[linux-2.6-block.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/workqueue.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>

/*
 * Theory of operations:
 * 1) We use a list, protected by a spinlock, to add
 *    new entries from both BH and non-BH context.
 * 2) In order to keep spinlock held for a small delay,
 *    we use a second list where are stored long lived
 *    entries, that are handled by the garbage collect thread
 *    fired by a workqueue.
 * 3) This list is guarded by a mutex,
 *    so that the gc_task and dst_dev_event() can be synchronized.
 */
#if RT_CACHE_DEBUG >= 2
static atomic_t			 dst_total = ATOMIC_INIT(0);
#endif

/*
 * We want to keep lock & list close together
 * to dirty as few cache lines as possible in __dst_free().
 * As this is not a very strong hint, we dont force an alignment on SMP.
 */
static struct {
	spinlock_t		lock;
	struct dst_entry 	*list;
	unsigned long		timer_inc;
	unsigned long		timer_expires;
} dst_garbage = {
	.lock = __SPIN_LOCK_UNLOCKED(dst_garbage.lock),
	.timer_inc = DST_GC_MAX,
};
static void dst_gc_task(struct work_struct *work);
static void ___dst_free(struct dst_entry * dst);

static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task);

static DEFINE_MUTEX(dst_gc_mutex);
/*
 * long lived entries are maintained in this list, guarded by dst_gc_mutex
 */
static struct dst_entry         *dst_busy_list;

static void dst_gc_task(struct work_struct *work)
{
	int    delayed = 0;
	int    work_performed = 0;
	unsigned long expires = ~0L;
	struct dst_entry *dst, *next, head;
	struct dst_entry *last = &head;
#if RT_CACHE_DEBUG >= 2
	ktime_t time_start = ktime_get();
	struct timespec elapsed;
#endif

	mutex_lock(&dst_gc_mutex);
	next = dst_busy_list;

loop:
	while ((dst = next) != NULL) {
		next = dst->next;
		prefetch(&next->next);
		if (likely(atomic_read(&dst->__refcnt))) {
			last->next = dst;
			last = dst;
			delayed++;
			continue;
		}
		work_performed++;

		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 = next;
			next = dst;
		}
	}

	spin_lock_bh(&dst_garbage.lock);
	next = dst_garbage.list;
	if (next) {
		dst_garbage.list = NULL;
		spin_unlock_bh(&dst_garbage.lock);
		goto loop;
	}
	last->next = NULL;
	dst_busy_list = head.next;
	if (!dst_busy_list)
		dst_garbage.timer_inc = DST_GC_MAX;
	else {
		/*
		 * if we freed less than 1/10 of delayed entries,
		 * we can sleep longer.
		 */
		if (work_performed <= delayed/10) {
			dst_garbage.timer_expires += dst_garbage.timer_inc;
			if (dst_garbage.timer_expires > DST_GC_MAX)
				dst_garbage.timer_expires = DST_GC_MAX;
			dst_garbage.timer_inc += DST_GC_INC;
		} else {
			dst_garbage.timer_inc = DST_GC_INC;
			dst_garbage.timer_expires = DST_GC_MIN;
		}
		expires = dst_garbage.timer_expires;
		/*
		 * if the next desired timer is more than 4 seconds in the future
		 * then round the timer to whole seconds
		 */
		if (expires > 4*HZ)
			expires = round_jiffies_relative(expires);
		schedule_delayed_work(&dst_gc_work, expires);
	}

	spin_unlock_bh(&dst_garbage.lock);
	mutex_unlock(&dst_gc_mutex);
#if RT_CACHE_DEBUG >= 2
	elapsed = ktime_to_timespec(ktime_sub(ktime_get(), time_start));
	printk(KERN_DEBUG "dst_total: %d delayed: %d work_perf: %d"
		" expires: %lu elapsed: %lu us\n",
		atomic_read(&dst_total), delayed, work_performed,
		expires,
		elapsed.tv_sec * USEC_PER_SEC + elapsed.tv_nsec / NSEC_PER_USEC);
#endif
}

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_garbage.lock);
	___dst_free(dst);
	dst->next = dst_garbage.list;
	dst_garbage.list = dst;
	if (dst_garbage.timer_inc > DST_GC_INC) {
		dst_garbage.timer_inc = DST_GC_INC;
		dst_garbage.timer_expires = DST_GC_MIN;
		schedule_delayed_work(&dst_gc_work, dst_garbage.timer_expires);
	}
	spin_unlock_bh(&dst_garbage.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 = init_net.loopback_dev;
		dev_hold(dst->dev);
		dev_put(dev);
		if (dst->neighbour && dst->neighbour->dev == dev) {
			dst->neighbour->dev = init_net.loopback_dev;
			dev_put(dev);
			dev_hold(dst->neighbour->dev);
		}
	}
}

