Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v4.6
   1/*
   2 *	Generic address resolution entity
   3 *
   4 *	Authors:
   5 *	Pedro Roque		<roque@di.fc.ul.pt>
   6 *	Alexey Kuznetsov	<kuznet@ms2.inr.ac.ru>
   7 *
   8 *	This program is free software; you can redistribute it and/or
   9 *      modify it under the terms of the GNU General Public License
  10 *      as published by the Free Software Foundation; either version
  11 *      2 of the License, or (at your option) any later version.
  12 *
  13 *	Fixes:
  14 *	Vitaly E. Lavrov	releasing NULL neighbor in neigh_add.
  15 *	Harald Welte		Add neighbour cache statistics like rtstat
  16 */
  17
  18#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20#include <linux/slab.h>
  21#include <linux/types.h>
  22#include <linux/kernel.h>
  23#include <linux/module.h>
  24#include <linux/socket.h>
  25#include <linux/netdevice.h>
  26#include <linux/proc_fs.h>
  27#ifdef CONFIG_SYSCTL
  28#include <linux/sysctl.h>
  29#endif
  30#include <linux/times.h>
  31#include <net/net_namespace.h>
  32#include <net/neighbour.h>
  33#include <net/dst.h>
  34#include <net/sock.h>
  35#include <net/netevent.h>
  36#include <net/netlink.h>
  37#include <linux/rtnetlink.h>
  38#include <linux/random.h>
  39#include <linux/string.h>
  40#include <linux/log2.h>
  41#include <linux/inetdevice.h>
  42#include <net/addrconf.h>
  43
  44#define DEBUG
  45#define NEIGH_DEBUG 1
  46#define neigh_dbg(level, fmt, ...)		\
  47do {						\
  48	if (level <= NEIGH_DEBUG)		\
  49		pr_debug(fmt, ##__VA_ARGS__);	\
  50} while (0)
  51
  52#define PNEIGH_HASHMASK		0xF
  53
  54static void neigh_timer_handler(unsigned long arg);
  55static void __neigh_notify(struct neighbour *n, int type, int flags);
  56static void neigh_update_notify(struct neighbour *neigh);
  57static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
 
 
  58
  59#ifdef CONFIG_PROC_FS
  60static const struct file_operations neigh_stat_seq_fops;
  61#endif
  62
  63/*
  64   Neighbour hash table buckets are protected with rwlock tbl->lock.
  65
  66   - All the scans/updates to hash buckets MUST be made under this lock.
  67   - NOTHING clever should be made under this lock: no callbacks
  68     to protocol backends, no attempts to send something to network.
  69     It will result in deadlocks, if backend/driver wants to use neighbour
  70     cache.
  71   - If the entry requires some non-trivial actions, increase
  72     its reference count and release table lock.
  73
  74   Neighbour entries are protected:
  75   - with reference count.
  76   - with rwlock neigh->lock
  77
  78   Reference count prevents destruction.
  79
  80   neigh->lock mainly serializes ll address data and its validity state.
  81   However, the same lock is used to protect another entry fields:
  82    - timer
  83    - resolution queue
  84
  85   Again, nothing clever shall be made under neigh->lock,
  86   the most complicated procedure, which we allow is dev->hard_header.
  87   It is supposed, that dev->hard_header is simplistic and does
  88   not make callbacks to neighbour tables.
  89 */
  90
  91static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
  92{
  93	kfree_skb(skb);
  94	return -ENETDOWN;
  95}
  96
  97static void neigh_cleanup_and_release(struct neighbour *neigh)
  98{
  99	if (neigh->parms->neigh_cleanup)
 100		neigh->parms->neigh_cleanup(neigh);
 101
 102	__neigh_notify(neigh, RTM_DELNEIGH, 0);
 
 103	neigh_release(neigh);
 104}
 105
 106/*
 107 * It is random distribution in the interval (1/2)*base...(3/2)*base.
 108 * It corresponds to default IPv6 settings and is not overridable,
 109 * because it is really reasonable choice.
 110 */
 111
 112unsigned long neigh_rand_reach_time(unsigned long base)
 113{
 114	return base ? (prandom_u32() % base) + (base >> 1) : 0;
 115}
 116EXPORT_SYMBOL(neigh_rand_reach_time);
 117
 118
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 119static int neigh_forced_gc(struct neigh_table *tbl)
 120{
 121	int shrunk = 0;
 122	int i;
 123	struct neigh_hash_table *nht;
 124
 125	NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 126
 127	write_lock_bh(&tbl->lock);
 128	nht = rcu_dereference_protected(tbl->nht,
 129					lockdep_is_held(&tbl->lock));
 130	for (i = 0; i < (1 << nht->hash_shift); i++) {
 131		struct neighbour *n;
 132		struct neighbour __rcu **np;
 133
 134		np = &nht->hash_buckets[i];
 135		while ((n = rcu_dereference_protected(*np,
 136					lockdep_is_held(&tbl->lock))) != NULL) {
 137			/* Neighbour record may be discarded if:
 138			 * - nobody refers to it.
 139			 * - it is not permanent
 140			 */
 141			write_lock(&n->lock);
 142			if (atomic_read(&n->refcnt) == 1 &&
 143			    !(n->nud_state & NUD_PERMANENT)) {
 144				rcu_assign_pointer(*np,
 145					rcu_dereference_protected(n->next,
 146						  lockdep_is_held(&tbl->lock)));
 147				n->dead = 1;
 148				shrunk	= 1;
 149				write_unlock(&n->lock);
 150				neigh_cleanup_and_release(n);
 151				continue;
 152			}
 153			write_unlock(&n->lock);
 154			np = &n->next;
 155		}
 156	}
 157
 158	tbl->last_flush = jiffies;
 159
 160	write_unlock_bh(&tbl->lock);
 161
 162	return shrunk;
 163}
 164
 165static void neigh_add_timer(struct neighbour *n, unsigned long when)
 166{
 167	neigh_hold(n);
 168	if (unlikely(mod_timer(&n->timer, when))) {
 169		printk("NEIGH: BUG, double timer add, state is %x\n",
 170		       n->nud_state);
 171		dump_stack();
 172	}
 173}
 174
 175static int neigh_del_timer(struct neighbour *n)
 176{
 177	if ((n->nud_state & NUD_IN_TIMER) &&
 178	    del_timer(&n->timer)) {
 179		neigh_release(n);
 180		return 1;
 181	}
 182	return 0;
 183}
 184
 185static void pneigh_queue_purge(struct sk_buff_head *list)
 186{
 187	struct sk_buff *skb;
 188
 189	while ((skb = skb_dequeue(list)) != NULL) {
 190		dev_put(skb->dev);
 191		kfree_skb(skb);
 192	}
 193}
 194
 195static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 196{
 197	int i;
 198	struct neigh_hash_table *nht;
 199
 200	nht = rcu_dereference_protected(tbl->nht,
 201					lockdep_is_held(&tbl->lock));
 202
 203	for (i = 0; i < (1 << nht->hash_shift); i++) {
 204		struct neighbour *n;
 205		struct neighbour __rcu **np = &nht->hash_buckets[i];
 206
 207		while ((n = rcu_dereference_protected(*np,
 208					lockdep_is_held(&tbl->lock))) != NULL) {
 209			if (dev && n->dev != dev) {
 210				np = &n->next;
 211				continue;
 212			}
 213			rcu_assign_pointer(*np,
 214				   rcu_dereference_protected(n->next,
 215						lockdep_is_held(&tbl->lock)));
 216			write_lock(&n->lock);
 217			neigh_del_timer(n);
 218			n->dead = 1;
 219
 220			if (atomic_read(&n->refcnt) != 1) {
 221				/* The most unpleasant situation.
 222				   We must destroy neighbour entry,
 223				   but someone still uses it.
 224
 225				   The destroy will be delayed until
 226				   the last user releases us, but
 227				   we must kill timers etc. and move
 228				   it to safe state.
 229				 */
 230				__skb_queue_purge(&n->arp_queue);
 231				n->arp_queue_len_bytes = 0;
 232				n->output = neigh_blackhole;
 233				if (n->nud_state & NUD_VALID)
 234					n->nud_state = NUD_NOARP;
 235				else
 236					n->nud_state = NUD_NONE;
 237				neigh_dbg(2, "neigh %p is stray\n", n);
 238			}
 239			write_unlock(&n->lock);
 240			neigh_cleanup_and_release(n);
 241		}
 242	}
 243}
 244
 245void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 246{
 247	write_lock_bh(&tbl->lock);
 248	neigh_flush_dev(tbl, dev);
 249	write_unlock_bh(&tbl->lock);
 250}
 251EXPORT_SYMBOL(neigh_changeaddr);
 252
 253int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 254{
 255	write_lock_bh(&tbl->lock);
 256	neigh_flush_dev(tbl, dev);
 257	pneigh_ifdown(tbl, dev);
 258	write_unlock_bh(&tbl->lock);
 259
 260	del_timer_sync(&tbl->proxy_timer);
 261	pneigh_queue_purge(&tbl->proxy_queue);
 262	return 0;
 263}
 264EXPORT_SYMBOL(neigh_ifdown);
 265
 266static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
 267{
 268	struct neighbour *n = NULL;
 269	unsigned long now = jiffies;
 270	int entries;
 271
 272	entries = atomic_inc_return(&tbl->entries) - 1;
 273	if (entries >= tbl->gc_thresh3 ||
 274	    (entries >= tbl->gc_thresh2 &&
 275	     time_after(now, tbl->last_flush + 5 * HZ))) {
 276		if (!neigh_forced_gc(tbl) &&
 277		    entries >= tbl->gc_thresh3) {
 278			net_info_ratelimited("%s: neighbor table overflow!\n",
 279					     tbl->id);
 280			NEIGH_CACHE_STAT_INC(tbl, table_fulls);
 281			goto out_entries;
 282		}
 283	}
 284
 285	n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
 286	if (!n)
 287		goto out_entries;
 288
 289	__skb_queue_head_init(&n->arp_queue);
 290	rwlock_init(&n->lock);
 291	seqlock_init(&n->ha_lock);
 292	n->updated	  = n->used = now;
 293	n->nud_state	  = NUD_NONE;
 294	n->output	  = neigh_blackhole;
 295	seqlock_init(&n->hh.hh_lock);
 296	n->parms	  = neigh_parms_clone(&tbl->parms);
 297	setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
 298
 299	NEIGH_CACHE_STAT_INC(tbl, allocs);
 300	n->tbl		  = tbl;
 301	atomic_set(&n->refcnt, 1);
 302	n->dead		  = 1;
 303out:
 304	return n;
 305
 306out_entries:
 307	atomic_dec(&tbl->entries);
 308	goto out;
 309}
 310
 311static void neigh_get_hash_rnd(u32 *x)
 312{
 313	get_random_bytes(x, sizeof(*x));
 314	*x |= 1;
 315}
 316
 317static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 318{
 319	size_t size = (1 << shift) * sizeof(struct neighbour *);
 320	struct neigh_hash_table *ret;
 321	struct neighbour __rcu **buckets;
 322	int i;
 323
 324	ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 325	if (!ret)
 326		return NULL;
 327	if (size <= PAGE_SIZE)
 328		buckets = kzalloc(size, GFP_ATOMIC);
 329	else
 330		buckets = (struct neighbour __rcu **)
 331			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 332					   get_order(size));
 333	if (!buckets) {
 334		kfree(ret);
 335		return NULL;
 336	}
 337	ret->hash_buckets = buckets;
 338	ret->hash_shift = shift;
 339	for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 340		neigh_get_hash_rnd(&ret->hash_rnd[i]);
 341	return ret;
 342}
 343
 344static void neigh_hash_free_rcu(struct rcu_head *head)
 345{
 346	struct neigh_hash_table *nht = container_of(head,
 347						    struct neigh_hash_table,
 348						    rcu);
 349	size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 350	struct neighbour __rcu **buckets = nht->hash_buckets;
 351
 352	if (size <= PAGE_SIZE)
 353		kfree(buckets);
 354	else
 355		free_pages((unsigned long)buckets, get_order(size));
 356	kfree(nht);
 357}
 358
 359static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 360						unsigned long new_shift)
 361{
 362	unsigned int i, hash;
 363	struct neigh_hash_table *new_nht, *old_nht;
 364
 365	NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 366
 367	old_nht = rcu_dereference_protected(tbl->nht,
 368					    lockdep_is_held(&tbl->lock));
 369	new_nht = neigh_hash_alloc(new_shift);
 370	if (!new_nht)
 371		return old_nht;
 372
 373	for (i = 0; i < (1 << old_nht->hash_shift); i++) {
 374		struct neighbour *n, *next;
 375
 376		for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
 377						   lockdep_is_held(&tbl->lock));
 378		     n != NULL;
 379		     n = next) {
 380			hash = tbl->hash(n->primary_key, n->dev,
 381					 new_nht->hash_rnd);
 382
 383			hash >>= (32 - new_nht->hash_shift);
 384			next = rcu_dereference_protected(n->next,
 385						lockdep_is_held(&tbl->lock));
 386
 387			rcu_assign_pointer(n->next,
 388					   rcu_dereference_protected(
 389						new_nht->hash_buckets[hash],
 390						lockdep_is_held(&tbl->lock)));
 391			rcu_assign_pointer(new_nht->hash_buckets[hash], n);
 392		}
 393	}
 394
 395	rcu_assign_pointer(tbl->nht, new_nht);
 396	call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
 397	return new_nht;
 398}
 399
 400struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 401			       struct net_device *dev)
 402{
 403	struct neighbour *n;
 404
 405	NEIGH_CACHE_STAT_INC(tbl, lookups);
 406
 407	rcu_read_lock_bh();
 408	n = __neigh_lookup_noref(tbl, pkey, dev);
 409	if (n) {
 410		if (!atomic_inc_not_zero(&n->refcnt))
 411			n = NULL;
 412		NEIGH_CACHE_STAT_INC(tbl, hits);
 413	}
 414
 415	rcu_read_unlock_bh();
 416	return n;
 417}
 418EXPORT_SYMBOL(neigh_lookup);
 419
 420struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 421				     const void *pkey)
 422{
 423	struct neighbour *n;
 424	int key_len = tbl->key_len;
 425	u32 hash_val;
 426	struct neigh_hash_table *nht;
 427
 428	NEIGH_CACHE_STAT_INC(tbl, lookups);
 429
 430	rcu_read_lock_bh();
 431	nht = rcu_dereference_bh(tbl->nht);
 432	hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
 433
 434	for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
 435	     n != NULL;
 436	     n = rcu_dereference_bh(n->next)) {
 437		if (!memcmp(n->primary_key, pkey, key_len) &&
 438		    net_eq(dev_net(n->dev), net)) {
 439			if (!atomic_inc_not_zero(&n->refcnt))
 440				n = NULL;
 441			NEIGH_CACHE_STAT_INC(tbl, hits);
 442			break;
 443		}
 444	}
 445
 446	rcu_read_unlock_bh();
 447	return n;
 448}
 449EXPORT_SYMBOL(neigh_lookup_nodev);
 450
 451struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
 452				 struct net_device *dev, bool want_ref)
 453{
 454	u32 hash_val;
 455	int key_len = tbl->key_len;
 456	int error;
 457	struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
 458	struct neigh_hash_table *nht;
 459
 460	if (!n) {
 461		rc = ERR_PTR(-ENOBUFS);
 462		goto out;
 463	}
 464
 465	memcpy(n->primary_key, pkey, key_len);
 466	n->dev = dev;
 467	dev_hold(dev);
 468
 469	/* Protocol specific setup. */
 470	if (tbl->constructor &&	(error = tbl->constructor(n)) < 0) {
 471		rc = ERR_PTR(error);
 472		goto out_neigh_release;
 473	}
 474
 475	if (dev->netdev_ops->ndo_neigh_construct) {
 476		error = dev->netdev_ops->ndo_neigh_construct(n);
 477		if (error < 0) {
 478			rc = ERR_PTR(error);
 479			goto out_neigh_release;
 480		}
 481	}
 482
 483	/* Device specific setup. */
 484	if (n->parms->neigh_setup &&
 485	    (error = n->parms->neigh_setup(n)) < 0) {
 486		rc = ERR_PTR(error);
 487		goto out_neigh_release;
 488	}
 489
 490	n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
 491
 492	write_lock_bh(&tbl->lock);
 493	nht = rcu_dereference_protected(tbl->nht,
 494					lockdep_is_held(&tbl->lock));
 495
 496	if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
 497		nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
 498
 499	hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 500
 501	if (n->parms->dead) {
 502		rc = ERR_PTR(-EINVAL);
 503		goto out_tbl_unlock;
 504	}
 505
 506	for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
 507					    lockdep_is_held(&tbl->lock));
 508	     n1 != NULL;
 509	     n1 = rcu_dereference_protected(n1->next,
 510			lockdep_is_held(&tbl->lock))) {
 511		if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
 512			if (want_ref)
 513				neigh_hold(n1);
 514			rc = n1;
 515			goto out_tbl_unlock;
 516		}
 517	}
 518
 519	n->dead = 0;
 520	if (want_ref)
 521		neigh_hold(n);
 522	rcu_assign_pointer(n->next,
 523			   rcu_dereference_protected(nht->hash_buckets[hash_val],
 524						     lockdep_is_held(&tbl->lock)));
 525	rcu_assign_pointer(nht->hash_buckets[hash_val], n);
 526	write_unlock_bh(&tbl->lock);
 527	neigh_dbg(2, "neigh %p is created\n", n);
 528	rc = n;
 529out:
 530	return rc;
 531out_tbl_unlock:
 532	write_unlock_bh(&tbl->lock);
 533out_neigh_release:
 534	neigh_release(n);
 535	goto out;
 536}
 537EXPORT_SYMBOL(__neigh_create);
 538
 539static u32 pneigh_hash(const void *pkey, int key_len)
 540{
 541	u32 hash_val = *(u32 *)(pkey + key_len - 4);
 542	hash_val ^= (hash_val >> 16);
 543	hash_val ^= hash_val >> 8;
 544	hash_val ^= hash_val >> 4;
 545	hash_val &= PNEIGH_HASHMASK;
 546	return hash_val;
 547}
 548
 549static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 550					      struct net *net,
 551					      const void *pkey,
 552					      int key_len,
 553					      struct net_device *dev)
 554{
 555	while (n) {
 556		if (!memcmp(n->key, pkey, key_len) &&
 557		    net_eq(pneigh_net(n), net) &&
 558		    (n->dev == dev || !n->dev))
 559			return n;
 560		n = n->next;
 561	}
 562	return NULL;
 563}
 564
 565struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 566		struct net *net, const void *pkey, struct net_device *dev)
 567{
 568	int key_len = tbl->key_len;
 569	u32 hash_val = pneigh_hash(pkey, key_len);
 570
 571	return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 572				 net, pkey, key_len, dev);
 573}
 574EXPORT_SYMBOL_GPL(__pneigh_lookup);
 575
 576struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 577				    struct net *net, const void *pkey,
 578				    struct net_device *dev, int creat)
 579{
 580	struct pneigh_entry *n;
 581	int key_len = tbl->key_len;
 582	u32 hash_val = pneigh_hash(pkey, key_len);
 583
 584	read_lock_bh(&tbl->lock);
 585	n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 586			      net, pkey, key_len, dev);
 587	read_unlock_bh(&tbl->lock);
 588
 589	if (n || !creat)
 590		goto out;
 591
 592	ASSERT_RTNL();
 593
 594	n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 595	if (!n)
 596		goto out;
 597
 598	write_pnet(&n->net, net);
 599	memcpy(n->key, pkey, key_len);
 600	n->dev = dev;
 601	if (dev)
 602		dev_hold(dev);
 603
 604	if (tbl->pconstructor && tbl->pconstructor(n)) {
 605		if (dev)
 606			dev_put(dev);
 607		kfree(n);
 608		n = NULL;
 609		goto out;
 610	}
 611
 612	write_lock_bh(&tbl->lock);
 613	n->next = tbl->phash_buckets[hash_val];
 614	tbl->phash_buckets[hash_val] = n;
 615	write_unlock_bh(&tbl->lock);
 616out:
 617	return n;
 618}
 619EXPORT_SYMBOL(pneigh_lookup);
 620
 621
 622int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 623		  struct net_device *dev)
 624{
 625	struct pneigh_entry *n, **np;
 626	int key_len = tbl->key_len;
 627	u32 hash_val = pneigh_hash(pkey, key_len);
 628
 629	write_lock_bh(&tbl->lock);
 630	for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 631	     np = &n->next) {
 632		if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 633		    net_eq(pneigh_net(n), net)) {
 634			*np = n->next;
 635			write_unlock_bh(&tbl->lock);
 636			if (tbl->pdestructor)
 637				tbl->pdestructor(n);
 638			if (n->dev)
 639				dev_put(n->dev);
 640			kfree(n);
 641			return 0;
 642		}
 643	}
 644	write_unlock_bh(&tbl->lock);
 645	return -ENOENT;
 646}
 647
 648static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 
 649{
 650	struct pneigh_entry *n, **np;
 651	u32 h;
 652
 653	for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 654		np = &tbl->phash_buckets[h];
 655		while ((n = *np) != NULL) {
 656			if (!dev || n->dev == dev) {
 657				*np = n->next;
 658				if (tbl->pdestructor)
 659					tbl->pdestructor(n);
 660				if (n->dev)
 661					dev_put(n->dev);
 662				kfree(n);
 663				continue;
 664			}
 665			np = &n->next;
 666		}
 667	}
 
 
 
 
 
 
 
 
 
 
 668	return -ENOENT;
 669}
 670
 671static void neigh_parms_destroy(struct neigh_parms *parms);
 672
 673static inline void neigh_parms_put(struct neigh_parms *parms)
 674{
 675	if (atomic_dec_and_test(&parms->refcnt))
 676		neigh_parms_destroy(parms);
 677}
 678
 679/*
 680 *	neighbour must already be out of the table;
 681 *
 682 */
 683void neigh_destroy(struct neighbour *neigh)
 684{
 685	struct net_device *dev = neigh->dev;
 686
 687	NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 688
 689	if (!neigh->dead) {
 690		pr_warn("Destroying alive neighbour %p\n", neigh);
 691		dump_stack();
 692		return;
 693	}
 694
 695	if (neigh_del_timer(neigh))
 696		pr_warn("Impossible event\n");
 697
 698	write_lock_bh(&neigh->lock);
 699	__skb_queue_purge(&neigh->arp_queue);
 700	write_unlock_bh(&neigh->lock);
 701	neigh->arp_queue_len_bytes = 0;
 702
 703	if (dev->netdev_ops->ndo_neigh_destroy)
 704		dev->netdev_ops->ndo_neigh_destroy(neigh);
 705
 706	dev_put(dev);
 707	neigh_parms_put(neigh->parms);
 708
 709	neigh_dbg(2, "neigh %p is destroyed\n", neigh);
 710
 711	atomic_dec(&neigh->tbl->entries);
 712	kfree_rcu(neigh, rcu);
 713}
 714EXPORT_SYMBOL(neigh_destroy);
 715
 716/* Neighbour state is suspicious;
 717   disable fast path.
 718
 719   Called with write_locked neigh.
 720 */
 721static void neigh_suspect(struct neighbour *neigh)
 722{
 723	neigh_dbg(2, "neigh %p is suspected\n", neigh);
 724
 725	neigh->output = neigh->ops->output;
 726}
 727
 728/* Neighbour state is OK;
 729   enable fast path.
 730
 731   Called with write_locked neigh.
 732 */
 733static void neigh_connect(struct neighbour *neigh)
 734{
 735	neigh_dbg(2, "neigh %p is connected\n", neigh);
 736
 737	neigh->output = neigh->ops->connected_output;
 738}
 739
 740static void neigh_periodic_work(struct work_struct *work)
 741{
 742	struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
 743	struct neighbour *n;
 744	struct neighbour __rcu **np;
 745	unsigned int i;
 746	struct neigh_hash_table *nht;
 747
 748	NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 749
 750	write_lock_bh(&tbl->lock);
 751	nht = rcu_dereference_protected(tbl->nht,
 752					lockdep_is_held(&tbl->lock));
 753
 754	/*
 755	 *	periodically recompute ReachableTime from random function
 756	 */
 757
 758	if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
 759		struct neigh_parms *p;
 760		tbl->last_rand = jiffies;
 761		list_for_each_entry(p, &tbl->parms_list, list)
 762			p->reachable_time =
 763				neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
 764	}
 765
 766	if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
 767		goto out;
 768
 769	for (i = 0 ; i < (1 << nht->hash_shift); i++) {
 770		np = &nht->hash_buckets[i];
 771
 772		while ((n = rcu_dereference_protected(*np,
 773				lockdep_is_held(&tbl->lock))) != NULL) {
 774			unsigned int state;
 775
 776			write_lock(&n->lock);
 777
 778			state = n->nud_state;
 779			if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 780				write_unlock(&n->lock);
 781				goto next_elt;
 782			}
 783
 784			if (time_before(n->used, n->confirmed))
 785				n->used = n->confirmed;
 786
 787			if (atomic_read(&n->refcnt) == 1 &&
 788			    (state == NUD_FAILED ||
 789			     time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
 790				*np = n->next;
 791				n->dead = 1;
 792				write_unlock(&n->lock);
 793				neigh_cleanup_and_release(n);
 794				continue;
 795			}
 796			write_unlock(&n->lock);
 797
 798next_elt:
 799			np = &n->next;
 800		}
 801		/*
 802		 * It's fine to release lock here, even if hash table
 803		 * grows while we are preempted.
 804		 */
 805		write_unlock_bh(&tbl->lock);
 806		cond_resched();
 807		write_lock_bh(&tbl->lock);
 808		nht = rcu_dereference_protected(tbl->nht,
 809						lockdep_is_held(&tbl->lock));
 810	}
 811out:
 812	/* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
 813	 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
 814	 * BASE_REACHABLE_TIME.
 815	 */
 816	queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
 817			      NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
 818	write_unlock_bh(&tbl->lock);
 819}
 820
 821static __inline__ int neigh_max_probes(struct neighbour *n)
 822{
 823	struct neigh_parms *p = n->parms;
 824	return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
 825	       (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
 826	        NEIGH_VAR(p, MCAST_PROBES));
 827}
 828
 829static void neigh_invalidate(struct neighbour *neigh)
 830	__releases(neigh->lock)
 831	__acquires(neigh->lock)
 832{
 833	struct sk_buff *skb;
 834
 835	NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 836	neigh_dbg(2, "neigh %p is failed\n", neigh);
 837	neigh->updated = jiffies;
 838
 839	/* It is very thin place. report_unreachable is very complicated
 840	   routine. Particularly, it can hit the same neighbour entry!
 841
 842	   So that, we try to be accurate and avoid dead loop. --ANK
 843	 */
 844	while (neigh->nud_state == NUD_FAILED &&
 845	       (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 846		write_unlock(&neigh->lock);
 847		neigh->ops->error_report(neigh, skb);
 848		write_lock(&neigh->lock);
 849	}
 850	__skb_queue_purge(&neigh->arp_queue);
 851	neigh->arp_queue_len_bytes = 0;
 852}
 853
 854static void neigh_probe(struct neighbour *neigh)
 855	__releases(neigh->lock)
 856{
 857	struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
 858	/* keep skb alive even if arp_queue overflows */
 859	if (skb)
 860		skb = skb_clone(skb, GFP_ATOMIC);
 861	write_unlock(&neigh->lock);
 862	neigh->ops->solicit(neigh, skb);
 
