[linux-2.6-block.git] / net / ipv4 / inetpeer.c

/*
 *		INETPEER - A storage for permanent information about peers
 *
 *  This source is covered by the GNU GPL, the same as all kernel sources.
 *
 *  Authors:	Andrey V. Savochkin <saw@msu.ru>
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/interrupt.h>
#include <linux/spinlock.h>
#include <linux/random.h>
#include <linux/timer.h>
#include <linux/time.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/net.h>
#include <linux/workqueue.h>
#include <net/ip.h>
#include <net/inetpeer.h>
#include <net/secure_seq.h>

/*
 *  Theory of operations.
 *  We keep one entry for each peer IP address.  The nodes contains long-living
 *  information about the peer which doesn't depend on routes.
 *
 *  Nodes are removed only when reference counter goes to 0.
 *  When it's happened the node may be removed when a sufficient amount of
 *  time has been passed since its last use.  The less-recently-used entry can
 *  also be removed if the pool is overloaded i.e. if the total amount of
 *  entries is greater-or-equal than the threshold.
 *
 *  Node pool is organised as an AVL tree.
 *  Such an implementation has been chosen not just for fun.  It's a way to
 *  prevent easy and efficient DoS attacks by creating hash collisions.  A huge
 *  amount of long living nodes in a single hash slot would significantly delay
 *  lookups performed with disabled BHs.
 *
 *  Serialisation issues.
 *  1.  Nodes may appear in the tree only with the pool lock held.
 *  2.  Nodes may disappear from the tree only with the pool lock held
 *      AND reference count being 0.
 *  3.  Global variable peer_total is modified under the pool lock.
 *  4.  struct inet_peer fields modification:
 *		avl_left, avl_right, avl_parent, avl_height: pool lock
 *		refcnt: atomically against modifications on other CPU;
 *		   usually under some other lock to prevent node disappearing
 *		daddr: unchangeable
 */

static struct kmem_cache *peer_cachep __read_mostly;

static LIST_HEAD(gc_list);
static const int gc_delay = 60 * HZ;
static struct delayed_work gc_work;
static DEFINE_SPINLOCK(gc_lock);

#define node_height(x) x->avl_height

#define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
#define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node)
static const struct inet_peer peer_fake_node = {
	.avl_left	= peer_avl_empty_rcu,
	.avl_right	= peer_avl_empty_rcu,
	.avl_height	= 0
};

void inet_peer_base_init(struct inet_peer_base *bp)
{
	bp->root = peer_avl_empty_rcu;
	seqlock_init(&bp->lock);
	bp->total = 0;
}
EXPORT_SYMBOL_GPL(inet_peer_base_init);

#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */

/* Exported for sysctl_net_ipv4.  */
int inet_peer_threshold __read_mostly = 65536 + 128;	/* start to throw entries more
					 * aggressively at this stage */
int inet_peer_minttl __read_mostly = 120 * HZ;	/* TTL under high load: 120 sec */
int inet_peer_maxttl __read_mostly = 10 * 60 * HZ;	/* usual time to live: 10 min */

static void inetpeer_gc_worker(struct work_struct *work)
{
	struct inet_peer *p, *n, *c;
	struct list_head list;

	spin_lock_bh(&gc_lock);
	list_replace_init(&gc_list, &list);
	spin_unlock_bh(&gc_lock);

	if (list_empty(&list))
		return;

	list_for_each_entry_safe(p, n, &list, gc_list) {

		if (need_resched())
			cond_resched();

		c = rcu_dereference_protected(p->avl_left, 1);
		if (c != peer_avl_empty) {
			list_add_tail(&c->gc_list, &list);
			p->avl_left = peer_avl_empty_rcu;
		}

		c = rcu_dereference_protected(p->avl_right, 1);
		if (c != peer_avl_empty) {
			list_add_tail(&c->gc_list, &list);
			p->avl_right = peer_avl_empty_rcu;
		}

		n = list_entry(p->gc_list.next, struct inet_peer, gc_list);

		if (!atomic_read(&p->refcnt)) {
			list_del(&p->gc_list);
			kmem_cache_free(peer_cachep, p);
		}
	}

	if (list_empty(&list))
		return;

	spin_lock_bh(&gc_lock);
	list_splice(&list, &gc_list);
	spin_unlock_bh(&gc_lock);

	schedule_delayed_work(&gc_work, gc_delay);
}

/* Called from ip_output.c:ip_init  */
void __init inet_initpeers(void)
{
	struct sysinfo si;

	/* Use the straight interface to information about memory. */
	si_meminfo(&si);
	/* The values below were suggested by Alexey Kuznetsov
	 * <kuznet@ms2.inr.ac.ru>.  I don't have any opinion about the values
	 * myself.  --SAW
	 */
	if (si.totalram <= (32768*1024)/PAGE_SIZE)
		inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
	if (si.totalram <= (16384*1024)/PAGE_SIZE)
		inet_peer_threshold >>= 1; /* about 512KB */
	if (si.totalram <= (8192*1024)/PAGE_SIZE)
		inet_peer_threshold >>= 2; /* about 128KB */

	peer_cachep = kmem_cache_create("inet_peer_cache",
			sizeof(struct inet_peer),
			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
			NULL);

	INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
}

static int addr_compare(const struct inetpeer_addr *a,
			const struct inetpeer_addr *b)
{
	int i, n = (a->family == AF_INET ? 1 : 4);

	for (i = 0; i < n; i++) {
		if (a->addr.a6[i] == b->addr.a6[i])
			continue;
		if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i])
			return -1;
		return 1;
	}

	return 0;
}

#define rcu_deref_locked(X, BASE)				\
	rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))

