tun.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *  TUN - Universal TUN/TAP device driver.
   4 *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
   5 *
   6 *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
   7 */
   8
   9/*
  10 *  Changes:
  11 *
  12 *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
  13 *    Add TUNSETLINK ioctl to set the link encapsulation
  14 *
  15 *  Mark Smith <markzzzsmith@yahoo.com.au>
  16 *    Use eth_random_addr() for tap MAC address.
  17 *
  18 *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
  19 *    Fixes in packet dropping, queue length setting and queue wakeup.
  20 *    Increased default tx queue length.
  21 *    Added ethtool API.
  22 *    Minor cleanups
  23 *
  24 *  Daniel Podlejski <underley@underley.eu.org>
  25 *    Modifications for 2.3.99-pre5 kernel.
  26 */
  27
  28#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30#define DRV_NAME	"tun"
  31#define DRV_VERSION	"1.6"
  32#define DRV_DESCRIPTION	"Universal TUN/TAP device driver"
  33#define DRV_COPYRIGHT	"(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
  34
  35#include <linux/module.h>
  36#include <linux/errno.h>
  37#include <linux/kernel.h>
  38#include <linux/sched/signal.h>
  39#include <linux/major.h>
  40#include <linux/slab.h>
  41#include <linux/poll.h>
  42#include <linux/fcntl.h>
  43#include <linux/init.h>
  44#include <linux/skbuff.h>
  45#include <linux/netdevice.h>
  46#include <linux/etherdevice.h>
  47#include <linux/miscdevice.h>
  48#include <linux/ethtool.h>
  49#include <linux/rtnetlink.h>
  50#include <linux/compat.h>
  51#include <linux/if.h>
  52#include <linux/if_arp.h>
  53#include <linux/if_ether.h>
  54#include <linux/if_tun.h>
  55#include <linux/if_vlan.h>
  56#include <linux/crc32.h>
  57#include <linux/math.h>
  58#include <linux/nsproxy.h>
  59#include <linux/virtio_net.h>
  60#include <linux/rcupdate.h>
  61#include <net/net_namespace.h>
  62#include <net/netns/generic.h>
  63#include <net/rtnetlink.h>
  64#include <net/sock.h>
  65#include <net/xdp.h>
  66#include <net/ip_tunnels.h>
  67#include <linux/seq_file.h>
  68#include <linux/uio.h>
  69#include <linux/skb_array.h>
  70#include <linux/bpf.h>
  71#include <linux/bpf_trace.h>
  72#include <linux/mutex.h>
  73#include <linux/ieee802154.h>
  74#include <uapi/linux/if_ltalk.h>
  75#include <uapi/linux/if_fddi.h>
  76#include <uapi/linux/if_hippi.h>
  77#include <uapi/linux/if_fc.h>
  78#include <net/ax25.h>
  79#include <net/rose.h>
  80#include <net/6lowpan.h>
  81#include <net/rps.h>
  82
  83#include <linux/uaccess.h>
  84#include <linux/proc_fs.h>
  85
  86#include "tun_vnet.h"
  87
  88static void tun_default_link_ksettings(struct net_device *dev,
  89				       struct ethtool_link_ksettings *cmd);
  90
  91#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  92
  93/* TUN device flags */
  94
  95/* IFF_ATTACH_QUEUE is never stored in device flags,
  96 * overload it to mean fasync when stored there.
  97 */
  98#define TUN_FASYNC	IFF_ATTACH_QUEUE
  99
 100#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
 101		      IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
 102
 103#define GOODCOPY_LEN 128
 104
 105#define FLT_EXACT_COUNT 8
 106struct tap_filter {
 107	unsigned int    count;    /* Number of addrs. Zero means disabled */
 108	u32             mask[2];  /* Mask of the hashed addrs */
 109	unsigned char	addr[FLT_EXACT_COUNT][ETH_ALEN];
 110};
 111
 112/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
 113 * to max number of VCPUs in guest. */
 114#define MAX_TAP_QUEUES 256
 115#define MAX_TAP_FLOWS  4096
 116
 117#define TUN_FLOW_EXPIRE (3 * HZ)
 118
 119/* A tun_file connects an open character device to a tuntap netdevice. It
 120 * also contains all socket related structures (except sock_fprog and tap_filter)
 121 * to serve as one transmit queue for tuntap device. The sock_fprog and
 122 * tap_filter were kept in tun_struct since they were used for filtering for the
 123 * netdevice not for a specific queue (at least I didn't see the requirement for
 124 * this).
 125 *
 126 * RCU usage:
 127 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
 128 * other can only be read while rcu_read_lock or rtnl_lock is held.
 129 */
 130struct tun_file {
 131	struct sock sk;
 132	struct socket socket;
 133	struct tun_struct __rcu *tun;
 134	struct fasync_struct *fasync;
 135	/* only used for fasnyc */
 136	unsigned int flags;
 137	union {
 138		u16 queue_index;
 139		unsigned int ifindex;
 140	};
 141	struct napi_struct napi;
 142	bool napi_enabled;
 143	bool napi_frags_enabled;
 144	struct mutex napi_mutex;	/* Protects access to the above napi */
 145	struct list_head next;
 146	struct tun_struct *detached;
 147	struct ptr_ring tx_ring;
 148	struct xdp_rxq_info xdp_rxq;
 149};
 150
 151struct tun_page {
 152	struct page *page;
 153	int count;
 154};
 155
 156struct tun_flow_entry {
 157	struct hlist_node hash_link;
 158	struct rcu_head rcu;
 159	struct tun_struct *tun;
 160
 161	u32 rxhash;
 162	u32 rps_rxhash;
 163	int queue_index;
 164	unsigned long updated ____cacheline_aligned_in_smp;
 165};
 166
 167#define TUN_NUM_FLOW_ENTRIES 1024
 168#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
 169
 170struct tun_prog {
 171	struct rcu_head rcu;
 172	struct bpf_prog *prog;
 173};
 174
 175/* Since the socket were moved to tun_file, to preserve the behavior of persist
 176 * device, socket filter, sndbuf and vnet header size were restore when the
 177 * file were attached to a persist device.
 178 */
 179struct tun_struct {
 180	struct tun_file __rcu	*tfiles[MAX_TAP_QUEUES];
 181	unsigned int            numqueues;
 182	unsigned int 		flags;
 183	kuid_t			owner;
 184	kgid_t			group;
 185
 186	struct net_device	*dev;
 187	netdev_features_t	set_features;
 188#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
 189			  NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4 | \
 190			  NETIF_F_GSO_UDP_TUNNEL | NETIF_F_GSO_UDP_TUNNEL_CSUM)
 191
 192	int			align;
 193	int			vnet_hdr_sz;
 194	int			sndbuf;
 195	struct tap_filter	txflt;
 196	struct sock_fprog	fprog;
 197	/* protected by rtnl lock */
 198	bool			filter_attached;
 199	u32			msg_enable;
 200	spinlock_t lock;
 201	struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
 202	struct timer_list flow_gc_timer;
 203	unsigned long ageing_time;
 204	unsigned int numdisabled;
 205	struct list_head disabled;
 206	void *security;
 207	u32 flow_count;
 208	u32 rx_batched;
 209	atomic_long_t rx_frame_errors;
 210	struct bpf_prog __rcu *xdp_prog;
 211	struct tun_prog __rcu *steering_prog;
 212	struct tun_prog __rcu *filter_prog;
 213	struct ethtool_link_ksettings link_ksettings;
 214	/* init args */
 215	struct file *file;
 216	struct ifreq *ifr;
 217};
 218
 219struct veth {
 220	__be16 h_vlan_proto;
 221	__be16 h_vlan_TCI;
 222};
 223
 224static void tun_flow_init(struct tun_struct *tun);
 225static void tun_flow_uninit(struct tun_struct *tun);
 226
 227static int tun_napi_receive(struct napi_struct *napi, int budget)
 228{
 229	struct tun_file *tfile = container_of(napi, struct tun_file, napi);
 230	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
 231	struct sk_buff_head process_queue;
 232	struct sk_buff *skb;
 233	int received = 0;
 234
 235	__skb_queue_head_init(&process_queue);
 236
 237	spin_lock(&queue->lock);
 238	skb_queue_splice_tail_init(queue, &process_queue);
 239	spin_unlock(&queue->lock);
 240
 241	while (received < budget && (skb = __skb_dequeue(&process_queue))) {
 242		napi_gro_receive(napi, skb);
 243		++received;
 244	}
 245
 246	if (!skb_queue_empty(&process_queue)) {
 247		spin_lock(&queue->lock);
 248		skb_queue_splice(&process_queue, queue);
 249		spin_unlock(&queue->lock);
 250	}
 251
 252	return received;
 253}
 254
 255static int tun_napi_poll(struct napi_struct *napi, int budget)
 256{
 257	unsigned int received;
 258
 259	received = tun_napi_receive(napi, budget);
 260
 261	if (received < budget)
 262		napi_complete_done(napi, received);
 263
 264	return received;
 265}
 266
 267static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
 268			  bool napi_en, bool napi_frags)
 269{
 270	tfile->napi_enabled = napi_en;
 271	tfile->napi_frags_enabled = napi_en && napi_frags;
 272	if (napi_en) {
 273		netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
 274		napi_enable(&tfile->napi);
 275	}
 276}
 277
 278static void tun_napi_enable(struct tun_file *tfile)
 279{
 280	if (tfile->napi_enabled)
 281		napi_enable(&tfile->napi);
 282}
 283
 284static void tun_napi_disable(struct tun_file *tfile)
 285{
 286	if (tfile->napi_enabled)
 287		napi_disable(&tfile->napi);
 288}
 289
 290static void tun_napi_del(struct tun_file *tfile)
 291{
 292	if (tfile->napi_enabled)
 293		netif_napi_del(&tfile->napi);
 294}
 295
 296static bool tun_napi_frags_enabled(const struct tun_file *tfile)
 297{
 298	return tfile->napi_frags_enabled;
 299}
 300
 301static inline u32 tun_hashfn(u32 rxhash)
 302{
 303	return rxhash & TUN_MASK_FLOW_ENTRIES;
 304}
 305
 306static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
 307{
 308	struct tun_flow_entry *e;
 309
 310	hlist_for_each_entry_rcu(e, head, hash_link) {
 311		if (e->rxhash == rxhash)
 312			return e;
 313	}
 314	return NULL;
 315}
 316
 317static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
 318					      struct hlist_head *head,
 319					      u32 rxhash, u16 queue_index)
 320{
 321	struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
 322
 323	if (e) {
 324		netif_info(tun, tx_queued, tun->dev,
 325			   "create flow: hash %u index %u\n",
 326			   rxhash, queue_index);
 327		e->updated = jiffies;
 328		e->rxhash = rxhash;
 329		e->rps_rxhash = 0;
 330		e->queue_index = queue_index;
 331		e->tun = tun;
 332		hlist_add_head_rcu(&e->hash_link, head);
 333		++tun->flow_count;
 334	}
 335	return e;
 336}
 337
 338static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
 339{
 340	netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
 341		   e->rxhash, e->queue_index);
 342	hlist_del_rcu(&e->hash_link);
 343	kfree_rcu(e, rcu);
 344	--tun->flow_count;
 345}
 346
 347static void tun_flow_flush(struct tun_struct *tun)
 348{
 349	int i;
 350
 351	spin_lock_bh(&tun->lock);
 352	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 353		struct tun_flow_entry *e;
 354		struct hlist_node *n;
 355
 356		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
 357			tun_flow_delete(tun, e);
 358	}
 359	spin_unlock_bh(&tun->lock);
 360}
 361
 362static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
 363{
 364	int i;
 365
 366	spin_lock_bh(&tun->lock);
 367	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 368		struct tun_flow_entry *e;
 369		struct hlist_node *n;
 370
 371		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 372			if (e->queue_index == queue_index)
 373				tun_flow_delete(tun, e);
 374		}
 375	}
 376	spin_unlock_bh(&tun->lock);
 377}
 378
 379static void tun_flow_cleanup(struct timer_list *t)
 380{
 381	struct tun_struct *tun = timer_container_of(tun, t, flow_gc_timer);
 382	unsigned long delay = tun->ageing_time;
 383	unsigned long next_timer = jiffies + delay;
 384	unsigned long count = 0;
 385	int i;
 386
 387	spin_lock(&tun->lock);
 388	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
 389		struct tun_flow_entry *e;
 390		struct hlist_node *n;
 391
 392		hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
 393			unsigned long this_timer;
 394
 395			this_timer = e->updated + delay;
 396			if (time_before_eq(this_timer, jiffies)) {
 397				tun_flow_delete(tun, e);
 398				continue;
 399			}
 400			count++;
 401			if (time_before(this_timer, next_timer))
 402				next_timer = this_timer;
 403		}
 404	}
 405
 406	if (count)
 407		mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
 408	spin_unlock(&tun->lock);
 409}
 410
 411static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
 412			    struct tun_file *tfile)
 413{
 414	struct hlist_head *head;
 415	struct tun_flow_entry *e;
 416	unsigned long delay = tun->ageing_time;
 417	u16 queue_index = tfile->queue_index;
 418
 419	head = &tun->flows[tun_hashfn(rxhash)];
 420
 421	rcu_read_lock();
 422
 423	e = tun_flow_find(head, rxhash);
 424	if (likely(e)) {
 425		/* TODO: keep queueing to old queue until it's empty? */
 426		if (READ_ONCE(e->queue_index) != queue_index)
 427			WRITE_ONCE(e->queue_index, queue_index);
 428		if (e->updated != jiffies)
 429			e->updated = jiffies;
 430		sock_rps_record_flow_hash(e->rps_rxhash);
 431	} else {
 432		spin_lock_bh(&tun->lock);
 433		if (!tun_flow_find(head, rxhash) &&
 434		    tun->flow_count < MAX_TAP_FLOWS)
 435			tun_flow_create(tun, head, rxhash, queue_index);
 436
 437		if (!timer_pending(&tun->flow_gc_timer))
 438			mod_timer(&tun->flow_gc_timer,
 439				  round_jiffies_up(jiffies + delay));
 440		spin_unlock_bh(&tun->lock);
 441	}
 442
 443	rcu_read_unlock();
 444}
 445
 446/* Save the hash received in the stack receive path and update the
 447 * flow_hash table accordingly.
 448 */
 449static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
 450{
 451	if (unlikely(e->rps_rxhash != hash))
 452		e->rps_rxhash = hash;
 453}
 454
 455/* We try to identify a flow through its rxhash. The reason that
 456 * we do not check rxq no. is because some cards(e.g 82599), chooses
 457 * the rxq based on the txq where the last packet of the flow comes. As
 458 * the userspace application move between processors, we may get a
 459 * different rxq no. here.