 863	atomic_inc(&neigh->probes);
 864	kfree_skb(skb);
 865}
 866
 867/* Called when a timer expires for a neighbour entry. */
 868
 869static void neigh_timer_handler(unsigned long arg)
 870{
 871	unsigned long now, next;
 872	struct neighbour *neigh = (struct neighbour *)arg;
 873	unsigned int state;
 874	int notify = 0;
 875
 876	write_lock(&neigh->lock);
 877
 878	state = neigh->nud_state;
 879	now = jiffies;
 880	next = now + HZ;
 881
 882	if (!(state & NUD_IN_TIMER))
 883		goto out;
 884
 885	if (state & NUD_REACHABLE) {
 886		if (time_before_eq(now,
 887				   neigh->confirmed + neigh->parms->reachable_time)) {
 888			neigh_dbg(2, "neigh %p is still alive\n", neigh);
 889			next = neigh->confirmed + neigh->parms->reachable_time;
 890		} else if (time_before_eq(now,
 891					  neigh->used +
 892					  NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
 893			neigh_dbg(2, "neigh %p is delayed\n", neigh);
 894			neigh->nud_state = NUD_DELAY;
 895			neigh->updated = jiffies;
 896			neigh_suspect(neigh);
 897			next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
 898		} else {
 899			neigh_dbg(2, "neigh %p is suspected\n", neigh);
 900			neigh->nud_state = NUD_STALE;
 901			neigh->updated = jiffies;
 902			neigh_suspect(neigh);
 903			notify = 1;
 904		}
 905	} else if (state & NUD_DELAY) {
 906		if (time_before_eq(now,
 907				   neigh->confirmed +
 908				   NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
 909			neigh_dbg(2, "neigh %p is now reachable\n", neigh);
 910			neigh->nud_state = NUD_REACHABLE;
 911			neigh->updated = jiffies;
 912			neigh_connect(neigh);
 913			notify = 1;
 914			next = neigh->confirmed + neigh->parms->reachable_time;
 915		} else {
 916			neigh_dbg(2, "neigh %p is probed\n", neigh);
 917			neigh->nud_state = NUD_PROBE;
 918			neigh->updated = jiffies;
 919			atomic_set(&neigh->probes, 0);
 920			notify = 1;
 921			next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
 922		}
 923	} else {
 924		/* NUD_PROBE|NUD_INCOMPLETE */
 925		next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
 926	}
 927
 928	if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 929	    atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 930		neigh->nud_state = NUD_FAILED;
 931		notify = 1;
 932		neigh_invalidate(neigh);
 933		goto out;
 934	}
 935
 936	if (neigh->nud_state & NUD_IN_TIMER) {
 937		if (time_before(next, jiffies + HZ/2))
 938			next = jiffies + HZ/2;
 939		if (!mod_timer(&neigh->timer, next))
 940			neigh_hold(neigh);
 941	}
 942	if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 943		neigh_probe(neigh);
 944	} else {
 945out:
 946		write_unlock(&neigh->lock);
 947	}
 948
 949	if (notify)
 950		neigh_update_notify(neigh);
 951
 952	neigh_release(neigh);
 953}
 954
 955int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
 956{
 957	int rc;
 958	bool immediate_probe = false;
 959
 960	write_lock_bh(&neigh->lock);
 961
 962	rc = 0;
 963	if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
 964		goto out_unlock_bh;
 965	if (neigh->dead)
 966		goto out_dead;
 967
 968	if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
 969		if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
 970		    NEIGH_VAR(neigh->parms, APP_PROBES)) {
 971			unsigned long next, now = jiffies;
 972
 973			atomic_set(&neigh->probes,
 974				   NEIGH_VAR(neigh->parms, UCAST_PROBES));
 975			neigh->nud_state     = NUD_INCOMPLETE;
 976			neigh->updated = now;
 977			next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
 978					 HZ/2);
 979			neigh_add_timer(neigh, next);
 980			immediate_probe = true;
 981		} else {
 982			neigh->nud_state = NUD_FAILED;
 983			neigh->updated = jiffies;
 984			write_unlock_bh(&neigh->lock);
 985
 986			kfree_skb(skb);
 987			return 1;
 988		}
 989	} else if (neigh->nud_state & NUD_STALE) {
 990		neigh_dbg(2, "neigh %p is delayed\n", neigh);
 991		neigh->nud_state = NUD_DELAY;
 992		neigh->updated = jiffies;
 993		neigh_add_timer(neigh, jiffies +
 994				NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
 995	}
 996
 997	if (neigh->nud_state == NUD_INCOMPLETE) {
 998		if (skb) {
 999			while (neigh->arp_queue_len_bytes + skb->truesize >
1000			       NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1001				struct sk_buff *buff;
1002
1003				buff = __skb_dequeue(&neigh->arp_queue);
1004				if (!buff)
1005					break;
1006				neigh->arp_queue_len_bytes -= buff->truesize;
1007				kfree_skb(buff);
1008				NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1009			}
1010			skb_dst_force(skb);
1011			__skb_queue_tail(&neigh->arp_queue, skb);
1012			neigh->arp_queue_len_bytes += skb->truesize;
1013		}
1014		rc = 1;
1015	}
1016out_unlock_bh:
1017	if (immediate_probe)
1018		neigh_probe(neigh);
1019	else
1020		write_unlock(&neigh->lock);
1021	local_bh_enable();
1022	return rc;
1023
1024out_dead:
1025	if (neigh->nud_state & NUD_STALE)
1026		goto out_unlock_bh;
1027	write_unlock_bh(&neigh->lock);
1028	kfree_skb(skb);
1029	return 1;
1030}
1031EXPORT_SYMBOL(__neigh_event_send);
1032
1033static void neigh_update_hhs(struct neighbour *neigh)
1034{
1035	struct hh_cache *hh;
1036	void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1037		= NULL;
1038
1039	if (neigh->dev->header_ops)
1040		update = neigh->dev->header_ops->cache_update;
1041
1042	if (update) {
1043		hh = &neigh->hh;
1044		if (hh->hh_len) {
1045			write_seqlock_bh(&hh->hh_lock);
1046			update(hh, neigh->dev, neigh->ha);
1047			write_sequnlock_bh(&hh->hh_lock);
1048		}
1049	}
1050}
1051
1052
1053
1054/* Generic update routine.
1055   -- lladdr is new lladdr or NULL, if it is not supplied.
1056   -- new    is new state.
1057   -- flags
1058	NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1059				if it is different.
1060	NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1061				lladdr instead of overriding it
1062				if it is different.
1063				It also allows to retain current state
1064				if lladdr is unchanged.
1065	NEIGH_UPDATE_F_ADMIN	means that the change is administrative.
1066
1067	NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1068				NTF_ROUTER flag.
1069	NEIGH_UPDATE_F_ISROUTER	indicates if the neighbour is known as
1070				a router.
1071
1072   Caller MUST hold reference count on the entry.
1073 */
1074
1075int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1076		 u32 flags)
1077{
1078	u8 old;
1079	int err;
1080	int notify = 0;
1081	struct net_device *dev;
1082	int update_isrouter = 0;
1083
1084	write_lock_bh(&neigh->lock);
1085
1086	dev    = neigh->dev;
1087	old    = neigh->nud_state;
1088	err    = -EPERM;
1089
1090	if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1091	    (old & (NUD_NOARP | NUD_PERMANENT)))
1092		goto out;
1093	if (neigh->dead)
1094		goto out;
1095
1096	if (!(new & NUD_VALID)) {
1097		neigh_del_timer(neigh);
1098		if (old & NUD_CONNECTED)
1099			neigh_suspect(neigh);
1100		neigh->nud_state = new;
1101		err = 0;
1102		notify = old & NUD_VALID;
1103		if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1104		    (new & NUD_FAILED)) {
1105			neigh_invalidate(neigh);
1106			notify = 1;
1107		}
1108		goto out;
1109	}
1110
1111	/* Compare new lladdr with cached one */
1112	if (!dev->addr_len) {
1113		/* First case: device needs no address. */
1114		lladdr = neigh->ha;
1115	} else if (lladdr) {
1116		/* The second case: if something is already cached
1117		   and a new address is proposed:
1118		   - compare new & old
1119		   - if they are different, check override flag
1120		 */
1121		if ((old & NUD_VALID) &&
1122		    !memcmp(lladdr, neigh->ha, dev->addr_len))
1123			lladdr = neigh->ha;
1124	} else {
1125		/* No address is supplied; if we know something,
1126		   use it, otherwise discard the request.
1127		 */
1128		err = -EINVAL;
1129		if (!(old & NUD_VALID))
1130			goto out;
1131		lladdr = neigh->ha;
1132	}
1133
1134	if (new & NUD_CONNECTED)
1135		neigh->confirmed = jiffies;
1136	neigh->updated = jiffies;
1137
1138	/* If entry was valid and address is not changed,
1139	   do not change entry state, if new one is STALE.
1140	 */
1141	err = 0;
1142	update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1143	if (old & NUD_VALID) {
1144		if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1145			update_isrouter = 0;
1146			if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1147			    (old & NUD_CONNECTED)) {
1148				lladdr = neigh->ha;
1149				new = NUD_STALE;
1150			} else
1151				goto out;
1152		} else {
1153			if (lladdr == neigh->ha && new == NUD_STALE &&
1154			    ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1155			     (old & NUD_CONNECTED))
1156			    )
1157				new = old;
1158		}
1159	}
1160
 
 
 
 
 
 
 
 
 
 
1161	if (new != old) {
1162		neigh_del_timer(neigh);
1163		if (new & NUD_PROBE)
1164			atomic_set(&neigh->probes, 0);
1165		if (new & NUD_IN_TIMER)
1166			neigh_add_timer(neigh, (jiffies +
1167						((new & NUD_REACHABLE) ?
1168						 neigh->parms->reachable_time :
1169						 0)));
1170		neigh->nud_state = new;
1171		notify = 1;
1172	}
1173
1174	if (lladdr != neigh->ha) {
1175		write_seqlock(&neigh->ha_lock);
1176		memcpy(&neigh->ha, lladdr, dev->addr_len);
1177		write_sequnlock(&neigh->ha_lock);
1178		neigh_update_hhs(neigh);
1179		if (!(new & NUD_CONNECTED))
1180			neigh->confirmed = jiffies -
1181				      (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1182		notify = 1;
1183	}
1184	if (new == old)
1185		goto out;
1186	if (new & NUD_CONNECTED)
1187		neigh_connect(neigh);
1188	else
1189		neigh_suspect(neigh);
1190	if (!(old & NUD_VALID)) {
1191		struct sk_buff *skb;
1192
1193		/* Again: avoid dead loop if something went wrong */
1194
1195		while (neigh->nud_state & NUD_VALID &&
1196		       (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1197			struct dst_entry *dst = skb_dst(skb);
1198			struct neighbour *n2, *n1 = neigh;
1199			write_unlock_bh(&neigh->lock);
1200
1201			rcu_read_lock();
1202
1203			/* Why not just use 'neigh' as-is?  The problem is that
1204			 * things such as shaper, eql, and sch_teql can end up
1205			 * using alternative, different, neigh objects to output
1206			 * the packet in the output path.  So what we need to do
1207			 * here is re-lookup the top-level neigh in the path so
1208			 * we can reinject the packet there.
1209			 */
1210			n2 = NULL;
1211			if (dst) {
1212				n2 = dst_neigh_lookup_skb(dst, skb);
1213				if (n2)
1214					n1 = n2;
1215			}
1216			n1->output(n1, skb);
1217			if (n2)
1218				neigh_release(n2);
1219			rcu_read_unlock();
1220
1221			write_lock_bh(&neigh->lock);
1222		}
1223		__skb_queue_purge(&neigh->arp_queue);
1224		neigh->arp_queue_len_bytes = 0;
1225	}
1226out:
1227	if (update_isrouter) {
1228		neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1229			(neigh->flags | NTF_ROUTER) :
1230			(neigh->flags & ~NTF_ROUTER);
1231	}
1232	write_unlock_bh(&neigh->lock);
1233
1234	if (notify)
1235		neigh_update_notify(neigh);
1236
1237	return err;
1238}
1239EXPORT_SYMBOL(neigh_update);
1240
1241/* Update the neigh to listen temporarily for probe responses, even if it is
1242 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1243 */
1244void __neigh_set_probe_once(struct neighbour *neigh)
1245{
1246	if (neigh->dead)
1247		return;
1248	neigh->updated = jiffies;
1249	if (!(neigh->nud_state & NUD_FAILED))
1250		return;
1251	neigh->nud_state = NUD_INCOMPLETE;
1252	atomic_set(&neigh->probes, neigh_max_probes(neigh));
1253	neigh_add_timer(neigh,
1254			jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1255}
1256EXPORT_SYMBOL(__neigh_set_probe_once);
1257
1258struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1259				 u8 *lladdr, void *saddr,
1260				 struct net_device *dev)
1261{
1262	struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1263						 lladdr || !dev->addr_len);
1264	if (neigh)
1265		neigh_update(neigh, lladdr, NUD_STALE,
1266			     NEIGH_UPDATE_F_OVERRIDE);
1267	return neigh;
1268}
1269EXPORT_SYMBOL(neigh_event_ns);
1270
1271/* called with read_lock_bh(&n->lock); */
1272static void neigh_hh_init(struct neighbour *n)
1273{
1274	struct net_device *dev = n->dev;
1275	__be16 prot = n->tbl->protocol;
1276	struct hh_cache	*hh = &n->hh;
1277
1278	write_lock_bh(&n->lock);
1279
1280	/* Only one thread can come in here and initialize the
1281	 * hh_cache entry.
1282	 */
1283	if (!hh->hh_len)
1284		dev->header_ops->cache(n, hh, prot);
1285
1286	write_unlock_bh(&n->lock);
1287}
1288
1289/* Slow and careful. */
1290
1291int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1292{
1293	int rc = 0;
1294
1295	if (!neigh_event_send(neigh, skb)) {
1296		int err;
1297		struct net_device *dev = neigh->dev;
1298		unsigned int seq;
1299
1300		if (dev->header_ops->cache && !neigh->hh.hh_len)
1301			neigh_hh_init(neigh);
1302
1303		do {
1304			__skb_pull(skb, skb_network_offset(skb));
1305			seq = read_seqbegin(&neigh->ha_lock);
1306			err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1307					      neigh->ha, NULL, skb->len);
1308		} while (read_seqretry(&neigh->ha_lock, seq));
1309
1310		if (err >= 0)
1311			rc = dev_queue_xmit(skb);
1312		else
1313			goto out_kfree_skb;
1314	}
1315out:
1316	return rc;
1317out_kfree_skb:
1318	rc = -EINVAL;
1319	kfree_skb(skb);
1320	goto out;
1321}
1322EXPORT_SYMBOL(neigh_resolve_output);
1323
1324/* As fast as possible without hh cache */
1325
1326int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1327{
1328	struct net_device *dev = neigh->dev;
1329	unsigned int seq;
1330	int err;
1331
1332	do {
1333		__skb_pull(skb, skb_network_offset(skb));
1334		seq = read_seqbegin(&neigh->ha_lock);
1335		err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1336				      neigh->ha, NULL, skb->len);
1337	} while (read_seqretry(&neigh->ha_lock, seq));
1338
1339	if (err >= 0)
1340		err = dev_queue_xmit(skb);
1341	else {
1342		err = -EINVAL;
1343		kfree_skb(skb);
1344	}
1345	return err;
1346}
1347EXPORT_SYMBOL(neigh_connected_output);
1348
1349int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1350{
1351	return dev_queue_xmit(skb);
1352}
1353EXPORT_SYMBOL(neigh_direct_output);
1354
1355static void neigh_proxy_process(unsigned long arg)
1356{
1357	struct neigh_table *tbl = (struct neigh_table *)arg;
1358	long sched_next = 0;
1359	unsigned long now = jiffies;
1360	struct sk_buff *skb, *n;
1361
1362	spin_lock(&tbl->proxy_queue.lock);
1363
1364	skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1365		long tdif = NEIGH_CB(skb)->sched_next - now;
1366
1367		if (tdif <= 0) {
1368			struct net_device *dev = skb->dev;
1369
1370			__skb_unlink(skb, &tbl->proxy_queue);
1371			if (tbl->proxy_redo && netif_running(dev)) {
1372				rcu_read_lock();
1373				tbl->proxy_redo(skb);
1374				rcu_read_unlock();
1375			} else {
1376				kfree_skb(skb);
1377			}
1378
1379			dev_put(dev);
1380		} else if (!sched_next || tdif < sched_next)
1381			sched_next = tdif;
1382	}
1383	del_timer(&tbl->proxy_timer);
1384	if (sched_next)
1385		mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1386	spin_unlock(&tbl->proxy_queue.lock);
1387}
1388
1389void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1390		    struct sk_buff *skb)
1391{
1392	unsigned long now = jiffies;
1393
1394	unsigned long sched_next = now + (prandom_u32() %
1395					  NEIGH_VAR(p, PROXY_DELAY));
1396
1397	if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1398		kfree_skb(skb);
1399		return;
1400	}
1401
1402	NEIGH_CB(skb)->sched_next = sched_next;
1403	NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1404
1405	spin_lock(&tbl->proxy_queue.lock);
1406	if (del_timer(&tbl->proxy_timer)) {
1407		if (time_before(tbl->proxy_timer.expires, sched_next))
1408			sched_next = tbl->proxy_timer.expires;
1409	}
1410	skb_dst_drop(skb);
1411	dev_hold(skb->dev);
1412	__skb_queue_tail(&tbl->proxy_queue, skb);
1413	mod_timer(&tbl->proxy_timer, sched_next);
1414	spin_unlock(&tbl->proxy_queue.lock);
1415}
1416EXPORT_SYMBOL(pneigh_enqueue);
1417
1418static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1419						      struct net *net, int ifindex)
1420{
1421	struct neigh_parms *p;
1422
1423	list_for_each_entry(p, &tbl->parms_list, list) {
1424		if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1425		    (!p->dev && !ifindex && net_eq(net, &init_net)))
1426			return p;
1427	}
1428
1429	return NULL;
1430}
1431
1432struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1433				      struct neigh_table *tbl)
1434{
1435	struct neigh_parms *p;
1436	struct net *net = dev_net(dev);
1437	const struct net_device_ops *ops = dev->netdev_ops;
1438
1439	p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1440	if (p) {
1441		p->tbl		  = tbl;
1442		atomic_set(&p->refcnt, 1);
1443		p->reachable_time =
1444				neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1445		dev_hold(dev);
1446		p->dev = dev;
1447		write_pnet(&p->net, net);
1448		p->sysctl_table = NULL;
1449
1450		if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1451			dev_put(dev);
1452			kfree(p);
1453			return NULL;
1454		}
1455
1456		write_lock_bh(&tbl->lock);
1457		list_add(&p->list, &tbl->parms.list);
1458		write_unlock_bh(&tbl->lock);
1459
1460		neigh_parms_data_state_cleanall(p);
1461	}
1462	return p;
1463}
1464EXPORT_SYMBOL(neigh_parms_alloc);
1465
1466static void neigh_rcu_free_parms(struct rcu_head *head)
1467{
1468	struct neigh_parms *parms =
1469		container_of(head, struct neigh_parms, rcu_head);
1470
1471	neigh_parms_put(parms);
1472}
1473
1474void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1475{
1476	if (!parms || parms == &tbl->parms)
1477		return;
1478	write_lock_bh(&tbl->lock);
1479	list_del(&parms->list);
1480	parms->dead = 1;
1481	write_unlock_bh(&tbl->lock);
1482	if (parms->dev)
1483		dev_put(parms->dev);
1484	call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1485}
1486EXPORT_SYMBOL(neigh_parms_release);
1487
1488static void neigh_parms_destroy(struct neigh_parms *parms)
1489{
1490	kfree(parms);
1491}
1492
1493static struct lock_class_key neigh_table_proxy_queue_class;
1494
1495static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1496
1497void neigh_table_init(int index, struct neigh_table *tbl)
1498{
1499	unsigned long now = jiffies;
1500	unsigned long phsize;
1501
1502	INIT_LIST_HEAD(&tbl->parms_list);
1503	list_add(&tbl->parms.list, &tbl->parms_list);
1504	write_pnet(&tbl->parms.net, &init_net);
1505	atomic_set(&tbl->parms.refcnt, 1);
1506	tbl->parms.reachable_time =
1507			  neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1508
1509	tbl->stats = alloc_percpu(struct neigh_statistics);
1510	if (!tbl->stats)
1511		panic("cannot create neighbour cache statistics");
1512
1513#ifdef CONFIG_PROC_FS
1514	if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1515			      &neigh_stat_seq_fops, tbl))
1516		panic("cannot create neighbour proc dir entry");
1517#endif
1518
1519	RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1520
1521	phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1522	tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1523
1524	if (!tbl->nht || !tbl->phash_buckets)
1525		panic("cannot allocate neighbour cache hashes");
1526
1527	if (!tbl->entry_size)
1528		tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1529					tbl->key_len, NEIGH_PRIV_ALIGN);
1530	else
1531		WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1532
1533	rwlock_init(&tbl->lock);
1534	INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1535	queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1536			tbl->parms.reachable_time);
1537	setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1538	skb_queue_head_init_class(&tbl->proxy_queue,
1539			&neigh_table_proxy_queue_class);
1540
1541	tbl->last_flush = now;
1542	tbl->last_rand	= now + tbl->parms.reachable_time * 20;
1543
1544	neigh_tables[index] = tbl;
1545}
1546EXPORT_SYMBOL(neigh_table_init);
1547
1548int neigh_table_clear(int index, struct neigh_table *tbl)
1549{
1550	neigh_tables[index] = NULL;
1551	/* It is not clean... Fix it to unload IPv6 module safely */
1552	cancel_delayed_work_sync(&tbl->gc_work);
1553	del_timer_sync(&tbl->proxy_timer);
1554	pneigh_queue_purge(&tbl->proxy_queue);
1555	neigh_ifdown(tbl, NULL);
1556	if (atomic_read(&tbl->entries))
1557		pr_crit("neighbour leakage\n");
1558
1559	call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1560		 neigh_hash_free_rcu);
1561	tbl->nht = NULL;
1562
1563	kfree(tbl->phash_buckets);
1564	tbl->phash_buckets = NULL;
1565
1566	remove_proc_entry(tbl->id, init_net.proc_net_stat);
1567
1568	free_percpu(tbl->stats);
1569	tbl->stats = NULL;
1570
1571	return 0;
1572}
1573EXPORT_SYMBOL(neigh_table_clear);
1574
1575static struct neigh_table *neigh_find_table(int family)
1576{
1577	struct neigh_table *tbl = NULL;
1578
1579	switch (family) {
1580	case AF_INET:
1581		tbl = neigh_tables[NEIGH_ARP_TABLE];
1582		break;
1583	case AF_INET6:
1584		tbl = neigh_tables[NEIGH_ND_TABLE];
1585		break;
1586	case AF_DECnet:
1587		tbl = neigh_tables[NEIGH_DN_TABLE];
1588		break;
1589	}
1590
1591	return tbl;
1592}
1593
1594static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
 
