register_netdevice() already called tun_free_netdev() */
2676         goto err_free_dev;
2677
2678 err_free_flow:
2679         tun_flow_uninit(tun);
2680         security_tun_dev_free_security(tun->security);
2681 err_free_stat:
2682         free_percpu(tun->pcpu_stats);
2683 err_free_dev:
2684         free_netdev(dev);
2685         return err;
2686 }
2687
2688 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2689                        struct ifreq *ifr)
2690 {
2691         tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2692
2693         strcpy(ifr->ifr_name, tun->dev->name);
2694
2695         ifr->ifr_flags = tun_flags(tun);
2696
2697 }
2698
2699 /* This is like a cut-down ethtool ops, except done via tun fd so no
2700  * privs required. */
2701 static int set_offload(struct tun_struct *tun, unsigned long arg)
2702 {
2703         netdev_features_t features = 0;
2704
2705         if (arg & TUN_F_CSUM) {
2706                 features |= NETIF_F_HW_CSUM;
2707                 arg &= ~TUN_F_CSUM;
2708
2709                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2710                         if (arg & TUN_F_TSO_ECN) {
2711                                 features |= NETIF_F_TSO_ECN;
2712                                 arg &= ~TUN_F_TSO_ECN;
2713                         }
2714                         if (arg & TUN_F_TSO4)
2715                                 features |= NETIF_F_TSO;
2716                         if (arg & TUN_F_TSO6)
2717                                 features |= NETIF_F_TSO6;
2718                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2719                 }
2720
2721                 arg &= ~TUN_F_UFO;
2722         }
2723
2724         /* This gives the user a way to test for new features in future by
2725          * trying to set them. */
2726         if (arg)
2727                 return -EINVAL;
2728
2729         tun->set_features = features;
2730         tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2731         tun->dev->wanted_features |= features;
2732         netdev_update_features(tun->dev);
2733
2734         return 0;
2735 }
2736
2737 static void tun_detach_filter(struct tun_struct *tun, int n)
2738 {
2739         int i;
2740         struct tun_file *tfile;
2741
2742         for (i = 0; i < n; i++) {
2743                 tfile = rtnl_dereference(tun->tfiles[i]);
2744                 lock_sock(tfile->socket.sk);
2745                 sk_detach_filter(tfile->socket.sk);
2746                 release_sock(tfile->socket.sk);
2747         }
2748
2749         tun->filter_attached = false;
2750 }
2751
2752 static int tun_attach_filter(struct tun_struct *tun)
2753 {
2754         int i, ret = 0;
2755         struct tun_file *tfile;
2756
2757         for (i = 0; i < tun->numqueues; i++) {
2758                 tfile = rtnl_dereference(tun->tfiles[i]);
2759                 lock_sock(tfile->socket.sk);
2760                 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2761                 release_sock(tfile->socket.sk);
2762                 if (ret) {
2763                         tun_detach_filter(tun, i);
2764                         return ret;
2765                 }
2766         }
2767
2768         tun->filter_attached = true;
2769         return ret;
2770 }
2771
2772 static void tun_set_sndbuf(struct tun_struct *tun)
2773 {
2774         struct tun_file *tfile;
2775         int i;
2776
2777         for (i = 0; i < tun->numqueues; i++) {
2778                 tfile = rtnl_dereference(tun->tfiles[i]);
2779                 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2780         }
2781 }
2782
2783 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2784 {
2785         struct tun_file *tfile = file->private_data;
2786         struct tun_struct *tun;
2787         int ret = 0;
2788
2789         rtnl_lock();
2790
2791         if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2792                 tun = tfile->detached;
2793                 if (!tun) {
2794                         ret = -EINVAL;
2795                         goto unlock;
2796                 }
2797                 ret = security_tun_dev_attach_queue(tun->security);
2798                 if (ret < 0)
2799                         goto unlock;
2800                 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
2801         } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2802                 tun = rtnl_dereference(tfile->tun);
2803                 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2804                         ret = -EINVAL;
2805                 else
2806                         __tun_detach(tfile, false);
2807         } else
2808                 ret = -EINVAL;
2809
2810 unlock:
2811         rtnl_unlock();
2812         return ret;
2813 }
2814
2815 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2816                         void __user *data)
2817 {
2818         struct bpf_prog *prog;
2819         int fd;
2820
2821         if (copy_from_user(&fd, data, sizeof(fd)))
2822                 return -EFAULT;
2823
2824         if (fd == -1) {
2825                 prog = NULL;
2826         } else {
2827                 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2828                 if (IS_ERR(prog))
2829                         return PTR_ERR(prog);
2830         }
2831
2832         return __tun_set_ebpf(tun, prog_p, prog);
2833 }
2834
2835 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2836                             unsigned long arg, int ifreq_len)
2837 {
2838         struct tun_file *tfile = file->private_data;
2839         struct tun_struct *tun;
2840         void __user* argp = (void __user*)arg;
2841         struct ifreq ifr;
2842         struct net *net;
2843         kuid_t owner;
2844         kgid_t group;
2845         int sndbuf;
2846         int vnet_hdr_sz;
2847         unsigned int ifindex;
2848         int le;
2849         int ret;
2850
2851         if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2852             (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
2853                 if (copy_from_user(&ifr, argp, ifreq_len))
2854                         return -EFAULT;
2855         } else {
2856                 memset(&ifr, 0, sizeof(ifr));
2857         }
2858         if (cmd == TUNGETFEATURES) {
2859                 /* Currently this just means: "what IFF flags are valid?".