/*
 * Called with local BH disabled and the pool lock held.
 */
#define lookup(_daddr, _stack, _base)				\
({								\
	struct inet_peer *u;					\
	struct inet_peer __rcu **v;				\
								\
	stackptr = _stack;					\
	*stackptr++ = &_base->root;				\
	for (u = rcu_deref_locked(_base->root, _base);		\
	     u != peer_avl_empty;) {				\
		int cmp = addr_compare(_daddr, &u->daddr);	\
		if (cmp == 0)					\
			break;					\
		if (cmp == -1)					\
			v = &u->avl_left;			\
		else						\
			v = &u->avl_right;			\
		*stackptr++ = v;				\
		u = rcu_deref_locked(*v, _base);		\
	}							\
	u;							\
})

/*
 * Called with rcu_read_lock()
 * Because we hold no lock against a writer, its quite possible we fall
 * in an endless loop.
 * But every pointer we follow is guaranteed to be valid thanks to RCU.
 * We exit from this function if number of links exceeds PEER_MAXDEPTH
 */
static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
				    struct inet_peer_base *base)
{
	struct inet_peer *u = rcu_dereference(base->root);
	int count = 0;

	while (u != peer_avl_empty) {
		int cmp = addr_compare(daddr, &u->daddr);
		if (cmp == 0) {
			/* Before taking a reference, check if this entry was
			 * deleted (refcnt=-1)
			 */
			if (!atomic_add_unless(&u->refcnt, 1, -1))
				u = NULL;
			return u;
		}
		if (cmp == -1)
			u = rcu_dereference(u->avl_left);
		else
			u = rcu_dereference(u->avl_right);
		if (unlikely(++count == PEER_MAXDEPTH))
			break;
	}
	return NULL;
}

/* Called with local BH disabled and the pool lock held. */
#define lookup_rightempty(start, base)				\
({								\
	struct inet_peer *u;					\
	struct inet_peer __rcu **v;				\
	*stackptr++ = &start->avl_left;				\
	v = &start->avl_left;					\
	for (u = rcu_deref_locked(*v, base);			\
	     u->avl_right != peer_avl_empty_rcu;) {		\
		v = &u->avl_right;				\
		*stackptr++ = v;				\
		u = rcu_deref_locked(*v, base);			\
	}							\
	u;							\
})

/* Called with local BH disabled and the pool lock held.
 * Variable names are the proof of operation correctness.
 * Look into mm/map_avl.c for more detail description of the ideas.
 */
static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
			       struct inet_peer __rcu ***stackend,
			       struct inet_peer_base *base)
{
	struct inet_peer __rcu **nodep;
	struct inet_peer *node, *l, *r;
	int lh, rh;

	while (stackend > stack) {
		nodep = *--stackend;
		node = rcu_deref_locked(*nodep, base);
		l = rcu_deref_locked(node->avl_left, base);
		r = rcu_deref_locked(node->avl_right, base);
		lh = node_height(l);
		rh = node_height(r);
		if (lh > rh + 1) { /* l: RH+2 */
			struct inet_peer *ll, *lr, *lrl, *lrr;
			int lrh;
			ll = rcu_deref_locked(l->avl_left, base);
			lr = rcu_deref_locked(l->avl_right, base);
			lrh = node_height(lr);
			if (lrh <= node_height(ll)) {	/* ll: RH+1 */
				RCU_INIT_POINTER(node->avl_left, lr);	/* lr: RH or RH+1 */
				RCU_INIT_POINTER(node->avl_right, r);	/* r: RH */
				node->avl_height = lrh + 1; /* RH+1 or RH+2 */
				RCU_INIT_POINTER(l->avl_left, ll);       /* ll: RH+1 */
				RCU_INIT_POINTER(l->avl_right, node);	/* node: RH+1 or RH+2 */
				l->avl_height = node->avl_height + 1;
				RCU_INIT_POINTER(*nodep, l);
			} else { /* ll: RH, lr: RH+1 */
				lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
				lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
				RCU_INIT_POINTER(node->avl_left, lrr);	/* lrr: RH or RH-1 */
				RCU_INIT_POINTER(node->avl_right, r);	/* r: RH */
				node->avl_height = rh + 1; /* node: RH+1 */
				RCU_INIT_POINTER(l->avl_left, ll);	/* ll: RH */
				RCU_INIT_POINTER(l->avl_right, lrl);	/* lrl: RH or RH-1 */
				l->avl_height = rh + 1;	/* l: RH+1 */
				RCU_INIT_POINTER(lr->avl_left, l);	/* l: RH+1 */
				RCU_INIT_POINTER(lr->avl_right, node);	/* node: RH+1 */
				lr->avl_height = rh + 2;
				RCU_INIT_POINTER(*nodep, lr);
			}
		} else if (rh > lh + 1) { /* r: LH+2 */
			struct inet_peer *rr, *rl, *rlr, *rll;
			int rlh;
			rr = rcu_deref_locked(r->avl_right, base);
			rl = rcu_deref_locked(r->avl_left, base);
			rlh = node_height(rl);
			if (rlh <= node_height(rr)) {	/* rr: LH+1 */
				RCU_INIT_POINTER(node->avl_right, rl);	/* rl: LH or LH+1 */
				RCU_INIT_POINTER(node->avl_left, l);	/* l: LH */
				node->avl_height = rlh + 1; /* LH+1 or LH+2 */
				RCU_INIT_POINTER(r->avl_right, rr);	/* rr: LH+1 */
				RCU_INIT_POINTER(r->avl_left, node);	/* node: LH+1 or LH+2 */
				r->avl_height = node->avl_height + 1;
				RCU_INIT_POINTER(*nodep, r);
			} else { /* rr: RH, rl: RH+1 */
				rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
				rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
				RCU_INIT_POINTER(node->avl_right, rll);	/* rll: LH or LH-1 */
				RCU_INIT_POINTER(node->avl_left, l);	/* l: LH */
				node->avl_height = lh + 1; /* node: LH+1 */
				RCU_INIT_POINTER(r->avl_right, rr);	/* rr: LH */
				RCU_INIT_POINTER(r->avl_left, rlr);	/* rlr: LH or LH-1 */
				r->avl_height = lh + 1;	/* r: LH+1 */
				RCU_INIT_POINTER(rl->avl_right, r);	/* r: LH+1 */
				RCU_INIT_POINTER(rl->avl_left, node);	/* node: LH+1 */
				rl->avl_height = lh + 2;
				RCU_INIT_POINTER(*nodep, rl);
			}
		} else {
			node->avl_height = (lh > rh ? lh : rh) + 1;
		}
	}
}

