[net-next-2.6.git] / net / core / dst.c

/*
 * net/core/dst.c	Protocol independent destination cache.
 *
 * Authors:		Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 *
 */

#include <linux/bitops.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/netdevice.h>
#include <linux/skbuff.h>
#include <linux/string.h>
#include <linux/types.h>
#include <net/net_namespace.h>

#include <net/dst.h>

/* Locking strategy:
 * 1) Garbage collection state of dead destination cache
 *    entries is protected by dst_lock.
 * 2) GC is run only from BH context, and is the only remover
 *    of entries.
 * 3) Entries are added to the garbage list from both BH
 *    and non-BH context, so local BH disabling is needed.
 * 4) All operations modify state, so a spinlock is used.
 */
static struct dst_entry 	*dst_garbage_list;
#if RT_CACHE_DEBUG >= 2
static atomic_t			 dst_total = ATOMIC_INIT(0);
#endif
static DEFINE_SPINLOCK(dst_lock);

static unsigned long dst_gc_timer_expires;
static unsigned long dst_gc_timer_inc = DST_GC_MAX;
static void dst_run_gc(unsigned long);
static void ___dst_free(struct dst_entry * dst);

static DEFINE_TIMER(dst_gc_timer, dst_run_gc, DST_GC_MIN, 0);

static void dst_run_gc(unsigned long dummy)
{
	int    delayed = 0;
	int    work_performed;
	struct dst_entry * dst, **dstp;

	if (!spin_trylock(&dst_lock)) {
		mod_timer(&dst_gc_timer, jiffies + HZ/10);
		return;
	}

	del_timer(&dst_gc_timer);
	dstp = &dst_garbage_list;
	work_performed = 0;
	while ((dst = *dstp) != NULL) {
		if (atomic_read(&dst->__refcnt)) {
			dstp = &dst->next;
			delayed++;
			continue;
		}
		*dstp = dst->next;
		work_performed = 1;

		dst = dst_destroy(dst);
		if (dst) {
			/* NOHASH and still referenced. Unless it is already
			 * on gc list, invalidate it and add to gc list.
			 *
			 * Note: this is temporary. Actually, NOHASH dst's
			 * must be obsoleted when parent is obsoleted.
			 * But we do not have state "obsoleted, but
			 * referenced by parent", so it is right.
			 */
			if (dst->obsolete > 1)
				continue;

			___dst_free(dst);
			dst->next = *dstp;
			*dstp = dst;
			dstp = &dst->next;
		}
	}
	if (!dst_garbage_list) {
		dst_gc_timer_inc = DST_GC_MAX;
		goto out;
	}
	if (!work_performed) {
		if ((dst_gc_timer_expires += dst_gc_timer_inc) > DST_GC_MAX)
			dst_gc_timer_expires = DST_GC_MAX;
		dst_gc_timer_inc += DST_GC_INC;
	} else {
		dst_gc_timer_inc = DST_GC_INC;
		dst_gc_timer_expires = DST_GC_MIN;
	}
#if RT_CACHE_DEBUG >= 2
	printk("dst_total: %d/%d %ld\n",
	       atomic_read(&dst_total), delayed,  dst_gc_timer_expires);
#endif
	/* if the next desired timer is more than 4 seconds in the future
	 * then round the timer to whole seconds
	 */
	if (dst_gc_timer_expires > 4*HZ)
		mod_timer(&dst_gc_timer,
			round_jiffies(jiffies + dst_gc_timer_expires));
	else
		mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);

out:
	spin_unlock(&dst_lock);
}

static int dst_discard(struct sk_buff *skb)
{
	kfree_skb(skb);
	return 0;
}

void * dst_alloc(struct dst_ops * ops)
{
	struct dst_entry * dst;

	if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) {
		if (ops->gc())
			return NULL;
	}
	dst = kmem_cache_zalloc(ops->kmem_cachep, GFP_ATOMIC);
	if (!dst)
		return NULL;
	atomic_set(&dst->__refcnt, 0);
	dst->ops = ops;
	dst->lastuse = jiffies;
	dst->path = dst;
	dst->input = dst->output = dst_discard;
#if RT_CACHE_DEBUG >= 2
	atomic_inc(&dst_total);
#endif
	atomic_inc(&ops->entries);
	return dst;
}

static void ___dst_free(struct dst_entry * dst)
{
	/* The first case (dev==NULL) is required, when
	   protocol module is unloaded.
	 */
	if (dst->dev == NULL || !(dst->dev->flags&IFF_UP)) {
		dst->input = dst->output = dst_discard;
	}
	dst->obsolete = 2;
}