2860                  * This is needed because we never checked for invalid flags on
2861                  * TUNSETIFF.
2862                  */
2863                 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2864                                 (unsigned int __user*)argp);
2865         } else if (cmd == TUNSETQUEUE)
2866                 return tun_set_queue(file, &ifr);
2867
2868         ret = 0;
2869         rtnl_lock();
2870
2871         tun = tun_get(tfile);
2872         net = sock_net(&tfile->sk);
2873         if (cmd == TUNSETIFF) {
2874                 ret = -EEXIST;
2875                 if (tun)
2876                         goto unlock;
2877
2878                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2879
2880                 ret = tun_set_iff(net, file, &ifr);
2881
2882                 if (ret)
2883                         goto unlock;
2884
2885                 if (copy_to_user(argp, &ifr, ifreq_len))
2886                         ret = -EFAULT;
2887                 goto unlock;
2888         }
2889         if (cmd == TUNSETIFINDEX) {
2890                 ret = -EPERM;
2891                 if (tun)
2892                         goto unlock;
2893
2894                 ret = -EFAULT;
2895                 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2896                         goto unlock;
2897
2898                 ret = 0;
2899                 tfile->ifindex = ifindex;
2900                 goto unlock;
2901         }
2902         if (cmd == SIOCGSKNS) {
2903                 ret = -EPERM;
2904                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2905                         goto unlock;
2906
2907                 ret = open_related_ns(&net->ns, get_net_ns);
2908                 goto unlock;
2909         }
2910
2911         ret = -EBADFD;
2912         if (!tun)
2913                 goto unlock;
2914
2915         tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2916
2917         ret = 0;
2918         switch (cmd) {
2919         case TUNGETIFF:
2920                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2921
2922                 if (tfile->detached)
2923                         ifr.ifr_flags |= IFF_DETACH_QUEUE;
2924                 if (!tfile->socket.sk->sk_filter)
2925                         ifr.ifr_flags |= IFF_NOFILTER;
2926
2927                 if (copy_to_user(argp, &ifr, ifreq_len))
2928                         ret = -EFAULT;
2929                 break;
2930
2931         case TUNSETNOCSUM:
2932                 /* Disable/Enable checksum */
2933
2934                 /* [unimplemented] */
2935                 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2936                           arg ? "disabled" : "enabled");
2937                 break;
2938
2939         case TUNSETPERSIST:
2940                 /* Disable/Enable persist mode. Keep an extra reference to the
2941                  * module to prevent the module being unprobed.