1595{
1596	struct net *net = sock_net(skb->sk);
1597	struct ndmsg *ndm;
1598	struct nlattr *dst_attr;
1599	struct neigh_table *tbl;
1600	struct neighbour *neigh;
1601	struct net_device *dev = NULL;
1602	int err = -EINVAL;
1603
1604	ASSERT_RTNL();
1605	if (nlmsg_len(nlh) < sizeof(*ndm))
1606		goto out;
1607
1608	dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1609	if (dst_attr == NULL)
1610		goto out;
1611
1612	ndm = nlmsg_data(nlh);
1613	if (ndm->ndm_ifindex) {
1614		dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1615		if (dev == NULL) {
1616			err = -ENODEV;
1617			goto out;
1618		}
1619	}
1620
1621	tbl = neigh_find_table(ndm->ndm_family);
1622	if (tbl == NULL)
1623		return -EAFNOSUPPORT;
1624
1625	if (nla_len(dst_attr) < tbl->key_len)
1626		goto out;
1627
1628	if (ndm->ndm_flags & NTF_PROXY) {
1629		err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1630		goto out;
1631	}
1632
1633	if (dev == NULL)
1634		goto out;
1635
1636	neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1637	if (neigh == NULL) {
1638		err = -ENOENT;
1639		goto out;
1640	}
1641
1642	err = neigh_update(neigh, NULL, NUD_FAILED,
1643			   NEIGH_UPDATE_F_OVERRIDE |
1644			   NEIGH_UPDATE_F_ADMIN);
 
 
1645	neigh_release(neigh);
 
 
1646
1647out:
1648	return err;
1649}
1650
1651static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
 
1652{
1653	int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1654	struct net *net = sock_net(skb->sk);
1655	struct ndmsg *ndm;
1656	struct nlattr *tb[NDA_MAX+1];
1657	struct neigh_table *tbl;
1658	struct net_device *dev = NULL;
1659	struct neighbour *neigh;
1660	void *dst, *lladdr;
1661	int err;
1662
1663	ASSERT_RTNL();
1664	err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1665	if (err < 0)
1666		goto out;
1667
1668	err = -EINVAL;
1669	if (tb[NDA_DST] == NULL)
1670		goto out;
1671
1672	ndm = nlmsg_data(nlh);
1673	if (ndm->ndm_ifindex) {
1674		dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1675		if (dev == NULL) {
1676			err = -ENODEV;
1677			goto out;
1678		}
1679
1680		if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1681			goto out;
1682	}
1683
1684	tbl = neigh_find_table(ndm->ndm_family);
1685	if (tbl == NULL)
1686		return -EAFNOSUPPORT;
1687
1688	if (nla_len(tb[NDA_DST]) < tbl->key_len)
1689		goto out;
1690	dst = nla_data(tb[NDA_DST]);
1691	lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1692
1693	if (ndm->ndm_flags & NTF_PROXY) {
1694		struct pneigh_entry *pn;
1695
1696		err = -ENOBUFS;
1697		pn = pneigh_lookup(tbl, net, dst, dev, 1);
1698		if (pn) {
1699			pn->flags = ndm->ndm_flags;
1700			err = 0;
1701		}
1702		goto out;
1703	}
1704
1705	if (dev == NULL)
1706		goto out;
1707
1708	neigh = neigh_lookup(tbl, dst, dev);
1709	if (neigh == NULL) {
1710		if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1711			err = -ENOENT;
1712			goto out;
1713		}
1714
1715		neigh = __neigh_lookup_errno(tbl, dst, dev);
1716		if (IS_ERR(neigh)) {
1717			err = PTR_ERR(neigh);
1718			goto out;
1719		}
1720	} else {
1721		if (nlh->nlmsg_flags & NLM_F_EXCL) {
1722			err = -EEXIST;
1723			neigh_release(neigh);
1724			goto out;
1725		}
1726
1727		if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1728			flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1729	}
1730
1731	if (ndm->ndm_flags & NTF_USE) {
1732		neigh_event_send(neigh, NULL);
1733		err = 0;
1734	} else
1735		err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
 
1736	neigh_release(neigh);
1737
1738out:
1739	return err;
1740}
1741
1742static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1743{
1744	struct nlattr *nest;
1745
1746	nest = nla_nest_start(skb, NDTA_PARMS);
1747	if (nest == NULL)
1748		return -ENOBUFS;
1749
1750	if ((parms->dev &&
1751	     nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1752	    nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1753	    nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1754			NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1755	    /* approximative value for deprecated QUEUE_LEN (in packets) */
1756	    nla_put_u32(skb, NDTPA_QUEUE_LEN,
1757			NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1758	    nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1759	    nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1760	    nla_put_u32(skb, NDTPA_UCAST_PROBES,
1761			NEIGH_VAR(parms, UCAST_PROBES)) ||
1762	    nla_put_u32(skb, NDTPA_MCAST_PROBES,
1763			NEIGH_VAR(parms, MCAST_PROBES)) ||
1764	    nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1765			NEIGH_VAR(parms, MCAST_REPROBES)) ||
1766	    nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
 
1767	    nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1768			  NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1769	    nla_put_msecs(skb, NDTPA_GC_STALETIME,
1770			  NEIGH_VAR(parms, GC_STALETIME)) ||
1771	    nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1772			  NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1773	    nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1774			  NEIGH_VAR(parms, RETRANS_TIME)) ||
1775	    nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1776			  NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1777	    nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1778			  NEIGH_VAR(parms, PROXY_DELAY)) ||
1779	    nla_put_msecs(skb, NDTPA_LOCKTIME,
1780			  NEIGH_VAR(parms, LOCKTIME)))
1781		goto nla_put_failure;
1782	return nla_nest_end(skb, nest);
1783
1784nla_put_failure:
1785	nla_nest_cancel(skb, nest);
1786	return -EMSGSIZE;
1787}
1788
1789static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1790			      u32 pid, u32 seq, int type, int flags)
1791{
1792	struct nlmsghdr *nlh;
1793	struct ndtmsg *ndtmsg;
1794
1795	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1796	if (nlh == NULL)
1797		return -EMSGSIZE;
1798
1799	ndtmsg = nlmsg_data(nlh);
1800
1801	read_lock_bh(&tbl->lock);
1802	ndtmsg->ndtm_family = tbl->family;
1803	ndtmsg->ndtm_pad1   = 0;
1804	ndtmsg->ndtm_pad2   = 0;
1805
1806	if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1807	    nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1808	    nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1809	    nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1810	    nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1811		goto nla_put_failure;
1812	{
1813		unsigned long now = jiffies;
1814		unsigned int flush_delta = now - tbl->last_flush;
1815		unsigned int rand_delta = now - tbl->last_rand;
1816		struct neigh_hash_table *nht;
1817		struct ndt_config ndc = {
1818			.ndtc_key_len		= tbl->key_len,
1819			.ndtc_entry_size	= tbl->entry_size,
1820			.ndtc_entries		= atomic_read(&tbl->entries),
1821			.ndtc_last_flush	= jiffies_to_msecs(flush_delta),
1822			.ndtc_last_rand		= jiffies_to_msecs(rand_delta),
1823			.ndtc_proxy_qlen	= tbl->proxy_queue.qlen,
1824		};
1825
1826		rcu_read_lock_bh();
1827		nht = rcu_dereference_bh(tbl->nht);
1828		ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1829		ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1830		rcu_read_unlock_bh();
1831
1832		if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1833			goto nla_put_failure;
1834	}
1835
1836	{
1837		int cpu;
1838		struct ndt_stats ndst;
1839
1840		memset(&ndst, 0, sizeof(ndst));
1841
1842		for_each_possible_cpu(cpu) {
1843			struct neigh_statistics	*st;
1844
1845			st = per_cpu_ptr(tbl->stats, cpu);
1846			ndst.ndts_allocs		+= st->allocs;
1847			ndst.ndts_destroys		+= st->destroys;
1848			ndst.ndts_hash_grows		+= st->hash_grows;
1849			ndst.ndts_res_failed		+= st->res_failed;
1850			ndst.ndts_lookups		+= st->lookups;
1851			ndst.ndts_hits			+= st->hits;
1852			ndst.ndts_rcv_probes_mcast	+= st->rcv_probes_mcast;
1853			ndst.ndts_rcv_probes_ucast	+= st->rcv_probes_ucast;
1854			ndst.ndts_periodic_gc_runs	+= st->periodic_gc_runs;
1855			ndst.ndts_forced_gc_runs	+= st->forced_gc_runs;
1856			ndst.ndts_table_fulls		+= st->table_fulls;
1857		}
1858
1859		if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
 
1860			goto nla_put_failure;
1861	}
1862
1863	BUG_ON(tbl->parms.dev);
1864	if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1865		goto nla_put_failure;
1866
1867	read_unlock_bh(&tbl->lock);
1868	nlmsg_end(skb, nlh);
1869	return 0;
1870
1871nla_put_failure:
1872	read_unlock_bh(&tbl->lock);
1873	nlmsg_cancel(skb, nlh);
1874	return -EMSGSIZE;
1875}
1876
1877static int neightbl_fill_param_info(struct sk_buff *skb,
1878				    struct neigh_table *tbl,
1879				    struct neigh_parms *parms,
1880				    u32 pid, u32 seq, int type,
1881				    unsigned int flags)
1882{
1883	struct ndtmsg *ndtmsg;
1884	struct nlmsghdr *nlh;
1885
1886	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1887	if (nlh == NULL)
1888		return -EMSGSIZE;
1889
1890	ndtmsg = nlmsg_data(nlh);
1891
1892	read_lock_bh(&tbl->lock);
1893	ndtmsg->ndtm_family = tbl->family;
1894	ndtmsg->ndtm_pad1   = 0;
1895	ndtmsg->ndtm_pad2   = 0;
1896
1897	if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1898	    neightbl_fill_parms(skb, parms) < 0)
1899		goto errout;
1900
1901	read_unlock_bh(&tbl->lock);
1902	nlmsg_end(skb, nlh);
1903	return 0;
1904errout:
1905	read_unlock_bh(&tbl->lock);
1906	nlmsg_cancel(skb, nlh);
1907	return -EMSGSIZE;
1908}
1909
1910static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1911	[NDTA_NAME]		= { .type = NLA_STRING },
1912	[NDTA_THRESH1]		= { .type = NLA_U32 },
1913	[NDTA_THRESH2]		= { .type = NLA_U32 },
1914	[NDTA_THRESH3]		= { .type = NLA_U32 },
1915	[NDTA_GC_INTERVAL]	= { .type = NLA_U64 },
1916	[NDTA_PARMS]		= { .type = NLA_NESTED },
1917};
1918
1919static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1920	[NDTPA_IFINDEX]			= { .type = NLA_U32 },
1921	[NDTPA_QUEUE_LEN]		= { .type = NLA_U32 },
1922	[NDTPA_PROXY_QLEN]		= { .type = NLA_U32 },
1923	[NDTPA_APP_PROBES]		= { .type = NLA_U32 },
1924	[NDTPA_UCAST_PROBES]		= { .type = NLA_U32 },
1925	[NDTPA_MCAST_PROBES]		= { .type = NLA_U32 },
1926	[NDTPA_MCAST_REPROBES]		= { .type = NLA_U32 },
1927	[NDTPA_BASE_REACHABLE_TIME]	= { .type = NLA_U64 },
1928	[NDTPA_GC_STALETIME]		= { .type = NLA_U64 },
1929	[NDTPA_DELAY_PROBE_TIME]	= { .type = NLA_U64 },
1930	[NDTPA_RETRANS_TIME]		= { .type = NLA_U64 },
1931	[NDTPA_ANYCAST_DELAY]		= { .type = NLA_U64 },
1932	[NDTPA_PROXY_DELAY]		= { .type = NLA_U64 },
1933	[NDTPA_LOCKTIME]		= { .type = NLA_U64 },
1934};
1935
1936static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
 
1937{
1938	struct net *net = sock_net(skb->sk);
1939	struct neigh_table *tbl;
1940	struct ndtmsg *ndtmsg;
1941	struct nlattr *tb[NDTA_MAX+1];
1942	bool found = false;
1943	int err, tidx;
1944
1945	err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1946			  nl_neightbl_policy);
1947	if (err < 0)
1948		goto errout;
1949
1950	if (tb[NDTA_NAME] == NULL) {
1951		err = -EINVAL;
1952		goto errout;
1953	}
1954
1955	ndtmsg = nlmsg_data(nlh);
1956
1957	for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1958		tbl = neigh_tables[tidx];
1959		if (!tbl)
1960			continue;
1961		if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1962			continue;
1963		if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1964			found = true;
1965			break;
1966		}
1967	}
1968
1969	if (!found)
1970		return -ENOENT;
1971
1972	/*
1973	 * We acquire tbl->lock to be nice to the periodic timers and
1974	 * make sure they always see a consistent set of values.
1975	 */
1976	write_lock_bh(&tbl->lock);
1977
1978	if (tb[NDTA_PARMS]) {
1979		struct nlattr *tbp[NDTPA_MAX+1];
1980		struct neigh_parms *p;
1981		int i, ifindex = 0;
1982
1983		err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1984				       nl_ntbl_parm_policy);
1985		if (err < 0)
1986			goto errout_tbl_lock;
1987
1988		if (tbp[NDTPA_IFINDEX])
1989			ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1990
1991		p = lookup_neigh_parms(tbl, net, ifindex);
1992		if (p == NULL) {
1993			err = -ENOENT;
1994			goto errout_tbl_lock;
1995		}
1996
1997		for (i = 1; i <= NDTPA_MAX; i++) {
1998			if (tbp[i] == NULL)
1999				continue;
2000
2001			switch (i) {
2002			case NDTPA_QUEUE_LEN:
2003				NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2004					      nla_get_u32(tbp[i]) *
2005					      SKB_TRUESIZE(ETH_FRAME_LEN));
2006				break;
2007			case NDTPA_QUEUE_LENBYTES:
2008				NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2009					      nla_get_u32(tbp[i]));
2010				break;
2011			case NDTPA_PROXY_QLEN:
2012				NEIGH_VAR_SET(p, PROXY_QLEN,
2013					      nla_get_u32(tbp[i]));
2014				break;
2015			case NDTPA_APP_PROBES:
2016				NEIGH_VAR_SET(p, APP_PROBES,
2017					      nla_get_u32(tbp[i]));
2018				break;
2019			case NDTPA_UCAST_PROBES:
2020				NEIGH_VAR_SET(p, UCAST_PROBES,
2021					      nla_get_u32(tbp[i]));
2022				break;
2023			case NDTPA_MCAST_PROBES:
2024				NEIGH_VAR_SET(p, MCAST_PROBES,
2025					      nla_get_u32(tbp[i]));
2026				break;
2027			case NDTPA_MCAST_REPROBES:
2028				NEIGH_VAR_SET(p, MCAST_REPROBES,
2029					      nla_get_u32(tbp[i]));
2030				break;
2031			case NDTPA_BASE_REACHABLE_TIME:
2032				NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2033					      nla_get_msecs(tbp[i]));
2034				/* update reachable_time as well, otherwise, the change will
2035				 * only be effective after the next time neigh_periodic_work
2036				 * decides to recompute it (can be multiple minutes)
2037				 */
2038				p->reachable_time =
2039					neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2040				break;
2041			case NDTPA_GC_STALETIME:
2042				NEIGH_VAR_SET(p, GC_STALETIME,
2043					      nla_get_msecs(tbp[i]));
2044				break;
2045			case NDTPA_DELAY_PROBE_TIME:
2046				NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2047					      nla_get_msecs(tbp[i]));
 
2048				break;
2049			case NDTPA_RETRANS_TIME:
2050				NEIGH_VAR_SET(p, RETRANS_TIME,
2051					      nla_get_msecs(tbp[i]));
2052				break;
2053			case NDTPA_ANYCAST_DELAY:
2054				NEIGH_VAR_SET(p, ANYCAST_DELAY,
2055					      nla_get_msecs(tbp[i]));
2056				break;
2057			case NDTPA_PROXY_DELAY:
2058				NEIGH_VAR_SET(p, PROXY_DELAY,
2059					      nla_get_msecs(tbp[i]));
2060				break;
2061			case NDTPA_LOCKTIME:
2062				NEIGH_VAR_SET(p, LOCKTIME,
2063					      nla_get_msecs(tbp[i]));
2064				break;
2065			}
2066		}
2067	}
2068
2069	err = -ENOENT;
2070	if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2071	     tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2072	    !net_eq(net, &init_net))
2073		goto errout_tbl_lock;
2074
2075	if (tb[NDTA_THRESH1])
2076		tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2077
2078	if (tb[NDTA_THRESH2])
2079		tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2080
2081	if (tb[NDTA_THRESH3])
2082		tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2083
2084	if (tb[NDTA_GC_INTERVAL])
2085		tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2086
2087	err = 0;
2088
2089errout_tbl_lock:
2090	write_unlock_bh(&tbl->lock);
2091errout:
2092	return err;
2093}
2094
2095static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2096{
2097	struct net *net = sock_net(skb->sk);
2098	int family, tidx, nidx = 0;
2099	int tbl_skip = cb->args[0];
2100	int neigh_skip = cb->args[1];
2101	struct neigh_table *tbl;
2102
2103	family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2104
2105	for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2106		struct neigh_parms *p;
2107
2108		tbl = neigh_tables[tidx];
2109		if (!tbl)
2110			continue;
2111
2112		if (tidx < tbl_skip || (family && tbl->family != family))
2113			continue;
2114
2115		if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2116				       cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2117				       NLM_F_MULTI) < 0)
2118			break;
2119
2120		nidx = 0;
2121		p = list_next_entry(&tbl->parms, list);
2122		list_for_each_entry_from(p, &tbl->parms_list, list) {
2123			if (!net_eq(neigh_parms_net(p), net))
2124				continue;
2125
2126			if (nidx < neigh_skip)
2127				goto next;
2128
2129			if (neightbl_fill_param_info(skb, tbl, p,
2130						     NETLINK_CB(cb->skb).portid,
2131						     cb->nlh->nlmsg_seq,
2132						     RTM_NEWNEIGHTBL,
2133						     NLM_F_MULTI) < 0)
2134				goto out;
2135		next:
2136			nidx++;
2137		}
2138
2139		neigh_skip = 0;
2140	}
2141out:
2142	cb->args[0] = tidx;
2143	cb->args[1] = nidx;
2144
2145	return skb->len;
2146}
2147
2148static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2149			   u32 pid, u32 seq, int type, unsigned int flags)
2150{
2151	unsigned long now = jiffies;
2152	struct nda_cacheinfo ci;
2153	struct nlmsghdr *nlh;
2154	struct ndmsg *ndm;
2155
2156	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2157	if (nlh == NULL)
2158		return -EMSGSIZE;
2159
2160	ndm = nlmsg_data(nlh);
2161	ndm->ndm_family	 = neigh->ops->family;
2162	ndm->ndm_pad1    = 0;
2163	ndm->ndm_pad2    = 0;
2164	ndm->ndm_flags	 = neigh->flags;
2165	ndm->ndm_type	 = neigh->type;
2166	ndm->ndm_ifindex = neigh->dev->ifindex;
2167
2168	if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2169		goto nla_put_failure;
2170
2171	read_lock_bh(&neigh->lock);
2172	ndm->ndm_state	 = neigh->nud_state;
2173	if (neigh->nud_state & NUD_VALID) {
2174		char haddr[MAX_ADDR_LEN];
2175
2176		neigh_ha_snapshot(haddr, neigh, neigh->dev);
2177		if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2178			read_unlock_bh(&neigh->lock);
2179			goto nla_put_failure;
2180		}
2181	}
2182
2183	ci.ndm_used	 = jiffies_to_clock_t(now - neigh->used);
2184	ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2185	ci.ndm_updated	 = jiffies_to_clock_t(now - neigh->updated);
2186	ci.ndm_refcnt	 = atomic_read(&neigh->refcnt) - 1;
2187	read_unlock_bh(&neigh->lock);
2188
2189	if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2190	    nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2191		goto nla_put_failure;
2192
2193	nlmsg_end(skb, nlh);
2194	return 0;
2195
2196nla_put_failure:
2197	nlmsg_cancel(skb, nlh);
2198	return -EMSGSIZE;
2199}
2200
2201static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2202			    u32 pid, u32 seq, int type, unsigned int flags,
2203			    struct neigh_table *tbl)
2204{
2205	struct nlmsghdr *nlh;
2206	struct ndmsg *ndm;
2207
2208	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2209	if (nlh == NULL)
2210		return -EMSGSIZE;
2211
2212	ndm = nlmsg_data(nlh);
2213	ndm->ndm_family	 = tbl->family;
2214	ndm->ndm_pad1    = 0;
2215	ndm->ndm_pad2    = 0;
2216	ndm->ndm_flags	 = pn->flags | NTF_PROXY;
2217	ndm->ndm_type	 = RTN_UNICAST;
2218	ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2219	ndm->ndm_state	 = NUD_NONE;
2220
2221	if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2222		goto nla_put_failure;
2223
2224	nlmsg_end(skb, nlh);
2225	return 0;
2226
2227nla_put_failure:
2228	nlmsg_cancel(skb, nlh);
2229	return -EMSGSIZE;
2230}
2231
2232static void neigh_update_notify(struct neighbour *neigh)
2233{
2234	call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2235	__neigh_notify(neigh, RTM_NEWNEIGH, 0);
2236}
2237
2238static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2239{
2240	struct net_device *master;
2241
2242	if (!master_idx)
2243		return false;
2244
2245	master = netdev_master_upper_dev_get(dev);
2246	if (!master || master->ifindex != master_idx)
2247		return true;
2248
2249	return false;
2250}
2251
2252static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2253{
2254	if (filter_idx && dev->ifindex != filter_idx)
2255		return true;
2256
2257	return false;
2258}
2259
2260static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2261			    struct netlink_callback *cb)
2262{
2263	struct net *net = sock_net(skb->sk);
2264	const struct nlmsghdr *nlh = cb->nlh;
2265	struct nlattr *tb[NDA_MAX + 1];
2266	struct neighbour *n;
2267	int rc, h, s_h = cb->args[1];
2268	int idx, s_idx = idx = cb->args[2];
2269	struct neigh_hash_table *nht;
2270	int filter_master_idx = 0, filter_idx = 0;
2271	unsigned int flags = NLM_F_MULTI;
2272	int err;
2273
2274	err = nlmsg_parse(nlh, sizeof(struct ndmsg), tb, NDA_MAX, NULL);
2275	if (!err) {
2276		if (tb[NDA_IFINDEX])
 
 
2277			filter_idx = nla_get_u32(tb[NDA_IFINDEX]);
2278
2279		if (tb[NDA_MASTER])
 