 460 */
 461static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 462{
 463	struct tun_flow_entry *e;
 464	u32 txq, numqueues;
 465
 466	numqueues = READ_ONCE(tun->numqueues);
 467
 468	txq = __skb_get_hash_symmetric(skb);
 469	e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
 470	if (e) {
 471		tun_flow_save_rps_rxhash(e, txq);
 472		txq = e->queue_index;
 473	} else {
 474		txq = reciprocal_scale(txq, numqueues);
 475	}
 476
 477	return txq;
 478}
 479
 480static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
 481{
 482	struct tun_prog *prog;
 483	u32 numqueues;
 484	u16 ret = 0;
 485
 486	numqueues = READ_ONCE(tun->numqueues);
 487	if (!numqueues)
 488		return 0;
 489
 490	prog = rcu_dereference(tun->steering_prog);
 491	if (prog)
 492		ret = bpf_prog_run_clear_cb(prog->prog, skb);
 493
 494	return ret % numqueues;
 495}
 496
 497static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
 498			    struct net_device *sb_dev)
 499{
 500	struct tun_struct *tun = netdev_priv(dev);
 501	u16 ret;
 502
 503	rcu_read_lock();
 504	if (rcu_dereference(tun->steering_prog))
 505		ret = tun_ebpf_select_queue(tun, skb);
 506	else
 507		ret = tun_automq_select_queue(tun, skb);
 508	rcu_read_unlock();
 509
 510	return ret;
 511}
 512
 513static inline bool tun_not_capable(struct tun_struct *tun)
 514{
 515	const struct cred *cred = current_cred();
 516	struct net *net = dev_net(tun->dev);
 517
 518	return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
 519		(gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
 520		!ns_capable(net->user_ns, CAP_NET_ADMIN);
 521}
 522
 523static void tun_set_real_num_queues(struct tun_struct *tun)
 524{
 525	netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
 526	netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
 527}
 528
 529static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
 530{
 531	tfile->detached = tun;
 532	list_add_tail(&tfile->next, &tun->disabled);
 533	++tun->numdisabled;
 534}
 535
 536static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 537{
 538	struct tun_struct *tun = tfile->detached;
 539
 540	tfile->detached = NULL;
 541	list_del_init(&tfile->next);
 542	--tun->numdisabled;
 543	return tun;
 544}
 545
 546void tun_ptr_free(void *ptr)
 547{
 548	if (!ptr)
 549		return;
 550	if (tun_is_xdp_frame(ptr)) {
 551		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
 552
 553		xdp_return_frame(xdpf);
 554	} else {
 555		__skb_array_destroy_skb(ptr);
 556	}
 557}
 558EXPORT_SYMBOL_GPL(tun_ptr_free);
 559
 560static void tun_queue_purge(struct tun_file *tfile)
 561{
 562	void *ptr;
 563
 564	while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
 565		tun_ptr_free(ptr);
 566
 567	skb_queue_purge(&tfile->sk.sk_write_queue);
 568	skb_queue_purge(&tfile->sk.sk_error_queue);
 569}
 570
 571static void __tun_detach(struct tun_file *tfile, bool clean)
 572{
 573	struct tun_file *ntfile;
 574	struct tun_struct *tun;
 575
 576	tun = rtnl_dereference(tfile->tun);
 577
 578	if (tun && clean) {
 579		if (!tfile->detached)
 580			tun_napi_disable(tfile);
 581		tun_napi_del(tfile);
 582	}
 583
 584	if (tun && !tfile->detached) {
 585		u16 index = tfile->queue_index;
 586		BUG_ON(index >= tun->numqueues);
 587
 588		rcu_assign_pointer(tun->tfiles[index],
 589				   tun->tfiles[tun->numqueues - 1]);
 590		ntfile = rtnl_dereference(tun->tfiles[index]);
 591		ntfile->queue_index = index;
 592		ntfile->xdp_rxq.queue_index = index;
 593		rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
 594				   NULL);
 595
 596		--tun->numqueues;
 597		if (clean) {
 598			RCU_INIT_POINTER(tfile->tun, NULL);
 599			sock_put(&tfile->sk);
 600		} else {
 601			tun_disable_queue(tun, tfile);
 602			tun_napi_disable(tfile);
 603		}
 604
 605		synchronize_net();
 606		tun_flow_delete_by_queue(tun, tun->numqueues + 1);
 607		/* Drop read queue */
 608		tun_queue_purge(tfile);
 609		tun_set_real_num_queues(tun);
 610	} else if (tfile->detached && clean) {
 611		tun = tun_enable_queue(tfile);
 612		sock_put(&tfile->sk);
 613	}
 614
 615	if (clean) {
 616		if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
 617			netif_carrier_off(tun->dev);
 618
 619			if (!(tun->flags & IFF_PERSIST) &&
 620			    tun->dev->reg_state == NETREG_REGISTERED)
 621				unregister_netdevice(tun->dev);
 622		}
 623		if (tun)
 624			xdp_rxq_info_unreg(&tfile->xdp_rxq);
 625		ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
 626	}
 627}
 628
 629static void tun_detach(struct tun_file *tfile, bool clean)
 630{
 631	struct tun_struct *tun;
 632	struct net_device *dev;
 633
 634	rtnl_lock();
 635	tun = rtnl_dereference(tfile->tun);
 636	dev = tun ? tun->dev : NULL;
 637	__tun_detach(tfile, clean);
 638	if (dev)
 639		netdev_state_change(dev);
 640	rtnl_unlock();
 641
 642	if (clean)
 643		sock_put(&tfile->sk);
 644}
 645
 646static void tun_detach_all(struct net_device *dev)
 647{
 648	struct tun_struct *tun = netdev_priv(dev);
 649	struct tun_file *tfile, *tmp;
 650	int i, n = tun->numqueues;
 651
 652	for (i = 0; i < n; i++) {
 653		tfile = rtnl_dereference(tun->tfiles[i]);
 654		BUG_ON(!tfile);
 655		tun_napi_disable(tfile);
 656		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 657		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 658		RCU_INIT_POINTER(tfile->tun, NULL);
 659		--tun->numqueues;
 660	}
 661	list_for_each_entry(tfile, &tun->disabled, next) {
 662		tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
 663		tfile->socket.sk->sk_data_ready(tfile->socket.sk);
 664		RCU_INIT_POINTER(tfile->tun, NULL);
 665	}
 666	BUG_ON(tun->numqueues != 0);
 667
 668	synchronize_net();
 669	for (i = 0; i < n; i++) {
 670		tfile = rtnl_dereference(tun->tfiles[i]);
 671		tun_napi_del(tfile);
 672		/* Drop read queue */
 673		tun_queue_purge(tfile);
 674		xdp_rxq_info_unreg(&tfile->xdp_rxq);
 675		sock_put(&tfile->sk);
 676	}
 677	list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
 678		tun_napi_del(tfile);
 679		tun_enable_queue(tfile);
 680		tun_queue_purge(tfile);
 681		xdp_rxq_info_unreg(&tfile->xdp_rxq);
 682		sock_put(&tfile->sk);
 683	}
 684	BUG_ON(tun->numdisabled != 0);
 685
 686	if (tun->flags & IFF_PERSIST)
 687		module_put(THIS_MODULE);
 688}
 689
 690static int tun_attach(struct tun_struct *tun, struct file *file,
 691		      bool skip_filter, bool napi, bool napi_frags,
 692		      bool publish_tun)
 693{
 694	struct tun_file *tfile = file->private_data;
 695	struct net_device *dev = tun->dev;
 696	int err;
 697
 698	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
 699	if (err < 0)
 700		goto out;
 701
 702	err = -EINVAL;
 703	if (rtnl_dereference(tfile->tun) && !tfile->detached)
 704		goto out;
 705
 706	err = -EBUSY;
 707	if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
 708		goto out;
 709
 710	err = -E2BIG;
 711	if (!tfile->detached &&
 712	    tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
 713		goto out;
 714
 715	err = 0;
 716
 717	/* Re-attach the filter to persist device */
 718	if (!skip_filter && (tun->filter_attached == true)) {
 719		lock_sock(tfile->socket.sk);
 720		err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
 721		release_sock(tfile->socket.sk);
 722		if (!err)
 723			goto out;
 724	}
 725
 726	if (!tfile->detached &&
 727	    ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
 728			    GFP_KERNEL, tun_ptr_free)) {
 729		err = -ENOMEM;
 730		goto out;
 731	}
 732
 733	tfile->queue_index = tun->numqueues;
 734	tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 735
 736	if (tfile->detached) {
 737		/* Re-attach detached tfile, updating XDP queue_index */
 738		WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
 739
 740		if (tfile->xdp_rxq.queue_index    != tfile->queue_index)
 741			tfile->xdp_rxq.queue_index = tfile->queue_index;
 742	} else {
 743		/* Setup XDP RX-queue info, for new tfile getting attached */
 744		err = xdp_rxq_info_reg(&tfile->xdp_rxq,
 745				       tun->dev, tfile->queue_index, 0);
 746		if (err < 0)
 747			goto out;
 748		err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
 749						 MEM_TYPE_PAGE_SHARED, NULL);
 750		if (err < 0) {
 751			xdp_rxq_info_unreg(&tfile->xdp_rxq);
 752			goto out;
 753		}
 754		err = 0;
 755	}
 756
 757	if (tfile->detached) {
 758		tun_enable_queue(tfile);
 759		tun_napi_enable(tfile);
 760	} else {
 761		sock_hold(&tfile->sk);
 762		tun_napi_init(tun, tfile, napi, napi_frags);
 763	}
 764
 765	if (rtnl_dereference(tun->xdp_prog))
 766		sock_set_flag(&tfile->sk, SOCK_XDP);
 767
 768	/* device is allowed to go away first, so no need to hold extra
 769	 * refcnt.
 770	 */
 771
 772	/* Publish tfile->tun and tun->tfiles only after we've fully
 773	 * initialized tfile; otherwise we risk using half-initialized
 774	 * object.
 775	 */
 776	if (publish_tun)
 777		rcu_assign_pointer(tfile->tun, tun);
 778	rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
 779	tun->numqueues++;
 780	tun_set_real_num_queues(tun);
 781out:
 782	return err;
 783}
 784
 785static struct tun_struct *tun_get(struct tun_file *tfile)
 786{
 787	struct tun_struct *tun;
 788
 789	rcu_read_lock();
 790	tun = rcu_dereference(tfile->tun);
 791	if (tun)
 792		dev_hold(tun->dev);
 793	rcu_read_unlock();
 794
 795	return tun;
 796}
 797
 798static void tun_put(struct tun_struct *tun)
 799{
 800	dev_put(tun->dev);
 801}
 802
 803/* TAP filtering */
 804static void addr_hash_set(u32 *mask, const u8 *addr)
 805{
 806	int n = ether_crc(ETH_ALEN, addr) >> 26;
 807	mask[n >> 5] |= (1 << (n & 31));
 808}
 809
 810static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
 811{
 812	int n = ether_crc(ETH_ALEN, addr) >> 26;
 813	return mask[n >> 5] & (1 << (n & 31));
 814}
 815
 816static int update_filter(struct tap_filter *filter, void __user *arg)
 817{
 818	struct { u8 u[ETH_ALEN]; } *addr;
 819	struct tun_filter uf;
 820	int err, alen, n, nexact;
 821
 822	if (copy_from_user(&uf, arg, sizeof(uf)))
 823		return -EFAULT;
 824
 825	if (!uf.count) {
 826		/* Disabled */
 827		filter->count = 0;
 828		return 0;
 829	}
 830
 831	alen = ETH_ALEN * uf.count;
 832	addr = memdup_user(arg + sizeof(uf), alen);
 833	if (IS_ERR(addr))
 834		return PTR_ERR(addr);
 835
 836	/* The filter is updated without holding any locks. Which is
 837	 * perfectly safe. We disable it first and in the worst
 838	 * case we'll accept a few undesired packets. */
 839	filter->count = 0;
 840	wmb();
 841
 842	/* Use first set of addresses as an exact filter */
 843	for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
 844		memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
 845
 846	nexact = n;
 847
 848	/* Remaining multicast addresses are hashed,
 849	 * unicast will leave the filter disabled. */
 850	memset(filter->mask, 0, sizeof(filter->mask));
 851	for (; n < uf.count; n++) {
 852		if (!is_multicast_ether_addr(addr[n].u)) {
 853			err = 0; /* no filter */
 854			goto free_addr;
 855		}
 856		addr_hash_set(filter->mask, addr[n].u);
 857	}
 858
 859	/* For ALLMULTI just set the mask to all ones.
 860	 * This overrides the mask populated above. */
 861	if ((uf.flags & TUN_FLT_ALLMULTI))
 862		memset(filter->mask, ~0, sizeof(filter->mask));
 863
 864	/* Now enable the filter */
 865	wmb();
 866	filter->count = nexact;
 867
 868	/* Return the number of exact filters */
 869	err = nexact;
 870free_addr:
 871	kfree(addr);
 872	return err;
 873}
 874
 875/* Returns: 0 - drop, !=0 - accept */
 876static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
 877{
 878	/* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
 879	 * at this point. */
 880	struct ethhdr *eh = (struct ethhdr *) skb->data;
 881	int i;
 882
 883	/* Exact match */
 884	for (i = 0; i < filter->count; i++)
 885		if (ether_addr_equal(eh->h_dest, filter->addr[i]))
 886			return 1;
 887
 888	/* Inexact match (multicast only) */
 889	if (is_multicast_ether_addr(eh->h_dest))
 890		return addr_hash_test(filter->mask, eh->h_dest);
 891
 892	return 0;
 893}
 894
 895/*
 896 * Checks whether the packet is accepted or not.