/* Called with local BH disabled and the pool lock held. */
#define link_to_pool(n, base)					\
do {								\
	n->avl_height = 1;					\
	n->avl_left = peer_avl_empty_rcu;			\
	n->avl_right = peer_avl_empty_rcu;			\
	/* lockless readers can catch us now */			\
	rcu_assign_pointer(**--stackptr, n);			\
	peer_avl_rebalance(stack, stackptr, base);		\
} while (0)

static void inetpeer_free_rcu(struct rcu_head *head)
{
	kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
}

static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
			     struct inet_peer __rcu **stack[PEER_MAXDEPTH])
{
	struct inet_peer __rcu ***stackptr, ***delp;

	if (lookup(&p->daddr, stack, base) != p)
		BUG();
	delp = stackptr - 1; /* *delp[0] == p */
	if (p->avl_left == peer_avl_empty_rcu) {
		*delp[0] = p->avl_right;
		--stackptr;
	} else {
		/* look for a node to insert instead of p */
		struct inet_peer *t;
		t = lookup_rightempty(p, base);
		BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
		**--stackptr = t->avl_left;
		/* t is removed, t->daddr > x->daddr for any
		 * x in p->avl_left subtree.
		 * Put t in the old place of p. */
		RCU_INIT_POINTER(*delp[0], t);
		t->avl_left = p->avl_left;
		t->avl_right = p->avl_right;
		t->avl_height = p->avl_height;
		BUG_ON(delp[1] != &p->avl_left);
		delp[1] = &t->avl_left; /* was &p->avl_left */
	}
	peer_avl_rebalance(stack, stackptr, base);
	base->total--;
	call_rcu(&p->rcu, inetpeer_free_rcu);
}

/* perform garbage collect on all items stacked during a lookup */
static int inet_peer_gc(struct inet_peer_base *base,
			struct inet_peer __rcu **stack[PEER_MAXDEPTH],
			struct inet_peer __rcu ***stackptr)
{
	struct inet_peer *p, *gchead = NULL;
	__u32 delta, ttl;
	int cnt = 0;

	if (base->total >= inet_peer_threshold)
		ttl = 0; /* be aggressive */
	else
		ttl = inet_peer_maxttl
				- (inet_peer_maxttl - inet_peer_minttl) / HZ *
					base->total / inet_peer_threshold * HZ;
	stackptr--; /* last stack slot is peer_avl_empty */
	while (stackptr > stack) {
		stackptr--;
		p = rcu_deref_locked(**stackptr, base);
		if (atomic_read(&p->refcnt) == 0) {
			smp_rmb();
			delta = (__u32)jiffies - p->dtime;
			if (delta >= ttl &&
			    atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
				p->gc_next = gchead;
				gchead = p;
			}
		}
	}
	while ((p = gchead) != NULL) {
		gchead = p->gc_next;
		cnt++;
		unlink_from_pool(p, base, stack);
	}
	return cnt;
}

struct inet_peer *inet_getpeer(struct inet_peer_base *base,
			       const struct inetpeer_addr *daddr,
			       int create)
{
	struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
	struct inet_peer *p;
	unsigned int sequence;
	int invalidated, gccnt = 0;

	/* Attempt a lockless lookup first.
	 * Because of a concurrent writer, we might not find an existing entry.
	 */
	rcu_read_lock();
	sequence = read_seqbegin(&base->lock);
	p = lookup_rcu(daddr, base);
	invalidated = read_seqretry(&base->lock, sequence);
	rcu_read_unlock();

	if (p)
		return p;

	/* If no writer did a change during our lookup, we can return early. */
	if (!create && !invalidated)
		return NULL;

	/* retry an exact lookup, taking the lock before.
	 * At least, nodes should be hot in our cache.
	 */
	write_seqlock_bh(&base->lock);
relookup:
	p = lookup(daddr, stack, base);
	if (p != peer_avl_empty) {
		atomic_inc(&p->refcnt);
		write_sequnlock_bh(&base->lock);
		return p;
	}
	if (!gccnt) {
		gccnt = inet_peer_gc(base, stack, stackptr);
		if (gccnt && create)
			goto relookup;
	}
	p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
	if (p) {
		p->daddr = *daddr;
		atomic_set(&p->refcnt, 1);
		atomic_set(&p->rid, 0);
		p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
		p->rate_tokens = 0;
		/* 60*HZ is arbitrary, but chosen enough high so that the first
		 * calculation of tokens is at its maximum.
		 */
		p->rate_last = jiffies - 60*HZ;
		INIT_LIST_HEAD(&p->gc_list);

		/* Link the node. */
		link_to_pool(p, base);
		base->total++;
	}
	write_sequnlock_bh(&base->lock);

	return p;
}
EXPORT_SYMBOL_GPL(inet_getpeer);

void inet_putpeer(struct inet_peer *p)
{
	p->dtime = (__u32)jiffies;
	smp_mb__before_atomic();
	atomic_dec(&p->refcnt);
}
EXPORT_SYMBOL_GPL(inet_putpeer);