void __dst_free(struct dst_entry * dst)
{
	spin_lock_bh(&dst_lock);
	___dst_free(dst);
	dst->next = dst_garbage_list;
	dst_garbage_list = dst;
	if (dst_gc_timer_inc > DST_GC_INC) {
		dst_gc_timer_inc = DST_GC_INC;
		dst_gc_timer_expires = DST_GC_MIN;
		mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);
	}
	spin_unlock_bh(&dst_lock);
}

struct dst_entry *dst_destroy(struct dst_entry * dst)
{
	struct dst_entry *child;
	struct neighbour *neigh;
	struct hh_cache *hh;

	smp_rmb();

again:
	neigh = dst->neighbour;
	hh = dst->hh;
	child = dst->child;

	dst->hh = NULL;
	if (hh && atomic_dec_and_test(&hh->hh_refcnt))
		kfree(hh);

	if (neigh) {
		dst->neighbour = NULL;
		neigh_release(neigh);
	}

	atomic_dec(&dst->ops->entries);

	if (dst->ops->destroy)
		dst->ops->destroy(dst);
	if (dst->dev)
		dev_put(dst->dev);
#if RT_CACHE_DEBUG >= 2
	atomic_dec(&dst_total);
#endif
	kmem_cache_free(dst->ops->kmem_cachep, dst);

	dst = child;
	if (dst) {
		int nohash = dst->flags & DST_NOHASH;

		if (atomic_dec_and_test(&dst->__refcnt)) {
			/* We were real parent of this dst, so kill child. */
			if (nohash)
				goto again;
		} else {
			/* Child is still referenced, return it for freeing. */
			if (nohash)
				return dst;
			/* Child is still in his hash table */
		}
	}
	return NULL;
}

/* Dirty hack. We did it in 2.2 (in __dst_free),
 * we have _very_ good reasons not to repeat
 * this mistake in 2.3, but we have no choice
 * now. _It_ _is_ _explicit_ _deliberate_
 * _race_ _condition_.
 *
 * Commented and originally written by Alexey.
 */
static inline void dst_ifdown(struct dst_entry *dst, struct net_device *dev,
			      int unregister)
{
	if (dst->ops->ifdown)
		dst->ops->ifdown(dst, dev, unregister);

	if (dev != dst->dev)
		return;

	if (!unregister) {
		dst->input = dst->output = dst_discard;
	} else {
		dst->dev = &loopback_dev;
		dev_hold(&loopback_dev);
		dev_put(dev);
		if (dst->neighbour && dst->neighbour->dev == dev) {
			dst->neighbour->dev = &loopback_dev;
			dev_put(dev);
			dev_hold(&loopback_dev);
		}
	}
}

static int dst_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
{
	struct net_device *dev = ptr;
	struct dst_entry *dst;

	if (dev->nd_net != &init_net)
		return NOTIFY_DONE;

	switch (event) {
	case NETDEV_UNREGISTER:
	case NETDEV_DOWN:
		spin_lock_bh(&dst_lock);
		for (dst = dst_garbage_list; dst; dst = dst->next) {
			dst_ifdown(dst, dev, event != NETDEV_DOWN);
		}
		spin_unlock_bh(&dst_lock);
		break;
	}
	return NOTIFY_DONE;
}

static struct notifier_block dst_dev_notifier = {
	.notifier_call	= dst_dev_event,
};