 897 * Returns: 0 - drop, !=0 - accept
 898 */
 899static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
 900{
 901	if (!filter->count)
 902		return 1;
 903
 904	return run_filter(filter, skb);
 905}
 906
 907/* Network device part of the driver */
 908
 909static const struct ethtool_ops tun_ethtool_ops;
 910
 911static int tun_net_init(struct net_device *dev)
 912{
 913	struct tun_struct *tun = netdev_priv(dev);
 914	struct ifreq *ifr = tun->ifr;
 915	int err;
 916
 917	spin_lock_init(&tun->lock);
 918
 919	err = security_tun_dev_alloc_security(&tun->security);
 920	if (err < 0)
 921		return err;
 922
 923	tun_flow_init(tun);
 924
 925	dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS;
 926	dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
 927			   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 928			   NETIF_F_HW_VLAN_STAG_TX;
 929	dev->hw_enc_features = dev->hw_features;
 930	dev->features = dev->hw_features;
 931	dev->vlan_features = dev->features &
 932			     ~(NETIF_F_HW_VLAN_CTAG_TX |
 933			       NETIF_F_HW_VLAN_STAG_TX);
 934	dev->lltx = true;
 935
 936	tun->flags = (tun->flags & ~TUN_FEATURES) |
 937		      (ifr->ifr_flags & TUN_FEATURES);
 938
 939	INIT_LIST_HEAD(&tun->disabled);
 940	err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
 941			 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
 942	if (err < 0) {
 943		tun_flow_uninit(tun);
 944		security_tun_dev_free_security(tun->security);
 945		return err;
 946	}
 947	return 0;
 948}
 949
 950/* Net device detach from fd. */
 951static void tun_net_uninit(struct net_device *dev)
 952{
 953	tun_detach_all(dev);
 954}
 955
 956/* Net device open. */
 957static int tun_net_open(struct net_device *dev)
 958{
 959	netif_tx_start_all_queues(dev);
 960
 961	return 0;
 962}
 963
 964/* Net device close. */
 965static int tun_net_close(struct net_device *dev)
 966{
 967	netif_tx_stop_all_queues(dev);
 968	return 0;
 969}
 970
 971/* Net device start xmit */
 972static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
 973{
 974#ifdef CONFIG_RPS
 975	if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
 976		/* Select queue was not called for the skbuff, so we extract the
 977		 * RPS hash and save it into the flow_table here.
 978		 */
 979		struct tun_flow_entry *e;
 980		__u32 rxhash;
 981
 982		rxhash = __skb_get_hash_symmetric(skb);
 983		e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
 984		if (e)
 985			tun_flow_save_rps_rxhash(e, rxhash);
 986	}
 987#endif
 988}
 989
 990static unsigned int run_ebpf_filter(struct tun_struct *tun,
 991				    struct sk_buff *skb,
 992				    int len)
 993{
 994	struct tun_prog *prog = rcu_dereference(tun->filter_prog);
 995
 996	if (prog)
 997		len = bpf_prog_run_clear_cb(prog->prog, skb);
 998
 999	return len;
1000}
1001
1002/* Net device start xmit */
1003static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1004{
1005	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1006	struct tun_struct *tun = netdev_priv(dev);
1007	int txq = skb->queue_mapping;
1008	struct netdev_queue *queue;
1009	struct tun_file *tfile;
1010	int len = skb->len;
1011
1012	rcu_read_lock();
1013	tfile = rcu_dereference(tun->tfiles[txq]);
1014
1015	/* Drop packet if interface is not attached */
1016	if (!tfile) {
1017		drop_reason = SKB_DROP_REASON_DEV_READY;
1018		goto drop;
1019	}
1020
1021	if (!rcu_dereference(tun->steering_prog))
1022		tun_automq_xmit(tun, skb);
1023
1024	netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1025
1026	/* Drop if the filter does not like it.
1027	 * This is a noop if the filter is disabled.
1028	 * Filter can be enabled only for the TAP devices. */
1029	if (!check_filter(&tun->txflt, skb)) {
1030		drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1031		goto drop;
1032	}
1033
1034	if (tfile->socket.sk->sk_filter &&
1035	    sk_filter_reason(tfile->socket.sk, skb, &drop_reason))
1036		goto drop;
1037
1038	len = run_ebpf_filter(tun, skb, len);
1039	if (len == 0) {
1040		drop_reason = SKB_DROP_REASON_TAP_FILTER;
1041		goto drop;
1042	}
1043
1044	if (pskb_trim(skb, len)) {
1045		drop_reason = SKB_DROP_REASON_NOMEM;
1046		goto drop;
1047	}
1048
1049	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1050		drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1051		goto drop;
1052	}
1053
1054	skb_tx_timestamp(skb);
1055
1056	/* Orphan the skb - required as we might hang on to it
1057	 * for indefinite time.
1058	 */
1059	skb_orphan(skb);
1060
1061	nf_reset_ct(skb);
1062
1063	if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1064		drop_reason = SKB_DROP_REASON_FULL_RING;
1065		goto drop;
1066	}
1067
1068	/* dev->lltx requires to do our own update of trans_start */
1069	queue = netdev_get_tx_queue(dev, txq);
1070	txq_trans_cond_update(queue);
1071
1072	/* Notify and wake up reader process */
1073	if (tfile->flags & TUN_FASYNC)
1074		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1075	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1076
1077	rcu_read_unlock();
1078	return NETDEV_TX_OK;
1079
1080drop:
1081	dev_core_stats_tx_dropped_inc(dev);
1082	skb_tx_error(skb);
1083	kfree_skb_reason(skb, drop_reason);
1084	rcu_read_unlock();
1085	return NET_XMIT_DROP;
1086}
1087
1088static void tun_net_mclist(struct net_device *dev)
1089{
1090	/*
1091	 * This callback is supposed to deal with mc filter in
1092	 * _rx_ path and has nothing to do with the _tx_ path.
1093	 * In rx path we always accept everything userspace gives us.
1094	 */
1095}
1096
1097static netdev_features_t tun_net_fix_features(struct net_device *dev,
1098	netdev_features_t features)
1099{
1100	struct tun_struct *tun = netdev_priv(dev);
1101
1102	return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1103}
1104
1105static void tun_set_headroom(struct net_device *dev, int new_hr)
1106{
1107	struct tun_struct *tun = netdev_priv(dev);
1108
1109	if (new_hr < NET_SKB_PAD)
1110		new_hr = NET_SKB_PAD;
1111
1112	tun->align = new_hr;
1113}
1114
1115static void
1116tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1117{
1118	struct tun_struct *tun = netdev_priv(dev);
1119
1120	dev_get_tstats64(dev, stats);
1121
1122	stats->rx_frame_errors +=
1123		(unsigned long)atomic_long_read(&tun->rx_frame_errors);
1124}
1125
1126static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1127		       struct netlink_ext_ack *extack)
1128{
1129	struct tun_struct *tun = netdev_priv(dev);
1130	struct tun_file *tfile;
1131	struct bpf_prog *old_prog;
1132	int i;
1133
1134	old_prog = rtnl_dereference(tun->xdp_prog);
1135	rcu_assign_pointer(tun->xdp_prog, prog);
1136	if (old_prog)
1137		bpf_prog_put(old_prog);
1138
1139	for (i = 0; i < tun->numqueues; i++) {
1140		tfile = rtnl_dereference(tun->tfiles[i]);
1141		if (prog)
1142			sock_set_flag(&tfile->sk, SOCK_XDP);
1143		else
1144			sock_reset_flag(&tfile->sk, SOCK_XDP);
1145	}
1146	list_for_each_entry(tfile, &tun->disabled, next) {
1147		if (prog)
1148			sock_set_flag(&tfile->sk, SOCK_XDP);
1149		else
1150			sock_reset_flag(&tfile->sk, SOCK_XDP);
1151	}
1152
1153	return 0;
1154}
1155
1156static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1157{
1158	switch (xdp->command) {
1159	case XDP_SETUP_PROG:
1160		return tun_xdp_set(dev, xdp->prog, xdp->extack);
1161	default:
1162		return -EINVAL;
1163	}
1164}
1165
1166static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1167{
1168	if (new_carrier) {
1169		struct tun_struct *tun = netdev_priv(dev);
1170
1171		if (!tun->numqueues)
1172			return -EPERM;
1173
1174		netif_carrier_on(dev);
1175	} else {
1176		netif_carrier_off(dev);
1177	}
1178	return 0;
1179}
1180
1181static const struct net_device_ops tun_netdev_ops = {
1182	.ndo_init		= tun_net_init,
1183	.ndo_uninit		= tun_net_uninit,
1184	.ndo_open		= tun_net_open,
1185	.ndo_stop		= tun_net_close,
1186	.ndo_start_xmit		= tun_net_xmit,
1187	.ndo_fix_features	= tun_net_fix_features,
1188	.ndo_select_queue	= tun_select_queue,
1189	.ndo_set_rx_headroom	= tun_set_headroom,
1190	.ndo_get_stats64	= tun_net_get_stats64,
1191	.ndo_change_carrier	= tun_net_change_carrier,
1192};
1193
1194static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1195{
1196	/* Notify and wake up reader process */
1197	if (tfile->flags & TUN_FASYNC)
1198		kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1199	tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1200}
1201
1202static int tun_xdp_xmit(struct net_device *dev, int n,
1203			struct xdp_frame **frames, u32 flags)
1204{
1205	struct tun_struct *tun = netdev_priv(dev);
1206	struct tun_file *tfile;
1207	u32 numqueues;
1208	int nxmit = 0;
1209	int i;
1210
1211	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1212		return -EINVAL;
1213
1214	rcu_read_lock();
1215
1216resample:
1217	numqueues = READ_ONCE(tun->numqueues);
1218	if (!numqueues) {
1219		rcu_read_unlock();
1220		return -ENXIO; /* Caller will free/return all frames */
1221	}
1222
1223	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1224					    numqueues]);
1225	if (unlikely(!tfile))
1226		goto resample;
1227
1228	spin_lock(&tfile->tx_ring.producer_lock);
1229	for (i = 0; i < n; i++) {
1230		struct xdp_frame *xdp = frames[i];
1231		/* Encode the XDP flag into lowest bit for consumer to differ
1232		 * XDP buffer from sk_buff.
1233		 */
1234		void *frame = tun_xdp_to_ptr(xdp);
1235
1236		if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1237			dev_core_stats_tx_dropped_inc(dev);
1238			break;
1239		}
1240		nxmit++;
1241	}
1242	spin_unlock(&tfile->tx_ring.producer_lock);
1243
1244	if (flags & XDP_XMIT_FLUSH)
1245		__tun_xdp_flush_tfile(tfile);
1246
1247	rcu_read_unlock();
1248	return nxmit;
1249}
1250
1251static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1252{
1253	struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1254	int nxmit;
1255
1256	if (unlikely(!frame))
1257		return -EOVERFLOW;
1258
1259	nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1260	if (!nxmit)
1261		xdp_return_frame_rx_napi(frame);
1262	return nxmit;
1263}
1264
1265static const struct net_device_ops tap_netdev_ops = {
1266	.ndo_init		= tun_net_init,
1267	.ndo_uninit		= tun_net_uninit,
1268	.ndo_open		= tun_net_open,
1269	.ndo_stop		= tun_net_close,
1270	.ndo_start_xmit		= tun_net_xmit,
1271	.ndo_fix_features	= tun_net_fix_features,
1272	.ndo_set_rx_mode	= tun_net_mclist,
1273	.ndo_set_mac_address	= eth_mac_addr,
1274	.ndo_validate_addr	= eth_validate_addr,
1275	.ndo_select_queue	= tun_select_queue,
1276	.ndo_features_check	= passthru_features_check,
1277	.ndo_set_rx_headroom	= tun_set_headroom,
1278	.ndo_bpf		= tun_xdp,
1279	.ndo_xdp_xmit		= tun_xdp_xmit,
1280	.ndo_change_carrier	= tun_net_change_carrier,
1281};
1282
1283static void tun_flow_init(struct tun_struct *tun)
1284{
1285	int i;
1286
1287	for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1288		INIT_HLIST_HEAD(&tun->flows[i]);
1289
1290	tun->ageing_time = TUN_FLOW_EXPIRE;
1291	timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1292	mod_timer(&tun->flow_gc_timer,
1293		  round_jiffies_up(jiffies + tun->ageing_time));
1294}
1295
1296static void tun_flow_uninit(struct tun_struct *tun)
1297{
1298	timer_delete_sync(&tun->flow_gc_timer);
1299	tun_flow_flush(tun);
1300}
1301
1302#define MIN_MTU 68
1303#define MAX_MTU 65535
1304
1305/* Initialize net device. */
1306static void tun_net_initialize(struct net_device *dev)
1307{
1308	struct tun_struct *tun = netdev_priv(dev);
1309
1310	switch (tun->flags & TUN_TYPE_MASK) {
1311	case IFF_TUN:
1312		dev->netdev_ops = &tun_netdev_ops;
1313		dev->header_ops = &ip_tunnel_header_ops;
1314
1315		/* Point-to-Point TUN Device */
1316		dev->hard_header_len = 0;
1317		dev->addr_len = 0;
1318		dev->mtu = 1500;
1319
1320		/* Zero header length */
1321		dev->type = ARPHRD_NONE;
1322		dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1323		break;
1324
1325	case IFF_TAP:
1326		dev->netdev_ops = &tap_netdev_ops;
1327		/* Ethernet TAP Device */
1328		ether_setup(dev);
1329		dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1330		dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1331
1332		eth_hw_addr_random(dev);
1333
1334		/* Currently tun does not support XDP, only tap does. */
1335		dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1336				    NETDEV_XDP_ACT_REDIRECT |
1337				    NETDEV_XDP_ACT_NDO_XMIT;
1338
1339		break;
1340	}
1341
1342	dev->min_mtu = MIN_MTU;
1343	dev->max_mtu = MAX_MTU - dev->hard_header_len;
1344}
1345
1346static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1347{
1348	struct sock *sk = tfile->socket.sk;
1349
1350	return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1351}
1352
1353/* Character device part */
1354
1355/* Poll */
1356static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1357{
1358	struct tun_file *tfile = file->private_data;
1359	struct tun_struct *tun = tun_get(tfile);
1360	struct sock *sk;
1361	__poll_t mask = 0;
1362
1363	if (!tun)
1364		return EPOLLERR;
1365
1366	sk = tfile->socket.sk;
1367
1368	poll_wait(file, sk_sleep(sk), wait);
1369
1370	if (!ptr_ring_empty(&tfile->tx_ring))
1371		mask |= EPOLLIN | EPOLLRDNORM;
1372
1373	/* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1374	 * guarantee EPOLLOUT to be raised by either here or
1375	 * tun_sock_write_space(). Then process could get notification
1376	 * after it writes to a down device and meets -EIO.