/*
 *	Check transmit rate limitation for given message.
 *	The rate information is held in the inet_peer entries now.
 *	This function is generic and could be used for other purposes
 *	too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
 *
 *	Note that the same inet_peer fields are modified by functions in
 *	route.c too, but these work for packet destinations while xrlim_allow
 *	works for icmp destinations. This means the rate limiting information
 *	for one "ip object" is shared - and these ICMPs are twice limited:
 *	by source and by destination.
 *
 *	RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
 *			  SHOULD allow setting of rate limits
 *
 * 	Shared between ICMPv4 and ICMPv6.
 */
#define XRLIM_BURST_FACTOR 6
bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
{
	unsigned long now, token;
	bool rc = false;

	if (!peer)
		return true;

	token = peer->rate_tokens;
	now = jiffies;
	token += now - peer->rate_last;
	peer->rate_last = now;
	if (token > XRLIM_BURST_FACTOR * timeout)
		token = XRLIM_BURST_FACTOR * timeout;
	if (token >= timeout) {
		token -= timeout;
		rc = true;
	}
	peer->rate_tokens = token;
	return rc;
}
EXPORT_SYMBOL(inet_peer_xrlim_allow);

static void inetpeer_inval_rcu(struct rcu_head *head)
{
	struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);

	spin_lock_bh(&gc_lock);
	list_add_tail(&p->gc_list, &gc_list);
	spin_unlock_bh(&gc_lock);

	schedule_delayed_work(&gc_work, gc_delay);
}

void inetpeer_invalidate_tree(struct inet_peer_base *base)
{
	struct inet_peer *root;

	write_seqlock_bh(&base->lock);

	root = rcu_deref_locked(base->root, base);
	if (root != peer_avl_empty) {
		base->root = peer_avl_empty_rcu;
		base->total = 0;
		call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
	}