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

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

	switch (event) {
	case NETDEV_UNREGISTER:
	case NETDEV_DOWN:
		mutex_lock(&dst_gc_mutex);
		for (dst = dst_busy_list; dst; dst = dst->next) {
			last = dst;
			dst_ifdown(dst, dev, event != NETDEV_DOWN);
		}

		spin_lock_bh(&dst_garbage.lock);
		dst = dst_garbage.list;
		dst_garbage.list = NULL;
		spin_unlock_bh(&dst_garbage.lock);

		if (last)
			last->next = dst;
		else
			dst_busy_list = dst;
		for (; dst; dst = dst->next) {
			dst_ifdown(dst, dev, event != NETDEV_DOWN);
		}
		mutex_unlock(&dst_gc_mutex);
		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>
86bba269	12	#include <linux/workqueue.h>
1da177e4 LT	13	#include <linux/mm.h>
	14	#include <linux/module.h>
	15	#include <linux/netdevice.h>
1da177e4 LT	16	#include <linux/skbuff.h>
	17	#include <linux/string.h>
	18	#include <linux/types.h>
e9dc8653	19	#include <net/net_namespace.h>
1da177e4 LT	20
	21	#include <net/dst.h>
	22
86bba269 ED	23	/*
	24	* Theory of operations:
	25	* 1) We use a list, protected by a spinlock, to add
	26	* new entries from both BH and non-BH context.
	27	* 2) In order to keep spinlock held for a small delay,
	28	* we use a second list where are stored long lived
	29	* entries, that are handled by the garbage collect thread
	30	* fired by a workqueue.
	31	* 3) This list is guarded by a mutex,
	32	* so that the gc_task and dst_dev_event() can be synchronized.
1da177e4	33	*/
4ec93edb	34	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	35	static atomic_t dst_total = ATOMIC_INIT(0);
1da177e4 LT	36	#endif
1da177e4	37
86bba269 ED	38	/*
	39	* We want to keep lock & list close together
	40	* to dirty as few cache lines as possible in __dst_free().
	41	* As this is not a very strong hint, we dont force an alignment on SMP.
	42	*/
	43	static struct {
	44	spinlock_t lock;
	45	struct dst_entry *list;
	46	unsigned long timer_inc;
	47	unsigned long timer_expires;
	48	} dst_garbage = {
	49	.lock = __SPIN_LOCK_UNLOCKED(dst_garbage.lock),
	50	.timer_inc = DST_GC_MAX,
	51	};
	52	static void dst_gc_task(struct work_struct *work);
1da177e4 LT	53	static void ___dst_free(struct dst_entry * dst);
1da177e4 LT	54
86bba269	55	static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task);
1da177e4	56
86bba269 ED	57	static DEFINE_MUTEX(dst_gc_mutex);
	58	/*
	59	* long lived entries are maintained in this list, guarded by dst_gc_mutex
	60	*/
	61	static struct dst_entry *dst_busy_list;
	62
	63	static void dst_gc_task(struct work_struct *work)
1da177e4 LT	64	{
1da177e4 LT	65	int delayed = 0;
86bba269 ED	66	int work_performed = 0;
	67	unsigned long expires = ~0L;
	68	struct dst_entry dst, next, head;
	69	struct dst_entry *last = &head;
	70	#if RT_CACHE_DEBUG >= 2
	71	ktime_t time_start = ktime_get();
	72	struct timespec elapsed;
	73	#endif
1da177e4	74
86bba269 ED	75	mutex_lock(&dst_gc_mutex);
86bba269 ED	76	next = dst_busy_list;
1da177e4	77
86bba269 ED	78	loop:
	79	while ((dst = next) != NULL) {
	80	next = dst->next;
	81	prefetch(&next->next);
	82	if (likely(atomic_read(&dst->__refcnt))) {
	83	last->next = dst;
	84	last = dst;
1da177e4 LT	85	delayed++;
	86	continue;
	87	}
86bba269	88	work_performed++;
1da177e4 LT	89
	90	dst = dst_destroy(dst);
	91	if (dst) {
	92	/* NOHASH and still referenced. Unless it is already
	93	* on gc list, invalidate it and add to gc list.
	94	*
	95	* Note: this is temporary. Actually, NOHASH dst's
	96	* must be obsoleted when parent is obsoleted.
	97	* But we do not have state "obsoleted, but
	98	* referenced by parent", so it is right.
	99	*/