1377	 */
1378	if (tun_sock_writeable(tun, tfile) ||
1379	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1380	     tun_sock_writeable(tun, tfile)))
1381		mask |= EPOLLOUT | EPOLLWRNORM;
1382
1383	if (tun->dev->reg_state != NETREG_REGISTERED)
1384		mask = EPOLLERR;
1385
1386	tun_put(tun);
1387	return mask;
1388}
1389
1390static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1391					    size_t len,
1392					    const struct iov_iter *it)
1393{
1394	struct sk_buff *skb;
1395	size_t linear;
1396	int err;
1397	int i;
1398
1399	if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1400	    len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1401		return ERR_PTR(-EMSGSIZE);
1402
1403	local_bh_disable();
1404	skb = napi_get_frags(&tfile->napi);
1405	local_bh_enable();
1406	if (!skb)
1407		return ERR_PTR(-ENOMEM);
1408
1409	linear = iov_iter_single_seg_count(it);
1410	err = __skb_grow(skb, linear);
1411	if (err)
1412		goto free;
1413
1414	skb->len = len;
1415	skb->data_len = len - linear;
1416	skb->truesize += skb->data_len;
1417
1418	for (i = 1; i < it->nr_segs; i++) {
1419		const struct iovec *iov = iter_iov(it) + i;
1420		size_t fragsz = iov->iov_len;
1421		struct page *page;
1422		void *frag;
1423
1424		if (fragsz == 0 || fragsz > PAGE_SIZE) {
1425			err = -EINVAL;
1426			goto free;
1427		}
1428		frag = netdev_alloc_frag(fragsz);
1429		if (!frag) {
1430			err = -ENOMEM;
1431			goto free;
1432		}
1433		page = virt_to_head_page(frag);
1434		skb_fill_page_desc(skb, i - 1, page,
1435				   frag - page_address(page), fragsz);
1436	}
1437
1438	return skb;
1439free:
1440	/* frees skb and all frags allocated with napi_alloc_frag() */
1441	napi_free_frags(&tfile->napi);
1442	return ERR_PTR(err);
1443}
1444
1445/* prepad is the amount to reserve at front.  len is length after that.
1446 * linear is a hint as to how much to copy (usually headers). */
1447static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1448				     size_t prepad, size_t len,
1449				     size_t linear, int noblock)
1450{
1451	struct sock *sk = tfile->socket.sk;
1452	struct sk_buff *skb;
1453	int err;
1454
1455	/* Under a page?  Don't bother with paged skb. */
1456	if (prepad + len < PAGE_SIZE)
1457		linear = len;
1458
1459	if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
1460		linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
1461	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1462				   &err, PAGE_ALLOC_COSTLY_ORDER);
1463	if (!skb)
1464		return ERR_PTR(err);
1465
1466	skb_reserve(skb, prepad);
1467	skb_put(skb, linear);
1468	skb->data_len = len - linear;
1469	skb->len += len - linear;
1470
1471	return skb;
1472}
1473
1474static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1475			   struct sk_buff *skb, int more)
1476{
1477	struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1478	struct sk_buff_head process_queue;
1479	u32 rx_batched = tun->rx_batched;
1480	bool rcv = false;
1481
1482	if (!rx_batched || (!more && skb_queue_empty(queue))) {
1483		local_bh_disable();
1484		skb_record_rx_queue(skb, tfile->queue_index);
1485		netif_receive_skb(skb);
1486		local_bh_enable();
1487		return;
1488	}
1489
1490	spin_lock(&queue->lock);
1491	if (!more || skb_queue_len(queue) == rx_batched) {
1492		__skb_queue_head_init(&process_queue);
1493		skb_queue_splice_tail_init(queue, &process_queue);
1494		rcv = true;
1495	} else {
1496		__skb_queue_tail(queue, skb);
1497	}
1498	spin_unlock(&queue->lock);
1499
1500	if (rcv) {
1501		struct sk_buff *nskb;
1502
1503		local_bh_disable();
1504		while ((nskb = __skb_dequeue(&process_queue))) {
1505			skb_record_rx_queue(nskb, tfile->queue_index);
1506			netif_receive_skb(nskb);
1507		}
1508		skb_record_rx_queue(skb, tfile->queue_index);
1509		netif_receive_skb(skb);
1510		local_bh_enable();
1511	}
1512}
1513
1514static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1515			      int len, int noblock, bool zerocopy)
1516{
1517	if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1518		return false;
1519
1520	if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1521		return false;
1522
1523	if (!noblock)
1524		return false;
1525
1526	if (zerocopy)
1527		return false;
1528
1529	if (SKB_DATA_ALIGN(len + TUN_RX_PAD + XDP_PACKET_HEADROOM) +
1530	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1531		return false;
1532
1533	return true;
1534}
1535
1536static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1537				       struct page_frag *alloc_frag, char *buf,
1538				       int buflen, int len, int pad,
1539				       int metasize)
1540{
1541	struct sk_buff *skb = build_skb(buf, buflen);
1542
1543	if (!skb)
1544		return ERR_PTR(-ENOMEM);
1545
1546	skb_reserve(skb, pad);
1547	skb_put(skb, len);
1548	if (metasize)
1549		skb_metadata_set(skb, metasize);
1550	skb_set_owner_w(skb, tfile->socket.sk);
1551
1552	get_page(alloc_frag->page);
1553	alloc_frag->offset += buflen;
1554
1555	return skb;
1556}
1557
1558static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1559		       struct xdp_buff *xdp, u32 act)
1560{
1561	int err;
1562
1563	switch (act) {
1564	case XDP_REDIRECT:
1565		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1566		if (err) {
1567			dev_core_stats_rx_dropped_inc(tun->dev);
1568			return err;
1569		}
1570		dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1571		break;
1572	case XDP_TX:
1573		err = tun_xdp_tx(tun->dev, xdp);
1574		if (err < 0) {
1575			dev_core_stats_rx_dropped_inc(tun->dev);
1576			return err;
1577		}
1578		dev_sw_netstats_rx_add(tun->dev, xdp->data_end - xdp->data);
1579		break;
1580	case XDP_PASS:
1581		break;
1582	default:
1583		bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1584		fallthrough;
1585	case XDP_ABORTED:
1586		trace_xdp_exception(tun->dev, xdp_prog, act);
1587		fallthrough;
1588	case XDP_DROP:
1589		dev_core_stats_rx_dropped_inc(tun->dev);
1590		break;
1591	}
1592
1593	return act;
1594}
1595
1596static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1597				     struct tun_file *tfile,
1598				     struct iov_iter *from,
1599				     struct virtio_net_hdr *hdr,
1600				     int len, int *skb_xdp)
1601{
1602	struct page_frag *alloc_frag = &current->task_frag;
1603	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
1604	struct bpf_prog *xdp_prog;
1605	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1606	char *buf;
1607	size_t copied;
1608	int pad = TUN_RX_PAD;
1609	int metasize = 0;
1610	int err = 0;
1611
1612	rcu_read_lock();
1613	xdp_prog = rcu_dereference(tun->xdp_prog);
1614	if (xdp_prog)
1615		pad += XDP_PACKET_HEADROOM;
1616	buflen += SKB_DATA_ALIGN(len + pad);
1617	rcu_read_unlock();
1618
1619	alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1620	if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1621		return ERR_PTR(-ENOMEM);
1622
1623	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1624	copied = copy_page_from_iter(alloc_frag->page,
1625				     alloc_frag->offset + pad,
1626				     len, from);
1627	if (copied != len)
1628		return ERR_PTR(-EFAULT);
1629
1630	/* There's a small window that XDP may be set after the check
1631	 * of xdp_prog above, this should be rare and for simplicity
1632	 * we do XDP on skb in case the headroom is not enough.
1633	 */
1634	if (hdr->gso_type || !xdp_prog) {
1635		*skb_xdp = 1;
1636		return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1637				       pad, metasize);
1638	}
1639
1640	*skb_xdp = 0;
1641
1642	local_bh_disable();
1643	rcu_read_lock();
1644	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
1645	xdp_prog = rcu_dereference(tun->xdp_prog);
1646	if (xdp_prog) {
1647		struct xdp_buff xdp;
1648		u32 act;
1649
1650		xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1651		xdp_prepare_buff(&xdp, buf, pad, len, true);
1652
1653		act = bpf_prog_run_xdp(xdp_prog, &xdp);
1654		if (act == XDP_REDIRECT || act == XDP_TX) {
1655			get_page(alloc_frag->page);
1656			alloc_frag->offset += buflen;
1657		}
1658		err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1659		if (err < 0) {
1660			if (act == XDP_REDIRECT || act == XDP_TX)
1661				put_page(alloc_frag->page);
1662			goto out;
1663		}
1664
1665		if (err == XDP_REDIRECT)
1666			xdp_do_flush();
1667		if (err != XDP_PASS)
1668			goto out;
1669
1670		pad = xdp.data - xdp.data_hard_start;
1671		len = xdp.data_end - xdp.data;
1672
1673		/* It is known that the xdp_buff was prepared with metadata
1674		 * support, so the metasize will never be negative.
1675		 */
1676		metasize = xdp.data - xdp.data_meta;
1677	}
1678	bpf_net_ctx_clear(bpf_net_ctx);
1679	rcu_read_unlock();
1680	local_bh_enable();
1681
1682	return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad,
1683			       metasize);
1684
1685out:
1686	bpf_net_ctx_clear(bpf_net_ctx);
1687	rcu_read_unlock();
1688	local_bh_enable();
1689	return NULL;
1690}
1691
1692/* Get packet from user space buffer */
1693static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1694			    void *msg_control, struct iov_iter *from,
1695			    int noblock, bool more)
1696{
1697	struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1698	struct sk_buff *skb;
1699	size_t total_len = iov_iter_count(from);
1700	size_t len = total_len, align = tun->align, linear;
1701	struct virtio_net_hdr_v1_hash_tunnel hdr;
1702	struct virtio_net_hdr *gso;
1703	int good_linear;
1704	int copylen;
1705	int hdr_len = 0;
1706	bool zerocopy = false;
1707	int err;
1708	u32 rxhash = 0;
1709	int skb_xdp = 1;
1710	bool frags = tun_napi_frags_enabled(tfile);
1711	enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1712	netdev_features_t features = 0;
1713
1714	/*
1715	 * Keep it easy and always zero the whole buffer, even if the
1716	 * tunnel-related field will be touched only when the feature
1717	 * is enabled and the hdr size id compatible.
1718	 */
1719	memset(&hdr, 0, sizeof(hdr));
1720	gso = (struct virtio_net_hdr *)&hdr;
1721
1722	if (!(tun->flags & IFF_NO_PI)) {
1723		if (len < sizeof(pi))
1724			return -EINVAL;
1725		len -= sizeof(pi);
1726
1727		if (!copy_from_iter_full(&pi, sizeof(pi), from))
1728			return -EFAULT;
1729	}
1730
1731	if (tun->flags & IFF_VNET_HDR) {
1732		int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1733
1734		features = tun_vnet_hdr_guest_features(vnet_hdr_sz);
1735		hdr_len = __tun_vnet_hdr_get(vnet_hdr_sz, tun->flags,
1736					     features, from, gso);
1737		if (hdr_len < 0)
1738			return hdr_len;
1739
1740		len -= vnet_hdr_sz;
1741	}
1742
1743	if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1744		align += NET_IP_ALIGN;
1745		if (unlikely(len < ETH_HLEN || (hdr_len && hdr_len < ETH_HLEN)))
1746			return -EINVAL;
1747	}
1748
1749	good_linear = SKB_MAX_HEAD(align);
1750
1751	if (msg_control) {
1752		struct iov_iter i = *from;
1753
1754		/* There are 256 bytes to be copied in skb, so there is
1755		 * enough room for skb expand head in case it is used.
1756		 * The rest of the buffer is mapped from userspace.
1757		 */
1758		copylen = min(hdr_len ? hdr_len : GOODCOPY_LEN, good_linear);
1759		linear = copylen;
1760		iov_iter_advance(&i, copylen);
1761		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1762			zerocopy = true;
1763	}
1764
1765	if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1766		/* For the packet that is not easy to be processed
1767		 * (e.g gso or jumbo packet), we will do it at after
1768		 * skb was created with generic XDP routine.
1769		 */
1770		skb = tun_build_skb(tun, tfile, from, gso, len, &skb_xdp);
1771		err = PTR_ERR_OR_ZERO(skb);
1772		if (err)
1773			goto drop;
1774		if (!skb)
1775			return total_len;
1776	} else {
1777		if (!zerocopy) {
1778			copylen = len;
1779			linear = min(hdr_len, good_linear);
1780		}
1781
1782		if (frags) {
1783			mutex_lock(&tfile->napi_mutex);
1784			skb = tun_napi_alloc_frags(tfile, copylen, from);
1785			/* tun_napi_alloc_frags() enforces a layout for the skb.
1786			 * If zerocopy is enabled, then this layout will be
1787			 * overwritten by zerocopy_sg_from_iter().
1788			 */
1789			zerocopy = false;
1790		} else {
1791			if (!linear)
1792				linear = min_t(size_t, good_linear, copylen);
1793
1794			skb = tun_alloc_skb(tfile, align, copylen, linear,
1795					    noblock);
1796		}
1797
1798		err = PTR_ERR_OR_ZERO(skb);
1799		if (err)
1800			goto drop;
1801
1802		if (zerocopy)
1803			err = zerocopy_sg_from_iter(skb, from);
1804		else
1805			err = skb_copy_datagram_from_iter(skb, 0, from, len);
1806
1807		if (err) {
1808			err = -EFAULT;
1809			drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1810			goto drop;
1811		}
1812	}
1813
1814	if (tun_vnet_hdr_tnl_to_skb(tun->flags, features, skb, &hdr)) {
1815		atomic_long_inc(&tun->rx_frame_errors);
1816		err = -EINVAL;
1817		goto free_skb;
1818	}
1819
1820	switch (tun->flags & TUN_TYPE_MASK) {
1821	case IFF_TUN:
1822		if (tun->flags & IFF_NO_PI) {
1823			u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1824
1825			switch (ip_version) {
1826			case 4:
1827				pi.proto = htons(ETH_P_IP);
1828				break;
1829			case 6:
1830				pi.proto = htons(ETH_P_IPV6);
1831				break;
1832			default:
1833				err = -EINVAL;
1834				goto drop;
1835			}
1836		}
1837
1838		skb_reset_mac_header(skb);
1839		skb->protocol = pi.proto;
1840		skb->dev = tun->dev;
1841		break;
1842	case IFF_TAP:
1843		if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1844			err = -ENOMEM;
1845			drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1846			goto drop;
1847		}
1848		skb->protocol = eth_type_trans(skb, tun->dev);
1849		break;
1850	}
1851
1852	/* copy skb_ubuf_info for callback when skb has no error */
1853	if (zerocopy) {
1854		skb_zcopy_init(skb, msg_control);
1855	} else if (msg_control) {
1856		struct ubuf_info *uarg = msg_control;
1857		uarg->ops->complete(NULL, uarg, false);
1858	}
1859
1860	skb_reset_network_header(skb);
1861	skb_probe_transport_header(skb);
1862	skb_record_rx_queue(skb, tfile->queue_index);
1863
1864	if (skb_xdp) {
1865		struct bpf_prog *xdp_prog;
1866		int ret;
1867
1868		local_bh_disable();
1869		rcu_read_lock();
1870		xdp_prog = rcu_dereference(tun->xdp_prog);
1871		if (xdp_prog) {
1872			ret = do_xdp_generic(xdp_prog, &skb);
1873			if (ret != XDP_PASS) {
1874				rcu_read_unlock();
1875				local_bh_enable();
1876				goto unlock_frags;
1877			}
1878
1879			if (frags && skb != tfile->napi.skb)
1880				tfile->napi.skb = skb;
1881		}
1882		rcu_read_unlock();
1883		local_bh_enable();
1884	}
1885
1886	/* Compute the costly rx hash only if needed for flow updates.