void __init dst_init(void)
{
	register_netdevice_notifier(&dst_dev_notifier);
}

EXPORT_SYMBOL(__dst_free);
EXPORT_SYMBOL(dst_alloc);
EXPORT_SYMBOL(dst_destroy);
Commit	Line	Data
1da177e4 LT	1	/*
	2	* net/core/dst.c Protocol independent destination cache.
	3	*
	4	* Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
	5	*
	6	*/
	7
	8	#include <linux/bitops.h>
	9	#include <linux/errno.h>
	10	#include <linux/init.h>
	11	#include <linux/kernel.h>
	12	#include <linux/mm.h>
	13	#include <linux/module.h>
	14	#include <linux/netdevice.h>
1da177e4 LT	15	#include <linux/skbuff.h>
	16	#include <linux/string.h>
	17	#include <linux/types.h>
e9dc8653	18	#include <net/net_namespace.h>
1da177e4 LT	19
	20	#include <net/dst.h>
	21
	22	/* Locking strategy:
	23	* 1) Garbage collection state of dead destination cache
	24	* entries is protected by dst_lock.
	25	* 2) GC is run only from BH context, and is the only remover
	26	* of entries.
	27	* 3) Entries are added to the garbage list from both BH
	28	* and non-BH context, so local BH disabling is needed.
	29	* 4) All operations modify state, so a spinlock is used.
	30	*/
	31	static struct dst_entry *dst_garbage_list;
4ec93edb	32	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	33	static atomic_t dst_total = ATOMIC_INIT(0);
	34	#endif
	35	static DEFINE_SPINLOCK(dst_lock);
	36
	37	static unsigned long dst_gc_timer_expires;
	38	static unsigned long dst_gc_timer_inc = DST_GC_MAX;
	39	static void dst_run_gc(unsigned long);
	40	static void ___dst_free(struct dst_entry * dst);
	41
8d06afab	42	static DEFINE_TIMER(dst_gc_timer, dst_run_gc, DST_GC_MIN, 0);
1da177e4 LT	43
	44	static void dst_run_gc(unsigned long dummy)
	45	{
	46	int delayed = 0;
f0098f78	47	int work_performed;
1da177e4 LT	48	struct dst_entry * dst, **dstp;
	49
	50	if (!spin_trylock(&dst_lock)) {
	51	mod_timer(&dst_gc_timer, jiffies + HZ/10);
	52	return;
	53	}
	54
1da177e4 LT	55	del_timer(&dst_gc_timer);
1da177e4 LT	56	dstp = &dst_garbage_list;
f0098f78	57	work_performed = 0;
1da177e4 LT	58	while ((dst = *dstp) != NULL) {
	59	if (atomic_read(&dst->__refcnt)) {
	60	dstp = &dst->next;
	61	delayed++;
	62	continue;
	63	}
	64	*dstp = dst->next;
f0098f78	65	work_performed = 1;
1da177e4 LT	66
	67	dst = dst_destroy(dst);
	68	if (dst) {
	69	/* NOHASH and still referenced. Unless it is already
	70	* on gc list, invalidate it and add to gc list.
	71	*
	72	* Note: this is temporary. Actually, NOHASH dst's
	73	* must be obsoleted when parent is obsoleted.
	74	* But we do not have state "obsoleted, but
	75	* referenced by parent", so it is right.
	76	*/
	77	if (dst->obsolete > 1)
	78	continue;
	79
	80	___dst_free(dst);
	81	dst->next = *dstp;
	82	*dstp = dst;
	83	dstp = &dst->next;
	84	}
	85	}
	86	if (!dst_garbage_list) {
	87	dst_gc_timer_inc = DST_GC_MAX;
	88	goto out;
	89	}
f0098f78 DL	90	if (!work_performed) {
	91	if ((dst_gc_timer_expires += dst_gc_timer_inc) > DST_GC_MAX)
	92	dst_gc_timer_expires = DST_GC_MAX;
	93	dst_gc_timer_inc += DST_GC_INC;
	94	} else {
	95	dst_gc_timer_inc = DST_GC_INC;
	96	dst_gc_timer_expires = DST_GC_MIN;
	97	}
1da177e4 LT	98	#if RT_CACHE_DEBUG >= 2
	99	printk("dst_total: %d/%d %ld\n",
	100	atomic_read(&dst_total), delayed, dst_gc_timer_expires);
	101	#endif
f5a6e01c AV	102	/* if the next desired timer is more than 4 seconds in the future
	103	* then round the timer to whole seconds
	104	*/
	105	if (dst_gc_timer_expires > 4*HZ)
	106	mod_timer(&dst_gc_timer,
	107	round_jiffies(jiffies + dst_gc_timer_expires));
	108	else
	109	mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);
1da177e4 LT	110
	111	out:
	112	spin_unlock(&dst_lock);
	113	}
	114
c4b1010f	115	static int dst_discard(struct sk_buff *skb)
1da177e4 LT	116	{
	117	kfree_skb(skb);
	118	return 0;
	119	}
	120
	121	void * dst_alloc(struct dst_ops * ops)
	122	{
	123	struct dst_entry * dst;
	124
	125	if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) {
	126	if (ops->gc())
	127	return NULL;
	128	}
c3762229	129	dst = kmem_cache_zalloc(ops->kmem_cachep, GFP_ATOMIC);
1da177e4 LT	130	if (!dst)
1da177e4 LT	131	return NULL;