2942                  */
2943                 if (arg && !(tun->flags & IFF_PERSIST)) {
2944                         tun->flags |= IFF_PERSIST;
2945                         __module_get(THIS_MODULE);
2946                 }
2947                 if (!arg && (tun->flags & IFF_PERSIST)) {
2948                         tun->flags &= ~IFF_PERSIST;
2949                         module_put(THIS_MODULE);
2950                 }
2951
2952                 tun_debug(KERN_INFO, tun, "persist %s\n",
2953                           arg ? "enabled" : "disabled");
2954                 break;
2955
2956         case TUNSETOWNER:
2957                 /* Set owner of the device */
2958                 owner = make_kuid(current_user_ns(), arg);
2959                 if (!uid_valid(owner)) {
2960                         ret = -EINVAL;
2961                         break;
2962                 }
2963                 tun->owner = owner;
2964                 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2965                           from_kuid(&init_user_ns, tun->owner));
2966                 break;
2967
2968         case TUNSETGROUP:
2969                 /* Set group of the device */
2970                 group = make_kgid(current_user_ns(), arg);
2971                 if (!gid_valid(group)) {
2972                         ret = -EINVAL;
2973                         break;
2974                 }
2975                 tun->group = group;
2976                 tun_debug(KERN_INFO, tun, "group set to %u\n",
2977                           from_kgid(&init_user_ns, tun->group));
2978                 break;
2979
2980         case TUNSETLINK:
2981                 /* Only allow setting the type when the interface is down */
2982                 if (tun->dev->flags & IFF_UP) {
2983                         tun_debug(KERN_INFO, tun,
2984                                   "Linktype set failed because interface is up\n");
2985                         ret = -EBUSY;
2986                 } else {
2987                         tun->dev->type = (int) arg;
2988                         tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2989                                   tun->dev->type);
2990                         ret = 0;
2991                 }
2992                 break;
2993
2994 #ifdef TUN_DEBUG
2995         case TUNSETDEBUG:
2996                 tun->debug = arg;
2997                 break;
2998 #endif
2999         case TUNSETOFFLOAD:
3000                 ret = set_offload(tun, arg);
3001                 break;
3002
3003         case TUNSETTXFILTER:
3004                 /* Can be set only for TAPs */
3005                 ret = -EINVAL;
3006                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3007                         break;
3008                 ret = update_filter(&tun->txflt, (void __user *)arg);
3009                 break;
3010
3011         case SIOCGIFHWADDR:
3012                 /* Get hw address */
3013                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3014                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3015                 if (copy_to_user(argp, &ifr, ifreq_len))
3016                         ret = -EFAULT;
3017                 break;
3018
3019         case SIOCSIFHWADDR:
3020                 /* Set hw address */
3021                 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3022                           ifr.ifr_hwaddr.sa_data);
3023
3024                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
3025                 break;
3026
3027         case TUNGETSNDBUF:
3028                 sndbuf = tfile->socket.sk->sk_sndbuf;
3029                 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3030                         ret = -EFAULT;
3031                 break;
3032
3033         case TUNSETSNDBUF:
3034                 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3035                         ret = -EFAULT;
3036                         break;
3037                 }
3038                 if (sndbuf <= 0) {
3039                         ret = -EINVAL;
3040                         break;
3041                 }
3042
3043                 tun->sndbuf = sndbuf;
3044                 tun_set_sndbuf(tun);
3045                 break;
3046
3047         case TUNGETVNETHDRSZ:
3048                 vnet_hdr_sz = tun->vnet_hdr_sz;
3049                 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3050                         ret = -EFAULT;
3051                 break;
3052
3053         case TUNSETVNETHDRSZ:
3054                 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3055                         ret = -EFAULT;
3056                         break;
3057                 }
3058                 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3059                         ret = -EINVAL;
3060                         break;