1887	 * We may get a very small possibility of OOO during switching, not
1888	 * worth to optimize.
1889	 */
1890	if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1891	    !tfile->detached)
1892		rxhash = __skb_get_hash_symmetric(skb);
1893
1894	rcu_read_lock();
1895	if (unlikely(!(tun->dev->flags & IFF_UP))) {
1896		err = -EIO;
1897		rcu_read_unlock();
1898		drop_reason = SKB_DROP_REASON_DEV_READY;
1899		goto drop;
1900	}
1901
1902	if (frags) {
1903		u32 headlen;
1904
1905		/* Exercise flow dissector code path. */
1906		skb_push(skb, ETH_HLEN);
1907		headlen = eth_get_headlen(tun->dev, skb->data,
1908					  skb_headlen(skb));
1909
1910		if (unlikely(headlen > skb_headlen(skb))) {
1911			WARN_ON_ONCE(1);
1912			err = -ENOMEM;
1913			dev_core_stats_rx_dropped_inc(tun->dev);
1914napi_busy:
1915			napi_free_frags(&tfile->napi);
1916			rcu_read_unlock();
1917			mutex_unlock(&tfile->napi_mutex);
1918			return err;
1919		}
1920
1921		if (likely(napi_schedule_prep(&tfile->napi))) {
1922			local_bh_disable();
1923			napi_gro_frags(&tfile->napi);
1924			napi_complete(&tfile->napi);
1925			local_bh_enable();
1926		} else {
1927			err = -EBUSY;
1928			goto napi_busy;
1929		}
1930		mutex_unlock(&tfile->napi_mutex);
1931	} else if (tfile->napi_enabled) {
1932		struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1933		int queue_len;
1934
1935		spin_lock_bh(&queue->lock);
1936
1937		if (unlikely(tfile->detached)) {
1938			spin_unlock_bh(&queue->lock);
1939			rcu_read_unlock();
1940			err = -EBUSY;
1941			goto free_skb;
1942		}
1943
1944		__skb_queue_tail(queue, skb);
1945		queue_len = skb_queue_len(queue);
1946		spin_unlock(&queue->lock);
1947
1948		if (!more || queue_len > NAPI_POLL_WEIGHT)
1949			napi_schedule(&tfile->napi);
1950
1951		local_bh_enable();
1952	} else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1953		tun_rx_batched(tun, tfile, skb, more);
1954	} else {
1955		netif_rx(skb);
1956	}
1957	rcu_read_unlock();
1958
1959	preempt_disable();
1960	dev_sw_netstats_rx_add(tun->dev, len);
1961	preempt_enable();
1962
1963	if (rxhash)
1964		tun_flow_update(tun, rxhash, tfile);
1965
1966	return total_len;
1967
1968drop:
1969	if (err != -EAGAIN)
1970		dev_core_stats_rx_dropped_inc(tun->dev);
1971
1972free_skb:
1973	if (!IS_ERR_OR_NULL(skb))
1974		kfree_skb_reason(skb, drop_reason);
1975
1976unlock_frags:
1977	if (frags) {
1978		tfile->napi.skb = NULL;
1979		mutex_unlock(&tfile->napi_mutex);
1980	}
1981
1982	return err ?: total_len;
1983}
1984
1985static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1986{
1987	struct file *file = iocb->ki_filp;
1988	struct tun_file *tfile = file->private_data;
1989	struct tun_struct *tun = tun_get(tfile);
1990	ssize_t result;
1991	int noblock = 0;
1992
1993	if (!tun)
1994		return -EBADFD;
1995
1996	if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1997		noblock = 1;
1998
1999	result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2000
2001	tun_put(tun);
2002	return result;
2003}
2004
2005static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2006				struct tun_file *tfile,
2007				struct xdp_frame *xdp_frame,
2008				struct iov_iter *iter)
2009{
2010	int vnet_hdr_sz = 0;
2011	size_t size = xdp_frame->len;
2012	ssize_t ret;
2013
2014	if (tun->flags & IFF_VNET_HDR) {
2015		struct virtio_net_hdr gso = { 0 };
2016
2017		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2018		ret = tun_vnet_hdr_put(vnet_hdr_sz, iter, &gso);
2019		if (ret)
2020			return ret;
2021	}
2022
2023	ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2024
2025	preempt_disable();
2026	dev_sw_netstats_tx_add(tun->dev, 1, ret);
2027	preempt_enable();
2028
2029	return ret;
2030}
2031
2032/* Put packet to the user space buffer */
2033static ssize_t tun_put_user(struct tun_struct *tun,
2034			    struct tun_file *tfile,
2035			    struct sk_buff *skb,
2036			    struct iov_iter *iter)
2037{
2038	struct tun_pi pi = { 0, skb->protocol };
2039	ssize_t total;
2040	int vlan_offset = 0;
2041	int vlan_hlen = 0;
2042	int vnet_hdr_sz = 0;
2043	int ret;
2044
2045	if (skb_vlan_tag_present(skb))
2046		vlan_hlen = VLAN_HLEN;
2047
2048	if (tun->flags & IFF_VNET_HDR)
2049		vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2050
2051	total = skb->len + vlan_hlen + vnet_hdr_sz;
2052
2053	if (!(tun->flags & IFF_NO_PI)) {
2054		if (iov_iter_count(iter) < sizeof(pi))
2055			return -EINVAL;
2056
2057		total += sizeof(pi);
2058		if (iov_iter_count(iter) < total) {
2059			/* Packet will be striped */
2060			pi.flags |= TUN_PKT_STRIP;
2061		}
2062
2063		if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2064			return -EFAULT;
2065	}
2066
2067	if (vnet_hdr_sz) {
2068		struct virtio_net_hdr_v1_hash_tunnel hdr;
2069		struct virtio_net_hdr *gso;
2070
2071		ret = tun_vnet_hdr_tnl_from_skb(tun->flags, tun->dev, skb,
2072						&hdr);
2073		if (ret)
2074			return ret;
2075
2076		/*
2077		 * Drop the packet if the configured header size is too small
2078		 * WRT the enabled offloads.
2079		 */
2080		gso = (struct virtio_net_hdr *)&hdr;
2081		ret = __tun_vnet_hdr_put(vnet_hdr_sz, tun->dev->features,
2082					 iter, gso);
2083		if (ret)
2084			return ret;
2085	}
2086
2087	if (vlan_hlen) {
2088		int ret;
2089		struct veth veth;
2090
2091		veth.h_vlan_proto = skb->vlan_proto;
2092		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2093
2094		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2095
2096		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2097		if (ret || !iov_iter_count(iter))
2098			goto done;
2099
2100		ret = copy_to_iter(&veth, sizeof(veth), iter);
2101		if (ret != sizeof(veth) || !iov_iter_count(iter))
2102			goto done;
2103	}
2104
2105	skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2106
2107done:
2108	/* caller is in process context, */
2109	preempt_disable();
2110	dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2111	preempt_enable();
2112
2113	return total;
2114}
2115
2116static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2117{
2118	DECLARE_WAITQUEUE(wait, current);
2119	void *ptr = NULL;
2120	int error = 0;
2121
2122	ptr = ptr_ring_consume(&tfile->tx_ring);
2123	if (ptr)
2124		goto out;
2125	if (noblock) {
2126		error = -EAGAIN;
2127		goto out;
2128	}
2129
2130	add_wait_queue(&tfile->socket.wq.wait, &wait);
2131
2132	while (1) {
2133		set_current_state(TASK_INTERRUPTIBLE);
2134		ptr = ptr_ring_consume(&tfile->tx_ring);
2135		if (ptr)
2136			break;
2137		if (signal_pending(current)) {
2138			error = -ERESTARTSYS;
2139			break;
2140		}
2141		if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2142			error = -EFAULT;
2143			break;
2144		}
2145
2146		schedule();
2147	}
2148
2149	__set_current_state(TASK_RUNNING);
2150	remove_wait_queue(&tfile->socket.wq.wait, &wait);
2151
2152out:
2153	*err = error;
2154	return ptr;
2155}
2156
2157static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2158			   struct iov_iter *to,
2159			   int noblock, void *ptr)
2160{
2161	ssize_t ret;
2162	int err;
2163
2164	if (!iov_iter_count(to)) {
2165		tun_ptr_free(ptr);
2166		return 0;
2167	}
2168
2169	if (!ptr) {
2170		/* Read frames from ring */
2171		ptr = tun_ring_recv(tfile, noblock, &err);
2172		if (!ptr)
2173			return err;
2174	}
2175
2176	if (tun_is_xdp_frame(ptr)) {
2177		struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2178
2179		ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2180		xdp_return_frame(xdpf);
2181	} else {
2182		struct sk_buff *skb = ptr;
2183
2184		ret = tun_put_user(tun, tfile, skb, to);
2185		if (unlikely(ret < 0))
2186			kfree_skb(skb);
2187		else
2188			consume_skb(skb);
2189	}
2190
2191	return ret;
2192}
2193
2194static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2195{
2196	struct file *file = iocb->ki_filp;
2197	struct tun_file *tfile = file->private_data;
2198	struct tun_struct *tun = tun_get(tfile);
2199	ssize_t len = iov_iter_count(to), ret;
2200	int noblock = 0;
2201
2202	if (!tun)
2203		return -EBADFD;
2204
2205	if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2206		noblock = 1;
2207
2208	ret = tun_do_read(tun, tfile, to, noblock, NULL);
2209	ret = min_t(ssize_t, ret, len);
2210	if (ret > 0)
2211		iocb->ki_pos = ret;
2212	tun_put(tun);
2213	return ret;
2214}
2215
2216static void tun_prog_free(struct rcu_head *rcu)
2217{
2218	struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2219
2220	bpf_prog_destroy(prog->prog);
2221	kfree(prog);
2222}
2223
2224static int __tun_set_ebpf(struct tun_struct *tun,
2225			  struct tun_prog __rcu **prog_p,
2226			  struct bpf_prog *prog)
2227{
2228	struct tun_prog *old, *new = NULL;
2229
2230	if (prog) {
2231		new = kmalloc(sizeof(*new), GFP_KERNEL);
2232		if (!new)
2233			return -ENOMEM;
2234		new->prog = prog;
2235	}
2236
2237	spin_lock_bh(&tun->lock);
2238	old = rcu_dereference_protected(*prog_p,
2239					lockdep_is_held(&tun->lock));
2240	rcu_assign_pointer(*prog_p, new);
2241	spin_unlock_bh(&tun->lock);
2242
2243	if (old)
2244		call_rcu(&old->rcu, tun_prog_free);
2245
2246	return 0;
2247}
2248
2249static void tun_free_netdev(struct net_device *dev)
2250{
2251	struct tun_struct *tun = netdev_priv(dev);
2252
2253	BUG_ON(!(list_empty(&tun->disabled)));
2254
2255	tun_flow_uninit(tun);
2256	security_tun_dev_free_security(tun->security);
2257	__tun_set_ebpf(tun, &tun->steering_prog, NULL);
2258	__tun_set_ebpf(tun, &tun->filter_prog, NULL);
2259}
2260
2261static void tun_setup(struct net_device *dev)
2262{
2263	struct tun_struct *tun = netdev_priv(dev);
2264
2265	tun->owner = INVALID_UID;
2266	tun->group = INVALID_GID;
2267	tun_default_link_ksettings(dev, &tun->link_ksettings);
2268
2269	dev->ethtool_ops = &tun_ethtool_ops;
2270	dev->needs_free_netdev = true;
2271	dev->priv_destructor = tun_free_netdev;
2272	/* We prefer our own queue length */
2273	dev->tx_queue_len = TUN_READQ_SIZE;
2274}
2275
2276/* Trivial set of netlink ops to allow deleting tun or tap
2277 * device with netlink.