 
2280			filter_master_idx = nla_get_u32(tb[NDA_MASTER]);
2281
2282		if (filter_idx || filter_master_idx)
2283			flags |= NLM_F_DUMP_FILTERED;
2284	}
2285
2286	rcu_read_lock_bh();
2287	nht = rcu_dereference_bh(tbl->nht);
2288
2289	for (h = s_h; h < (1 << nht->hash_shift); h++) {
2290		if (h > s_h)
2291			s_idx = 0;
2292		for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2293		     n != NULL;
2294		     n = rcu_dereference_bh(n->next)) {
2295			if (!net_eq(dev_net(n->dev), net))
2296				continue;
2297			if (neigh_ifindex_filtered(n->dev, filter_idx))
2298				continue;
2299			if (neigh_master_filtered(n->dev, filter_master_idx))
2300				continue;
2301			if (idx < s_idx)
2302				goto next;
2303			if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2304					    cb->nlh->nlmsg_seq,
2305					    RTM_NEWNEIGH,
2306					    flags) < 0) {
2307				rc = -1;
2308				goto out;
2309			}
2310next:
2311			idx++;
2312		}
2313	}
2314	rc = skb->len;
2315out:
2316	rcu_read_unlock_bh();
2317	cb->args[1] = h;
2318	cb->args[2] = idx;
2319	return rc;
2320}
2321
2322static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2323			     struct netlink_callback *cb)
2324{
2325	struct pneigh_entry *n;
2326	struct net *net = sock_net(skb->sk);
2327	int rc, h, s_h = cb->args[3];
2328	int idx, s_idx = idx = cb->args[4];
2329
2330	read_lock_bh(&tbl->lock);
2331
2332	for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2333		if (h > s_h)
2334			s_idx = 0;
2335		for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2336			if (pneigh_net(n) != net)
2337				continue;
2338			if (idx < s_idx)
2339				goto next;
2340			if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2341					    cb->nlh->nlmsg_seq,
2342					    RTM_NEWNEIGH,
2343					    NLM_F_MULTI, tbl) < 0) {
2344				read_unlock_bh(&tbl->lock);
2345				rc = -1;
2346				goto out;
2347			}
2348		next:
2349			idx++;
2350		}
2351	}
2352
2353	read_unlock_bh(&tbl->lock);
2354	rc = skb->len;
2355out:
2356	cb->args[3] = h;
2357	cb->args[4] = idx;
2358	return rc;
2359
2360}
2361
2362static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2363{
2364	struct neigh_table *tbl;
2365	int t, family, s_t;
2366	int proxy = 0;
2367	int err;
2368
2369	family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2370
2371	/* check for full ndmsg structure presence, family member is
2372	 * the same for both structures
2373	 */
2374	if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2375	    ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2376		proxy = 1;
2377
2378	s_t = cb->args[0];
2379
2380	for (t = 0; t < NEIGH_NR_TABLES; t++) {
2381		tbl = neigh_tables[t];
2382
2383		if (!tbl)
2384			continue;
2385		if (t < s_t || (family && tbl->family != family))
2386			continue;
2387		if (t > s_t)
2388			memset(&cb->args[1], 0, sizeof(cb->args) -
2389						sizeof(cb->args[0]));
2390		if (proxy)
2391			err = pneigh_dump_table(tbl, skb, cb);
2392		else
2393			err = neigh_dump_table(tbl, skb, cb);
2394		if (err < 0)
2395			break;
2396	}
2397
2398	cb->args[0] = t;
2399	return skb->len;
2400}
2401
2402void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2403{
2404	int chain;
2405	struct neigh_hash_table *nht;
2406
2407	rcu_read_lock_bh();
2408	nht = rcu_dereference_bh(tbl->nht);
2409
2410	read_lock(&tbl->lock); /* avoid resizes */
2411	for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2412		struct neighbour *n;
2413
2414		for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2415		     n != NULL;
2416		     n = rcu_dereference_bh(n->next))
2417			cb(n, cookie);
2418	}
2419	read_unlock(&tbl->lock);
2420	rcu_read_unlock_bh();
2421}
2422EXPORT_SYMBOL(neigh_for_each);
2423
2424/* The tbl->lock must be held as a writer and BH disabled. */
2425void __neigh_for_each_release(struct neigh_table *tbl,
2426			      int (*cb)(struct neighbour *))
2427{
2428	int chain;
2429	struct neigh_hash_table *nht;
2430
2431	nht = rcu_dereference_protected(tbl->nht,
2432					lockdep_is_held(&tbl->lock));
2433	for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2434		struct neighbour *n;
2435		struct neighbour __rcu **np;
2436
2437		np = &nht->hash_buckets[chain];
2438		while ((n = rcu_dereference_protected(*np,
2439					lockdep_is_held(&tbl->lock))) != NULL) {
2440			int release;
2441
2442			write_lock(&n->lock);
2443			release = cb(n);
2444			if (release) {
2445				rcu_assign_pointer(*np,
2446					rcu_dereference_protected(n->next,
2447						lockdep_is_held(&tbl->lock)));
2448				n->dead = 1;
2449			} else
2450				np = &n->next;
2451			write_unlock(&n->lock);
2452			if (release)
2453				neigh_cleanup_and_release(n);
2454		}
2455	}
2456}
2457EXPORT_SYMBOL(__neigh_for_each_release);
2458
2459int neigh_xmit(int index, struct net_device *dev,
2460	       const void *addr, struct sk_buff *skb)
2461{
2462	int err = -EAFNOSUPPORT;
2463	if (likely(index < NEIGH_NR_TABLES)) {
2464		struct neigh_table *tbl;
2465		struct neighbour *neigh;
2466
2467		tbl = neigh_tables[index];
2468		if (!tbl)
2469			goto out;
 
2470		neigh = __neigh_lookup_noref(tbl, addr, dev);
2471		if (!neigh)
2472			neigh = __neigh_create(tbl, addr, dev, false);
2473		err = PTR_ERR(neigh);
2474		if (IS_ERR(neigh))
 
2475			goto out_kfree_skb;
 
2476		err = neigh->output(neigh, skb);
 
2477	}
2478	else if (index == NEIGH_LINK_TABLE) {
2479		err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2480				      addr, NULL, skb->len);
2481		if (err < 0)
2482			goto out_kfree_skb;
2483		err = dev_queue_xmit(skb);
2484	}
2485out:
2486	return err;
2487out_kfree_skb:
2488	kfree_skb(skb);
2489	goto out;
2490}
2491EXPORT_SYMBOL(neigh_xmit);
2492
2493#ifdef CONFIG_PROC_FS
2494
2495static struct neighbour *neigh_get_first(struct seq_file *seq)
2496{
2497	struct neigh_seq_state *state = seq->private;
2498	struct net *net = seq_file_net(seq);
2499	struct neigh_hash_table *nht = state->nht;
2500	struct neighbour *n = NULL;
2501	int bucket = state->bucket;
2502
2503	state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2504	for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2505		n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2506
2507		while (n) {
2508			if (!net_eq(dev_net(n->dev), net))
2509				goto next;
2510			if (state->neigh_sub_iter) {
2511				loff_t fakep = 0;
2512				void *v;
2513
2514				v = state->neigh_sub_iter(state, n, &fakep);
2515				if (!v)
2516					goto next;
2517			}
2518			if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2519				break;
2520			if (n->nud_state & ~NUD_NOARP)
2521				break;
2522next:
2523			n = rcu_dereference_bh(n->next);
2524		}
2525
2526		if (n)
2527			break;
2528	}
2529	state->bucket = bucket;
2530
2531	return n;
2532}
2533
2534static struct neighbour *neigh_get_next(struct seq_file *seq,
2535					struct neighbour *n,
2536					loff_t *pos)
2537{
2538	struct neigh_seq_state *state = seq->private;
2539	struct net *net = seq_file_net(seq);
2540	struct neigh_hash_table *nht = state->nht;
2541
2542	if (state->neigh_sub_iter) {
2543		void *v = state->neigh_sub_iter(state, n, pos);
2544		if (v)
2545			return n;
2546	}
2547	n = rcu_dereference_bh(n->next);
2548
2549	while (1) {
2550		while (n) {
2551			if (!net_eq(dev_net(n->dev), net))
2552				goto next;
2553			if (state->neigh_sub_iter) {
2554				void *v = state->neigh_sub_iter(state, n, pos);
2555				if (v)
2556					return n;
2557				goto next;
2558			}
2559			if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2560				break;
2561
2562			if (n->nud_state & ~NUD_NOARP)
2563				break;
2564next:
2565			n = rcu_dereference_bh(n->next);
2566		}
2567
2568		if (n)
2569			break;
2570
2571		if (++state->bucket >= (1 << nht->hash_shift))
2572			break;
2573
2574		n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2575	}
2576
2577	if (n && pos)
2578		--(*pos);
2579	return n;
2580}
2581
2582static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2583{
2584	struct neighbour *n = neigh_get_first(seq);
2585
2586	if (n) {
2587		--(*pos);
2588		while (*pos) {
2589			n = neigh_get_next(seq, n, pos);
2590			if (!n)
2591				break;
2592		}
2593	}
2594	return *pos ? NULL : n;
2595}
2596
2597static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2598{
2599	struct neigh_seq_state *state = seq->private;
2600	struct net *net = seq_file_net(seq);
2601	struct neigh_table *tbl = state->tbl;
2602	struct pneigh_entry *pn = NULL;
2603	int bucket = state->bucket;
2604
2605	state->flags |= NEIGH_SEQ_IS_PNEIGH;
2606	for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2607		pn = tbl->phash_buckets[bucket];
2608		while (pn && !net_eq(pneigh_net(pn), net))
2609			pn = pn->next;
2610		if (pn)
2611			break;
2612	}
2613	state->bucket = bucket;
2614
2615	return pn;
2616}
2617
2618static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2619					    struct pneigh_entry *pn,
2620					    loff_t *pos)
2621{
2622	struct neigh_seq_state *state = seq->private;
2623	struct net *net = seq_file_net(seq);
2624	struct neigh_table *tbl = state->tbl;
2625
2626	do {
2627		pn = pn->next;
2628	} while (pn && !net_eq(pneigh_net(pn), net));
2629
2630	while (!pn) {
2631		if (++state->bucket > PNEIGH_HASHMASK)
2632			break;
2633		pn = tbl->phash_buckets[state->bucket];
2634		while (pn && !net_eq(pneigh_net(pn), net))
2635			pn = pn->next;
2636		if (pn)
2637			break;
2638	}
2639
2640	if (pn && pos)
2641		--(*pos);
2642
2643	return pn;
2644}
2645
2646static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2647{
2648	struct pneigh_entry *pn = pneigh_get_first(seq);
2649
2650	if (pn) {
2651		--(*pos);
2652		while (*pos) {
2653			pn = pneigh_get_next(seq, pn, pos);
2654			if (!pn)
2655				break;
2656		}
2657	}
2658	return *pos ? NULL : pn;
2659}
2660
2661static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2662{
2663	struct neigh_seq_state *state = seq->private;
2664	void *rc;
2665	loff_t idxpos = *pos;
2666
2667	rc = neigh_get_idx(seq, &idxpos);
2668	if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2669		rc = pneigh_get_idx(seq, &idxpos);
2670
2671	return rc;
2672}
2673
2674void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2675	__acquires(rcu_bh)
2676{
2677	struct neigh_seq_state *state = seq->private;
2678
2679	state->tbl = tbl;
2680	state->bucket = 0;
2681	state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2682
2683	rcu_read_lock_bh();
2684	state->nht = rcu_dereference_bh(tbl->nht);
2685
2686	return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2687}
2688EXPORT_SYMBOL(neigh_seq_start);
2689
2690void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2691{
2692	struct neigh_seq_state *state;
2693	void *rc;
2694
2695	if (v == SEQ_START_TOKEN) {
2696		rc = neigh_get_first(seq);
2697		goto out;
2698	}
2699
2700	state = seq->private;
2701	if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2702		rc = neigh_get_next(seq, v, NULL);
2703		if (rc)
2704			goto out;
2705		if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2706			rc = pneigh_get_first(seq);
2707	} else {
2708		BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2709		rc = pneigh_get_next(seq, v, NULL);
2710	}
2711out:
2712	++(*pos);
2713	return rc;
2714}
2715EXPORT_SYMBOL(neigh_seq_next);
2716
2717void neigh_seq_stop(struct seq_file *seq, void *v)
2718	__releases(rcu_bh)
2719{
2720	rcu_read_unlock_bh();
2721}
2722EXPORT_SYMBOL(neigh_seq_stop);
2723
2724/* statistics via seq_file */
2725
2726static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2727{
2728	struct neigh_table *tbl = seq->private;
2729	int cpu;
2730
2731	if (*pos == 0)
2732		return SEQ_START_TOKEN;
2733
2734	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2735		if (!cpu_possible(cpu))
2736			continue;
2737		*pos = cpu+1;
2738		return per_cpu_ptr(tbl->stats, cpu);
2739	}
2740	return NULL;
2741}
2742
2743static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2744{
2745	struct neigh_table *tbl = seq->private;
2746	int cpu;
2747
2748	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2749		if (!cpu_possible(cpu))
2750			continue;
2751		*pos = cpu+1;
2752		return per_cpu_ptr(tbl->stats, cpu);
2753	}
2754	return NULL;
2755}
2756
2757static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2758{
2759
2760}
2761
2762static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2763{
2764	struct neigh_table *tbl = seq->private;
2765	struct neigh_statistics *st = v;
2766
2767	if (v == SEQ_START_TOKEN) {
2768		seq_printf(seq, "entries  allocs destroys hash_grows  lookups hits  res_failed  rcv_probes_mcast rcv_probes_ucast  periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
2769		return 0;
2770	}
2771
2772	seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2773			"%08lx %08lx  %08lx %08lx %08lx %08lx\n",
2774		   atomic_read(&tbl->entries),
2775
2776		   st->allocs,
2777		   st->destroys,
2778		   st->hash_grows,
2779
2780		   st->lookups,
2781		   st->hits,
2782
2783		   st->res_failed,
2784
2785		   st->rcv_probes_mcast,
2786		   st->rcv_probes_ucast,
2787
2788		   st->periodic_gc_runs,
2789		   st->forced_gc_runs,
2790		   st->unres_discards,
2791		   st->table_fulls
2792		   );
2793
2794	return 0;
2795}
2796
2797static const struct seq_operations neigh_stat_seq_ops = {
2798	.start	= neigh_stat_seq_start,
2799	.next	= neigh_stat_seq_next,
2800	.stop	= neigh_stat_seq_stop,
2801	.show	= neigh_stat_seq_show,
2802};
2803
2804static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2805{
2806	int ret = seq_open(file, &neigh_stat_seq_ops);
2807
2808	if (!ret) {
2809		struct seq_file *sf = file->private_data;
2810		sf->private = PDE_DATA(inode);
2811	}
2812	return ret;
2813};
2814
2815static const struct file_operations neigh_stat_seq_fops = {
2816	.owner	 = THIS_MODULE,
2817	.open 	 = neigh_stat_seq_open,
2818	.read	 = seq_read,
2819	.llseek	 = seq_lseek,
2820	.release = seq_release,
2821};
2822
2823#endif /* CONFIG_PROC_FS */
2824
2825static inline size_t neigh_nlmsg_size(void)
2826{
2827	return NLMSG_ALIGN(sizeof(struct ndmsg))
2828	       + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2829	       + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2830	       + nla_total_size(sizeof(struct nda_cacheinfo))
2831	       + nla_total_size(4); /* NDA_PROBES */
2832}
2833
2834static void __neigh_notify(struct neighbour *n, int type, int flags)
 
2835{
2836	struct net *net = dev_net(n->dev);
2837	struct sk_buff *skb;
2838	int err = -ENOBUFS;
2839
2840	skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2841	if (skb == NULL)
2842		goto errout;
2843
2844	err = neigh_fill_info(skb, n, 0, 0, type, flags);
2845	if (err < 0) {
2846		/* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2847		WARN_ON(err == -EMSGSIZE);
2848		kfree_skb(skb);
2849		goto errout;
2850	}
2851	rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2852	return;
2853errout:
2854	if (err < 0)
2855		rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2856}
2857
2858void neigh_app_ns(struct neighbour *n)
2859{
2860	__neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2861}
2862EXPORT_SYMBOL(neigh_app_ns);
2863
2864#ifdef CONFIG_SYSCTL
2865static int zero;
2866static int int_max = INT_MAX;
2867static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2868
2869static int proc_unres_qlen(struct ctl_table *ctl, int write,
2870			   void __user *buffer, size_t *lenp, loff_t *ppos)
2871{
2872	int size, ret;
2873	struct ctl_table tmp = *ctl;
2874
2875	tmp.extra1 = &zero;
2876	tmp.extra2 = &unres_qlen_max;
2877	tmp.data = &size;
2878
2879	size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2880	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2881
2882	if (write && !ret)
2883		*(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2884	return ret;
2885}
2886
2887static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2888						   int family)
2889{
2890	switch (family) {
2891	case AF_INET:
2892		return __in_dev_arp_parms_get_rcu(dev);
2893	case AF_INET6:
2894		return __in6_dev_nd_parms_get_rcu(dev);
2895	}
2896	return NULL;
2897}
2898
2899static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2900				  int index)
2901{
2902	struct net_device *dev;
2903	int family = neigh_parms_family(p);
2904
2905	rcu_read_lock();
2906	for_each_netdev_rcu(net, dev) {
2907		struct neigh_parms *dst_p =
2908				neigh_get_dev_parms_rcu(dev, family);
2909
2910		if (dst_p && !test_bit(index, dst_p->data_state))
2911			dst_p->data[index] = p->data[index];
2912	}
2913	rcu_read_unlock();
2914}
2915
2916static void neigh_proc_update(struct ctl_table *ctl, int write)
2917{
2918	struct net_device *dev = ctl->extra1;
2919	struct neigh_parms *p = ctl->extra2;
2920	struct net *net = neigh_parms_net(p);
2921	int index = (int *) ctl->data - p->data;
2922
2923	if (!write)
2924		return;
2925
2926	set_bit(index, p->data_state);
 
 
2927	if (!dev) /* NULL dev means this is default value */
2928		neigh_copy_dflt_parms(net, p, index);
2929}
2930
2931static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2932					   void __user *buffer,
2933					   size_t *lenp, loff_t *ppos)
2934{
2935	struct ctl_table tmp = *ctl;
2936	int ret;
2937
2938	tmp.extra1 = &zero;
2939	tmp.extra2 = &int_max;
2940
2941	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2942	neigh_proc_update(ctl, write);
2943	return ret;
2944}
2945
2946int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2947			void __user *buffer, size_t *lenp, loff_t *ppos)
2948{
2949	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2950
2951	neigh_proc_update(ctl, write);
2952	return ret;
2953}
2954EXPORT_SYMBOL(neigh_proc_dointvec);
2955
2956int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2957				void __user *buffer,
2958				size_t *lenp, loff_t *ppos)
2959{
2960	int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2961
2962	neigh_proc_update(ctl, write);
2963	return ret;
2964}
2965EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2966
2967static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2968					      void __user *buffer,
2969					      size_t *lenp, loff_t *ppos)
2970{
2971	int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2972
2973	neigh_proc_update(ctl, write);
2974	return ret;
2975}
2976
2977int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2978				   void __user *buffer,
2979				   size_t *lenp, loff_t *ppos)
2980{
2981	int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2982
2983	neigh_proc_update(ctl, write);
2984	return ret;
2985}
2986EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2987
2988static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2989					  void __user *buffer,
2990					  size_t *lenp, loff_t *ppos)
2991{
2992	int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2993
2994	neigh_proc_update(ctl, write);
2995	return ret;
2996}
2997
2998static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
2999					  void __user *buffer,
3000					  size_t *lenp, loff_t *ppos)
3001{
3002	struct neigh_parms *p = ctl->extra2;
3003	int ret;
3004
3005	if (strcmp(ctl->procname, "base_reachable_time") == 0)
3006		ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3007	else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3008		ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3009	else
3010		ret = -1;
3011
3012	if (write && ret == 0) {
3013		/* update reachable_time as well, otherwise, the change will
3014		 * only be effective after the next time neigh_periodic_work
3015		 * decides to recompute it
3016		 */
3017		p->reachable_time =
3018			neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3019	}
3020	return ret;
3021}
3022
3023#define NEIGH_PARMS_DATA_OFFSET(index)	\
3024	(&((struct neigh_parms *) 0)->data[index])
3025
3026#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3027	[NEIGH_VAR_ ## attr] = { \
3028		.procname	= name, \
3029		.data		= NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3030		.maxlen		= sizeof(int), \
3031		.mode		= mval, \
3032		.proc_handler	= proc, \
3033	}
3034
3035#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3036	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3037
3038#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3039	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3040
3041#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3042	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3043
3044#define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
3045	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3046
3047#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3048	NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3049
3050#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3051	NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3052
3053static struct neigh_sysctl_table {
3054	struct ctl_table_header *sysctl_header;
3055	struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3056} neigh_sysctl_template __read_mostly = {
3057	.neigh_vars = {
3058		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3059		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3060		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3061		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3062		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3063		NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3064		NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3065		NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3066		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3067		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3068		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3069		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3070		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3071		NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3072		NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3073		NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3074		[NEIGH_VAR_GC_INTERVAL] = {
3075			.procname	= "gc_interval",
3076			.maxlen		= sizeof(int),
3077			.mode		= 0644,
3078			.proc_handler	= proc_dointvec_jiffies,
3079		},
3080		[NEIGH_VAR_GC_THRESH1] = {
3081			.procname	= "gc_thresh1",
3082			.maxlen		= sizeof(int),
3083			.mode		= 0644,
3084			.extra1 	= &zero,
3085			.extra2		= &int_max,
3086			.proc_handler	= proc_dointvec_minmax,
3087		},
3088		[NEIGH_VAR_GC_THRESH2] = {
3089			.procname	= "gc_thresh2",
3090			.maxlen		= sizeof(int),
3091			.mode		= 0644,
3092			.extra1 	= &zero,
3093			.extra2		= &int_max,
3094			.proc_handler	= proc_dointvec_minmax,
3095		},
3096		[NEIGH_VAR_GC_THRESH3] = {
3097			.procname	= "gc_thresh3",
3098			.maxlen		= sizeof(int),
3099			.mode		= 0644,
3100			.extra1 	= &zero,
3101			.extra2		= &int_max,
3102			.proc_handler	= proc_dointvec_minmax,
3103		},
3104		{},
3105	},
3106};
3107
3108int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3109			  proc_handler *handler)
3110{
3111	int i;
3112	struct neigh_sysctl_table *t;
3113	const char *dev_name_source;
3114	char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3115	char *p_name;
3116
3117	t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3118	if (!t)
3119		goto err;
3120
3121	for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3122		t->neigh_vars[i].data += (long) p;
3123		t->neigh_vars[i].extra1 = dev;
3124		t->neigh_vars[i].extra2 = p;
3125	}
3126
3127	if (dev) {
3128		dev_name_source = dev->name;
3129		/* Terminate the table early */
3130		memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3131		       sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3132	} else {
3133		struct neigh_table *tbl = p->tbl;
3134		dev_name_source = "default";
3135		t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3136		t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3137		t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3138		t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3139	}
3140
3141	if (handler) {
3142		/* RetransTime */
3143		t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3144		/* ReachableTime */
3145		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3146		/* RetransTime (in milliseconds)*/
3147		t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3148		/* ReachableTime (in milliseconds) */
3149		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3150	} else {
3151		/* Those handlers will update p->reachable_time after
3152		 * base_reachable_time(_ms) is set to ensure the new timer starts being
3153		 * applied after the next neighbour update instead of waiting for
3154		 * neigh_periodic_work to update its value (can be multiple minutes)
3155		 * So any handler that replaces them should do this as well
3156		 */
3157		/* ReachableTime */
3158		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3159			neigh_proc_base_reachable_time;
3160		/* ReachableTime (in milliseconds) */
3161		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3162			neigh_proc_base_reachable_time;
3163	}
3164
3165	/* Don't export sysctls to unprivileged users */
3166	if (neigh_parms_net(p)->user_ns != &init_user_ns)
3167		t->neigh_vars[0].procname = NULL;
3168
3169	switch (neigh_parms_family(p)) {
3170	case AF_INET:
3171	      p_name = "ipv4";
3172	      break;
3173	case AF_INET6:
3174	      p_name = "ipv6";
3175	      break;
3176	default:
3177	      BUG();
3178	}
3179
3180	snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3181		p_name, dev_name_source);
3182	t->sysctl_header =
3183		register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3184	if (!t->sysctl_header)
3185		goto free;
3186
3187	p->sysctl_table = t;
3188	return 0;
3189
3190free:
3191	kfree(t);
3192err:
3193	return -ENOBUFS;
3194}
3195EXPORT_SYMBOL(neigh_sysctl_register);
3196
3197void neigh_sysctl_unregister(struct neigh_parms *p)
3198{
3199	if (p->sysctl_table) {
3200		struct neigh_sysctl_table *t = p->sysctl_table;
3201		p->sysctl_table = NULL;
3202		unregister_net_sysctl_table(t->sysctl_header);
3203		kfree(t);
3204	}
3205}
3206EXPORT_SYMBOL(neigh_sysctl_unregister);
3207
3208#endif	/* CONFIG_SYSCTL */
3209
3210static int __init neigh_init(void)
3211{
3212	rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3213	rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3214	rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3215
3216	rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3217		      NULL);
3218	rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3219
3220	return 0;
3221}
3222
3223subsys_initcall(neigh_init);
3224
v4.17
   1/*
   2 *	Generic address resolution entity
   3 *
   4 *	Authors:
   5 *	Pedro Roque		<roque@di.fc.ul.pt>
   6 *	Alexey Kuznetsov	<kuznet@ms2.inr.ac.ru>
   7 *
   8 *	This program is free software; you can redistribute it and/or
   9 *      modify it under the terms of the GNU General Public License
  10 *      as published by the Free Software Foundation; either version
  11 *      2 of the License, or (at your option) any later version.
  12 *
  13 *	Fixes:
  14 *	Vitaly E. Lavrov	releasing NULL neighbor in neigh_add.
  15 *	Harald Welte		Add neighbour cache statistics like rtstat
  16 */
  17
  18#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20#include <linux/slab.h>
  21#include <linux/types.h>
  22#include <linux/kernel.h>
  23#include <linux/module.h>
  24#include <linux/socket.h>
  25#include <linux/netdevice.h>
  26#include <linux/proc_fs.h>
  27#ifdef CONFIG_SYSCTL
  28#include <linux/sysctl.h>
  29#endif
  30#include <linux/times.h>
  31#include <net/net_namespace.h>
  32#include <net/neighbour.h>
  33#include <net/dst.h>
  34#include <net/sock.h>
  35#include <net/netevent.h>
  36#include <net/netlink.h>
  37#include <linux/rtnetlink.h>
  38#include <linux/random.h>
  39#include <linux/string.h>
  40#include <linux/log2.h>
  41#include <linux/inetdevice.h>
  42#include <net/addrconf.h>
  43
  44#define DEBUG
  45#define NEIGH_DEBUG 1
  46#define neigh_dbg(level, fmt, ...)		\
  47do {						\
  48	if (level <= NEIGH_DEBUG)		\
  49		pr_debug(fmt, ##__VA_ARGS__);	\
  50} while (0)
  51
  52#define PNEIGH_HASHMASK		0xF
  53
  54static void neigh_timer_handler(struct timer_list *t);
  55static void __neigh_notify(struct neighbour *n, int type, int flags,
  56			   u32 pid);
  57static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
  58static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
  59				    struct net_device *dev);
  60
  61#ifdef CONFIG_PROC_FS
  62static const struct file_operations neigh_stat_seq_fops;
  63#endif
  64
  65/*
  66   Neighbour hash table buckets are protected with rwlock tbl->lock.
  67
  68   - All the scans/updates to hash buckets MUST be made under this lock.
  69   - NOTHING clever should be made under this lock: no callbacks
  70     to protocol backends, no attempts to send something to network.
  71     It will result in deadlocks, if backend/driver wants to use neighbour
  72     cache.
  73   - If the entry requires some non-trivial actions, increase
  74     its reference count and release table lock.
  75
  76   Neighbour entries are protected:
  77   - with reference count.
  78   - with rwlock neigh->lock
  79
  80   Reference count prevents destruction.
  81
  82   neigh->lock mainly serializes ll address data and its validity state.
  83   However, the same lock is used to protect another entry fields:
  84    - timer
  85    - resolution queue
  86
  87   Again, nothing clever shall be made under neigh->lock,
  88   the most complicated procedure, which we allow is dev->hard_header.
  89   It is supposed, that dev->hard_header is simplistic and does
  90   not make callbacks to neighbour tables.
  91 */
  92
  93static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
  94{
  95	kfree_skb(skb);
  96	return -ENETDOWN;
  97}
  98
  99static void neigh_cleanup_and_release(struct neighbour *neigh)
 100{
 101	if (neigh->parms->neigh_cleanup)
 102		neigh->parms->neigh_cleanup(neigh);
 103
 104	__neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
 105	call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
 106	neigh_release(neigh);
 107}
 108
 109/*
 110 * It is random distribution in the interval (1/2)*base...(3/2)*base.
 111 * It corresponds to default IPv6 settings and is not overridable,
 112 * because it is really reasonable choice.
 113 */
 114
 115unsigned long neigh_rand_reach_time(unsigned long base)
 116{
 117	return base ? (prandom_u32() % base) + (base >> 1) : 0;
 118}
 119EXPORT_SYMBOL(neigh_rand_reach_time);
 120
 121
 122static bool neigh_del(struct neighbour *n, __u8 state,
 123		      struct neighbour __rcu **np, struct neigh_table *tbl)
 124{
 125	bool retval = false;
 126
 127	write_lock(&n->lock);
 128	if (refcount_read(&n->refcnt) == 1 && !(n->nud_state & state)) {
 129		struct neighbour *neigh;
 130
 131		neigh = rcu_dereference_protected(n->next,
 132						  lockdep_is_held(&tbl->lock));
 133		rcu_assign_pointer(*np, neigh);
 134		n->dead = 1;
 135		retval = true;
 136	}
 137	write_unlock(&n->lock);
 138	if (retval)
 139		neigh_cleanup_and_release(n);
 140	return retval;
 141}
 142
 143bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
 144{
 145	struct neigh_hash_table *nht;
 146	void *pkey = ndel->primary_key;
 147	u32 hash_val;
 148	struct neighbour *n;
 149	struct neighbour __rcu **np;
 150
 151	nht = rcu_dereference_protected(tbl->nht,
 152					lockdep_is_held(&tbl->lock));
 153	hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
 154	hash_val = hash_val >> (32 - nht->hash_shift);
 155
 156	np = &nht->hash_buckets[hash_val];
 157	while ((n = rcu_dereference_protected(*np,
 158					      lockdep_is_held(&tbl->lock)))) {
 159		if (n == ndel)
 160			return neigh_del(n, 0, np, tbl);
 161		np = &n->next;
 162	}
 163	return false;
 164}
 165
 166static int neigh_forced_gc(struct neigh_table *tbl)
 167{
 168	int shrunk = 0;
 169	int i;
 170	struct neigh_hash_table *nht;
 171
 172	NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
 173
 174	write_lock_bh(&tbl->lock);
 175	nht = rcu_dereference_protected(tbl->nht,
 176					lockdep_is_held(&tbl->lock));
 177	for (i = 0; i < (1 << nht->hash_shift); i++) {
 178		struct neighbour *n;
 179		struct neighbour __rcu **np;
 180
 181		np = &nht->hash_buckets[i];
 182		while ((n = rcu_dereference_protected(*np,
 183					lockdep_is_held(&tbl->lock))) != NULL) {
 184			/* Neighbour record may be discarded if:
 185			 * - nobody refers to it.
 186			 * - it is not permanent
 187			 */
 188			if (neigh_del(n, NUD_PERMANENT, np, tbl)) {
 189				shrunk = 1;
 