3061                 }
3062
3063                 tun->vnet_hdr_sz = vnet_hdr_sz;
3064                 break;
3065
3066         case TUNGETVNETLE:
3067                 le = !!(tun->flags & TUN_VNET_LE);
3068                 if (put_user(le, (int __user *)argp))
3069                         ret = -EFAULT;
3070                 break;
3071
3072         case TUNSETVNETLE:
3073                 if (get_user(le, (int __user *)argp)) {
3074                         ret = -EFAULT;
3075                         break;
3076                 }
3077                 if (le)
3078                         tun->flags |= TUN_VNET_LE;
3079                 else
3080                         tun->flags &= ~TUN_VNET_LE;
3081                 break;
3082
3083         case TUNGETVNETBE:
3084                 ret = tun_get_vnet_be(tun, argp);
3085                 break;
3086
3087         case TUNSETVNETBE:
3088                 ret = tun_set_vnet_be(tun, argp);
3089                 break;
3090
3091         case TUNATTACHFILTER:
3092                 /* Can be set only for TAPs */
3093                 ret = -EINVAL;
3094                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3095                         break;
3096                 ret = -EFAULT;
3097                 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3098                         break;
3099
3100                 ret = tun_attach_filter(tun);
3101                 break;
3102
3103         case TUNDETACHFILTER:
3104                 /* Can be set only for TAPs */
3105                 ret = -EINVAL;
3106                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3107                         break;
3108                 ret = 0;
3109                 tun_detach_filter(tun, tun->numqueues);
3110                 break;
3111
3112         case TUNGETFILTER:
3113                 ret = -EINVAL;
3114                 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3115                         break;
3116                 ret = -EFAULT;
3117                 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3118                         break;
3119                 ret = 0;
3120                 break;
3121
3122         case TUNSETSTEERINGEBPF:
3123                 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3124                 break;
3125
3126         case TUNSETFILTEREBPF:
3127                 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3128                 break;
3129
3130         default:
3131                 ret = -EINVAL;
3132                 break;
3133         }
3134
3135 unlock:
3136         rtnl_unlock();
3137         if (tun)
3138                 tun_put(tun);
3139         return ret;
3140 }
3141
3142 static long tun_chr_ioctl(struct file *file,
3143                           unsigned int cmd, unsigned long arg)
3144 {
3145         return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3146 }
3147
3148 #ifdef CONFIG_COMPAT
3149 static long tun_chr_compat_ioctl(struct file *file,
3150                          unsigned int cmd, unsigned long arg)
3151 {
3152         switch (cmd) {
3153         case TUNSETIFF:
3154         case TUNGETIFF:
3155         case TUNSETTXFILTER:
3156         case TUNGETSNDBUF:
3157         case TUNSETSNDBUF:
3158         case SIOCGIFHWADDR:
3159         case SIOCSIFHWADDR:
3160                 arg = (unsigned long)compat_ptr(arg);
3161                 break;
3162         default:
3163                 arg = (compat_ulong_t)arg;
3164                 break;
3165         }
3166
3167         /*
3168          * compat_ifreq is shorter than ifreq, so we must not access beyond
3169          * the end of that structure. All fields that are used in this
3170          * driver are compatible though, we don't need to convert the
3171          * contents.
3172          */
3173         return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3174 }
3175 #endif /* CONFIG_COMPAT */
3176
3177 static int tun_chr_fasync(int fd, struct file *file, int on)
3178 {
3179         struct tun_file *tfile = file->private_data;
3180         int ret;
3181
3182         if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3183                 goto out;
3184
3185         if (on) {
3186                 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
3187                 tfile->flags |= TUN_FASYNC;
3188         } else
3189                 tfile->flags &= ~TUN_FASYNC;
3190         ret = 0;
3191 out:
3192         return ret;
3193 }
3194
3195 static int tun_chr_open(struct inode *inode, struct file * file)
3196 {
3197         struct net *net = current->nsproxy->net_ns;
3198         struct tun_file *tfile;
3199
3200         DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3201
3202         tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3203                                             &tun_proto, 0);