2278 */
2279static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2280			struct netlink_ext_ack *extack)
2281{
2282	NL_SET_ERR_MSG(extack,
2283		       "tun/tap creation via rtnetlink is not supported.");
2284	return -EOPNOTSUPP;
2285}
2286
2287static size_t tun_get_size(const struct net_device *dev)
2288{
2289	BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2290	BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2291
2292	return nla_total_size(sizeof(uid_t)) + /* OWNER */
2293	       nla_total_size(sizeof(gid_t)) + /* GROUP */
2294	       nla_total_size(sizeof(u8)) + /* TYPE */
2295	       nla_total_size(sizeof(u8)) + /* PI */
2296	       nla_total_size(sizeof(u8)) + /* VNET_HDR */
2297	       nla_total_size(sizeof(u8)) + /* PERSIST */
2298	       nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2299	       nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2300	       nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2301	       0;
2302}
2303
2304static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2305{
2306	struct tun_struct *tun = netdev_priv(dev);
2307
2308	if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2309		goto nla_put_failure;
2310	if (uid_valid(tun->owner) &&
2311	    nla_put_u32(skb, IFLA_TUN_OWNER,
2312			from_kuid_munged(current_user_ns(), tun->owner)))
2313		goto nla_put_failure;
2314	if (gid_valid(tun->group) &&
2315	    nla_put_u32(skb, IFLA_TUN_GROUP,
2316			from_kgid_munged(current_user_ns(), tun->group)))
2317		goto nla_put_failure;
2318	if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2319		goto nla_put_failure;
2320	if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2321		goto nla_put_failure;
2322	if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2323		goto nla_put_failure;
2324	if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2325		       !!(tun->flags & IFF_MULTI_QUEUE)))
2326		goto nla_put_failure;
2327	if (tun->flags & IFF_MULTI_QUEUE) {
2328		if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2329			goto nla_put_failure;
2330		if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2331				tun->numdisabled))
2332			goto nla_put_failure;
2333	}
2334
2335	return 0;
2336
2337nla_put_failure:
2338	return -EMSGSIZE;
2339}
2340
2341static struct rtnl_link_ops tun_link_ops __read_mostly = {
2342	.kind		= DRV_NAME,
2343	.priv_size	= sizeof(struct tun_struct),
2344	.setup		= tun_setup,
2345	.validate	= tun_validate,
2346	.get_size       = tun_get_size,
2347	.fill_info      = tun_fill_info,
2348};
2349
2350static void tun_sock_write_space(struct sock *sk)
2351{
2352	struct tun_file *tfile;
2353	wait_queue_head_t *wqueue;
2354
2355	if (!sock_writeable(sk))
2356		return;
2357
2358	if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2359		return;
2360
2361	wqueue = sk_sleep(sk);
2362	if (wqueue && waitqueue_active(wqueue))
2363		wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2364						EPOLLWRNORM | EPOLLWRBAND);
2365
2366	tfile = container_of(sk, struct tun_file, sk);
2367	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2368}
2369
2370static void tun_put_page(struct tun_page *tpage)
2371{
2372	if (tpage->page)
2373		__page_frag_cache_drain(tpage->page, tpage->count);
2374}
2375
2376static int tun_xdp_one(struct tun_struct *tun,
2377		       struct tun_file *tfile,
2378		       struct xdp_buff *xdp, int *flush,
2379		       struct tun_page *tpage)
2380{
2381	unsigned int datasize = xdp->data_end - xdp->data;
2382	struct virtio_net_hdr *gso = xdp->data_hard_start;
2383	struct virtio_net_hdr_v1_hash_tunnel *tnl_hdr;
2384	struct bpf_prog *xdp_prog;
2385	struct sk_buff *skb = NULL;
2386	struct sk_buff_head *queue;
2387	netdev_features_t features;
2388	u32 rxhash = 0, act;
2389	int buflen = xdp->frame_sz;
2390	int metasize = 0;
2391	int ret = 0;
2392	bool skb_xdp = false;
2393	struct page *page;
2394
2395	if (unlikely(datasize < ETH_HLEN))
2396		return -EINVAL;
2397
2398	xdp_prog = rcu_dereference(tun->xdp_prog);
2399	if (xdp_prog) {
2400		if (gso->gso_type) {
2401			skb_xdp = true;
2402			goto build;
2403		}
2404
2405		xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2406
2407		act = bpf_prog_run_xdp(xdp_prog, xdp);
2408		ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2409		if (ret < 0) {
2410			put_page(virt_to_head_page(xdp->data));
2411			return ret;
2412		}
2413
2414		switch (ret) {
2415		case XDP_REDIRECT:
2416			*flush = true;
2417			fallthrough;
2418		case XDP_TX:
2419			return 0;
2420		case XDP_PASS:
2421			break;
2422		default:
2423			page = virt_to_head_page(xdp->data);
2424			if (tpage->page == page) {
2425				++tpage->count;
2426			} else {
2427				tun_put_page(tpage);
2428				tpage->page = page;
2429				tpage->count = 1;
2430			}
2431			return 0;
2432		}
2433	}
2434
2435build:
2436	skb = build_skb(xdp->data_hard_start, buflen);
2437	if (!skb) {
2438		ret = -ENOMEM;
2439		goto out;
2440	}
2441
2442	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2443	skb_put(skb, xdp->data_end - xdp->data);
2444
2445	/* The externally provided xdp_buff may have no metadata support, which
2446	 * is marked by xdp->data_meta being xdp->data + 1. This will lead to a
2447	 * metasize of -1 and is the reason why the condition checks for > 0.
2448	 */
2449	metasize = xdp->data - xdp->data_meta;
2450	if (metasize > 0)
2451		skb_metadata_set(skb, metasize);
2452
2453	features = tun_vnet_hdr_guest_features(READ_ONCE(tun->vnet_hdr_sz));
2454	tnl_hdr = (struct virtio_net_hdr_v1_hash_tunnel *)gso;
2455	if (tun_vnet_hdr_tnl_to_skb(tun->flags, features, skb, tnl_hdr)) {
2456		atomic_long_inc(&tun->rx_frame_errors);
2457		kfree_skb(skb);
2458		ret = -EINVAL;
2459		goto out;
2460	}
2461
2462	skb->protocol = eth_type_trans(skb, tun->dev);
2463	skb_reset_network_header(skb);
2464	skb_probe_transport_header(skb);
2465	skb_record_rx_queue(skb, tfile->queue_index);
2466
2467	if (skb_xdp) {
2468		ret = do_xdp_generic(xdp_prog, &skb);
2469		if (ret != XDP_PASS) {
2470			ret = 0;
2471			goto out;
2472		}
2473	}
2474
2475	if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2476	    !tfile->detached)
2477		rxhash = __skb_get_hash_symmetric(skb);
2478
2479	if (tfile->napi_enabled) {
2480		queue = &tfile->sk.sk_write_queue;
2481		spin_lock(&queue->lock);
2482
2483		if (unlikely(tfile->detached)) {
2484			spin_unlock(&queue->lock);
2485			kfree_skb(skb);
2486			return -EBUSY;
2487		}
2488
2489		__skb_queue_tail(queue, skb);
2490		spin_unlock(&queue->lock);
2491		ret = 1;
2492	} else {
2493		netif_receive_skb(skb);
2494		ret = 0;
2495	}
2496
2497	/* No need to disable preemption here since this function is
2498	 * always called with bh disabled
2499	 */
2500	dev_sw_netstats_rx_add(tun->dev, datasize);
2501
2502	if (rxhash)
2503		tun_flow_update(tun, rxhash, tfile);
2504
2505out:
2506	return ret;
2507}
2508
2509static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2510{
2511	int ret, i;
2512	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2513	struct tun_struct *tun = tun_get(tfile);
2514	struct tun_msg_ctl *ctl = m->msg_control;
2515	struct xdp_buff *xdp;
2516
2517	if (!tun)
2518		return -EBADFD;
2519
2520	if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2521	    ctl && ctl->type == TUN_MSG_PTR) {
2522		struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
2523		struct tun_page tpage;
2524		int n = ctl->num;
2525		int flush = 0, queued = 0;
2526
2527		memset(&tpage, 0, sizeof(tpage));
2528
2529		local_bh_disable();
2530		rcu_read_lock();
2531		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
2532
2533		for (i = 0; i < n; i++) {
2534			xdp = &((struct xdp_buff *)ctl->ptr)[i];
2535			ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2536			if (ret > 0)
2537				queued += ret;
2538		}
2539
2540		if (flush)
2541			xdp_do_flush();
2542
2543		if (tfile->napi_enabled && queued > 0)
2544			napi_schedule(&tfile->napi);
2545
2546		bpf_net_ctx_clear(bpf_net_ctx);
2547		rcu_read_unlock();
2548		local_bh_enable();
2549
2550		tun_put_page(&tpage);
2551
2552		ret = total_len;
2553		goto out;
2554	}
2555
2556	ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2557			   m->msg_flags & MSG_DONTWAIT,
2558			   m->msg_flags & MSG_MORE);
2559out:
2560	tun_put(tun);
2561	return ret;
2562}
2563
2564static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2565		       int flags)
2566{
2567	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2568	struct tun_struct *tun = tun_get(tfile);
2569	void *ptr = m->msg_control;
2570	int ret;
2571
2572	if (!tun) {
2573		ret = -EBADFD;
2574		goto out_free;
2575	}
2576
2577	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2578		ret = -EINVAL;
2579		goto out_put_tun;
2580	}
2581	if (flags & MSG_ERRQUEUE) {
2582		ret = sock_recv_errqueue(sock->sk, m, total_len,
2583					 SOL_PACKET, TUN_TX_TIMESTAMP);
2584		goto out;
2585	}
2586	ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2587	if (ret > (ssize_t)total_len) {
2588		m->msg_flags |= MSG_TRUNC;
2589		ret = flags & MSG_TRUNC ? ret : total_len;
2590	}
2591out:
2592	tun_put(tun);
2593	return ret;
2594
2595out_put_tun:
2596	tun_put(tun);
2597out_free:
2598	tun_ptr_free(ptr);
2599	return ret;
2600}
2601
2602static int tun_ptr_peek_len(void *ptr)
2603{
2604	if (likely(ptr)) {
2605		if (tun_is_xdp_frame(ptr)) {
2606			struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2607
2608			return xdpf->len;
2609		}
2610		return __skb_array_len_with_tag(ptr);
2611	} else {
2612		return 0;
2613	}
2614}
2615
2616static int tun_peek_len(struct socket *sock)
2617{
2618	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2619	struct tun_struct *tun;
2620	int ret = 0;
2621
2622	tun = tun_get(tfile);
2623	if (!tun)
2624		return 0;
2625
2626	ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2627	tun_put(tun);
2628
2629	return ret;
2630}
2631
2632/* Ops structure to mimic raw sockets with tun */
2633static const struct proto_ops tun_socket_ops = {
2634	.peek_len = tun_peek_len,
2635	.sendmsg = tun_sendmsg,
2636	.recvmsg = tun_recvmsg,
2637};
2638
2639static struct proto tun_proto = {
2640	.name		= "tun",
2641	.owner		= THIS_MODULE,
2642	.obj_size	= sizeof(struct tun_file),
2643};
2644
2645static int tun_flags(struct tun_struct *tun)
2646{
2647	return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2648}
2649
2650static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2651			      char *buf)
2652{
2653	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2654	return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2655}
2656
2657static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2658			  char *buf)
2659{
2660	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2661	return uid_valid(tun->owner)?
2662		sysfs_emit(buf, "%u\n",
2663			   from_kuid_munged(current_user_ns(), tun->owner)) :
2664		sysfs_emit(buf, "-1\n");
2665}
2666
2667static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2668			  char *buf)
2669{
2670	struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2671	return gid_valid(tun->group) ?
2672		sysfs_emit(buf, "%u\n",
2673			   from_kgid_munged(current_user_ns(), tun->group)) :
2674		sysfs_emit(buf, "-1\n");
2675}
2676
2677static DEVICE_ATTR_RO(tun_flags);
2678static DEVICE_ATTR_RO(owner);
2679static DEVICE_ATTR_RO(group);
2680
2681static struct attribute *tun_dev_attrs[] = {
2682	&dev_attr_tun_flags.attr,
2683	&dev_attr_owner.attr,
2684	&dev_attr_group.attr,
2685	NULL
2686};
2687
2688static const struct attribute_group tun_attr_group = {
2689	.attrs = tun_dev_attrs
2690};
2691
2692static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2693{
2694	struct tun_struct *tun;
2695	struct tun_file *tfile = file->private_data;
2696	struct net_device *dev;
2697	int err;
2698
2699	if (tfile->detached)
2700		return -EINVAL;
2701
2702	if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2703		if (!capable(CAP_NET_ADMIN))
2704			return -EPERM;
2705
2706		if (!(ifr->ifr_flags & IFF_NAPI) ||
2707		    (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2708			return -EINVAL;
2709	}
2710
2711	dev = __dev_get_by_name(net, ifr->ifr_name);
2712	if (dev) {
2713		if (ifr->ifr_flags & IFF_TUN_EXCL)
2714			return -EBUSY;
2715		if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2716			tun = netdev_priv(dev);
2717		else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2718			tun = netdev_priv(dev);
2719		else
2720			return -EINVAL;
2721
2722		if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2723		    !!(tun->flags & IFF_MULTI_QUEUE))
2724			return -EINVAL;
2725
2726		if (tun_not_capable(tun))
2727			return -EPERM;
2728		err = security_tun_dev_open(tun->security);
2729		if (err < 0)
2730			return err;
2731
2732		err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2733				 ifr->ifr_flags & IFF_NAPI,
2734				 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2735		if (err < 0)
2736			return err;
2737
2738		if (tun->flags & IFF_MULTI_QUEUE &&
2739		    (tun->numqueues + tun->numdisabled > 1)) {
2740			/* One or more queue has already been attached, no need
2741			 * to initialize the device again.
2742			 */
2743			netdev_state_change(dev);
2744			return 0;
2745		}
2746
2747		tun->flags = (tun->flags & ~TUN_FEATURES) |
2748			      (ifr->ifr_flags & TUN_FEATURES);
2749
2750		netdev_state_change(dev);
2751	} else {
2752		char *name;
2753		unsigned long flags = 0;
2754		int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2755			     MAX_TAP_QUEUES : 1;
2756
2757		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2758			return -EPERM;
2759		err = security_tun_dev_create();
2760		if (err < 0)
2761			return err;
2762
2763		/* Set dev type */
2764		if (ifr->ifr_flags & IFF_TUN) {
2765			/* TUN device */
2766			flags |= IFF_TUN;
2767			name = "tun%d";
2768		} else if (ifr->ifr_flags & IFF_TAP) {
2769			/* TAP device */
2770			flags |= IFF_TAP;
2771			name = "tap%d";
2772		} else
2773			return -EINVAL;
2774
2775		if (*ifr->ifr_name)
2776			name = ifr->ifr_name;
2777
2778		dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2779				       NET_NAME_UNKNOWN, tun_setup, queues,
2780				       queues);
2781
2782		if (!dev)
2783			return -ENOMEM;
2784
2785		dev_net_set(dev, net);
2786		dev->rtnl_link_ops = &tun_link_ops;
2787		dev->ifindex = tfile->ifindex;
2788		dev->sysfs_groups[0] = &tun_attr_group;
2789
2790		tun = netdev_priv(dev);
2791		tun->dev = dev;
2792		tun->flags = flags;
2793		tun->txflt.count = 0;
2794		tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2795
2796		tun->align = NET_SKB_PAD;
2797		tun->filter_attached = false;
2798		tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2799		tun->rx_batched = 0;
2800		RCU_INIT_POINTER(tun->steering_prog, NULL);
2801
2802		tun->ifr = ifr;
2803		tun->file = file;
2804
2805		tun_net_initialize(dev);
2806
2807		err = register_netdevice(tun->dev);
2808		if (err < 0) {
2809			free_netdev(dev);
2810			return err;
2811		}
2812		/* free_netdev() won't check refcnt, to avoid race
2813		 * with dev_put() we need publish tun after registration.