 
 
 
 
 
 
 
 190				continue;
 191			}
 
 192			np = &n->next;
 193		}
 194	}
 195
 196	tbl->last_flush = jiffies;
 197
 198	write_unlock_bh(&tbl->lock);
 199
 200	return shrunk;
 201}
 202
 203static void neigh_add_timer(struct neighbour *n, unsigned long when)
 204{
 205	neigh_hold(n);
 206	if (unlikely(mod_timer(&n->timer, when))) {
 207		printk("NEIGH: BUG, double timer add, state is %x\n",
 208		       n->nud_state);
 209		dump_stack();
 210	}
 211}
 212
 213static int neigh_del_timer(struct neighbour *n)
 214{
 215	if ((n->nud_state & NUD_IN_TIMER) &&
 216	    del_timer(&n->timer)) {
 217		neigh_release(n);
 218		return 1;
 219	}
 220	return 0;
 221}
 222
 223static void pneigh_queue_purge(struct sk_buff_head *list)
 224{
 225	struct sk_buff *skb;
 226
 227	while ((skb = skb_dequeue(list)) != NULL) {
 228		dev_put(skb->dev);
 229		kfree_skb(skb);
 230	}
 231}
 232
 233static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
 234{
 235	int i;
 236	struct neigh_hash_table *nht;
 237
 238	nht = rcu_dereference_protected(tbl->nht,
 239					lockdep_is_held(&tbl->lock));
 240
 241	for (i = 0; i < (1 << nht->hash_shift); i++) {
 242		struct neighbour *n;
 243		struct neighbour __rcu **np = &nht->hash_buckets[i];
 244
 245		while ((n = rcu_dereference_protected(*np,
 246					lockdep_is_held(&tbl->lock))) != NULL) {
 247			if (dev && n->dev != dev) {
 248				np = &n->next;
 249				continue;
 250			}
 251			rcu_assign_pointer(*np,
 252				   rcu_dereference_protected(n->next,
 253						lockdep_is_held(&tbl->lock)));
 254			write_lock(&n->lock);
 255			neigh_del_timer(n);
 256			n->dead = 1;
 257
 258			if (refcount_read(&n->refcnt) != 1) {
 259				/* The most unpleasant situation.
 260				   We must destroy neighbour entry,
 261				   but someone still uses it.
 262
 263				   The destroy will be delayed until
 264				   the last user releases us, but
 265				   we must kill timers etc. and move
 266				   it to safe state.
 267				 */
 268				__skb_queue_purge(&n->arp_queue);
 269				n->arp_queue_len_bytes = 0;
 270				n->output = neigh_blackhole;
 271				if (n->nud_state & NUD_VALID)
 272					n->nud_state = NUD_NOARP;
 273				else
 274					n->nud_state = NUD_NONE;
 275				neigh_dbg(2, "neigh %p is stray\n", n);
 276			}
 277			write_unlock(&n->lock);
 278			neigh_cleanup_and_release(n);
 279		}
 280	}
 281}
 282
 283void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
 284{
 285	write_lock_bh(&tbl->lock);
 286	neigh_flush_dev(tbl, dev);
 287	write_unlock_bh(&tbl->lock);
 288}
 289EXPORT_SYMBOL(neigh_changeaddr);
 290
 291int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
 292{
 293	write_lock_bh(&tbl->lock);
 294	neigh_flush_dev(tbl, dev);
 295	pneigh_ifdown_and_unlock(tbl, dev);
 
 296
 297	del_timer_sync(&tbl->proxy_timer);
 298	pneigh_queue_purge(&tbl->proxy_queue);
 299	return 0;
 300}
 301EXPORT_SYMBOL(neigh_ifdown);
 302
 303static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
 304{
 305	struct neighbour *n = NULL;
 306	unsigned long now = jiffies;
 307	int entries;
 308
 309	entries = atomic_inc_return(&tbl->entries) - 1;
 310	if (entries >= tbl->gc_thresh3 ||
 311	    (entries >= tbl->gc_thresh2 &&
 312	     time_after(now, tbl->last_flush + 5 * HZ))) {
 313		if (!neigh_forced_gc(tbl) &&
 314		    entries >= tbl->gc_thresh3) {
 315			net_info_ratelimited("%s: neighbor table overflow!\n",
 316					     tbl->id);
 317			NEIGH_CACHE_STAT_INC(tbl, table_fulls);
 318			goto out_entries;
 319		}
 320	}
 321
 322	n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
 323	if (!n)
 324		goto out_entries;
 325
 326	__skb_queue_head_init(&n->arp_queue);
 327	rwlock_init(&n->lock);
 328	seqlock_init(&n->ha_lock);
 329	n->updated	  = n->used = now;
 330	n->nud_state	  = NUD_NONE;
 331	n->output	  = neigh_blackhole;
 332	seqlock_init(&n->hh.hh_lock);
 333	n->parms	  = neigh_parms_clone(&tbl->parms);
 334	timer_setup(&n->timer, neigh_timer_handler, 0);
 335
 336	NEIGH_CACHE_STAT_INC(tbl, allocs);
 337	n->tbl		  = tbl;
 338	refcount_set(&n->refcnt, 1);
 339	n->dead		  = 1;
 340out:
 341	return n;
 342
 343out_entries:
 344	atomic_dec(&tbl->entries);
 345	goto out;
 346}
 347
 348static void neigh_get_hash_rnd(u32 *x)
 349{
 350	*x = get_random_u32() | 1;
 
 351}
 352
 353static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 354{
 355	size_t size = (1 << shift) * sizeof(struct neighbour *);
 356	struct neigh_hash_table *ret;
 357	struct neighbour __rcu **buckets;
 358	int i;
 359
 360	ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
 361	if (!ret)
 362		return NULL;
 363	if (size <= PAGE_SIZE)
 364		buckets = kzalloc(size, GFP_ATOMIC);
 365	else
 366		buckets = (struct neighbour __rcu **)
 367			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 368					   get_order(size));
 369	if (!buckets) {
 370		kfree(ret);
 371		return NULL;
 372	}
 373	ret->hash_buckets = buckets;
 374	ret->hash_shift = shift;
 375	for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 376		neigh_get_hash_rnd(&ret->hash_rnd[i]);
 377	return ret;
 378}
 379
 380static void neigh_hash_free_rcu(struct rcu_head *head)
 381{
 382	struct neigh_hash_table *nht = container_of(head,
 383						    struct neigh_hash_table,
 384						    rcu);
 385	size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 386	struct neighbour __rcu **buckets = nht->hash_buckets;
 387
 388	if (size <= PAGE_SIZE)
 389		kfree(buckets);
 390	else
 391		free_pages((unsigned long)buckets, get_order(size));
 392	kfree(nht);
 393}
 394
 395static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 396						unsigned long new_shift)
 397{
 398	unsigned int i, hash;
 399	struct neigh_hash_table *new_nht, *old_nht;
 400
 401	NEIGH_CACHE_STAT_INC(tbl, hash_grows);
 402
 403	old_nht = rcu_dereference_protected(tbl->nht,
 404					    lockdep_is_held(&tbl->lock));
 405	new_nht = neigh_hash_alloc(new_shift);
 406	if (!new_nht)
 407		return old_nht;
 408
 409	for (i = 0; i < (1 << old_nht->hash_shift); i++) {
 410		struct neighbour *n, *next;
 411
 412		for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
 413						   lockdep_is_held(&tbl->lock));
 414		     n != NULL;
 415		     n = next) {
 416			hash = tbl->hash(n->primary_key, n->dev,
 417					 new_nht->hash_rnd);
 418
 419			hash >>= (32 - new_nht->hash_shift);
 420			next = rcu_dereference_protected(n->next,
 421						lockdep_is_held(&tbl->lock));
 422
 423			rcu_assign_pointer(n->next,
 424					   rcu_dereference_protected(
 425						new_nht->hash_buckets[hash],
 426						lockdep_is_held(&tbl->lock)));
 427			rcu_assign_pointer(new_nht->hash_buckets[hash], n);
 428		}
 429	}
 430
 431	rcu_assign_pointer(tbl->nht, new_nht);
 432	call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
 433	return new_nht;
 434}
 435
 436struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
 437			       struct net_device *dev)
 438{
 439	struct neighbour *n;
 440
 441	NEIGH_CACHE_STAT_INC(tbl, lookups);
 442
 443	rcu_read_lock_bh();
 444	n = __neigh_lookup_noref(tbl, pkey, dev);
 445	if (n) {
 446		if (!refcount_inc_not_zero(&n->refcnt))
 447			n = NULL;
 448		NEIGH_CACHE_STAT_INC(tbl, hits);
 449	}
 450
 451	rcu_read_unlock_bh();
 452	return n;
 453}
 454EXPORT_SYMBOL(neigh_lookup);
 455
 456struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
 457				     const void *pkey)
 458{
 459	struct neighbour *n;
 460	unsigned int key_len = tbl->key_len;
 461	u32 hash_val;
 462	struct neigh_hash_table *nht;
 463
 464	NEIGH_CACHE_STAT_INC(tbl, lookups);
 465
 466	rcu_read_lock_bh();
 467	nht = rcu_dereference_bh(tbl->nht);
 468	hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
 469
 470	for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
 471	     n != NULL;
 472	     n = rcu_dereference_bh(n->next)) {
 473		if (!memcmp(n->primary_key, pkey, key_len) &&
 474		    net_eq(dev_net(n->dev), net)) {
 475			if (!refcount_inc_not_zero(&n->refcnt))
 476				n = NULL;
 477			NEIGH_CACHE_STAT_INC(tbl, hits);
 478			break;
 479		}
 480	}
 481
 482	rcu_read_unlock_bh();
 483	return n;
 484}
 485EXPORT_SYMBOL(neigh_lookup_nodev);
 486
 487struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
 488				 struct net_device *dev, bool want_ref)
 489{
 490	u32 hash_val;
 491	unsigned int key_len = tbl->key_len;
 492	int error;
 493	struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
 494	struct neigh_hash_table *nht;
 495
 496	if (!n) {
 497		rc = ERR_PTR(-ENOBUFS);
 498		goto out;
 499	}
 500
 501	memcpy(n->primary_key, pkey, key_len);
 502	n->dev = dev;
 503	dev_hold(dev);
 504
 505	/* Protocol specific setup. */
 506	if (tbl->constructor &&	(error = tbl->constructor(n)) < 0) {
 507		rc = ERR_PTR(error);
 508		goto out_neigh_release;
 509	}
 510
 511	if (dev->netdev_ops->ndo_neigh_construct) {
 512		error = dev->netdev_ops->ndo_neigh_construct(dev, n);
 513		if (error < 0) {
 514			rc = ERR_PTR(error);
 515			goto out_neigh_release;
 516		}
 517	}
 518
 519	/* Device specific setup. */
 520	if (n->parms->neigh_setup &&
 521	    (error = n->parms->neigh_setup(n)) < 0) {
 522		rc = ERR_PTR(error);
 523		goto out_neigh_release;
 524	}
 525
 526	n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
 527
 528	write_lock_bh(&tbl->lock);
 529	nht = rcu_dereference_protected(tbl->nht,
 530					lockdep_is_held(&tbl->lock));
 531
 532	if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
 533		nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
 534
 535	hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
 536
 537	if (n->parms->dead) {
 538		rc = ERR_PTR(-EINVAL);
 539		goto out_tbl_unlock;
 540	}
 541
 542	for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
 543					    lockdep_is_held(&tbl->lock));
 544	     n1 != NULL;
 545	     n1 = rcu_dereference_protected(n1->next,
 546			lockdep_is_held(&tbl->lock))) {
 547		if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
 548			if (want_ref)
 549				neigh_hold(n1);
 550			rc = n1;
 551			goto out_tbl_unlock;
 552		}
 553	}
 554
 555	n->dead = 0;
 556	if (want_ref)
 557		neigh_hold(n);
 558	rcu_assign_pointer(n->next,
 559			   rcu_dereference_protected(nht->hash_buckets[hash_val],
 560						     lockdep_is_held(&tbl->lock)));
 561	rcu_assign_pointer(nht->hash_buckets[hash_val], n);
 562	write_unlock_bh(&tbl->lock);
 563	neigh_dbg(2, "neigh %p is created\n", n);
 564	rc = n;
 565out:
 566	return rc;
 567out_tbl_unlock:
 568	write_unlock_bh(&tbl->lock);
 569out_neigh_release:
 570	neigh_release(n);
 571	goto out;
 572}
 573EXPORT_SYMBOL(__neigh_create);
 574
 575static u32 pneigh_hash(const void *pkey, unsigned int key_len)
 576{
 577	u32 hash_val = *(u32 *)(pkey + key_len - 4);
 578	hash_val ^= (hash_val >> 16);
 579	hash_val ^= hash_val >> 8;
 580	hash_val ^= hash_val >> 4;
 581	hash_val &= PNEIGH_HASHMASK;
 582	return hash_val;
 583}
 584
 585static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
 586					      struct net *net,
 587					      const void *pkey,
 588					      unsigned int key_len,
 589					      struct net_device *dev)
 590{
 591	while (n) {
 592		if (!memcmp(n->key, pkey, key_len) &&
 593		    net_eq(pneigh_net(n), net) &&
 594		    (n->dev == dev || !n->dev))
 595			return n;
 596		n = n->next;
 597	}
 598	return NULL;
 599}
 600
 601struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
 602		struct net *net, const void *pkey, struct net_device *dev)
 603{
 604	unsigned int key_len = tbl->key_len;
 605	u32 hash_val = pneigh_hash(pkey, key_len);
 606
 607	return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 608				 net, pkey, key_len, dev);
 609}
 610EXPORT_SYMBOL_GPL(__pneigh_lookup);
 611
 612struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
 613				    struct net *net, const void *pkey,
 614				    struct net_device *dev, int creat)
 615{
 616	struct pneigh_entry *n;
 617	unsigned int key_len = tbl->key_len;
 618	u32 hash_val = pneigh_hash(pkey, key_len);
 619
 620	read_lock_bh(&tbl->lock);
 621	n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
 622			      net, pkey, key_len, dev);
 623	read_unlock_bh(&tbl->lock);
 624
 625	if (n || !creat)
 626		goto out;
 627
 628	ASSERT_RTNL();
 629
 630	n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
 631	if (!n)
 632		goto out;
 633
 634	write_pnet(&n->net, net);
 635	memcpy(n->key, pkey, key_len);
 636	n->dev = dev;
 637	if (dev)
 638		dev_hold(dev);
 639
 640	if (tbl->pconstructor && tbl->pconstructor(n)) {
 641		if (dev)
 642			dev_put(dev);
 643		kfree(n);
 644		n = NULL;
 645		goto out;
 646	}
 647
 648	write_lock_bh(&tbl->lock);
 649	n->next = tbl->phash_buckets[hash_val];
 650	tbl->phash_buckets[hash_val] = n;
 651	write_unlock_bh(&tbl->lock);
 652out:
 653	return n;
 654}
 655EXPORT_SYMBOL(pneigh_lookup);
 656
 657
 658int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
 659		  struct net_device *dev)
 660{
 661	struct pneigh_entry *n, **np;
 662	unsigned int key_len = tbl->key_len;
 663	u32 hash_val = pneigh_hash(pkey, key_len);
 664
 665	write_lock_bh(&tbl->lock);
 666	for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
 667	     np = &n->next) {
 668		if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
 669		    net_eq(pneigh_net(n), net)) {
 670			*np = n->next;
 671			write_unlock_bh(&tbl->lock);
 672			if (tbl->pdestructor)
 673				tbl->pdestructor(n);
 674			if (n->dev)
 675				dev_put(n->dev);
 676			kfree(n);
 677			return 0;
 678		}
 679	}
 680	write_unlock_bh(&tbl->lock);
 681	return -ENOENT;
 682}
 683
 684static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
 685				    struct net_device *dev)
 686{
 687	struct pneigh_entry *n, **np, *freelist = NULL;
 688	u32 h;
 689
 690	for (h = 0; h <= PNEIGH_HASHMASK; h++) {
 691		np = &tbl->phash_buckets[h];
 692		while ((n = *np) != NULL) {
 693			if (!dev || n->dev == dev) {
 694				*np = n->next;
 695				n->next = freelist;
 696				freelist = n;
 
 
 