	write_sequnlock_bh(&base->lock);
}
EXPORT_SYMBOL(inetpeer_invalidate_tree);
Commit	Line	Data
1da177e4 LT	1	/*
	2	* INETPEER - A storage for permanent information about peers
	3	*
	4	* This source is covered by the GNU GPL, the same as all kernel sources.
	5	*
1da177e4 LT	6	* Authors: Andrey V. Savochkin <saw@msu.ru>
	7	*/
	8
	9	#include <linux/module.h>
	10	#include <linux/types.h>
	11	#include <linux/slab.h>
	12	#include <linux/interrupt.h>
	13	#include <linux/spinlock.h>
	14	#include <linux/random.h>
1da177e4 LT	15	#include <linux/timer.h>
	16	#include <linux/time.h>
	17	#include <linux/kernel.h>
	18	#include <linux/mm.h>
	19	#include <linux/net.h>
5faa5df1	20	#include <linux/workqueue.h>
20380731	21	#include <net/ip.h>
1da177e4	22	#include <net/inetpeer.h>
6e5714ea	23	#include <net/secure_seq.h>
1da177e4 LT	24
	25	/*
	26	* Theory of operations.
	27	* We keep one entry for each peer IP address. The nodes contains long-living
	28	* information about the peer which doesn't depend on routes.
1da177e4	29	*
1da177e4 LT	30	* Nodes are removed only when reference counter goes to 0.
	31	* When it's happened the node may be removed when a sufficient amount of
	32	* time has been passed since its last use. The less-recently-used entry can
	33	* also be removed if the pool is overloaded i.e. if the total amount of
	34	* entries is greater-or-equal than the threshold.
	35	*
	36	* Node pool is organised as an AVL tree.
	37	* Such an implementation has been chosen not just for fun. It's a way to
	38	* prevent easy and efficient DoS attacks by creating hash collisions. A huge
	39	* amount of long living nodes in a single hash slot would significantly delay
	40	* lookups performed with disabled BHs.
	41	*
	42	* Serialisation issues.
aa1039e7 ED	43	* 1. Nodes may appear in the tree only with the pool lock held.
aa1039e7 ED	44	* 2. Nodes may disappear from the tree only with the pool lock held
1da177e4	45	* AND reference count being 0.
4b9d9be8 ED	46	* 3. Global variable peer_total is modified under the pool lock.
4b9d9be8 ED	47	* 4. struct inet_peer fields modification:
1da177e4	48	* avl_left, avl_right, avl_parent, avl_height: pool lock
1da177e4 LT	49	* refcnt: atomically against modifications on other CPU;
1da177e4 LT	50	* usually under some other lock to prevent node disappearing
582a72da	51	* daddr: unchangeable
1da177e4 LT	52	*/
1da177e4 LT	53
e18b890b	54	static struct kmem_cache *peer_cachep __read_mostly;
1da177e4	55
5faa5df1 SK	56	static LIST_HEAD(gc_list);
	57	static const int gc_delay = 60 * HZ;
	58	static struct delayed_work gc_work;
	59	static DEFINE_SPINLOCK(gc_lock);
	60
1da177e4	61	#define node_height(x) x->avl_height
d6cc1d64 ED	62
d6cc1d64 ED	63	#define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
b914c4ea	64	#define peer_avl_empty_rcu ((struct inet_peer __rcu __force *)&peer_fake_node)
d6cc1d64	65	static const struct inet_peer peer_fake_node = {
b914c4ea ED	66	.avl_left = peer_avl_empty_rcu,
b914c4ea ED	67	.avl_right = peer_avl_empty_rcu,
1da177e4 LT	68	.avl_height = 0
1da177e4 LT	69	};
d6cc1d64	70
c3426b47 DM	71	void inet_peer_base_init(struct inet_peer_base *bp)
	72	{
	73	bp->root = peer_avl_empty_rcu;
	74	seqlock_init(&bp->lock);
	75	bp->total = 0;
	76	}
	77	EXPORT_SYMBOL_GPL(inet_peer_base_init);
021e9299	78
1da177e4 LT	79	#define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
1da177e4 LT	80
1da177e4	81	/* Exported for sysctl_net_ipv4. */
243bbcaa	82	int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
1da177e4	83	* aggressively at this stage */
243bbcaa ED	84	int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
243bbcaa ED	85	int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
1da177e4	86
5faa5df1 SK	87	static void inetpeer_gc_worker(struct work_struct *work)
5faa5df1 SK	88	{
da557374	89	struct inet_peer p, n, *c;
851bdd11	90	struct list_head list;
5faa5df1 SK	91
	92	spin_lock_bh(&gc_lock);
	93	list_replace_init(&gc_list, &list);
	94	spin_unlock_bh(&gc_lock);
	95
	96	if (list_empty(&list))
	97	return;
	98
	99	list_for_each_entry_safe(p, n, &list, gc_list) {
	100
da557374	101	if (need_resched())
5faa5df1 SK	102	cond_resched();
5faa5df1 SK	103
da557374 ED	104	c = rcu_dereference_protected(p->avl_left, 1);
	105	if (c != peer_avl_empty) {
	106	list_add_tail(&c->gc_list, &list);
	107	p->avl_left = peer_avl_empty_rcu;
5faa5df1 SK	108	}
5faa5df1 SK	109
da557374 ED	110	c = rcu_dereference_protected(p->avl_right, 1);
	111	if (c != peer_avl_empty) {
	112	list_add_tail(&c->gc_list, &list);
	113	p->avl_right = peer_avl_empty_rcu;
5faa5df1 SK	114	}
	115
	116	n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
	117
	118	if (!atomic_read(&p->refcnt)) {
	119	list_del(&p->gc_list);
	120	kmem_cache_free(peer_cachep, p);
	121	}
	122	}
	123
	124	if (list_empty(&list))
	125	return;
	126
	127	spin_lock_bh(&gc_lock);
	128	list_splice(&list, &gc_list);
	129	spin_unlock_bh(&gc_lock);
	130
	131	schedule_delayed_work(&gc_work, gc_delay);
	132	}
1da177e4 LT	133
	134	/* Called from ip_output.c:ip_init */
	135	void __init inet_initpeers(void)
	136	{
	137	struct sysinfo si;
	138
	139	/* Use the straight interface to information about memory. */
	140	si_meminfo(&si);
	141	/* The values below were suggested by Alexey Kuznetsov