2814		 */
2815		rcu_assign_pointer(tfile->tun, tun);
2816	}
2817
2818	if (ifr->ifr_flags & IFF_NO_CARRIER)
2819		netif_carrier_off(tun->dev);
2820	else
2821		netif_carrier_on(tun->dev);
2822
2823	/* Make sure persistent devices do not get stuck in
2824	 * xoff state.
2825	 */
2826	if (netif_running(tun->dev))
2827		netif_tx_wake_all_queues(tun->dev);
2828
2829	strscpy(ifr->ifr_name, tun->dev->name);
2830	return 0;
2831}
2832
2833static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2834{
2835	strscpy(ifr->ifr_name, tun->dev->name);
2836
2837	ifr->ifr_flags = tun_flags(tun);
2838
2839}
2840
2841#define PLAIN_GSO (NETIF_F_GSO_UDP_L4 | NETIF_F_TSO | NETIF_F_TSO6)
2842
2843/* This is like a cut-down ethtool ops, except done via tun fd so no
2844 * privs required. */
2845static int set_offload(struct tun_struct *tun, unsigned long arg)
2846{
2847	netdev_features_t features = 0;
2848
2849	if (arg & TUN_F_CSUM) {
2850		features |= NETIF_F_HW_CSUM;
2851		arg &= ~TUN_F_CSUM;
2852
2853		if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2854			if (arg & TUN_F_TSO_ECN) {
2855				features |= NETIF_F_TSO_ECN;
2856				arg &= ~TUN_F_TSO_ECN;
2857			}
2858			if (arg & TUN_F_TSO4)
2859				features |= NETIF_F_TSO;
2860			if (arg & TUN_F_TSO6)
2861				features |= NETIF_F_TSO6;
2862			arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2863		}
2864
2865		arg &= ~TUN_F_UFO;
2866
2867		/* TODO: for now USO4 and USO6 should work simultaneously */
2868		if (arg & TUN_F_USO4 && arg & TUN_F_USO6) {
2869			features |= NETIF_F_GSO_UDP_L4;
2870			arg &= ~(TUN_F_USO4 | TUN_F_USO6);
2871		}
2872
2873		/*
2874		 * Tunnel offload is allowed only if some plain offload is
2875		 * available, too.
2876		 */
2877		if (features & PLAIN_GSO && arg & TUN_F_UDP_TUNNEL_GSO) {
2878			features |= NETIF_F_GSO_UDP_TUNNEL;
2879			if (arg & TUN_F_UDP_TUNNEL_GSO_CSUM)
2880				features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2881			arg &= ~(TUN_F_UDP_TUNNEL_GSO |
2882				 TUN_F_UDP_TUNNEL_GSO_CSUM);
2883		}
2884	}
2885
2886	/* This gives the user a way to test for new features in future by
2887	 * trying to set them. */
2888	if (arg)
2889		return -EINVAL;
2890
2891	tun->set_features = features;
2892	tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2893	tun->dev->wanted_features |= features;
2894	netdev_update_features(tun->dev);
2895
2896	return 0;
2897}
2898
2899static void tun_detach_filter(struct tun_struct *tun, int n)
2900{
2901	int i;
2902	struct tun_file *tfile;
2903
2904	for (i = 0; i < n; i++) {
2905		tfile = rtnl_dereference(tun->tfiles[i]);
2906		lock_sock(tfile->socket.sk);
2907		sk_detach_filter(tfile->socket.sk);
2908		release_sock(tfile->socket.sk);
2909	}
2910
2911	tun->filter_attached = false;
2912}
2913
2914static int tun_attach_filter(struct tun_struct *tun)
2915{
2916	int i, ret = 0;
2917	struct tun_file *tfile;
2918
2919	for (i = 0; i < tun->numqueues; i++) {
2920		tfile = rtnl_dereference(tun->tfiles[i]);
2921		lock_sock(tfile->socket.sk);
2922		ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2923		release_sock(tfile->socket.sk);
2924		if (ret) {
2925			tun_detach_filter(tun, i);
2926			return ret;
2927		}
2928	}
2929
2930	tun->filter_attached = true;
2931	return ret;
2932}
2933
2934static void tun_set_sndbuf(struct tun_struct *tun)
2935{
2936	struct tun_file *tfile;
2937	int i;
2938
2939	for (i = 0; i < tun->numqueues; i++) {
2940		tfile = rtnl_dereference(tun->tfiles[i]);
2941		tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2942	}
2943}
2944
2945static int tun_set_queue(struct file *file, struct ifreq *ifr)
2946{
2947	struct tun_file *tfile = file->private_data;
2948	struct tun_struct *tun;
2949	int ret = 0;
2950
2951	rtnl_lock();
2952
2953	if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2954		tun = tfile->detached;
2955		if (!tun) {
2956			ret = -EINVAL;
2957			goto unlock;
2958		}
2959		ret = security_tun_dev_attach_queue(tun->security);
2960		if (ret < 0)
2961			goto unlock;
2962		ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2963				 tun->flags & IFF_NAPI_FRAGS, true);
2964	} else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2965		tun = rtnl_dereference(tfile->tun);
2966		if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2967			ret = -EINVAL;
2968		else
2969			__tun_detach(tfile, false);
2970	} else
2971		ret = -EINVAL;
2972
2973	if (ret >= 0)
2974		netdev_state_change(tun->dev);
2975
2976unlock:
2977	rtnl_unlock();
2978	return ret;
2979}
2980
2981static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2982			void __user *data)
2983{
2984	struct bpf_prog *prog;
2985	int fd;
2986
2987	if (copy_from_user(&fd, data, sizeof(fd)))
2988		return -EFAULT;
2989
2990	if (fd == -1) {
2991		prog = NULL;
2992	} else {
2993		prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2994		if (IS_ERR(prog))
2995			return PTR_ERR(prog);
2996	}
2997
2998	return __tun_set_ebpf(tun, prog_p, prog);
2999}
3000
3001/* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3002static unsigned char tun_get_addr_len(unsigned short type)
3003{
3004	switch (type) {
3005	case ARPHRD_IP6GRE:
3006	case ARPHRD_TUNNEL6:
3007		return sizeof(struct in6_addr);
3008	case ARPHRD_IPGRE:
3009	case ARPHRD_TUNNEL:
3010	case ARPHRD_SIT:
3011		return 4;
3012	case ARPHRD_ETHER:
3013		return ETH_ALEN;
3014	case ARPHRD_IEEE802154:
3015	case ARPHRD_IEEE802154_MONITOR:
3016		return IEEE802154_EXTENDED_ADDR_LEN;
3017	case ARPHRD_PHONET_PIPE:
3018	case ARPHRD_PPP:
3019	case ARPHRD_NONE:
3020		return 0;
3021	case ARPHRD_6LOWPAN:
3022		return EUI64_ADDR_LEN;
3023	case ARPHRD_FDDI:
3024		return FDDI_K_ALEN;
3025	case ARPHRD_HIPPI:
3026		return HIPPI_ALEN;
3027	case ARPHRD_IEEE802:
3028		return FC_ALEN;
3029	case ARPHRD_ROSE:
3030		return ROSE_ADDR_LEN;
3031	case ARPHRD_NETROM:
3032		return AX25_ADDR_LEN;
3033	case ARPHRD_LOCALTLK:
3034		return LTALK_ALEN;
3035	default:
3036		return 0;
3037	}
3038}
3039
3040static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3041			    unsigned long arg, int ifreq_len)
3042{
3043	struct tun_file *tfile = file->private_data;
3044	struct net *net = sock_net(&tfile->sk);
3045	struct tun_struct *tun;
3046	void __user* argp = (void __user*)arg;
3047	unsigned int carrier;
3048	struct ifreq ifr;
3049	kuid_t owner;
3050	kgid_t group;
3051	int ifindex;
3052	int sndbuf;
3053	int ret;
3054	bool do_notify = false;
3055
3056	if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3057	    (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3058		if (copy_from_user(&ifr, argp, ifreq_len))
3059			return -EFAULT;
3060	} else {
3061		memset(&ifr, 0, sizeof(ifr));
3062	}
3063	if (cmd == TUNGETFEATURES) {
3064		/* Currently this just means: "what IFF flags are valid?".
3065		 * This is needed because we never checked for invalid flags on
3066		 * TUNSETIFF.
3067		 */
3068		return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3069				TUN_FEATURES, (unsigned int __user*)argp);
3070	} else if (cmd == TUNSETQUEUE) {
3071		return tun_set_queue(file, &ifr);
3072	} else if (cmd == SIOCGSKNS) {
3073		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3074			return -EPERM;
3075		return open_related_ns(&net->ns, get_net_ns);
3076	}
3077
3078	rtnl_lock();
3079
3080	tun = tun_get(tfile);
3081	if (cmd == TUNSETIFF) {
3082		ret = -EEXIST;
3083		if (tun)
3084			goto unlock;
3085
3086		ifr.ifr_name[IFNAMSIZ-1] = '\0';
3087
3088		ret = tun_set_iff(net, file, &ifr);
3089
3090		if (ret)
3091			goto unlock;
3092
3093		if (copy_to_user(argp, &ifr, ifreq_len))
3094			ret = -EFAULT;
3095		goto unlock;
3096	}
3097	if (cmd == TUNSETIFINDEX) {
3098		ret = -EPERM;
3099		if (tun)
3100			goto unlock;
3101
3102		ret = -EFAULT;
3103		if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3104			goto unlock;
3105		ret = -EINVAL;
3106		if (ifindex < 0)
3107			goto unlock;
3108		ret = 0;
3109		tfile->ifindex = ifindex;
3110		goto unlock;
3111	}
3112
3113	ret = -EBADFD;
3114	if (!tun)
3115		goto unlock;
3116
3117	netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3118
3119	net = dev_net(tun->dev);
3120	ret = 0;
3121	switch (cmd) {
3122	case TUNGETIFF:
3123		tun_get_iff(tun, &ifr);
3124
3125		if (tfile->detached)
3126			ifr.ifr_flags |= IFF_DETACH_QUEUE;
3127		if (!tfile->socket.sk->sk_filter)
3128			ifr.ifr_flags |= IFF_NOFILTER;
3129
3130		if (copy_to_user(argp, &ifr, ifreq_len))
3131			ret = -EFAULT;
3132		break;
3133
3134	case TUNSETNOCSUM:
3135		/* Disable/Enable checksum */
3136
3137		/* [unimplemented] */
3138		netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3139			   arg ? "disabled" : "enabled");
3140		break;
3141
3142	case TUNSETPERSIST:
3143		/* Disable/Enable persist mode. Keep an extra reference to the
3144		 * module to prevent the module being unprobed.
3145		 */
3146		if (arg && !(tun->flags & IFF_PERSIST)) {
3147			tun->flags |= IFF_PERSIST;
3148			__module_get(THIS_MODULE);
3149			do_notify = true;
3150		}
3151		if (!arg && (tun->flags & IFF_PERSIST)) {
3152			tun->flags &= ~IFF_PERSIST;
3153			module_put(THIS_MODULE);
3154			do_notify = true;
3155		}
3156
3157		netif_info(tun, drv, tun->dev, "persist %s\n",
3158			   arg ? "enabled" : "disabled");
3159		break;
3160
3161	case TUNSETOWNER:
3162		/* Set owner of the device */
3163		owner = make_kuid(current_user_ns(), arg);
3164		if (!uid_valid(owner)) {
3165			ret = -EINVAL;
3166			break;
3167		}
3168		tun->owner = owner;
3169		do_notify = true;
3170		netif_info(tun, drv, tun->dev, "owner set to %u\n",
3171			   from_kuid(&init_user_ns, tun->owner));
3172		break;
3173
3174	case TUNSETGROUP:
3175		/* Set group of the device */
3176		group = make_kgid(current_user_ns(), arg);
3177		if (!gid_valid(group)) {
3178			ret = -EINVAL;
3179			break;
3180		}
3181		tun->group = group;
3182		do_notify = true;
3183		netif_info(tun, drv, tun->dev, "group set to %u\n",
3184			   from_kgid(&init_user_ns, tun->group));
3185		break;
3186
3187	case TUNSETLINK:
3188		/* Only allow setting the type when the interface is down */
3189		if (tun->dev->flags & IFF_UP) {
3190			netif_info(tun, drv, tun->dev,
3191				   "Linktype set failed because interface is up\n");
3192			ret = -EBUSY;
3193		} else {
3194			ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3195						       tun->dev);
3196			ret = notifier_to_errno(ret);
3197			if (ret) {
3198				netif_info(tun, drv, tun->dev,
3199					   "Refused to change device type\n");
3200				break;
3201			}
3202			tun->dev->type = (int) arg;
3203			tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3204			netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3205				   tun->dev->type);
3206			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3207						 tun->dev);
3208		}
3209		break;
3210
3211	case TUNSETDEBUG:
3212		tun->msg_enable = (u32)arg;
3213		break;
3214
3215	case TUNSETOFFLOAD:
3216		ret = set_offload(tun, arg);
3217		break;
3218
3219	case TUNSETTXFILTER:
3220		/* Can be set only for TAPs */
3221		ret = -EINVAL;
3222		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3223			break;
3224		ret = update_filter(&tun->txflt, (void __user *)arg);
3225		break;
3226
3227	case SIOCGIFHWADDR:
3228		/* Get hw address */
3229		netif_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3230		if (copy_to_user(argp, &ifr, ifreq_len))
3231			ret = -EFAULT;
3232		break;
3233
3234	case SIOCSIFHWADDR:
3235		/* Set hw address */
3236		if (tun->dev->addr_len > sizeof(ifr.ifr_hwaddr)) {
3237			ret = -EINVAL;
3238			break;
3239		}
3240		ret = dev_set_mac_address_user(tun->dev,
3241					       (struct sockaddr_storage *)&ifr.ifr_hwaddr,
3242					       NULL);
3243		break;
3244
3245	case TUNGETSNDBUF:
3246		sndbuf = tfile->socket.sk->sk_sndbuf;
3247		if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3248			ret = -EFAULT;
3249		break;
3250
3251	case TUNSETSNDBUF:
3252		if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3253			ret = -EFAULT;
3254			break;
3255		}
3256		if (sndbuf <= 0) {
3257			ret = -EINVAL;
3258			break;
3259		}
3260
3261		tun->sndbuf = sndbuf;
3262		tun_set_sndbuf(tun);
3263		break;
3264
3265	case TUNATTACHFILTER:
3266		/* Can be set only for TAPs */
3267		ret = -EINVAL;
3268		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3269			break;
3270		ret = -EFAULT;
3271		if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3272			break;
3273
3274		ret = tun_attach_filter(tun);
3275		break;
3276
3277	case TUNDETACHFILTER:
3278		/* Can be set only for TAPs */
3279		ret = -EINVAL;
3280		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3281			break;
3282		ret = 0;
3283		tun_detach_filter(tun, tun->numqueues);
3284		break;
3285
3286	case TUNGETFILTER:
3287		ret = -EINVAL;
3288		if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3289			break;
3290		ret = -EFAULT;
3291		if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3292			break;
3293		ret = 0;
3294		break;
3295
3296	case TUNSETSTEERINGEBPF:
3297		ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3298		break;
3299
3300	case TUNSETFILTEREBPF:
3301		ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3302		break;
3303
3304	case TUNSETCARRIER:
3305		ret = -EFAULT;
3306		if (copy_from_user(&carrier, argp, sizeof(carrier)))
3307			goto unlock;
3308
3309		ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3310		break;
3311
3312	case TUNGETDEVNETNS:
3313		ret = -EPERM;
3314		if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3315			goto unlock;
3316		ret = open_related_ns(&net->ns, get_net_ns);
3317		break;
3318
3319	default:
3320		ret = tun_vnet_ioctl(&tun->vnet_hdr_sz, &tun->flags, cmd, argp);
3321		break;
3322	}
3323
3324	if (do_notify)
3325		netdev_state_change(tun->dev);
3326
3327unlock:
3328	rtnl_unlock();
3329	if (tun)
3330		tun_put(tun);
3331	return ret;
3332}
3333
3334static long tun_chr_ioctl(struct file *file,
3335			  unsigned int cmd, unsigned long arg)
3336{
3337	return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3338}
3339
3340#ifdef CONFIG_COMPAT
3341static long tun_chr_compat_ioctl(struct file *file,
3342			 unsigned int cmd, unsigned long arg)
3343{
3344	switch (cmd) {
3345	case TUNSETIFF:
3346	case TUNGETIFF:
3347	case TUNSETTXFILTER:
3348	case TUNGETSNDBUF:
3349	case TUNSETSNDBUF:
3350	case SIOCGIFHWADDR:
3351	case SIOCSIFHWADDR:
3352		arg = (unsigned long)compat_ptr(arg);
3353		break;
3354	default:
3355		arg = (compat_ulong_t)arg;
3356		break;
3357	}
3358
3359	/*
3360	 * compat_ifreq is shorter than ifreq, so we must not access beyond
3361	 * the end of that structure. All fields that are used in this
3362	 * driver are compatible though, we don't need to convert the
3363	 * contents.