 697				continue;
 698			}
 699			np = &n->next;
 700		}
 701	}
 702	write_unlock_bh(&tbl->lock);
 703	while ((n = freelist)) {
 704		freelist = n->next;
 705		n->next = NULL;
 706		if (tbl->pdestructor)
 707			tbl->pdestructor(n);
 708		if (n->dev)
 709			dev_put(n->dev);
 710		kfree(n);
 711	}
 712	return -ENOENT;
 713}
 714
 715static void neigh_parms_destroy(struct neigh_parms *parms);
 716
 717static inline void neigh_parms_put(struct neigh_parms *parms)
 718{
 719	if (refcount_dec_and_test(&parms->refcnt))
 720		neigh_parms_destroy(parms);
 721}
 722
 723/*
 724 *	neighbour must already be out of the table;
 725 *
 726 */
 727void neigh_destroy(struct neighbour *neigh)
 728{
 729	struct net_device *dev = neigh->dev;
 730
 731	NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
 732
 733	if (!neigh->dead) {
 734		pr_warn("Destroying alive neighbour %p\n", neigh);
 735		dump_stack();
 736		return;
 737	}
 738
 739	if (neigh_del_timer(neigh))
 740		pr_warn("Impossible event\n");
 741
 742	write_lock_bh(&neigh->lock);
 743	__skb_queue_purge(&neigh->arp_queue);
 744	write_unlock_bh(&neigh->lock);
 745	neigh->arp_queue_len_bytes = 0;
 746
 747	if (dev->netdev_ops->ndo_neigh_destroy)
 748		dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
 749
 750	dev_put(dev);
 751	neigh_parms_put(neigh->parms);
 752
 753	neigh_dbg(2, "neigh %p is destroyed\n", neigh);
 754
 755	atomic_dec(&neigh->tbl->entries);
 756	kfree_rcu(neigh, rcu);
 757}
 758EXPORT_SYMBOL(neigh_destroy);
 759
 760/* Neighbour state is suspicious;
 761   disable fast path.
 762
 763   Called with write_locked neigh.
 764 */
 765static void neigh_suspect(struct neighbour *neigh)
 766{
 767	neigh_dbg(2, "neigh %p is suspected\n", neigh);
 768
 769	neigh->output = neigh->ops->output;
 770}
 771
 772/* Neighbour state is OK;
 773   enable fast path.
 774
 775   Called with write_locked neigh.
 776 */
 777static void neigh_connect(struct neighbour *neigh)
 778{
 779	neigh_dbg(2, "neigh %p is connected\n", neigh);
 780
 781	neigh->output = neigh->ops->connected_output;
 782}
 783
 784static void neigh_periodic_work(struct work_struct *work)
 785{
 786	struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
 787	struct neighbour *n;
 788	struct neighbour __rcu **np;
 789	unsigned int i;
 790	struct neigh_hash_table *nht;
 791
 792	NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
 793
 794	write_lock_bh(&tbl->lock);
 795	nht = rcu_dereference_protected(tbl->nht,
 796					lockdep_is_held(&tbl->lock));
 797
 798	/*
 799	 *	periodically recompute ReachableTime from random function
 800	 */
 801
 802	if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
 803		struct neigh_parms *p;
 804		tbl->last_rand = jiffies;
 805		list_for_each_entry(p, &tbl->parms_list, list)
 806			p->reachable_time =
 807				neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
 808	}
 809
 810	if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
 811		goto out;
 812
 813	for (i = 0 ; i < (1 << nht->hash_shift); i++) {
 814		np = &nht->hash_buckets[i];
 815
 816		while ((n = rcu_dereference_protected(*np,
 817				lockdep_is_held(&tbl->lock))) != NULL) {
 818			unsigned int state;
 819
 820			write_lock(&n->lock);
 821
 822			state = n->nud_state;
 823			if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
 824				write_unlock(&n->lock);
 825				goto next_elt;
 826			}
 827
 828			if (time_before(n->used, n->confirmed))
 829				n->used = n->confirmed;
 830
 831			if (refcount_read(&n->refcnt) == 1 &&
 832			    (state == NUD_FAILED ||
 833			     time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
 834				*np = n->next;
 835				n->dead = 1;
 836				write_unlock(&n->lock);
 837				neigh_cleanup_and_release(n);
 838				continue;
 839			}
 840			write_unlock(&n->lock);
 841
 842next_elt:
 843			np = &n->next;
 844		}
 845		/*
 846		 * It's fine to release lock here, even if hash table
 847		 * grows while we are preempted.
 848		 */
 849		write_unlock_bh(&tbl->lock);
 850		cond_resched();
 851		write_lock_bh(&tbl->lock);
 852		nht = rcu_dereference_protected(tbl->nht,
 853						lockdep_is_held(&tbl->lock));
 854	}
 855out:
 856	/* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
 857	 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
 858	 * BASE_REACHABLE_TIME.
 859	 */
 860	queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
 861			      NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
 862	write_unlock_bh(&tbl->lock);
 863}
 864
 865static __inline__ int neigh_max_probes(struct neighbour *n)
 866{
 867	struct neigh_parms *p = n->parms;
 868	return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
 869	       (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
 870	        NEIGH_VAR(p, MCAST_PROBES));
 871}
 872
 873static void neigh_invalidate(struct neighbour *neigh)
 874	__releases(neigh->lock)
 875	__acquires(neigh->lock)
 876{
 877	struct sk_buff *skb;
 878
 879	NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
 880	neigh_dbg(2, "neigh %p is failed\n", neigh);
 881	neigh->updated = jiffies;
 882
 883	/* It is very thin place. report_unreachable is very complicated
 884	   routine. Particularly, it can hit the same neighbour entry!
 885
 886	   So that, we try to be accurate and avoid dead loop. --ANK
 887	 */
 888	while (neigh->nud_state == NUD_FAILED &&
 889	       (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
 890		write_unlock(&neigh->lock);
 891		neigh->ops->error_report(neigh, skb);
 892		write_lock(&neigh->lock);
 893	}
 894	__skb_queue_purge(&neigh->arp_queue);
 895	neigh->arp_queue_len_bytes = 0;
 896}
 897
 898static void neigh_probe(struct neighbour *neigh)
 899	__releases(neigh->lock)
 900{
 901	struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
 902	/* keep skb alive even if arp_queue overflows */
 903	if (skb)
 904		skb = skb_clone(skb, GFP_ATOMIC);
 905	write_unlock(&neigh->lock);
 906	if (neigh->ops->solicit)
 907		neigh->ops->solicit(neigh, skb);
 908	atomic_inc(&neigh->probes);
 909	kfree_skb(skb);
 910}
 911
 912/* Called when a timer expires for a neighbour entry. */
 913
 914static void neigh_timer_handler(struct timer_list *t)
 915{
 916	unsigned long now, next;
 917	struct neighbour *neigh = from_timer(neigh, t, timer);
 918	unsigned int state;
 919	int notify = 0;
 920
 921	write_lock(&neigh->lock);
 922
 923	state = neigh->nud_state;
 924	now = jiffies;
 925	next = now + HZ;
 926
 927	if (!(state & NUD_IN_TIMER))
 928		goto out;
 929
 930	if (state & NUD_REACHABLE) {
 931		if (time_before_eq(now,
 932				   neigh->confirmed + neigh->parms->reachable_time)) {
 933			neigh_dbg(2, "neigh %p is still alive\n", neigh);
 934			next = neigh->confirmed + neigh->parms->reachable_time;
 935		} else if (time_before_eq(now,
 936					  neigh->used +
 937					  NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
 938			neigh_dbg(2, "neigh %p is delayed\n", neigh);
 939			neigh->nud_state = NUD_DELAY;
 940			neigh->updated = jiffies;
 941			neigh_suspect(neigh);
 942			next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
 943		} else {
 944			neigh_dbg(2, "neigh %p is suspected\n", neigh);
 945			neigh->nud_state = NUD_STALE;
 946			neigh->updated = jiffies;
 947			neigh_suspect(neigh);
 948			notify = 1;
 949		}
 950	} else if (state & NUD_DELAY) {
 951		if (time_before_eq(now,
 952				   neigh->confirmed +
 953				   NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
 954			neigh_dbg(2, "neigh %p is now reachable\n", neigh);
 955			neigh->nud_state = NUD_REACHABLE;
 956			neigh->updated = jiffies;
 957			neigh_connect(neigh);
 958			notify = 1;
 959			next = neigh->confirmed + neigh->parms->reachable_time;
 960		} else {
 961			neigh_dbg(2, "neigh %p is probed\n", neigh);
 962			neigh->nud_state = NUD_PROBE;
 963			neigh->updated = jiffies;
 964			atomic_set(&neigh->probes, 0);
 965			notify = 1;
 966			next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
 967		}
 968	} else {
 969		/* NUD_PROBE|NUD_INCOMPLETE */
 970		next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
 971	}
 972
 973	if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
 974	    atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
 975		neigh->nud_state = NUD_FAILED;
 976		notify = 1;
 977		neigh_invalidate(neigh);
 978		goto out;
 979	}
 980
 981	if (neigh->nud_state & NUD_IN_TIMER) {
 982		if (time_before(next, jiffies + HZ/2))
 983			next = jiffies + HZ/2;
 984		if (!mod_timer(&neigh->timer, next))
 985			neigh_hold(neigh);
 986	}
 987	if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
 988		neigh_probe(neigh);
 989	} else {
 990out:
 991		write_unlock(&neigh->lock);
 992	}
 993
 994	if (notify)
 995		neigh_update_notify(neigh, 0);
 996
 997	neigh_release(neigh);
 998}
 999
1000int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1001{
1002	int rc;
1003	bool immediate_probe = false;
1004
1005	write_lock_bh(&neigh->lock);
1006
1007	rc = 0;
1008	if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1009		goto out_unlock_bh;
1010	if (neigh->dead)
1011		goto out_dead;
1012
1013	if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1014		if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1015		    NEIGH_VAR(neigh->parms, APP_PROBES)) {
1016			unsigned long next, now = jiffies;
1017
1018			atomic_set(&neigh->probes,
1019				   NEIGH_VAR(neigh->parms, UCAST_PROBES));
1020			neigh->nud_state     = NUD_INCOMPLETE;
1021			neigh->updated = now;
1022			next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1023					 HZ/2);
1024			neigh_add_timer(neigh, next);
1025			immediate_probe = true;
1026		} else {
1027			neigh->nud_state = NUD_FAILED;
1028			neigh->updated = jiffies;
1029			write_unlock_bh(&neigh->lock);
1030
1031			kfree_skb(skb);
1032			return 1;
1033		}
1034	} else if (neigh->nud_state & NUD_STALE) {
1035		neigh_dbg(2, "neigh %p is delayed\n", neigh);
1036		neigh->nud_state = NUD_DELAY;
1037		neigh->updated = jiffies;
1038		neigh_add_timer(neigh, jiffies +
1039				NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1040	}
1041
1042	if (neigh->nud_state == NUD_INCOMPLETE) {
1043		if (skb) {
1044			while (neigh->arp_queue_len_bytes + skb->truesize >
1045			       NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1046				struct sk_buff *buff;
1047
1048				buff = __skb_dequeue(&neigh->arp_queue);
1049				if (!buff)
1050					break;
1051				neigh->arp_queue_len_bytes -= buff->truesize;
1052				kfree_skb(buff);
1053				NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1054			}
1055			skb_dst_force(skb);
1056			__skb_queue_tail(&neigh->arp_queue, skb);
1057			neigh->arp_queue_len_bytes += skb->truesize;
1058		}
1059		rc = 1;
1060	}
1061out_unlock_bh:
1062	if (immediate_probe)
1063		neigh_probe(neigh);
1064	else
1065		write_unlock(&neigh->lock);
1066	local_bh_enable();
1067	return rc;
1068
1069out_dead:
1070	if (neigh->nud_state & NUD_STALE)
1071		goto out_unlock_bh;
1072	write_unlock_bh(&neigh->lock);
1073	kfree_skb(skb);
1074	return 1;
1075}
1076EXPORT_SYMBOL(__neigh_event_send);
1077
1078static void neigh_update_hhs(struct neighbour *neigh)
1079{
1080	struct hh_cache *hh;
1081	void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1082		= NULL;
1083
1084	if (neigh->dev->header_ops)
1085		update = neigh->dev->header_ops->cache_update;
1086
1087	if (update) {
1088		hh = &neigh->hh;
1089		if (hh->hh_len) {
1090			write_seqlock_bh(&hh->hh_lock);
1091			update(hh, neigh->dev, neigh->ha);
1092			write_sequnlock_bh(&hh->hh_lock);
1093		}
1094	}
1095}
1096
1097
1098
1099/* Generic update routine.
1100   -- lladdr is new lladdr or NULL, if it is not supplied.
1101   -- new    is new state.
1102   -- flags
1103	NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1104				if it is different.
1105	NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1106				lladdr instead of overriding it
1107				if it is different.
 
 
1108	NEIGH_UPDATE_F_ADMIN	means that the change is administrative.
1109
1110	NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1111				NTF_ROUTER flag.
1112	NEIGH_UPDATE_F_ISROUTER	indicates if the neighbour is known as
1113				a router.
1114
1115   Caller MUST hold reference count on the entry.
1116 */
1117
1118int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1119		 u32 flags, u32 nlmsg_pid)
1120{
1121	u8 old;
1122	int err;
1123	int notify = 0;
1124	struct net_device *dev;
1125	int update_isrouter = 0;
1126
1127	write_lock_bh(&neigh->lock);
1128
1129	dev    = neigh->dev;
1130	old    = neigh->nud_state;
1131	err    = -EPERM;
1132
1133	if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1134	    (old & (NUD_NOARP | NUD_PERMANENT)))
1135		goto out;
1136	if (neigh->dead)
1137		goto out;
1138
1139	if (!(new & NUD_VALID)) {
1140		neigh_del_timer(neigh);
1141		if (old & NUD_CONNECTED)
1142			neigh_suspect(neigh);
1143		neigh->nud_state = new;
1144		err = 0;
1145		notify = old & NUD_VALID;
1146		if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1147		    (new & NUD_FAILED)) {
1148			neigh_invalidate(neigh);
1149			notify = 1;
1150		}
1151		goto out;
1152	}
1153
1154	/* Compare new lladdr with cached one */
1155	if (!dev->addr_len) {
1156		/* First case: device needs no address. */
1157		lladdr = neigh->ha;
1158	} else if (lladdr) {
1159		/* The second case: if something is already cached
1160		   and a new address is proposed:
1161		   - compare new & old
1162		   - if they are different, check override flag
1163		 */
1164		if ((old & NUD_VALID) &&
1165		    !memcmp(lladdr, neigh->ha, dev->addr_len))
1166			lladdr = neigh->ha;
1167	} else {
1168		/* No address is supplied; if we know something,
1169		   use it, otherwise discard the request.
1170		 */
1171		err = -EINVAL;
1172		if (!(old & NUD_VALID))
1173			goto out;
1174		lladdr = neigh->ha;
1175	}
1176
 
 
 
 
1177	/* If entry was valid and address is not changed,
1178	   do not change entry state, if new one is STALE.
1179	 */
1180	err = 0;
1181	update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1182	if (old & NUD_VALID) {
1183		if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1184			update_isrouter = 0;
1185			if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1186			    (old & NUD_CONNECTED)) {
1187				lladdr = neigh->ha;
1188				new = NUD_STALE;
1189			} else
1190				goto out;
1191		} else {
1192			if (lladdr == neigh->ha && new == NUD_STALE &&
1193			    !(flags & NEIGH_UPDATE_F_ADMIN))
 