	142	* <kuznet@ms2.inr.ac.ru>. I don't have any opinion about the values
	143	* myself. --SAW
	144	*/
	145	if (si.totalram <= (32768*1024)/PAGE_SIZE)
	146	inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
	147	if (si.totalram <= (16384*1024)/PAGE_SIZE)
	148	inet_peer_threshold >>= 1; /* about 512KB */
	149	if (si.totalram <= (8192*1024)/PAGE_SIZE)
	150	inet_peer_threshold >>= 2; /* about 128KB */
	151
	152	peer_cachep = kmem_cache_create("inet_peer_cache",
	153	sizeof(struct inet_peer),
317fe0e6	154	0, SLAB_HWCACHE_ALIGN \| SLAB_PANIC,
20c2df83	155	NULL);
1da177e4	156
203b42f7	157	INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
1da177e4 LT	158	}
1da177e4 LT	159
8790ca17 DM	160	static int addr_compare(const struct inetpeer_addr *a,
8790ca17 DM	161	const struct inetpeer_addr *b)
02663045 DM	162	{
	163	int i, n = (a->family == AF_INET ? 1 : 4);
	164
	165	for (i = 0; i < n; i++) {
7a71ed89	166	if (a->addr.a6[i] == b->addr.a6[i])
02663045	167	continue;
747465ef	168	if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i])
02663045 DM	169	return -1;
	170	return 1;
	171	}
	172
	173	return 0;
	174	}
	175
65e8354e ED	176	#define rcu_deref_locked(X, BASE) \
	177	rcu_dereference_protected(X, lockdep_is_held(&(BASE)->lock.lock))
	178
243bbcaa ED	179	/*
243bbcaa ED	180	* Called with local BH disabled and the pool lock held.
243bbcaa	181	*/
98158f5a	182	#define lookup(_daddr, _stack, _base) \
1da177e4	183	({ \
b914c4ea ED	184	struct inet_peer *u; \
b914c4ea ED	185	struct inet_peer __rcu **v; \
aa1039e7 ED	186	\
aa1039e7 ED	187	stackptr = _stack; \
98158f5a	188	*stackptr++ = &_base->root; \
65e8354e	189	for (u = rcu_deref_locked(_base->root, _base); \
0c9a67d2	190	u != peer_avl_empty;) { \
02663045 DM	191	int cmp = addr_compare(_daddr, &u->daddr); \
02663045 DM	192	if (cmp == 0) \
1da177e4	193	break; \
02663045	194	if (cmp == -1) \
1da177e4 LT	195	v = &u->avl_left; \
	196	else \
	197	v = &u->avl_right; \
aa1039e7	198	*stackptr++ = v; \
65e8354e	199	u = rcu_deref_locked(*v, _base); \
1da177e4 LT	200	} \
	201	u; \
	202	})
	203
aa1039e7	204	/*
7b46ac4e	205	* Called with rcu_read_lock()
aa1039e7 ED	206	* Because we hold no lock against a writer, its quite possible we fall
	207	* in an endless loop.
	208	* But every pointer we follow is guaranteed to be valid thanks to RCU.
	209	* We exit from this function if number of links exceeds PEER_MAXDEPTH
	210	*/
7b46ac4e	211	static struct inet_peer lookup_rcu(const struct inetpeer_addr daddr,
4b9d9be8	212	struct inet_peer_base *base)
aa1039e7	213	{
7b46ac4e	214	struct inet_peer *u = rcu_dereference(base->root);
aa1039e7 ED	215	int count = 0;
	216
	217	while (u != peer_avl_empty) {
02663045 DM	218	int cmp = addr_compare(daddr, &u->daddr);
02663045 DM	219	if (cmp == 0) {
5f2f8920	220	/* Before taking a reference, check if this entry was
4b9d9be8	221	* deleted (refcnt=-1)
5f2f8920	222	*/
4b9d9be8	223	if (!atomic_add_unless(&u->refcnt, 1, -1))
aa1039e7 ED	224	u = NULL;
	225	return u;
	226	}
02663045	227	if (cmp == -1)
7b46ac4e	228	u = rcu_dereference(u->avl_left);
aa1039e7	229	else
7b46ac4e	230	u = rcu_dereference(u->avl_right);
aa1039e7 ED	231	if (unlikely(++count == PEER_MAXDEPTH))
	232	break;
	233	}
	234	return NULL;
	235	}
	236
	237	/* Called with local BH disabled and the pool lock held. */
98158f5a	238	#define lookup_rightempty(start, base) \
1da177e4	239	({ \
b914c4ea ED	240	struct inet_peer *u; \
b914c4ea ED	241	struct inet_peer __rcu **v; \
1da177e4 LT	242	*stackptr++ = &start->avl_left; \
1da177e4 LT	243	v = &start->avl_left; \
65e8354e	244	for (u = rcu_deref_locked(*v, base); \
0c9a67d2	245	u->avl_right != peer_avl_empty_rcu;) { \
1da177e4 LT	246	v = &u->avl_right; \
1da177e4 LT	247	*stackptr++ = v; \
65e8354e	248	u = rcu_deref_locked(*v, base); \
1da177e4 LT	249	} \
	250	u; \
	251	})
	252
aa1039e7	253	/* Called with local BH disabled and the pool lock held.
1da177e4	254	* Variable names are the proof of operation correctness.
aa1039e7 ED	255	* Look into mm/map_avl.c for more detail description of the ideas.
aa1039e7 ED	256	*/
b914c4ea	257	static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
98158f5a DM	258	struct inet_peer __rcu ***stackend,
98158f5a DM	259	struct inet_peer_base *base)
1da177e4	260	{
b914c4ea ED	261	struct inet_peer __rcu **nodep;
b914c4ea ED	262	struct inet_peer node, l, *r;
1da177e4 LT	263	int lh, rh;
	264
	265	while (stackend > stack) {
	266	nodep = *--stackend;
65e8354e ED	267	node = rcu_deref_locked(*nodep, base);
	268	l = rcu_deref_locked(node->avl_left, base);
	269	r = rcu_deref_locked(node->avl_right, base);
1da177e4 LT	270	lh = node_height(l);
	271	rh = node_height(r);
	272	if (lh > rh + 1) { /* l: RH+2 */
	273	struct inet_peer ll, lr, lrl, lrr;
	274	int lrh;
65e8354e ED	275	ll = rcu_deref_locked(l->avl_left, base);
65e8354e ED	276	lr = rcu_deref_locked(l->avl_right, base);
1da177e4 LT	277	lrh = node_height(lr);
1da177e4 LT	278	if (lrh <= node_height(ll)) { /* ll: RH+1 */
b914c4ea ED	279	RCU_INIT_POINTER(node->avl_left, lr); /* lr: RH or RH+1 */
b914c4ea ED	280	RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
1da177e4	281	node->avl_height = lrh + 1; /* RH+1 or RH+2 */
b914c4ea ED	282	RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH+1 */