3204         if (!tfile)
3205                 return -ENOMEM;
3206         RCU_INIT_POINTER(tfile->tun, NULL);
3207         tfile->flags = 0;
3208         tfile->ifindex = 0;
3209
3210         init_waitqueue_head(&tfile->wq.wait);
3211         RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3212
3213         tfile->socket.file = file;
3214         tfile->socket.ops = &tun_socket_ops;
3215
3216         sock_init_data(&tfile->socket, &tfile->sk);
3217
3218         tfile->sk.sk_write_space = tun_sock_write_space;
3219         tfile->sk.sk_sndbuf = INT_MAX;
3220
3221         file->private_data = tfile;
3222         INIT_LIST_HEAD(&tfile->next);
3223
3224         sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3225
3226         memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
3227
3228         return 0;
3229 }
3230
3231 static int tun_chr_close(struct inode *inode, struct file *file)
3232 {
3233         struct tun_file *tfile = file->private_data;
3234
3235         tun_detach(tfile, true);
3236
3237         return 0;
3238 }
3239
3240 #ifdef CONFIG_PROC_FS
3241 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3242 {
3243         struct tun_file *tfile = file->private_data;
3244         struct tun_struct *tun;
3245         struct ifreq ifr;
3246
3247         memset(&ifr, 0, sizeof(ifr));
3248
3249         rtnl_lock();
3250         tun = tun_get(tfile);
3251         if (tun)
3252                 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3253         rtnl_unlock();
3254
3255         if (tun)
3256                 tun_put(tun);
3257
3258         seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3259 }
3260 #endif
3261
3262 static const struct file_operations tun_fops = {
3263         .owner  = THIS_MODULE,
3264         .llseek = no_llseek,
3265         .read_iter  = tun_chr_read_iter,
3266         .write_iter = tun_chr_write_iter,
3267         .poll   = tun_chr_poll,
3268         .unlocked_ioctl = tun_chr_ioctl,
3269 #ifdef CONFIG_COMPAT
3270         .compat_ioctl = tun_chr_compat_ioctl,
3271 #endif
3272         .open   = tun_chr_open,
3273         .release = tun_chr_close,
3274         .fasync = tun_chr_fasync,
3275 #ifdef CONFIG_PROC_FS
3276         .show_fdinfo = tun_chr_show_fdinfo,
3277 #endif
3278 };
3279
3280 static struct miscdevice tun_miscdev = {
3281         .minor = TUN_MINOR,
3282         .name = "tun",
3283         .nodename = "net/tun",
3284         .fops = &tun_fops,
3285 };
3286
3287 /* ethtool interface */
3288
3289 static int tun_get_link_ksettings(struct net_device *dev,
3290                                   struct ethtool_link_ksettings *cmd)
3291 {
3292         ethtool_link_ksettings_zero_link_mode(cmd, supported);
3293         ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3294         cmd->base.speed         = SPEED_10;
3295         cmd->base.duplex        = DUPLEX_FULL;
3296         cmd->base.port          = PORT_TP;
3297         cmd->base.phy_address   = 0;
3298         cmd->base.autoneg       = AUTONEG_DISABLE;
3299         return 0;
3300 }
3301
3302 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3303 {
3304         struct tun_struct *tun = netdev_priv(dev);
3305
3306         strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3307         strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3308
3309         switch (tun->flags & TUN_TYPE_MASK) {
3310         case IFF_TUN:
3311                 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3312                 break;
3313         case IFF_TAP:
3314                 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3315                 break;
3316         }
3317 }
3318
3319 static u32 tun_get_msglevel(struct net_device *dev)
3320 {
3321 #ifdef TUN_DEBUG
3322         struct tun_struct *tun = netdev_priv(dev);
3323         return tun->debug;
3324 #else
3325         return -EOPNOTSUPP;
3326 #endif
3327 }
3328
3329 static void tun_set_msglevel(struct net_device *dev, u32 value)
3330 {
3331 #ifdef TUN_DEBUG
3332         struct tun_struct *tun = netdev_priv(dev);
3333         tun->debug = value;
3334 #endif
3335 }
3336
3337 static int tun_get_coalesce(struct net_device *dev,
3338                             struct ethtool_coalesce *ec)
3339 {
3340         struct tun_struct *tun = netdev_priv(dev);
3341
3342         ec->rx_max_coalesced_frames = tun->rx_batched;
3343
3344         return 0;
3345 }
3346
3347 static int tun_set_coalesce(struct net_device *dev,
3348                             struct ethtool_coalesce *ec)
3349 {
3350         struct tun_struct *tun = netdev_priv(dev);
3351
3352         if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3353                 tun->rx_batched = NAPI_POLL_WEIGHT;
3354         else
3355                 tun->rx_batched = ec->rx_max_coalesced_frames;
3356