3364	 */
3365	return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3366}
3367#endif /* CONFIG_COMPAT */
3368
3369static int tun_chr_fasync(int fd, struct file *file, int on)
3370{
3371	struct tun_file *tfile = file->private_data;
3372	int ret;
3373
3374	if (on) {
3375		ret = file_f_owner_allocate(file);
3376		if (ret)
3377			goto out;
3378	}
3379
3380	if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3381		goto out;
3382
3383	if (on) {
3384		__f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3385		tfile->flags |= TUN_FASYNC;
3386	} else
3387		tfile->flags &= ~TUN_FASYNC;
3388	ret = 0;
3389out:
3390	return ret;
3391}
3392
3393static int tun_chr_open(struct inode *inode, struct file * file)
3394{
3395	struct net *net = current->nsproxy->net_ns;
3396	struct tun_file *tfile;
3397
3398	tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3399					    &tun_proto, 0);
3400	if (!tfile)
3401		return -ENOMEM;
3402	if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3403		sk_free(&tfile->sk);
3404		return -ENOMEM;
3405	}
3406
3407	mutex_init(&tfile->napi_mutex);
3408	RCU_INIT_POINTER(tfile->tun, NULL);
3409	tfile->flags = 0;
3410	tfile->ifindex = 0;
3411
3412	init_waitqueue_head(&tfile->socket.wq.wait);
3413
3414	tfile->socket.file = file;
3415	tfile->socket.ops = &tun_socket_ops;
3416
3417	sock_init_data_uid(&tfile->socket, &tfile->sk, current_fsuid());
3418
3419	tfile->sk.sk_write_space = tun_sock_write_space;
3420	tfile->sk.sk_sndbuf = INT_MAX;
3421
3422	file->private_data = tfile;
3423	INIT_LIST_HEAD(&tfile->next);
3424
3425	sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3426
3427	/* tun groks IOCB_NOWAIT just fine, mark it as such */
3428	file->f_mode |= FMODE_NOWAIT;
3429	return 0;
3430}
3431
3432static int tun_chr_close(struct inode *inode, struct file *file)
3433{
3434	struct tun_file *tfile = file->private_data;
3435
3436	tun_detach(tfile, true);
3437
3438	return 0;
3439}
3440
3441#ifdef CONFIG_PROC_FS
3442static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3443{
3444	struct tun_file *tfile = file->private_data;
3445	struct tun_struct *tun;
3446	struct ifreq ifr;
3447
3448	memset(&ifr, 0, sizeof(ifr));
3449
3450	rtnl_lock();
3451	tun = tun_get(tfile);
3452	if (tun)
3453		tun_get_iff(tun, &ifr);
3454	rtnl_unlock();
3455
3456	if (tun)
3457		tun_put(tun);
3458
3459	seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3460}
3461#endif
3462
3463static const struct file_operations tun_fops = {
3464	.owner	= THIS_MODULE,
3465	.read_iter  = tun_chr_read_iter,
3466	.write_iter = tun_chr_write_iter,
3467	.poll	= tun_chr_poll,
3468	.unlocked_ioctl	= tun_chr_ioctl,
3469#ifdef CONFIG_COMPAT
3470	.compat_ioctl = tun_chr_compat_ioctl,
3471#endif
3472	.open	= tun_chr_open,
3473	.release = tun_chr_close,
3474	.fasync = tun_chr_fasync,
3475#ifdef CONFIG_PROC_FS
3476	.show_fdinfo = tun_chr_show_fdinfo,
3477#endif
3478};
3479
3480static struct miscdevice tun_miscdev = {
3481	.minor = TUN_MINOR,
3482	.name = "tun",
3483	.nodename = "net/tun",
3484	.fops = &tun_fops,
3485};
3486
3487/* ethtool interface */
3488
3489static void tun_default_link_ksettings(struct net_device *dev,
3490				       struct ethtool_link_ksettings *cmd)
3491{
3492	ethtool_link_ksettings_zero_link_mode(cmd, supported);
3493	ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3494	cmd->base.speed		= SPEED_10000;
3495	cmd->base.duplex	= DUPLEX_FULL;
3496	cmd->base.port		= PORT_TP;
3497	cmd->base.phy_address	= 0;
3498	cmd->base.autoneg	= AUTONEG_DISABLE;
3499}
3500
3501static int tun_get_link_ksettings(struct net_device *dev,
3502				  struct ethtool_link_ksettings *cmd)
3503{
3504	struct tun_struct *tun = netdev_priv(dev);
3505
3506	memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3507	return 0;
3508}
3509
3510static int tun_set_link_ksettings(struct net_device *dev,
3511				  const struct ethtool_link_ksettings *cmd)
3512{
3513	struct tun_struct *tun = netdev_priv(dev);
3514
3515	memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3516	return 0;
3517}
3518
3519static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3520{
3521	struct tun_struct *tun = netdev_priv(dev);
3522
3523	strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3524	strscpy(info->version, DRV_VERSION, sizeof(info->version));
3525
3526	switch (tun->flags & TUN_TYPE_MASK) {
3527	case IFF_TUN:
3528		strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3529		break;
3530	case IFF_TAP:
3531		strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3532		break;
3533	}
3534}
3535
3536static u32 tun_get_msglevel(struct net_device *dev)
3537{
3538	struct tun_struct *tun = netdev_priv(dev);
3539
3540	return tun->msg_enable;
3541}
3542
3543static void tun_set_msglevel(struct net_device *dev, u32 value)
3544{
3545	struct tun_struct *tun = netdev_priv(dev);
3546
3547	tun->msg_enable = value;
3548}
3549
3550static int tun_get_coalesce(struct net_device *dev,
3551			    struct ethtool_coalesce *ec,
3552			    struct kernel_ethtool_coalesce *kernel_coal,
3553			    struct netlink_ext_ack *extack)
3554{
3555	struct tun_struct *tun = netdev_priv(dev);
3556
3557	ec->rx_max_coalesced_frames = tun->rx_batched;
3558
3559	return 0;
3560}
3561
3562static int tun_set_coalesce(struct net_device *dev,
3563			    struct ethtool_coalesce *ec,
3564			    struct kernel_ethtool_coalesce *kernel_coal,
3565			    struct netlink_ext_ack *extack)
3566{
3567	struct tun_struct *tun = netdev_priv(dev);
3568
3569	if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3570		tun->rx_batched = NAPI_POLL_WEIGHT;
3571	else
3572		tun->rx_batched = ec->rx_max_coalesced_frames;
3573
3574	return 0;
3575}
3576
3577static void tun_get_channels(struct net_device *dev,
3578			     struct ethtool_channels *channels)
3579{
3580	struct tun_struct *tun = netdev_priv(dev);
3581
3582	channels->combined_count = tun->numqueues;
3583	channels->max_combined = tun->flags & IFF_MULTI_QUEUE ? MAX_TAP_QUEUES : 1;
3584}
3585
3586static const struct ethtool_ops tun_ethtool_ops = {
3587	.supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3588	.get_drvinfo	= tun_get_drvinfo,
3589	.get_msglevel	= tun_get_msglevel,
3590	.set_msglevel	= tun_set_msglevel,
3591	.get_link	= ethtool_op_get_link,
3592	.get_channels   = tun_get_channels,
3593	.get_ts_info	= ethtool_op_get_ts_info,
3594	.get_coalesce   = tun_get_coalesce,
3595	.set_coalesce   = tun_set_coalesce,
3596	.get_link_ksettings = tun_get_link_ksettings,
3597	.set_link_ksettings = tun_set_link_ksettings,
3598};
3599
3600static int tun_queue_resize(struct tun_struct *tun)
3601{
3602	struct net_device *dev = tun->dev;
3603	struct tun_file *tfile;
3604	struct ptr_ring **rings;
3605	int n = tun->numqueues + tun->numdisabled;
3606	int ret, i;
3607
3608	rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3609	if (!rings)
3610		return -ENOMEM;
3611
3612	for (i = 0; i < tun->numqueues; i++) {
3613		tfile = rtnl_dereference(tun->tfiles[i]);
3614		rings[i] = &tfile->tx_ring;
3615	}
3616	list_for_each_entry(tfile, &tun->disabled, next)
3617		rings[i++] = &tfile->tx_ring;
3618
3619	ret = ptr_ring_resize_multiple_bh(rings, n,
3620					  dev->tx_queue_len, GFP_KERNEL,
3621					  tun_ptr_free);
3622
3623	kfree(rings);
3624	return ret;
3625}
3626
3627static int tun_device_event(struct notifier_block *unused,
3628			    unsigned long event, void *ptr)
3629{
3630	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3631	struct tun_struct *tun = netdev_priv(dev);
3632	int i;
3633
3634	if (dev->rtnl_link_ops != &tun_link_ops)
3635		return NOTIFY_DONE;
3636
3637	switch (event) {
3638	case NETDEV_CHANGE_TX_QUEUE_LEN:
3639		if (tun_queue_resize(tun))
3640			return NOTIFY_BAD;
3641		break;
3642	case NETDEV_UP:
3643		for (i = 0; i < tun->numqueues; i++) {
3644			struct tun_file *tfile;
3645
3646			tfile = rtnl_dereference(tun->tfiles[i]);
3647			tfile->socket.sk->sk_write_space(tfile->socket.sk);
3648		}
3649		break;
3650	default:
3651		break;
3652	}
3653
3654	return NOTIFY_DONE;
3655}
3656
3657static struct notifier_block tun_notifier_block __read_mostly = {
3658	.notifier_call	= tun_device_event,
3659};
3660
3661static int __init tun_init(void)
3662{
3663	int ret = 0;
3664
3665	pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3666
3667	ret = rtnl_link_register(&tun_link_ops);
3668	if (ret) {
3669		pr_err("Can't register link_ops\n");
3670		goto err_linkops;
3671	}
3672
3673	ret = misc_register(&tun_miscdev);
3674	if (ret) {
3675		pr_err("Can't register misc device %d\n", TUN_MINOR);
3676		goto err_misc;
3677	}
3678
3679	ret = register_netdevice_notifier(&tun_notifier_block);
3680	if (ret) {
3681		pr_err("Can't register netdevice notifier\n");
3682		goto err_notifier;
3683	}
3684
3685	return  0;
3686
3687err_notifier:
3688	misc_deregister(&tun_miscdev);
3689err_misc:
3690	rtnl_link_unregister(&tun_link_ops);
3691err_linkops:
3692	return ret;
3693}
3694
3695static void __exit tun_cleanup(void)
3696{
3697	misc_deregister(&tun_miscdev);
3698	rtnl_link_unregister(&tun_link_ops);
3699	unregister_netdevice_notifier(&tun_notifier_block);
3700}
3701
3702/* Get an underlying socket object from tun file.  Returns error unless file is
3703 * attached to a device.  The returned object works like a packet socket, it
3704 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
3705 * holding a reference to the file for as long as the socket is in use. */
3706struct socket *tun_get_socket(struct file *file)
3707{
3708	struct tun_file *tfile;
3709	if (file->f_op != &tun_fops)
3710		return ERR_PTR(-EINVAL);
3711	tfile = file->private_data;
3712	if (!tfile)
3713		return ERR_PTR(-EBADFD);
3714	return &tfile->socket;
3715}
3716EXPORT_SYMBOL_GPL(tun_get_socket);
3717
3718struct ptr_ring *tun_get_tx_ring(struct file *file)
3719{
3720	struct tun_file *tfile;
3721
3722	if (file->f_op != &tun_fops)
3723		return ERR_PTR(-EINVAL);
3724	tfile = file->private_data;
3725	if (!tfile)
3726		return ERR_PTR(-EBADFD);
3727	return &tfile->tx_ring;
3728}
3729EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3730
3731module_init(tun_init);
3732module_exit(tun_cleanup);
3733MODULE_DESCRIPTION(DRV_DESCRIPTION);
3734MODULE_AUTHOR(DRV_COPYRIGHT);
3735MODULE_LICENSE("GPL");
3736MODULE_ALIAS_MISCDEV(TUN_MINOR);
3737MODULE_ALIAS("devname:net/tun");
3738MODULE_IMPORT_NS("NETDEV_INTERNAL");