b914c4ea ED	283	RCU_INIT_POINTER(l->avl_right, node); /* node: RH+1 or RH+2 */
1da177e4	284	l->avl_height = node->avl_height + 1;
b914c4ea	285	RCU_INIT_POINTER(*nodep, l);
1da177e4	286	} else { /* ll: RH, lr: RH+1 */
65e8354e ED	287	lrl = rcu_deref_locked(lr->avl_left, base);/* lrl: RH or RH-1 */
65e8354e ED	288	lrr = rcu_deref_locked(lr->avl_right, base);/* lrr: RH or RH-1 */
b914c4ea ED	289	RCU_INIT_POINTER(node->avl_left, lrr); /* lrr: RH or RH-1 */
b914c4ea ED	290	RCU_INIT_POINTER(node->avl_right, r); /* r: RH */
1da177e4	291	node->avl_height = rh + 1; /* node: RH+1 */
b914c4ea ED	292	RCU_INIT_POINTER(l->avl_left, ll); /* ll: RH */
b914c4ea ED	293	RCU_INIT_POINTER(l->avl_right, lrl); /* lrl: RH or RH-1 */
1da177e4	294	l->avl_height = rh + 1; /* l: RH+1 */
b914c4ea ED	295	RCU_INIT_POINTER(lr->avl_left, l); /* l: RH+1 */
b914c4ea ED	296	RCU_INIT_POINTER(lr->avl_right, node); /* node: RH+1 */
1da177e4	297	lr->avl_height = rh + 2;
b914c4ea	298	RCU_INIT_POINTER(*nodep, lr);
1da177e4 LT	299	}
	300	} else if (rh > lh + 1) { /* r: LH+2 */
	301	struct inet_peer rr, rl, rlr, rll;
	302	int rlh;
65e8354e ED	303	rr = rcu_deref_locked(r->avl_right, base);
65e8354e ED	304	rl = rcu_deref_locked(r->avl_left, base);
1da177e4 LT	305	rlh = node_height(rl);
1da177e4 LT	306	if (rlh <= node_height(rr)) { /* rr: LH+1 */
b914c4ea ED	307	RCU_INIT_POINTER(node->avl_right, rl); /* rl: LH or LH+1 */
b914c4ea ED	308	RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
1da177e4	309	node->avl_height = rlh + 1; /* LH+1 or LH+2 */
b914c4ea ED	310	RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH+1 */
b914c4ea ED	311	RCU_INIT_POINTER(r->avl_left, node); /* node: LH+1 or LH+2 */
1da177e4	312	r->avl_height = node->avl_height + 1;
b914c4ea	313	RCU_INIT_POINTER(*nodep, r);
1da177e4	314	} else { /* rr: RH, rl: RH+1 */
65e8354e ED	315	rlr = rcu_deref_locked(rl->avl_right, base);/* rlr: LH or LH-1 */
65e8354e ED	316	rll = rcu_deref_locked(rl->avl_left, base);/* rll: LH or LH-1 */
b914c4ea ED	317	RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
b914c4ea ED	318	RCU_INIT_POINTER(node->avl_left, l); /* l: LH */
1da177e4	319	node->avl_height = lh + 1; /* node: LH+1 */
b914c4ea ED	320	RCU_INIT_POINTER(r->avl_right, rr); /* rr: LH */
b914c4ea ED	321	RCU_INIT_POINTER(r->avl_left, rlr); /* rlr: LH or LH-1 */
1da177e4	322	r->avl_height = lh + 1; /* r: LH+1 */
b914c4ea ED	323	RCU_INIT_POINTER(rl->avl_right, r); /* r: LH+1 */
b914c4ea ED	324	RCU_INIT_POINTER(rl->avl_left, node); /* node: LH+1 */
1da177e4	325	rl->avl_height = lh + 2;
b914c4ea	326	RCU_INIT_POINTER(*nodep, rl);
1da177e4 LT	327	}
	328	} else {
	329	node->avl_height = (lh > rh ? lh : rh) + 1;
	330	}
	331	}
	332	}
	333
aa1039e7	334	/* Called with local BH disabled and the pool lock held. */
98158f5a	335	#define link_to_pool(n, base) \
1da177e4 LT	336	do { \
1da177e4 LT	337	n->avl_height = 1; \
b914c4ea ED	338	n->avl_left = peer_avl_empty_rcu; \
	339	n->avl_right = peer_avl_empty_rcu; \
	340	/* lockless readers can catch us now */ \
	341	rcu_assign_pointer(**--stackptr, n); \
98158f5a	342	peer_avl_rebalance(stack, stackptr, base); \
d6cc1d64	343	} while (0)
1da177e4	344
aa1039e7 ED	345	static void inetpeer_free_rcu(struct rcu_head *head)
	346	{
	347	kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
	348	}
	349
66944e1c ED	350	static void unlink_from_pool(struct inet_peer p, struct inet_peer_base base,
66944e1c ED	351	struct inet_peer __rcu **stack[PEER_MAXDEPTH])
1da177e4	352	{
4b9d9be8 ED	353	struct inet_peer __rcu *stackptr, *delp;
	354
	355	if (lookup(&p->daddr, stack, base) != p)
	356	BUG();
	357	delp = stackptr - 1; /* delp[0] == p /
	358	if (p->avl_left == peer_avl_empty_rcu) {
	359	*delp[0] = p->avl_right;
	360	--stackptr;
	361	} else {
	362	/* look for a node to insert instead of p */
	363	struct inet_peer *t;
	364	t = lookup_rightempty(p, base);
	365	BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
	366	**--stackptr = t->avl_left;
	367	/* t is removed, t->daddr > x->daddr for any
	368	* x in p->avl_left subtree.
	369	* Put t in the old place of p. */
	370	RCU_INIT_POINTER(*delp[0], t);
	371	t->avl_left = p->avl_left;
	372	t->avl_right = p->avl_right;
	373	t->avl_height = p->avl_height;
	374	BUG_ON(delp[1] != &p->avl_left);
	375	delp[1] = &t->avl_left; /* was &p->avl_left */
1da177e4	376	}
4b9d9be8 ED	377	peer_avl_rebalance(stack, stackptr, base);
	378	base->total--;
	379	call_rcu(&p->rcu, inetpeer_free_rcu);
1da177e4 LT	380	}
1da177e4 LT	381
4b9d9be8 ED	382	/* perform garbage collect on all items stacked during a lookup */
	383	static int inet_peer_gc(struct inet_peer_base *base,
	384	struct inet_peer __rcu **stack[PEER_MAXDEPTH],
	385	struct inet_peer __rcu ***stackptr)
98158f5a	386	{
4b9d9be8 ED	387	struct inet_peer p, gchead = NULL;
	388	__u32 delta, ttl;
	389	int cnt = 0;
d71209de	390
4b9d9be8 ED	391	if (base->total >= inet_peer_threshold)
	392	ttl = 0; /* be aggressive */
	393	else
	394	ttl = inet_peer_maxttl
	395	- (inet_peer_maxttl - inet_peer_minttl) / HZ *
	396	base->total / inet_peer_threshold * HZ;
	397	stackptr--; /* last stack slot is peer_avl_empty */
	398	while (stackptr > stack) {
	399	stackptr--;
	400	p = rcu_deref_locked(**stackptr, base);