1da177e4 LT	132	atomic_set(&dst->__refcnt, 0);
	133	dst->ops = ops;
	134	dst->lastuse = jiffies;
	135	dst->path = dst;
c4b1010f	136	dst->input = dst->output = dst_discard;
4ec93edb	137	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	138	atomic_inc(&dst_total);
	139	#endif
	140	atomic_inc(&ops->entries);
	141	return dst;
	142	}
	143
	144	static void ___dst_free(struct dst_entry * dst)
	145	{
	146	/* The first case (dev==NULL) is required, when
	147	protocol module is unloaded.
	148	*/
	149	if (dst->dev == NULL \|\| !(dst->dev->flags&IFF_UP)) {
c4b1010f	150	dst->input = dst->output = dst_discard;
1da177e4 LT	151	}
	152	dst->obsolete = 2;
	153	}
	154
	155	void __dst_free(struct dst_entry * dst)
	156	{
	157	spin_lock_bh(&dst_lock);
	158	___dst_free(dst);
	159	dst->next = dst_garbage_list;
	160	dst_garbage_list = dst;
	161	if (dst_gc_timer_inc > DST_GC_INC) {
	162	dst_gc_timer_inc = DST_GC_INC;
	163	dst_gc_timer_expires = DST_GC_MIN;
	164	mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);
	165	}
	166	spin_unlock_bh(&dst_lock);
	167	}
	168
	169	struct dst_entry dst_destroy(struct dst_entry dst)
	170	{
	171	struct dst_entry *child;
	172	struct neighbour *neigh;
	173	struct hh_cache *hh;
	174
	175	smp_rmb();
	176
	177	again:
	178	neigh = dst->neighbour;
	179	hh = dst->hh;
	180	child = dst->child;
	181
	182	dst->hh = NULL;
	183	if (hh && atomic_dec_and_test(&hh->hh_refcnt))
	184	kfree(hh);
	185
	186	if (neigh) {
	187	dst->neighbour = NULL;
	188	neigh_release(neigh);
	189	}
	190
	191	atomic_dec(&dst->ops->entries);
	192
	193	if (dst->ops->destroy)
	194	dst->ops->destroy(dst);
	195	if (dst->dev)
	196	dev_put(dst->dev);
4ec93edb	197	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	198	atomic_dec(&dst_total);
	199	#endif
	200	kmem_cache_free(dst->ops->kmem_cachep, dst);
	201
	202	dst = child;
	203	if (dst) {
6775cab9 HX	204	int nohash = dst->flags & DST_NOHASH;
6775cab9 HX	205
1da177e4 LT	206	if (atomic_dec_and_test(&dst->__refcnt)) {
1da177e4 LT	207	/* We were real parent of this dst, so kill child. */
6775cab9	208	if (nohash)
1da177e4 LT	209	goto again;
	210	} else {
	211	/* Child is still referenced, return it for freeing. */
6775cab9	212	if (nohash)
1da177e4 LT	213	return dst;
	214	/* Child is still in his hash table */
	215	}
	216	}
	217	return NULL;
	218	}
	219
	220	/* Dirty hack. We did it in 2.2 (in __dst_free),
	221	* we have _very_ good reasons not to repeat
	222	* this mistake in 2.3, but we have no choice
	223	* now. _It_ _is_ _explicit_ _deliberate_
	224	* _race_ _condition_.
	225	*
	226	* Commented and originally written by Alexey.
	227	*/
	228	static inline void dst_ifdown(struct dst_entry dst, struct net_device dev,
	229	int unregister)
	230	{
	231	if (dst->ops->ifdown)
	232	dst->ops->ifdown(dst, dev, unregister);
	233
	234	if (dev != dst->dev)
	235	return;
	236
	237	if (!unregister) {
c4b1010f	238	dst->input = dst->output = dst_discard;
1da177e4 LT	239	} else {
	240	dst->dev = &loopback_dev;
	241	dev_hold(&loopback_dev);
	242	dev_put(dev);
	243	if (dst->neighbour && dst->neighbour->dev == dev) {
	244	dst->neighbour->dev = &loopback_dev;
	245	dev_put(dev);
	246	dev_hold(&loopback_dev);
	247	}
	248	}
	249	}
	250
	251	static int dst_dev_event(struct notifier_block this, unsigned long event, void ptr)
	252	{
	253	struct net_device *dev = ptr;
	254	struct dst_entry *dst;
	255
e9dc8653 EB	256	if (dev->nd_net != &init_net)
	257	return NOTIFY_DONE;
	258
1da177e4 LT	259	switch (event) {
	260	case NETDEV_UNREGISTER:
	261	case NETDEV_DOWN:
	262	spin_lock_bh(&dst_lock);
	263	for (dst = dst_garbage_list; dst; dst = dst->next) {
	264	dst_ifdown(dst, dev, event != NETDEV_DOWN);
	265	}
	266	spin_unlock_bh(&dst_lock);
	267	break;
	268	}
	269	return NOTIFY_DONE;
	270	}
	271
	272	static struct notifier_block dst_dev_notifier = {
	273	.notifier_call = dst_dev_event,
	274	};
	275
	276	void __init dst_init(void)
	277	{
	278	register_netdevice_notifier(&dst_dev_notifier);
	279	}
	280
	281	EXPORT_SYMBOL(__dst_free);
	282	EXPORT_SYMBOL(dst_alloc);
	283	EXPORT_SYMBOL(dst_destroy);