 
1194				new = old;
1195		}
1196	}
1197
1198	/* Update timestamps only once we know we will make a change to the
1199	 * neighbour entry. Otherwise we risk to move the locktime window with
1200	 * noop updates and ignore relevant ARP updates.
1201	 */
1202	if (new != old || lladdr != neigh->ha) {
1203		if (new & NUD_CONNECTED)
1204			neigh->confirmed = jiffies;
1205		neigh->updated = jiffies;
1206	}
1207
1208	if (new != old) {
1209		neigh_del_timer(neigh);
1210		if (new & NUD_PROBE)
1211			atomic_set(&neigh->probes, 0);
1212		if (new & NUD_IN_TIMER)
1213			neigh_add_timer(neigh, (jiffies +
1214						((new & NUD_REACHABLE) ?
1215						 neigh->parms->reachable_time :
1216						 0)));
1217		neigh->nud_state = new;
1218		notify = 1;
1219	}
1220
1221	if (lladdr != neigh->ha) {
1222		write_seqlock(&neigh->ha_lock);
1223		memcpy(&neigh->ha, lladdr, dev->addr_len);
1224		write_sequnlock(&neigh->ha_lock);
1225		neigh_update_hhs(neigh);
1226		if (!(new & NUD_CONNECTED))
1227			neigh->confirmed = jiffies -
1228				      (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1229		notify = 1;
1230	}
1231	if (new == old)
1232		goto out;
1233	if (new & NUD_CONNECTED)
1234		neigh_connect(neigh);
1235	else
1236		neigh_suspect(neigh);
1237	if (!(old & NUD_VALID)) {
1238		struct sk_buff *skb;
1239
1240		/* Again: avoid dead loop if something went wrong */
1241
1242		while (neigh->nud_state & NUD_VALID &&
1243		       (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1244			struct dst_entry *dst = skb_dst(skb);
1245			struct neighbour *n2, *n1 = neigh;
1246			write_unlock_bh(&neigh->lock);
1247
1248			rcu_read_lock();
1249
1250			/* Why not just use 'neigh' as-is?  The problem is that
1251			 * things such as shaper, eql, and sch_teql can end up
1252			 * using alternative, different, neigh objects to output
1253			 * the packet in the output path.  So what we need to do
1254			 * here is re-lookup the top-level neigh in the path so
1255			 * we can reinject the packet there.
1256			 */
1257			n2 = NULL;
1258			if (dst) {
1259				n2 = dst_neigh_lookup_skb(dst, skb);
1260				if (n2)
1261					n1 = n2;
1262			}
1263			n1->output(n1, skb);
1264			if (n2)
1265				neigh_release(n2);
1266			rcu_read_unlock();
1267
1268			write_lock_bh(&neigh->lock);
1269		}
1270		__skb_queue_purge(&neigh->arp_queue);
1271		neigh->arp_queue_len_bytes = 0;
1272	}
1273out:
1274	if (update_isrouter) {
1275		neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1276			(neigh->flags | NTF_ROUTER) :
1277			(neigh->flags & ~NTF_ROUTER);
1278	}
1279	write_unlock_bh(&neigh->lock);
1280
1281	if (notify)
1282		neigh_update_notify(neigh, nlmsg_pid);
1283
1284	return err;
1285}
1286EXPORT_SYMBOL(neigh_update);
1287
1288/* Update the neigh to listen temporarily for probe responses, even if it is
1289 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1290 */
1291void __neigh_set_probe_once(struct neighbour *neigh)
1292{
1293	if (neigh->dead)
1294		return;
1295	neigh->updated = jiffies;
1296	if (!(neigh->nud_state & NUD_FAILED))
1297		return;
1298	neigh->nud_state = NUD_INCOMPLETE;
1299	atomic_set(&neigh->probes, neigh_max_probes(neigh));
1300	neigh_add_timer(neigh,
1301			jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1302}
1303EXPORT_SYMBOL(__neigh_set_probe_once);
1304
1305struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1306				 u8 *lladdr, void *saddr,
1307				 struct net_device *dev)
1308{
1309	struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1310						 lladdr || !dev->addr_len);
1311	if (neigh)
1312		neigh_update(neigh, lladdr, NUD_STALE,
1313			     NEIGH_UPDATE_F_OVERRIDE, 0);
1314	return neigh;
1315}
1316EXPORT_SYMBOL(neigh_event_ns);
1317
1318/* called with read_lock_bh(&n->lock); */
1319static void neigh_hh_init(struct neighbour *n)
1320{
1321	struct net_device *dev = n->dev;
1322	__be16 prot = n->tbl->protocol;
1323	struct hh_cache	*hh = &n->hh;
1324
1325	write_lock_bh(&n->lock);
1326
1327	/* Only one thread can come in here and initialize the
1328	 * hh_cache entry.
1329	 */
1330	if (!hh->hh_len)
1331		dev->header_ops->cache(n, hh, prot);
1332
1333	write_unlock_bh(&n->lock);
1334}
1335
1336/* Slow and careful. */
1337
1338int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1339{
1340	int rc = 0;
1341
1342	if (!neigh_event_send(neigh, skb)) {
1343		int err;
1344		struct net_device *dev = neigh->dev;
1345		unsigned int seq;
1346
1347		if (dev->header_ops->cache && !neigh->hh.hh_len)
1348			neigh_hh_init(neigh);
1349
1350		do {
1351			__skb_pull(skb, skb_network_offset(skb));
1352			seq = read_seqbegin(&neigh->ha_lock);
1353			err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1354					      neigh->ha, NULL, skb->len);
1355		} while (read_seqretry(&neigh->ha_lock, seq));
1356
1357		if (err >= 0)
1358			rc = dev_queue_xmit(skb);
1359		else
1360			goto out_kfree_skb;
1361	}
1362out:
1363	return rc;
1364out_kfree_skb:
1365	rc = -EINVAL;
1366	kfree_skb(skb);
1367	goto out;
1368}
1369EXPORT_SYMBOL(neigh_resolve_output);
1370
1371/* As fast as possible without hh cache */
1372
1373int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1374{
1375	struct net_device *dev = neigh->dev;
1376	unsigned int seq;
1377	int err;
1378
1379	do {
1380		__skb_pull(skb, skb_network_offset(skb));
1381		seq = read_seqbegin(&neigh->ha_lock);
1382		err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1383				      neigh->ha, NULL, skb->len);
1384	} while (read_seqretry(&neigh->ha_lock, seq));
1385
1386	if (err >= 0)
1387		err = dev_queue_xmit(skb);
1388	else {
1389		err = -EINVAL;
1390		kfree_skb(skb);
1391	}
1392	return err;
1393}
1394EXPORT_SYMBOL(neigh_connected_output);
1395
1396int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1397{
1398	return dev_queue_xmit(skb);
1399}
1400EXPORT_SYMBOL(neigh_direct_output);
1401
1402static void neigh_proxy_process(struct timer_list *t)
1403{
1404	struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1405	long sched_next = 0;
1406	unsigned long now = jiffies;
1407	struct sk_buff *skb, *n;
1408
1409	spin_lock(&tbl->proxy_queue.lock);
1410
1411	skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1412		long tdif = NEIGH_CB(skb)->sched_next - now;
1413
1414		if (tdif <= 0) {
1415			struct net_device *dev = skb->dev;
1416
1417			__skb_unlink(skb, &tbl->proxy_queue);
1418			if (tbl->proxy_redo && netif_running(dev)) {
1419				rcu_read_lock();
1420				tbl->proxy_redo(skb);
1421				rcu_read_unlock();
1422			} else {
1423				kfree_skb(skb);
1424			}
1425
1426			dev_put(dev);
1427		} else if (!sched_next || tdif < sched_next)
1428			sched_next = tdif;
1429	}
1430	del_timer(&tbl->proxy_timer);
1431	if (sched_next)
1432		mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1433	spin_unlock(&tbl->proxy_queue.lock);
1434}
1435
1436void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1437		    struct sk_buff *skb)
1438{
1439	unsigned long now = jiffies;
1440
1441	unsigned long sched_next = now + (prandom_u32() %
1442					  NEIGH_VAR(p, PROXY_DELAY));
1443
1444	if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1445		kfree_skb(skb);
1446		return;
1447	}
1448
1449	NEIGH_CB(skb)->sched_next = sched_next;
1450	NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1451
1452	spin_lock(&tbl->proxy_queue.lock);
1453	if (del_timer(&tbl->proxy_timer)) {
1454		if (time_before(tbl->proxy_timer.expires, sched_next))
1455			sched_next = tbl->proxy_timer.expires;
1456	}
1457	skb_dst_drop(skb);
1458	dev_hold(skb->dev);
1459	__skb_queue_tail(&tbl->proxy_queue, skb);
1460	mod_timer(&tbl->proxy_timer, sched_next);
1461	spin_unlock(&tbl->proxy_queue.lock);
1462}
1463EXPORT_SYMBOL(pneigh_enqueue);
1464
1465static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1466						      struct net *net, int ifindex)
1467{
1468	struct neigh_parms *p;
1469
1470	list_for_each_entry(p, &tbl->parms_list, list) {
1471		if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1472		    (!p->dev && !ifindex && net_eq(net, &init_net)))
1473			return p;
1474	}
1475
1476	return NULL;
1477}
1478
1479struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1480				      struct neigh_table *tbl)
1481{
1482	struct neigh_parms *p;
1483	struct net *net = dev_net(dev);
1484	const struct net_device_ops *ops = dev->netdev_ops;
1485
1486	p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1487	if (p) {
1488		p->tbl		  = tbl;
1489		refcount_set(&p->refcnt, 1);
1490		p->reachable_time =
1491				neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1492		dev_hold(dev);
1493		p->dev = dev;
1494		write_pnet(&p->net, net);
1495		p->sysctl_table = NULL;
1496
1497		if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1498			dev_put(dev);
1499			kfree(p);
1500			return NULL;
1501		}
1502
1503		write_lock_bh(&tbl->lock);
1504		list_add(&p->list, &tbl->parms.list);
1505		write_unlock_bh(&tbl->lock);
1506
1507		neigh_parms_data_state_cleanall(p);
1508	}
1509	return p;
1510}
1511EXPORT_SYMBOL(neigh_parms_alloc);
1512
1513static void neigh_rcu_free_parms(struct rcu_head *head)
1514{
1515	struct neigh_parms *parms =
1516		container_of(head, struct neigh_parms, rcu_head);
1517
1518	neigh_parms_put(parms);
1519}
1520
1521void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1522{
1523	if (!parms || parms == &tbl->parms)
1524		return;
1525	write_lock_bh(&tbl->lock);
1526	list_del(&parms->list);
1527	parms->dead = 1;
1528	write_unlock_bh(&tbl->lock);
1529	if (parms->dev)
1530		dev_put(parms->dev);
1531	call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1532}
1533EXPORT_SYMBOL(neigh_parms_release);
1534
1535static void neigh_parms_destroy(struct neigh_parms *parms)
1536{
1537	kfree(parms);
1538}
1539
1540static struct lock_class_key neigh_table_proxy_queue_class;
1541
1542static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1543
1544void neigh_table_init(int index, struct neigh_table *tbl)
1545{
1546	unsigned long now = jiffies;
1547	unsigned long phsize;
1548
1549	INIT_LIST_HEAD(&tbl->parms_list);
1550	list_add(&tbl->parms.list, &tbl->parms_list);
1551	write_pnet(&tbl->parms.net, &init_net);
1552	refcount_set(&tbl->parms.refcnt, 1);
1553	tbl->parms.reachable_time =
1554			  neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1555
1556	tbl->stats = alloc_percpu(struct neigh_statistics);
1557	if (!tbl->stats)
1558		panic("cannot create neighbour cache statistics");
1559
1560#ifdef CONFIG_PROC_FS
1561	if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1562			      &neigh_stat_seq_fops, tbl))
1563		panic("cannot create neighbour proc dir entry");
1564#endif
1565
1566	RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1567
1568	phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1569	tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1570
1571	if (!tbl->nht || !tbl->phash_buckets)
1572		panic("cannot allocate neighbour cache hashes");
1573
1574	if (!tbl->entry_size)
1575		tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1576					tbl->key_len, NEIGH_PRIV_ALIGN);
1577	else
1578		WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1579
1580	rwlock_init(&tbl->lock);
1581	INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1582	queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1583			tbl->parms.reachable_time);
1584	timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1585	skb_queue_head_init_class(&tbl->proxy_queue,
1586			&neigh_table_proxy_queue_class);
1587
1588	tbl->last_flush = now;
1589	tbl->last_rand	= now + tbl->parms.reachable_time * 20;
1590
1591	neigh_tables[index] = tbl;
1592}
1593EXPORT_SYMBOL(neigh_table_init);
1594
1595int neigh_table_clear(int index, struct neigh_table *tbl)
1596{
1597	neigh_tables[index] = NULL;
1598	/* It is not clean... Fix it to unload IPv6 module safely */
1599	cancel_delayed_work_sync(&tbl->gc_work);
1600	del_timer_sync(&tbl->proxy_timer);
1601	pneigh_queue_purge(&tbl->proxy_queue);
1602	neigh_ifdown(tbl, NULL);
1603	if (atomic_read(&tbl->entries))
1604		pr_crit("neighbour leakage\n");
1605
1606	call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1607		 neigh_hash_free_rcu);
1608	tbl->nht = NULL;
1609
1610	kfree(tbl->phash_buckets);
1611	tbl->phash_buckets = NULL;
1612
1613	remove_proc_entry(tbl->id, init_net.proc_net_stat);
1614
1615	free_percpu(tbl->stats);
1616	tbl->stats = NULL;
1617
1618	return 0;
1619}
1620EXPORT_SYMBOL(neigh_table_clear);
1621
1622static struct neigh_table *neigh_find_table(int family)
1623{
1624	struct neigh_table *tbl = NULL;
1625
1626	switch (family) {
1627	case AF_INET:
1628		tbl = neigh_tables[NEIGH_ARP_TABLE];
1629		break;
1630	case AF_INET6:
1631		tbl = neigh_tables[NEIGH_ND_TABLE];
1632		break;
1633	case AF_DECnet:
1634		tbl = neigh_tables[NEIGH_DN_TABLE];
1635		break;
1636	}
1637
1638	return tbl;
1639}
1640
1641static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1642			struct netlink_ext_ack *extack)
1643{
1644	struct net *net = sock_net(skb->sk);
1645	struct ndmsg *ndm;
1646	struct nlattr *dst_attr;
1647	struct neigh_table *tbl;
1648	struct neighbour *neigh;
1649	struct net_device *dev = NULL;
1650	int err = -EINVAL;
1651
1652	ASSERT_RTNL();
1653	if (nlmsg_len(nlh) < sizeof(*ndm))
1654		goto out;
1655
1656	dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1657	if (dst_attr == NULL)
1658		goto out;
1659
1660	ndm = nlmsg_data(nlh);
1661	if (ndm->ndm_ifindex) {
1662		dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1663		if (dev == NULL) {
1664			err = -ENODEV;
1665			goto out;
1666		}
1667	}
1668
1669	tbl = neigh_find_table(ndm->ndm_family);
1670	if (tbl == NULL)
1671		return -EAFNOSUPPORT;
1672
1673	if (nla_len(dst_attr) < (int)tbl->key_len)
1674		goto out;
1675
1676	if (ndm->ndm_flags & NTF_PROXY) {
1677		err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1678		goto out;
1679	}
1680
1681	if (dev == NULL)
1682		goto out;
1683
1684	neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1685	if (neigh == NULL) {
1686		err = -ENOENT;
1687		goto out;
1688	}
1689
1690	err = neigh_update(neigh, NULL, NUD_FAILED,
1691			   NEIGH_UPDATE_F_OVERRIDE |
1692			   NEIGH_UPDATE_F_ADMIN,
1693			   NETLINK_CB(skb).portid);
1694	write_lock_bh(&tbl->lock);
1695	neigh_release(neigh);
1696	neigh_remove_one(neigh, tbl);
1697	write_unlock_bh(&tbl->lock);
1698
1699out:
1700	return err;
1701}
1702
1703static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1704		     struct netlink_ext_ack *extack)
1705{
1706	int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1707	struct net *net = sock_net(skb->sk);
1708	struct ndmsg *ndm;
1709	struct nlattr *tb[NDA_MAX+1];
1710	struct neigh_table *tbl;
1711	struct net_device *dev = NULL;
1712	struct neighbour *neigh;
1713	void *dst, *lladdr;
1714	int err;
1715
1716	ASSERT_RTNL();
1717	err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL, extack);
1718	if (err < 0)
1719		goto out;
1720
1721	err = -EINVAL;
1722	if (tb[NDA_DST] == NULL)
1723		goto out;
1724
1725	ndm = nlmsg_data(nlh);
1726	if (ndm->ndm_ifindex) {
1727		dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1728		if (dev == NULL) {
1729			err = -ENODEV;
1730			goto out;
1731		}
1732
1733		if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1734			goto out;
1735	}
1736
1737	tbl = neigh_find_table(ndm->ndm_family);
1738	if (tbl == NULL)
1739		return -EAFNOSUPPORT;
1740
1741	if (nla_len(tb[NDA_DST]) < (int)tbl->key_len)
1742		goto out;
1743	dst = nla_data(tb[NDA_DST]);
1744	lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1745
1746	if (ndm->ndm_flags & NTF_PROXY) {
1747		struct pneigh_entry *pn;
1748
1749		err = -ENOBUFS;
1750		pn = pneigh_lookup(tbl, net, dst, dev, 1);
1751		if (pn) {
1752			pn->flags = ndm->ndm_flags;
1753			err = 0;
1754		}
1755		goto out;
1756	}
1757
1758	if (dev == NULL)
1759		goto out;
1760
1761	neigh = neigh_lookup(tbl, dst, dev);
1762	if (neigh == NULL) {
1763		if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1764			err = -ENOENT;
1765			goto out;
1766		}
1767
1768		neigh = __neigh_lookup_errno(tbl, dst, dev);
1769		if (IS_ERR(neigh)) {
1770			err = PTR_ERR(neigh);
1771			goto out;
1772		}
1773	} else {
1774		if (nlh->nlmsg_flags & NLM_F_EXCL) {
1775			err = -EEXIST;
1776			neigh_release(neigh);
1777			goto out;
1778		}
1779
1780		if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1781			flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1782	}
1783
1784	if (ndm->ndm_flags & NTF_USE) {
1785		neigh_event_send(neigh, NULL);
1786		err = 0;
1787	} else
1788		err = neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1789				   NETLINK_CB(skb).portid);
1790	neigh_release(neigh);
1791
1792out:
1793	return err;
1794}
1795
1796static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1797{
1798	struct nlattr *nest;
1799
1800	nest = nla_nest_start(skb, NDTA_PARMS);
1801	if (nest == NULL)
1802		return -ENOBUFS;
1803
1804	if ((parms->dev &&
1805	     nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1806	    nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1807	    nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1808			NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1809	    /* approximative value for deprecated QUEUE_LEN (in packets) */
1810	    nla_put_u32(skb, NDTPA_QUEUE_LEN,
1811			NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1812	    nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1813	    nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1814	    nla_put_u32(skb, NDTPA_UCAST_PROBES,
1815			NEIGH_VAR(parms, UCAST_PROBES)) ||
1816	    nla_put_u32(skb, NDTPA_MCAST_PROBES,
1817			NEIGH_VAR(parms, MCAST_PROBES)) ||
1818	    nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1819			NEIGH_VAR(parms, MCAST_REPROBES)) ||
1820	    nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
1821			  NDTPA_PAD) ||
1822	    nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1823			  NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
1824	    nla_put_msecs(skb, NDTPA_GC_STALETIME,
1825			  NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
1826	    nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1827			  NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
1828	    nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1829			  NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
1830	    nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1831			  NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
1832	    nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1833			  NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
1834	    nla_put_msecs(skb, NDTPA_LOCKTIME,
1835			  NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
1836		goto nla_put_failure;
1837	return nla_nest_end(skb, nest);
1838
1839nla_put_failure:
1840	nla_nest_cancel(skb, nest);
1841	return -EMSGSIZE;
1842}
1843
1844static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1845			      u32 pid, u32 seq, int type, int flags)
1846{
1847	struct nlmsghdr *nlh;
1848	struct ndtmsg *ndtmsg;
1849
1850	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1851	if (nlh == NULL)
1852		return -EMSGSIZE;
1853
1854	ndtmsg = nlmsg_data(nlh);
1855
1856	read_lock_bh(&tbl->lock);
1857	ndtmsg->ndtm_family = tbl->family;
1858	ndtmsg->ndtm_pad1   = 0;
1859	ndtmsg->ndtm_pad2   = 0;
1860
1861	if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1862	    nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
1863	    nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1864	    nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1865	    nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1866		goto nla_put_failure;
1867	{
1868		unsigned long now = jiffies;
1869		unsigned int flush_delta = now - tbl->last_flush;
1870		unsigned int rand_delta = now - tbl->last_rand;
1871		struct neigh_hash_table *nht;
1872		struct ndt_config ndc = {
1873			.ndtc_key_len		= tbl->key_len,
1874			.ndtc_entry_size	= tbl->entry_size,
1875			.ndtc_entries		= atomic_read(&tbl->entries),
1876			.ndtc_last_flush	= jiffies_to_msecs(flush_delta),
1877			.ndtc_last_rand		= jiffies_to_msecs(rand_delta),
1878			.ndtc_proxy_qlen	= tbl->proxy_queue.qlen,
1879		};
1880
1881		rcu_read_lock_bh();
1882		nht = rcu_dereference_bh(tbl->nht);
1883		ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1884		ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1885		rcu_read_unlock_bh();
1886
1887		if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1888			goto nla_put_failure;
1889	}
1890
1891	{
1892		int cpu;
1893		struct ndt_stats ndst;
1894
1895		memset(&ndst, 0, sizeof(ndst));
1896
1897		for_each_possible_cpu(cpu) {
1898			struct neigh_statistics	*st;
1899
1900			st = per_cpu_ptr(tbl->stats, cpu);
1901			ndst.ndts_allocs		+= st->allocs;
1902			ndst.ndts_destroys		+= st->destroys;
1903			ndst.ndts_hash_grows		+= st->hash_grows;
1904			ndst.ndts_res_failed		+= st->res_failed;
1905			ndst.ndts_lookups		+= st->lookups;
1906			ndst.ndts_hits			+= st->hits;
1907			ndst.ndts_rcv_probes_mcast	+= st->rcv_probes_mcast;
1908			ndst.ndts_rcv_probes_ucast	+= st->rcv_probes_ucast;
1909			ndst.ndts_periodic_gc_runs	+= st->periodic_gc_runs;
1910			ndst.ndts_forced_gc_runs	+= st->forced_gc_runs;
1911			ndst.ndts_table_fulls		+= st->table_fulls;
1912		}
1913
1914		if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
1915				  NDTA_PAD))
1916			goto nla_put_failure;
1917	}
1918
1919	BUG_ON(tbl->parms.dev);
1920	if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1921		goto nla_put_failure;
1922
1923	read_unlock_bh(&tbl->lock);
1924	nlmsg_end(skb, nlh);
1925	return 0;
1926
1927nla_put_failure:
1928	read_unlock_bh(&tbl->lock);
1929	nlmsg_cancel(skb, nlh);
1930	return -EMSGSIZE;
1931}
1932
1933static int neightbl_fill_param_info(struct sk_buff *skb,
1934				    struct neigh_table *tbl,
1935				    struct neigh_parms *parms,
1936				    u32 pid, u32 seq, int type,
1937				    unsigned int flags)
1938{
1939	struct ndtmsg *ndtmsg;
1940	struct nlmsghdr *nlh;
1941
1942	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1943	if (nlh == NULL)
1944		return -EMSGSIZE;
1945
1946	ndtmsg = nlmsg_data(nlh);
1947
1948	read_lock_bh(&tbl->lock);
1949	ndtmsg->ndtm_family = tbl->family;
1950	ndtmsg->ndtm_pad1   = 0;
1951	ndtmsg->ndtm_pad2   = 0;
1952
1953	if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1954	    neightbl_fill_parms(skb, parms) < 0)
1955		goto errout;
1956
1957	read_unlock_bh(&tbl->lock);
1958	nlmsg_end(skb, nlh);
1959	return 0;
1960errout:
1961	read_unlock_bh(&tbl->lock);
1962	nlmsg_cancel(skb, nlh);
1963	return -EMSGSIZE;
1964}
1965
1966static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1967	[NDTA_NAME]		= { .type = NLA_STRING },
1968	[NDTA_THRESH1]		= { .type = NLA_U32 },
1969	[NDTA_THRESH2]		= { .type = NLA_U32 },
1970	[NDTA_THRESH3]		= { .type = NLA_U32 },
1971	[NDTA_GC_INTERVAL]	= { .type = NLA_U64 },
1972	[NDTA_PARMS]		= { .type = NLA_NESTED },
1973};
1974
1975static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1976	[NDTPA_IFINDEX]			= { .type = NLA_U32 },
1977	[NDTPA_QUEUE_LEN]		= { .type = NLA_U32 },
1978	[NDTPA_PROXY_QLEN]		= { .type = NLA_U32 },
1979	[NDTPA_APP_PROBES]		= { .type = NLA_U32 },
1980	[NDTPA_UCAST_PROBES]		= { .type = NLA_U32 },
1981	[NDTPA_MCAST_PROBES]		= { .type = NLA_U32 },
1982	[NDTPA_MCAST_REPROBES]		= { .type = NLA_U32 },
1983	[NDTPA_BASE_REACHABLE_TIME]	= { .type = NLA_U64 },
1984	[NDTPA_GC_STALETIME]		= { .type = NLA_U64 },
1985	[NDTPA_DELAY_PROBE_TIME]	= { .type = NLA_U64 },
1986	[NDTPA_RETRANS_TIME]		= { .type = NLA_U64 },
1987	[NDTPA_ANYCAST_DELAY]		= { .type = NLA_U64 },
1988	[NDTPA_PROXY_DELAY]		= { .type = NLA_U64 },
1989	[NDTPA_LOCKTIME]		= { .type = NLA_U64 },
1990};
1991
1992static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
1993			struct netlink_ext_ack *extack)
1994{
1995	struct net *net = sock_net(skb->sk);
1996	struct neigh_table *tbl;
1997	struct ndtmsg *ndtmsg;
1998	struct nlattr *tb[NDTA_MAX+1];
1999	bool found = false;
2000	int err, tidx;
2001
2002	err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2003			  nl_neightbl_policy, extack);
2004	if (err < 0)
2005		goto errout;
2006
2007	if (tb[NDTA_NAME] == NULL) {
2008		err = -EINVAL;
2009		goto errout;
2010	}
2011
2012	ndtmsg = nlmsg_data(nlh);
2013
2014	for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2015		tbl = neigh_tables[tidx];
2016		if (!tbl)
2017			continue;
2018		if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2019			continue;
2020		if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2021			found = true;
2022			break;
2023		}
2024	}
2025
2026	if (!found)
2027		return -ENOENT;
2028
2029	/*
2030	 * We acquire tbl->lock to be nice to the periodic timers and
2031	 * make sure they always see a consistent set of values.
2032	 */
2033	write_lock_bh(&tbl->lock);
2034
2035	if (tb[NDTA_PARMS]) {
2036		struct nlattr *tbp[NDTPA_MAX+1];
2037		struct neigh_parms *p;
2038		int i, ifindex = 0;
2039
2040		err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2041				       nl_ntbl_parm_policy, extack);
2042		if (err < 0)
2043			goto errout_tbl_lock;
2044
2045		if (tbp[NDTPA_IFINDEX])
2046			ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2047
2048		p = lookup_neigh_parms(tbl, net, ifindex);
2049		if (p == NULL) {
2050			err = -ENOENT;
2051			goto errout_tbl_lock;
2052		}
2053
2054		for (i = 1; i <= NDTPA_MAX; i++) {
2055			if (tbp[i] == NULL)
2056				continue;
2057
2058			switch (i) {
2059			case NDTPA_QUEUE_LEN:
2060				NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2061					      nla_get_u32(tbp[i]) *
2062					      SKB_TRUESIZE(ETH_FRAME_LEN));
2063				break;
2064			case NDTPA_QUEUE_LENBYTES:
2065				NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2066					      nla_get_u32(tbp[i]));
2067				break;
2068			case NDTPA_PROXY_QLEN:
2069				NEIGH_VAR_SET(p, PROXY_QLEN,
2070					      nla_get_u32(tbp[i]));
2071				break;
2072			case NDTPA_APP_PROBES:
2073				NEIGH_VAR_SET(p, APP_PROBES,
2074					      nla_get_u32(tbp[i]));
2075				break;
2076			case NDTPA_UCAST_PROBES:
2077				NEIGH_VAR_SET(p, UCAST_PROBES,
2078					      nla_get_u32(tbp[i]));
2079				break;
2080			case NDTPA_MCAST_PROBES:
2081				NEIGH_VAR_SET(p, MCAST_PROBES,
2082					      nla_get_u32(tbp[i]));
2083				break;
2084			case NDTPA_MCAST_REPROBES:
2085				NEIGH_VAR_SET(p, MCAST_REPROBES,
2086					      nla_get_u32(tbp[i]));
2087				break;
2088			case NDTPA_BASE_REACHABLE_TIME:
2089				NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2090					      nla_get_msecs(tbp[i]));
2091				/* update reachable_time as well, otherwise, the change will
2092				 * only be effective after the next time neigh_periodic_work
2093				 * decides to recompute it (can be multiple minutes)
2094				 */
2095				p->reachable_time =
2096					neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2097				break;
2098			case NDTPA_GC_STALETIME:
2099				NEIGH_VAR_SET(p, GC_STALETIME,
2100					      nla_get_msecs(tbp[i]));
2101				break;
2102			case NDTPA_DELAY_PROBE_TIME:
2103				NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2104					      nla_get_msecs(tbp[i]));
2105				call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2106				break;
2107			case NDTPA_RETRANS_TIME:
2108				NEIGH_VAR_SET(p, RETRANS_TIME,
2109					      nla_get_msecs(tbp[i]));
2110				break;
2111			case NDTPA_ANYCAST_DELAY:
2112				NEIGH_VAR_SET(p, ANYCAST_DELAY,
2113					      nla_get_msecs(tbp[i]));
2114				break;
2115			case NDTPA_PROXY_DELAY:
2116				NEIGH_VAR_SET(p, PROXY_DELAY,
2117					      nla_get_msecs(tbp[i]));
2118				break;
2119			case NDTPA_LOCKTIME:
2120				NEIGH_VAR_SET(p, LOCKTIME,
2121					      nla_get_msecs(tbp[i]));
2122				break;
2123			}
2124		}
2125	}
2126
2127	err = -ENOENT;
2128	if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2129	     tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2130	    !net_eq(net, &init_net))
2131		goto errout_tbl_lock;
2132
2133	if (tb[NDTA_THRESH1])
2134		tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2135
2136	if (tb[NDTA_THRESH2])
2137		tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2138
2139	if (tb[NDTA_THRESH3])
2140		tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2141
2142	if (tb[NDTA_GC_INTERVAL])
2143		tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2144
2145	err = 0;
2146
2147errout_tbl_lock:
2148	write_unlock_bh(&tbl->lock);
2149errout:
2150	return err;
2151}
2152
2153static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2154{
2155	struct net *net = sock_net(skb->sk);
2156	int family, tidx, nidx = 0;
2157	int tbl_skip = cb->args[0];
2158	int neigh_skip = cb->args[1];
2159	struct neigh_table *tbl;
2160
2161	family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2162
2163	for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2164		struct neigh_parms *p;
2165
2166		tbl = neigh_tables[tidx];
2167		if (!tbl)
2168			continue;
2169
2170		if (tidx < tbl_skip || (family && tbl->family != family))
2171			continue;
2172
2173		if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2174				       cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2175				       NLM_F_MULTI) < 0)
2176			break;
2177
2178		nidx = 0;
2179		p = list_next_entry(&tbl->parms, list);
2180		list_for_each_entry_from(p, &tbl->parms_list, list) {
2181			if (!net_eq(neigh_parms_net(p), net))
2182				continue;
2183
2184			if (nidx < neigh_skip)
2185				goto next;
2186
2187			if (neightbl_fill_param_info(skb, tbl, p,
2188						     NETLINK_CB(cb->skb).portid,
2189						     cb->nlh->nlmsg_seq,
2190						     RTM_NEWNEIGHTBL,
2191						     NLM_F_MULTI) < 0)
2192				goto out;
2193		next:
2194			nidx++;
2195		}
2196
2197		neigh_skip = 0;
2198	}
2199out:
2200	cb->args[0] = tidx;
2201	cb->args[1] = nidx;
2202
2203	return skb->len;
2204}
2205
2206static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2207			   u32 pid, u32 seq, int type, unsigned int flags)
2208{
2209	unsigned long now = jiffies;
2210	struct nda_cacheinfo ci;
2211	struct nlmsghdr *nlh;
2212	struct ndmsg *ndm;
2213
2214	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2215	if (nlh == NULL)
2216		return -EMSGSIZE;
2217
2218	ndm = nlmsg_data(nlh);
2219	ndm->ndm_family	 = neigh->ops->family;
2220	ndm->ndm_pad1    = 0;
2221	ndm->ndm_pad2    = 0;
2222	ndm->ndm_flags	 = neigh->flags;
2223	ndm->ndm_type	 = neigh->type;
2224	ndm->ndm_ifindex = neigh->dev->ifindex;
2225
2226	if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2227		goto nla_put_failure;
2228
2229	read_lock_bh(&neigh->lock);
2230	ndm->ndm_state	 = neigh->nud_state;
2231	if (neigh->nud_state & NUD_VALID) {
2232		char haddr[MAX_ADDR_LEN];
2233
2234		neigh_ha_snapshot(haddr, neigh, neigh->dev);
2235		if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2236			read_unlock_bh(&neigh->lock);
2237			goto nla_put_failure;
2238		}
2239	}
2240
2241	ci.ndm_used	 = jiffies_to_clock_t(now - neigh->used);
2242	ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2243	ci.ndm_updated	 = jiffies_to_clock_t(now - neigh->updated);
2244	ci.ndm_refcnt	 = refcount_read(&neigh->refcnt) - 1;
2245	read_unlock_bh(&neigh->lock);
2246
2247	if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2248	    nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2249		goto nla_put_failure;
2250
2251	nlmsg_end(skb, nlh);
2252	return 0;
2253
2254nla_put_failure:
2255	nlmsg_cancel(skb, nlh);
2256	return -EMSGSIZE;
2257}
2258
2259static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2260			    u32 pid, u32 seq, int type, unsigned int flags,
2261			    struct neigh_table *tbl)
2262{
2263	struct nlmsghdr *nlh;
2264	struct ndmsg *ndm;
2265
2266	nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2267	if (nlh == NULL)
2268		return -EMSGSIZE;
2269
2270	ndm = nlmsg_data(nlh);
2271	ndm->ndm_family	 = tbl->family;
2272	ndm->ndm_pad1    = 0;
2273	ndm->ndm_pad2    = 0;
2274	ndm->ndm_flags	 = pn->flags | NTF_PROXY;
2275	ndm->ndm_type	 = RTN_UNICAST;
2276	ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2277	ndm->ndm_state	 = NUD_NONE;
2278
2279	if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2280		goto nla_put_failure;
2281
2282	nlmsg_end(skb, nlh);
2283	return 0;
2284
2285nla_put_failure:
2286	nlmsg_cancel(skb, nlh);
2287	return -EMSGSIZE;
2288}
2289
2290static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2291{
2292	call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2293	__neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2294}
2295
2296static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2297{
2298	struct net_device *master;
2299
2300	if (!master_idx)
2301		return false;
2302
2303	master = netdev_master_upper_dev_get(dev);
2304	if (!master || master->ifindex != master_idx)
2305		return true;
2306
2307	return false;
2308}
2309
2310static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2311{
2312	if (filter_idx && dev->ifindex != filter_idx)
2313		return true;
2314
2315	return false;
2316}
2317
2318static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2319			    struct netlink_callback *cb)
2320{
2321	struct net *net = sock_net(skb->sk);
2322	const struct nlmsghdr *nlh = cb->nlh;
2323	struct nlattr *tb[NDA_MAX + 1];
2324	struct neighbour *n;
2325	int rc, h, s_h = cb->args[1];
2326	int idx, s_idx = idx = cb->args[2];
2327	struct neigh_hash_table *nht;
2328	int filter_master_idx = 0, filter_idx = 0;
2329	unsigned int flags = NLM_F_MULTI;
2330	int err;
2331
2332	err = nlmsg_parse(nlh, sizeof(struct ndmsg), tb, NDA_MAX, NULL, NULL);
2333	if (!err) {
2334		if (tb[NDA_IFINDEX]) {
2335			if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
2336				return -EINVAL;
2337			filter_idx = nla_get_u32(tb[NDA_IFINDEX]);
2338		}
2339		if (tb[NDA_MASTER]) {
2340			if (nla_len(tb[NDA_MASTER]) != sizeof(u32))
2341				return -EINVAL;
2342			filter_master_idx = nla_get_u32(tb[NDA_MASTER]);
2343		}
2344		if (filter_idx || filter_master_idx)
2345			flags |= NLM_F_DUMP_FILTERED;
2346	}
2347
2348	rcu_read_lock_bh();
2349	nht = rcu_dereference_bh(tbl->nht);
2350
2351	for (h = s_h; h < (1 << nht->hash_shift); h++) {
2352		if (h > s_h)
2353			s_idx = 0;
2354		for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2355		     n != NULL;
2356		     n = rcu_dereference_bh(n->next)) {
2357			if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2358				goto next;
2359			if (neigh_ifindex_filtered(n->dev, filter_idx) ||
2360			    neigh_master_filtered(n->dev, filter_master_idx))
 
 
 