3357         return 0;
3358 }
3359
3360 static const struct ethtool_ops tun_ethtool_ops = {
3361         .get_drvinfo    = tun_get_drvinfo,
3362         .get_msglevel   = tun_get_msglevel,
3363         .set_msglevel   = tun_set_msglevel,
3364         .get_link       = ethtool_op_get_link,
3365         .get_ts_info    = ethtool_op_get_ts_info,
3366         .get_coalesce   = tun_get_coalesce,
3367         .set_coalesce   = tun_set_coalesce,
3368         .get_link_ksettings = tun_get_link_ksettings,
3369 };
3370
3371 static int tun_queue_resize(struct tun_struct *tun)
3372 {
3373         struct net_device *dev = tun->dev;
3374         struct tun_file *tfile;
3375         struct ptr_ring **rings;
3376         int n = tun->numqueues + tun->numdisabled;
3377         int ret, i;
3378
3379         rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3380         if (!rings)
3381                 return -ENOMEM;
3382
3383         for (i = 0; i < tun->numqueues; i++) {
3384                 tfile = rtnl_dereference(tun->tfiles[i]);
3385                 rings[i] = &tfile->tx_ring;
3386         }
3387         list_for_each_entry(tfile, &tun->disabled, next)
3388                 rings[i++] = &tfile->tx_ring;
3389
3390         ret = ptr_ring_resize_multiple(rings, n,
3391                                        dev->tx_queue_len, GFP_KERNEL,
3392                                        tun_ptr_free);
3393
3394         kfree(rings);
3395         return ret;
3396 }
3397
3398 static int tun_device_event(struct notifier_block *unused,
3399                             unsigned long event, void *ptr)
3400 {
3401         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3402         struct tun_struct *tun = netdev_priv(dev);
3403
3404         if (dev->rtnl_link_ops != &tun_link_ops)
3405                 return NOTIFY_DONE;
3406
3407         switch (event) {
3408         case NETDEV_CHANGE_TX_QUEUE_LEN:
3409                 if (tun_queue_resize(tun))
3410                         return NOTIFY_BAD;
3411                 break;
3412         default:
3413                 break;
3414         }
3415
3416         return NOTIFY_DONE;
3417 }
3418
3419 static struct notifier_block tun_notifier_block __read_mostly = {
3420         .notifier_call  = tun_device_event,
3421 };
3422
3423 static int __init tun_init(void)
3424 {
3425         int ret = 0;
3426
3427         pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3428
3429         ret = rtnl_link_register(&tun_link_ops);
3430         if (ret) {
3431                 pr_err("Can't register link_ops\n");
3432                 goto err_linkops;
3433         }
3434
3435         ret = misc_register(&tun_miscdev);
3436         if (ret) {
3437                 pr_err("Can't register misc device %d\n", TUN_MINOR);
3438                 goto err_misc;
3439         }
3440
3441         ret = register_netdevice_notifier(&tun_notifier_block);
3442         if (ret) {
3443                 pr_err("Can't register netdevice notifier\n");
3444                 goto err_notifier;
3445         }
3446
3447         return  0;
3448
3449 err_notifier:
3450         misc_deregister(&tun_miscdev);
3451 err_misc:
3452         rtnl_link_unregister(&tun_link_ops);
3453 err_linkops:
3454         return ret;
3455 }
3456
3457 static void tun_cleanup(void)
3458 {
3459         misc_deregister(&tun_miscdev);
3460         rtnl_link_unregister(&tun_link_ops);
3461         unregister_netdevice_notifier(&tun_notifier_block);
3462 }
3463
3464 /* Get an underlying socket object from tun file.  Returns error unless file is
3465  * attached to a device.  The returned object works like a packet socket, it
3466  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3467  * holding a reference to the file for as long as the socket is in use. */
3468 struct socket *tun_get_socket(struct file *file)
3469 {
3470         struct tun_file *tfile;
3471         if (file->f_op != &tun_fops)
3472                 return ERR_PTR(-EINVAL);
3473         tfile = file->private_data;
3474         if (!tfile)
3475                 return ERR_PTR(-EBADFD);
3476         return &tfile->socket;
3477 }
3478 EXPORT_SYMBOL_GPL(tun_get_socket);
3479
3480 struct ptr_ring *tun_get_tx_ring(struct file *file)
3481 {
3482         struct tun_file *tfile;
3483
3484         if (file->f_op != &tun_fops)
3485                 return ERR_PTR(-EINVAL);
3486         tfile = file->private_data;
3487         if (!tfile)
3488                 return ERR_PTR(-EBADFD);
3489         return &tfile->tx_ring;
3490 }
3491 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3492
3493 module_init(tun_init);
3494 module_exit(tun_cleanup);
3495 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3496 MODULE_AUTHOR(DRV_COPYRIGHT);
3497 MODULE_LICENSE("GPL");
3498 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3499 MODULE_ALIAS("devname:net/tun");