6d1a3e04 ED	401	if (atomic_read(&p->refcnt) == 0) {
	402	smp_rmb();
	403	delta = (__u32)jiffies - p->dtime;
	404	if (delta >= ttl &&
	405	atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
	406	p->gc_next = gchead;
	407	gchead = p;
	408	}
1da177e4	409	}
1da177e4	410	}
4b9d9be8 ED	411	while ((p = gchead) != NULL) {
	412	gchead = p->gc_next;
	413	cnt++;
	414	unlink_from_pool(p, base, stack);
	415	}
	416	return cnt;
1da177e4 LT	417	}
1da177e4 LT	418
c0efc887	419	struct inet_peer inet_getpeer(struct inet_peer_base base,
c8a627ed G	420	const struct inetpeer_addr *daddr,
c8a627ed G	421	int create)
1da177e4	422	{
b914c4ea	423	struct inet_peer __rcu stack[PEER_MAXDEPTH], *stackptr;
98158f5a	424	struct inet_peer *p;
65e8354e	425	unsigned int sequence;
4b9d9be8	426	int invalidated, gccnt = 0;
1da177e4	427
4b9d9be8	428	/* Attempt a lockless lookup first.
aa1039e7 ED	429	* Because of a concurrent writer, we might not find an existing entry.
aa1039e7 ED	430	*/
7b46ac4e	431	rcu_read_lock();
65e8354e	432	sequence = read_seqbegin(&base->lock);
4b9d9be8	433	p = lookup_rcu(daddr, base);
65e8354e	434	invalidated = read_seqretry(&base->lock, sequence);
7b46ac4e	435	rcu_read_unlock();
aa1039e7	436
4b9d9be8	437	if (p)
aa1039e7	438	return p;
1da177e4	439
65e8354e ED	440	/* If no writer did a change during our lookup, we can return early. */
	441	if (!create && !invalidated)
	442	return NULL;
	443
aa1039e7 ED	444	/* retry an exact lookup, taking the lock before.
	445	* At least, nodes should be hot in our cache.
	446	*/
65e8354e	447	write_seqlock_bh(&base->lock);
4b9d9be8	448	relookup:
02663045	449	p = lookup(daddr, stack, base);
1da177e4	450	if (p != peer_avl_empty) {
4b9d9be8	451	atomic_inc(&p->refcnt);
65e8354e	452	write_sequnlock_bh(&base->lock);
4b9d9be8 ED	453	return p;
	454	}
	455	if (!gccnt) {
	456	gccnt = inet_peer_gc(base, stack, stackptr);
	457	if (gccnt && create)
	458	goto relookup;
1da177e4	459	}
aa1039e7 ED	460	p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
aa1039e7 ED	461	if (p) {
b534ecf1	462	p->daddr = *daddr;
aa1039e7 ED	463	atomic_set(&p->refcnt, 1);
aa1039e7 ED	464	atomic_set(&p->rid, 0);
144001bd	465	p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
92d86829	466	p->rate_tokens = 0;
bc9259a8 ND	467	/* 60*HZ is arbitrary, but chosen enough high so that the first
	468	* calculation of tokens is at its maximum.
	469	*/
	470	p->rate_last = jiffies - 60*HZ;
5faa5df1	471	INIT_LIST_HEAD(&p->gc_list);
aa1039e7 ED	472
aa1039e7 ED	473	/* Link the node. */
98158f5a DM	474	link_to_pool(p, base);
98158f5a DM	475	base->total++;
aa1039e7	476	}
65e8354e	477	write_sequnlock_bh(&base->lock);
1da177e4	478
1da177e4 LT	479	return p;
1da177e4 LT	480	}
b3419363	481	EXPORT_SYMBOL_GPL(inet_getpeer);
98158f5a	482
4663afe2 ED	483	void inet_putpeer(struct inet_peer *p)
4663afe2 ED	484	{
4b9d9be8	485	p->dtime = (__u32)jiffies;
4e857c58	486	smp_mb__before_atomic();
4b9d9be8	487	atomic_dec(&p->refcnt);
4663afe2	488	}
b3419363	489	EXPORT_SYMBOL_GPL(inet_putpeer);
92d86829 DM	490
	491	/*
	492	* Check transmit rate limitation for given message.
	493	* The rate information is held in the inet_peer entries now.
	494	* This function is generic and could be used for other purposes
	495	* too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
	496	*
	497	* Note that the same inet_peer fields are modified by functions in
	498	* route.c too, but these work for packet destinations while xrlim_allow
	499	* works for icmp destinations. This means the rate limiting information
	500	* for one "ip object" is shared - and these ICMPs are twice limited:
	501	* by source and by destination.
	502	*
	503	* RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
	504	* SHOULD allow setting of rate limits
	505	*
	506	* Shared between ICMPv4 and ICMPv6.
	507	*/
	508	#define XRLIM_BURST_FACTOR 6
	509	bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
	510	{
	511	unsigned long now, token;
	512	bool rc = false;
	513
	514	if (!peer)
	515	return true;
	516
	517	token = peer->rate_tokens;
	518	now = jiffies;
	519	token += now - peer->rate_last;
	520	peer->rate_last = now;
	521	if (token > XRLIM_BURST_FACTOR * timeout)
	522	token = XRLIM_BURST_FACTOR * timeout;
	523	if (token >= timeout) {
	524	token -= timeout;
	525	rc = true;
	526	}
	527	peer->rate_tokens = token;
	528	return rc;
	529	}
	530	EXPORT_SYMBOL(inet_peer_xrlim_allow);
5faa5df1	531
55432d2b ED	532	static void inetpeer_inval_rcu(struct rcu_head *head)
	533	{
	534	struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);
	535
	536	spin_lock_bh(&gc_lock);
	537	list_add_tail(&p->gc_list, &gc_list);
	538	spin_unlock_bh(&gc_lock);
	539
	540	schedule_delayed_work(&gc_work, gc_delay);
	541	}
	542
56a6b248	543	void inetpeer_invalidate_tree(struct inet_peer_base *base)
5faa5df1	544	{
da557374	545	struct inet_peer *root;
5faa5df1 SK	546
	547	write_seqlock_bh(&base->lock);
	548
da557374 ED	549	root = rcu_deref_locked(base->root, base);
	550	if (root != peer_avl_empty) {
	551	base->root = peer_avl_empty_rcu;
5faa5df1	552	base->total = 0;
da557374	553	call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
5faa5df1 SK	554	}
5faa5df1 SK	555
5faa5df1 SK	556	write_sequnlock_bh(&base->lock);
	557	}
	558	EXPORT_SYMBOL(inetpeer_invalidate_tree);