2361				goto next;
2362			if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2363					    cb->nlh->nlmsg_seq,
2364					    RTM_NEWNEIGH,
2365					    flags) < 0) {
2366				rc = -1;
2367				goto out;
2368			}
2369next:
2370			idx++;
2371		}
2372	}
2373	rc = skb->len;
2374out:
2375	rcu_read_unlock_bh();
2376	cb->args[1] = h;
2377	cb->args[2] = idx;
2378	return rc;
2379}
2380
2381static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2382			     struct netlink_callback *cb)
2383{
2384	struct pneigh_entry *n;
2385	struct net *net = sock_net(skb->sk);
2386	int rc, h, s_h = cb->args[3];
2387	int idx, s_idx = idx = cb->args[4];
2388
2389	read_lock_bh(&tbl->lock);
2390
2391	for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2392		if (h > s_h)
2393			s_idx = 0;
2394		for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2395			if (idx < s_idx || pneigh_net(n) != net)
 
 
2396				goto next;
2397			if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2398					    cb->nlh->nlmsg_seq,
2399					    RTM_NEWNEIGH,
2400					    NLM_F_MULTI, tbl) < 0) {
2401				read_unlock_bh(&tbl->lock);
2402				rc = -1;
2403				goto out;
2404			}
2405		next:
2406			idx++;
2407		}
2408	}
2409
2410	read_unlock_bh(&tbl->lock);
2411	rc = skb->len;
2412out:
2413	cb->args[3] = h;
2414	cb->args[4] = idx;
2415	return rc;
2416
2417}
2418
2419static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2420{
2421	struct neigh_table *tbl;
2422	int t, family, s_t;
2423	int proxy = 0;
2424	int err;
2425
2426	family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2427
2428	/* check for full ndmsg structure presence, family member is
2429	 * the same for both structures
2430	 */
2431	if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2432	    ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2433		proxy = 1;
2434
2435	s_t = cb->args[0];
2436
2437	for (t = 0; t < NEIGH_NR_TABLES; t++) {
2438		tbl = neigh_tables[t];
2439
2440		if (!tbl)
2441			continue;
2442		if (t < s_t || (family && tbl->family != family))
2443			continue;
2444		if (t > s_t)
2445			memset(&cb->args[1], 0, sizeof(cb->args) -
2446						sizeof(cb->args[0]));
2447		if (proxy)
2448			err = pneigh_dump_table(tbl, skb, cb);
2449		else
2450			err = neigh_dump_table(tbl, skb, cb);
2451		if (err < 0)
2452			break;
2453	}
2454
2455	cb->args[0] = t;
2456	return skb->len;
2457}
2458
2459void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2460{
2461	int chain;
2462	struct neigh_hash_table *nht;
2463
2464	rcu_read_lock_bh();
2465	nht = rcu_dereference_bh(tbl->nht);
2466
2467	read_lock(&tbl->lock); /* avoid resizes */
2468	for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2469		struct neighbour *n;
2470
2471		for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2472		     n != NULL;
2473		     n = rcu_dereference_bh(n->next))
2474			cb(n, cookie);
2475	}
2476	read_unlock(&tbl->lock);
2477	rcu_read_unlock_bh();
2478}
2479EXPORT_SYMBOL(neigh_for_each);
2480
2481/* The tbl->lock must be held as a writer and BH disabled. */
2482void __neigh_for_each_release(struct neigh_table *tbl,
2483			      int (*cb)(struct neighbour *))
2484{
2485	int chain;
2486	struct neigh_hash_table *nht;
2487
2488	nht = rcu_dereference_protected(tbl->nht,
2489					lockdep_is_held(&tbl->lock));
2490	for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2491		struct neighbour *n;
2492		struct neighbour __rcu **np;
2493
2494		np = &nht->hash_buckets[chain];
2495		while ((n = rcu_dereference_protected(*np,
2496					lockdep_is_held(&tbl->lock))) != NULL) {
2497			int release;
2498
2499			write_lock(&n->lock);
2500			release = cb(n);
2501			if (release) {
2502				rcu_assign_pointer(*np,
2503					rcu_dereference_protected(n->next,
2504						lockdep_is_held(&tbl->lock)));
2505				n->dead = 1;
2506			} else
2507				np = &n->next;
2508			write_unlock(&n->lock);
2509			if (release)
2510				neigh_cleanup_and_release(n);
2511		}
2512	}
2513}
2514EXPORT_SYMBOL(__neigh_for_each_release);
2515
2516int neigh_xmit(int index, struct net_device *dev,
2517	       const void *addr, struct sk_buff *skb)
2518{
2519	int err = -EAFNOSUPPORT;
2520	if (likely(index < NEIGH_NR_TABLES)) {
2521		struct neigh_table *tbl;
2522		struct neighbour *neigh;
2523
2524		tbl = neigh_tables[index];
2525		if (!tbl)
2526			goto out;
2527		rcu_read_lock_bh();
2528		neigh = __neigh_lookup_noref(tbl, addr, dev);
2529		if (!neigh)
2530			neigh = __neigh_create(tbl, addr, dev, false);
2531		err = PTR_ERR(neigh);
2532		if (IS_ERR(neigh)) {
2533			rcu_read_unlock_bh();
2534			goto out_kfree_skb;
2535		}
2536		err = neigh->output(neigh, skb);
2537		rcu_read_unlock_bh();
2538	}
2539	else if (index == NEIGH_LINK_TABLE) {
2540		err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2541				      addr, NULL, skb->len);
2542		if (err < 0)
2543			goto out_kfree_skb;
2544		err = dev_queue_xmit(skb);
2545	}
2546out:
2547	return err;
2548out_kfree_skb:
2549	kfree_skb(skb);
2550	goto out;
2551}
2552EXPORT_SYMBOL(neigh_xmit);
2553
2554#ifdef CONFIG_PROC_FS
2555
2556static struct neighbour *neigh_get_first(struct seq_file *seq)
2557{
2558	struct neigh_seq_state *state = seq->private;
2559	struct net *net = seq_file_net(seq);
2560	struct neigh_hash_table *nht = state->nht;
2561	struct neighbour *n = NULL;
2562	int bucket = state->bucket;
2563
2564	state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2565	for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2566		n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2567
2568		while (n) {
2569			if (!net_eq(dev_net(n->dev), net))
2570				goto next;
2571			if (state->neigh_sub_iter) {
2572				loff_t fakep = 0;
2573				void *v;
2574
2575				v = state->neigh_sub_iter(state, n, &fakep);
2576				if (!v)
2577					goto next;
2578			}
2579			if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2580				break;
2581			if (n->nud_state & ~NUD_NOARP)
2582				break;
2583next:
2584			n = rcu_dereference_bh(n->next);
2585		}
2586
2587		if (n)
2588			break;
2589	}
2590	state->bucket = bucket;
2591
2592	return n;
2593}
2594
2595static struct neighbour *neigh_get_next(struct seq_file *seq,
2596					struct neighbour *n,
2597					loff_t *pos)
2598{
2599	struct neigh_seq_state *state = seq->private;
2600	struct net *net = seq_file_net(seq);
2601	struct neigh_hash_table *nht = state->nht;
2602
2603	if (state->neigh_sub_iter) {
2604		void *v = state->neigh_sub_iter(state, n, pos);
2605		if (v)
2606			return n;
2607	}
2608	n = rcu_dereference_bh(n->next);
2609
2610	while (1) {
2611		while (n) {
2612			if (!net_eq(dev_net(n->dev), net))
2613				goto next;
2614			if (state->neigh_sub_iter) {
2615				void *v = state->neigh_sub_iter(state, n, pos);
2616				if (v)
2617					return n;
2618				goto next;
2619			}
2620			if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2621				break;
2622
2623			if (n->nud_state & ~NUD_NOARP)
2624				break;
2625next:
2626			n = rcu_dereference_bh(n->next);
2627		}
2628
2629		if (n)
2630			break;
2631
2632		if (++state->bucket >= (1 << nht->hash_shift))
2633			break;
2634
2635		n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2636	}
2637
2638	if (n && pos)
2639		--(*pos);
2640	return n;
2641}
2642
2643static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2644{
2645	struct neighbour *n = neigh_get_first(seq);
2646
2647	if (n) {
2648		--(*pos);
2649		while (*pos) {
2650			n = neigh_get_next(seq, n, pos);
2651			if (!n)
2652				break;
2653		}
2654	}
2655	return *pos ? NULL : n;
2656}
2657
2658static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2659{
2660	struct neigh_seq_state *state = seq->private;
2661	struct net *net = seq_file_net(seq);
2662	struct neigh_table *tbl = state->tbl;
2663	struct pneigh_entry *pn = NULL;
2664	int bucket = state->bucket;
2665
2666	state->flags |= NEIGH_SEQ_IS_PNEIGH;
2667	for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2668		pn = tbl->phash_buckets[bucket];
2669		while (pn && !net_eq(pneigh_net(pn), net))
2670			pn = pn->next;
2671		if (pn)
2672			break;
2673	}
2674	state->bucket = bucket;
2675
2676	return pn;
2677}
2678
2679static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2680					    struct pneigh_entry *pn,
2681					    loff_t *pos)
2682{
2683	struct neigh_seq_state *state = seq->private;
2684	struct net *net = seq_file_net(seq);
2685	struct neigh_table *tbl = state->tbl;
2686
2687	do {
2688		pn = pn->next;
2689	} while (pn && !net_eq(pneigh_net(pn), net));
2690
2691	while (!pn) {
2692		if (++state->bucket > PNEIGH_HASHMASK)
2693			break;
2694		pn = tbl->phash_buckets[state->bucket];
2695		while (pn && !net_eq(pneigh_net(pn), net))
2696			pn = pn->next;
2697		if (pn)
2698			break;
2699	}
2700
2701	if (pn && pos)
2702		--(*pos);
2703
2704	return pn;
2705}
2706
2707static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2708{
2709	struct pneigh_entry *pn = pneigh_get_first(seq);
2710
2711	if (pn) {
2712		--(*pos);
2713		while (*pos) {
2714			pn = pneigh_get_next(seq, pn, pos);
2715			if (!pn)
2716				break;
2717		}
2718	}
2719	return *pos ? NULL : pn;
2720}
2721
2722static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2723{
2724	struct neigh_seq_state *state = seq->private;
2725	void *rc;
2726	loff_t idxpos = *pos;
2727
2728	rc = neigh_get_idx(seq, &idxpos);
2729	if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2730		rc = pneigh_get_idx(seq, &idxpos);
2731
2732	return rc;
2733}
2734
2735void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2736	__acquires(rcu_bh)
2737{
2738	struct neigh_seq_state *state = seq->private;
2739
2740	state->tbl = tbl;
2741	state->bucket = 0;
2742	state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2743
2744	rcu_read_lock_bh();
2745	state->nht = rcu_dereference_bh(tbl->nht);
2746
2747	return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2748}
2749EXPORT_SYMBOL(neigh_seq_start);
2750
2751void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2752{
2753	struct neigh_seq_state *state;
2754	void *rc;
2755
2756	if (v == SEQ_START_TOKEN) {
2757		rc = neigh_get_first(seq);
2758		goto out;
2759	}
2760
2761	state = seq->private;
2762	if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2763		rc = neigh_get_next(seq, v, NULL);
2764		if (rc)
2765			goto out;
2766		if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2767			rc = pneigh_get_first(seq);
2768	} else {
2769		BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2770		rc = pneigh_get_next(seq, v, NULL);
2771	}
2772out:
2773	++(*pos);
2774	return rc;
2775}
2776EXPORT_SYMBOL(neigh_seq_next);
2777
2778void neigh_seq_stop(struct seq_file *seq, void *v)
2779	__releases(rcu_bh)
2780{
2781	rcu_read_unlock_bh();
2782}
2783EXPORT_SYMBOL(neigh_seq_stop);
2784
2785/* statistics via seq_file */
2786
2787static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2788{
2789	struct neigh_table *tbl = seq->private;
2790	int cpu;
2791
2792	if (*pos == 0)
2793		return SEQ_START_TOKEN;
2794
2795	for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2796		if (!cpu_possible(cpu))
2797			continue;
2798		*pos = cpu+1;
2799		return per_cpu_ptr(tbl->stats, cpu);
2800	}
2801	return NULL;
2802}
2803
2804static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2805{
2806	struct neigh_table *tbl = seq->private;
2807	int cpu;
2808
2809	for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2810		if (!cpu_possible(cpu))
2811			continue;
2812		*pos = cpu+1;
2813		return per_cpu_ptr(tbl->stats, cpu);
2814	}
2815	return NULL;
2816}
2817
2818static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2819{
2820
2821}
2822
2823static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2824{
2825	struct neigh_table *tbl = seq->private;
2826	struct neigh_statistics *st = v;
2827
2828	if (v == SEQ_START_TOKEN) {
2829		seq_printf(seq, "entries  allocs destroys hash_grows  lookups hits  res_failed  rcv_probes_mcast rcv_probes_ucast  periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
2830		return 0;
2831	}
2832
2833	seq_printf(seq, "%08x  %08lx %08lx %08lx  %08lx %08lx  %08lx  "
2834			"%08lx %08lx  %08lx %08lx %08lx %08lx\n",
2835		   atomic_read(&tbl->entries),
2836
2837		   st->allocs,
2838		   st->destroys,
2839		   st->hash_grows,
2840
2841		   st->lookups,
2842		   st->hits,
2843
2844		   st->res_failed,
2845
2846		   st->rcv_probes_mcast,
2847		   st->rcv_probes_ucast,
2848
2849		   st->periodic_gc_runs,
2850		   st->forced_gc_runs,
2851		   st->unres_discards,
2852		   st->table_fulls
2853		   );
2854
2855	return 0;
2856}
2857
2858static const struct seq_operations neigh_stat_seq_ops = {
2859	.start	= neigh_stat_seq_start,
2860	.next	= neigh_stat_seq_next,
2861	.stop	= neigh_stat_seq_stop,
2862	.show	= neigh_stat_seq_show,
2863};
2864
2865static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2866{
2867	int ret = seq_open(file, &neigh_stat_seq_ops);
2868
2869	if (!ret) {
2870		struct seq_file *sf = file->private_data;
2871		sf->private = PDE_DATA(inode);
2872	}
2873	return ret;
2874};
2875
2876static const struct file_operations neigh_stat_seq_fops = {
 
2877	.open 	 = neigh_stat_seq_open,
2878	.read	 = seq_read,
2879	.llseek	 = seq_lseek,
2880	.release = seq_release,
2881};
2882
2883#endif /* CONFIG_PROC_FS */
2884
2885static inline size_t neigh_nlmsg_size(void)
2886{
2887	return NLMSG_ALIGN(sizeof(struct ndmsg))
2888	       + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2889	       + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2890	       + nla_total_size(sizeof(struct nda_cacheinfo))
2891	       + nla_total_size(4); /* NDA_PROBES */
2892}
2893
2894static void __neigh_notify(struct neighbour *n, int type, int flags,
2895			   u32 pid)
2896{
2897	struct net *net = dev_net(n->dev);
2898	struct sk_buff *skb;
2899	int err = -ENOBUFS;
2900
2901	skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2902	if (skb == NULL)
2903		goto errout;
2904
2905	err = neigh_fill_info(skb, n, pid, 0, type, flags);
2906	if (err < 0) {
2907		/* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2908		WARN_ON(err == -EMSGSIZE);
2909		kfree_skb(skb);
2910		goto errout;
2911	}
2912	rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2913	return;
2914errout:
2915	if (err < 0)
2916		rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2917}
2918
2919void neigh_app_ns(struct neighbour *n)
2920{
2921	__neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
2922}
2923EXPORT_SYMBOL(neigh_app_ns);
2924
2925#ifdef CONFIG_SYSCTL
2926static int zero;
2927static int int_max = INT_MAX;
2928static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2929
2930static int proc_unres_qlen(struct ctl_table *ctl, int write,
2931			   void __user *buffer, size_t *lenp, loff_t *ppos)
2932{
2933	int size, ret;
2934	struct ctl_table tmp = *ctl;
2935
2936	tmp.extra1 = &zero;
2937	tmp.extra2 = &unres_qlen_max;
2938	tmp.data = &size;
2939
2940	size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2941	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2942
2943	if (write && !ret)
2944		*(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2945	return ret;
2946}
2947
2948static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2949						   int family)
2950{
2951	switch (family) {
2952	case AF_INET:
2953		return __in_dev_arp_parms_get_rcu(dev);
2954	case AF_INET6:
2955		return __in6_dev_nd_parms_get_rcu(dev);
2956	}
2957	return NULL;
2958}
2959
2960static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2961				  int index)
2962{
2963	struct net_device *dev;
2964	int family = neigh_parms_family(p);
2965
2966	rcu_read_lock();
2967	for_each_netdev_rcu(net, dev) {
2968		struct neigh_parms *dst_p =
2969				neigh_get_dev_parms_rcu(dev, family);
2970
2971		if (dst_p && !test_bit(index, dst_p->data_state))
2972			dst_p->data[index] = p->data[index];
2973	}
2974	rcu_read_unlock();
2975}
2976
2977static void neigh_proc_update(struct ctl_table *ctl, int write)
2978{
2979	struct net_device *dev = ctl->extra1;
2980	struct neigh_parms *p = ctl->extra2;
2981	struct net *net = neigh_parms_net(p);
2982	int index = (int *) ctl->data - p->data;
2983
2984	if (!write)
2985		return;
2986
2987	set_bit(index, p->data_state);
2988	if (index == NEIGH_VAR_DELAY_PROBE_TIME)
2989		call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2990	if (!dev) /* NULL dev means this is default value */
2991		neigh_copy_dflt_parms(net, p, index);
2992}
2993
2994static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2995					   void __user *buffer,
2996					   size_t *lenp, loff_t *ppos)
2997{
2998	struct ctl_table tmp = *ctl;
2999	int ret;
3000
3001	tmp.extra1 = &zero;
3002	tmp.extra2 = &int_max;
3003
3004	ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3005	neigh_proc_update(ctl, write);
3006	return ret;
3007}
3008
3009int neigh_proc_dointvec(struct ctl_table *ctl, int write,
3010			void __user *buffer, size_t *lenp, loff_t *ppos)
3011{
3012	int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3013
3014	neigh_proc_update(ctl, write);
3015	return ret;
3016}
3017EXPORT_SYMBOL(neigh_proc_dointvec);
3018
3019int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
3020				void __user *buffer,
3021				size_t *lenp, loff_t *ppos)
3022{
3023	int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3024
3025	neigh_proc_update(ctl, write);
3026	return ret;
3027}
3028EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3029
3030static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3031					      void __user *buffer,
3032					      size_t *lenp, loff_t *ppos)
3033{
3034	int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3035
3036	neigh_proc_update(ctl, write);
3037	return ret;
3038}
3039
3040int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3041				   void __user *buffer,
3042				   size_t *lenp, loff_t *ppos)
3043{
3044	int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3045
3046	neigh_proc_update(ctl, write);
3047	return ret;
3048}
3049EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3050
3051static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3052					  void __user *buffer,
3053					  size_t *lenp, loff_t *ppos)
3054{
3055	int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3056
3057	neigh_proc_update(ctl, write);
3058	return ret;
3059}
3060
3061static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3062					  void __user *buffer,
3063					  size_t *lenp, loff_t *ppos)
3064{
3065	struct neigh_parms *p = ctl->extra2;
3066	int ret;
3067
3068	if (strcmp(ctl->procname, "base_reachable_time") == 0)
3069		ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3070	else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3071		ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3072	else
3073		ret = -1;
3074
3075	if (write && ret == 0) {
3076		/* update reachable_time as well, otherwise, the change will
3077		 * only be effective after the next time neigh_periodic_work
3078		 * decides to recompute it
3079		 */
3080		p->reachable_time =
3081			neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3082	}
3083	return ret;
3084}
3085
3086#define NEIGH_PARMS_DATA_OFFSET(index)	\
3087	(&((struct neigh_parms *) 0)->data[index])
3088
3089#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3090	[NEIGH_VAR_ ## attr] = { \
3091		.procname	= name, \
3092		.data		= NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3093		.maxlen		= sizeof(int), \
3094		.mode		= mval, \
3095		.proc_handler	= proc, \
3096	}
3097
3098#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3099	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3100
3101#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3102	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3103
3104#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3105	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3106
3107#define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
3108	NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3109
3110#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3111	NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3112
3113#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3114	NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3115
3116static struct neigh_sysctl_table {
3117	struct ctl_table_header *sysctl_header;
3118	struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3119} neigh_sysctl_template __read_mostly = {
3120	.neigh_vars = {
3121		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3122		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3123		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3124		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3125		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3126		NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3127		NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3128		NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3129		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3130		NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3131		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3132		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3133		NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3134		NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3135		NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3136		NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3137		[NEIGH_VAR_GC_INTERVAL] = {
3138			.procname	= "gc_interval",
3139			.maxlen		= sizeof(int),
3140			.mode		= 0644,
3141			.proc_handler	= proc_dointvec_jiffies,
3142		},
3143		[NEIGH_VAR_GC_THRESH1] = {
3144			.procname	= "gc_thresh1",
3145			.maxlen		= sizeof(int),
3146			.mode		= 0644,
3147			.extra1 	= &zero,
3148			.extra2		= &int_max,
3149			.proc_handler	= proc_dointvec_minmax,
3150		},
3151		[NEIGH_VAR_GC_THRESH2] = {
3152			.procname	= "gc_thresh2",
3153			.maxlen		= sizeof(int),
3154			.mode		= 0644,
3155			.extra1 	= &zero,
3156			.extra2		= &int_max,
3157			.proc_handler	= proc_dointvec_minmax,
3158		},
3159		[NEIGH_VAR_GC_THRESH3] = {
3160			.procname	= "gc_thresh3",
3161			.maxlen		= sizeof(int),
3162			.mode		= 0644,
3163			.extra1 	= &zero,
3164			.extra2		= &int_max,
3165			.proc_handler	= proc_dointvec_minmax,
3166		},
3167		{},
3168	},
3169};
3170
3171int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3172			  proc_handler *handler)
3173{
3174	int i;
3175	struct neigh_sysctl_table *t;
3176	const char *dev_name_source;
3177	char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3178	char *p_name;
3179
3180	t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3181	if (!t)
3182		goto err;
3183
3184	for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3185		t->neigh_vars[i].data += (long) p;
3186		t->neigh_vars[i].extra1 = dev;
3187		t->neigh_vars[i].extra2 = p;
3188	}
3189
3190	if (dev) {
3191		dev_name_source = dev->name;
3192		/* Terminate the table early */
3193		memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3194		       sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3195	} else {
3196		struct neigh_table *tbl = p->tbl;
3197		dev_name_source = "default";
3198		t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3199		t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3200		t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3201		t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3202	}
3203
3204	if (handler) {
3205		/* RetransTime */
3206		t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3207		/* ReachableTime */
3208		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3209		/* RetransTime (in milliseconds)*/
3210		t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3211		/* ReachableTime (in milliseconds) */
3212		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3213	} else {
3214		/* Those handlers will update p->reachable_time after
3215		 * base_reachable_time(_ms) is set to ensure the new timer starts being
3216		 * applied after the next neighbour update instead of waiting for
3217		 * neigh_periodic_work to update its value (can be multiple minutes)
3218		 * So any handler that replaces them should do this as well
3219		 */
3220		/* ReachableTime */
3221		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3222			neigh_proc_base_reachable_time;
3223		/* ReachableTime (in milliseconds) */
3224		t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3225			neigh_proc_base_reachable_time;
3226	}
3227
3228	/* Don't export sysctls to unprivileged users */
3229	if (neigh_parms_net(p)->user_ns != &init_user_ns)
3230		t->neigh_vars[0].procname = NULL;
3231
3232	switch (neigh_parms_family(p)) {
3233	case AF_INET:
3234	      p_name = "ipv4";
3235	      break;
3236	case AF_INET6:
3237	      p_name = "ipv6";
3238	      break;
3239	default:
3240	      BUG();
3241	}
3242
3243	snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3244		p_name, dev_name_source);
3245	t->sysctl_header =
3246		register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3247	if (!t->sysctl_header)
3248		goto free;
3249
3250	p->sysctl_table = t;
3251	return 0;
3252
3253free:
3254	kfree(t);
3255err:
3256	return -ENOBUFS;
3257}
3258EXPORT_SYMBOL(neigh_sysctl_register);
3259
3260void neigh_sysctl_unregister(struct neigh_parms *p)
3261{
3262	if (p->sysctl_table) {
3263		struct neigh_sysctl_table *t = p->sysctl_table;
3264		p->sysctl_table = NULL;
3265		unregister_net_sysctl_table(t->sysctl_header);
3266		kfree(t);
3267	}
3268}
3269EXPORT_SYMBOL(neigh_sysctl_unregister);
3270
3271#endif	/* CONFIG_SYSCTL */
3272
3273static int __init neigh_init(void)
3274{
3275	rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3276	rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3277	rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, 0);
3278
3279	rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3280		      0);
3281	rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3282
3283	return 0;
3284}
3285
3286subsys_initcall(neigh_init);
3287