	100	if (dst->obsolete > 1)
	101	continue;
	102
	103	___dst_free(dst);
86bba269 ED	104	dst->next = next;
86bba269 ED	105	next = dst;
1da177e4 LT	106	}
1da177e4 LT	107	}
86bba269 ED	108
	109	spin_lock_bh(&dst_garbage.lock);
	110	next = dst_garbage.list;
	111	if (next) {
	112	dst_garbage.list = NULL;
	113	spin_unlock_bh(&dst_garbage.lock);
	114	goto loop;
1da177e4	115	}
86bba269 ED	116	last->next = NULL;
	117	dst_busy_list = head.next;
	118	if (!dst_busy_list)
	119	dst_garbage.timer_inc = DST_GC_MAX;
	120	else {
	121	/*
	122	* if we freed less than 1/10 of delayed entries,
	123	* we can sleep longer.
	124	*/
	125	if (work_performed <= delayed/10) {
	126	dst_garbage.timer_expires += dst_garbage.timer_inc;
	127	if (dst_garbage.timer_expires > DST_GC_MAX)
	128	dst_garbage.timer_expires = DST_GC_MAX;
	129	dst_garbage.timer_inc += DST_GC_INC;
	130	} else {
	131	dst_garbage.timer_inc = DST_GC_INC;
	132	dst_garbage.timer_expires = DST_GC_MIN;
	133	}
	134	expires = dst_garbage.timer_expires;
	135	/*
	136	* if the next desired timer is more than 4 seconds in the future
	137	* then round the timer to whole seconds
	138	*/
	139	if (expires > 4*HZ)
	140	expires = round_jiffies_relative(expires);
	141	schedule_delayed_work(&dst_gc_work, expires);
f0098f78	142	}
86bba269 ED	143
	144	spin_unlock_bh(&dst_garbage.lock);
	145	mutex_unlock(&dst_gc_mutex);
1da177e4	146	#if RT_CACHE_DEBUG >= 2
86bba269 ED	147	elapsed = ktime_to_timespec(ktime_sub(ktime_get(), time_start));
	148	printk(KERN_DEBUG "dst_total: %d delayed: %d work_perf: %d"
	149	" expires: %lu elapsed: %lu us\n",
	150	atomic_read(&dst_total), delayed, work_performed,
	151	expires,
	152	elapsed.tv_sec * USEC_PER_SEC + elapsed.tv_nsec / NSEC_PER_USEC);
1da177e4	153	#endif
1da177e4 LT	154	}
1da177e4 LT	155
c4b1010f	156	static int dst_discard(struct sk_buff *skb)
1da177e4 LT	157	{
	158	kfree_skb(skb);
	159	return 0;
	160	}
	161
	162	void * dst_alloc(struct dst_ops * ops)
	163	{
	164	struct dst_entry * dst;
	165
	166	if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) {
	167	if (ops->gc())
	168	return NULL;
	169	}
c3762229	170	dst = kmem_cache_zalloc(ops->kmem_cachep, GFP_ATOMIC);
1da177e4 LT	171	if (!dst)
1da177e4 LT	172	return NULL;
1da177e4 LT	173	atomic_set(&dst->__refcnt, 0);
	174	dst->ops = ops;
	175	dst->lastuse = jiffies;
	176	dst->path = dst;
c4b1010f	177	dst->input = dst->output = dst_discard;
4ec93edb	178	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	179	atomic_inc(&dst_total);
	180	#endif
	181	atomic_inc(&ops->entries);
	182	return dst;
	183	}
	184
	185	static void ___dst_free(struct dst_entry * dst)
	186	{
	187	/* The first case (dev==NULL) is required, when
	188	protocol module is unloaded.
	189	*/
	190	if (dst->dev == NULL \|\| !(dst->dev->flags&IFF_UP)) {
c4b1010f	191	dst->input = dst->output = dst_discard;
1da177e4 LT	192	}
	193	dst->obsolete = 2;
	194	}
	195
	196	void __dst_free(struct dst_entry * dst)
	197	{
86bba269	198	spin_lock_bh(&dst_garbage.lock);
1da177e4	199	___dst_free(dst);
86bba269 ED	200	dst->next = dst_garbage.list;
	201	dst_garbage.list = dst;
	202	if (dst_garbage.timer_inc > DST_GC_INC) {
	203	dst_garbage.timer_inc = DST_GC_INC;
	204	dst_garbage.timer_expires = DST_GC_MIN;
	205	schedule_delayed_work(&dst_gc_work, dst_garbage.timer_expires);
1da177e4	206	}
86bba269	207	spin_unlock_bh(&dst_garbage.lock);
1da177e4 LT	208	}
	209
	210	struct dst_entry dst_destroy(struct dst_entry dst)
	211	{
	212	struct dst_entry *child;
	213	struct neighbour *neigh;
	214	struct hh_cache *hh;
	215
	216	smp_rmb();
	217
	218	again:
	219	neigh = dst->neighbour;
	220	hh = dst->hh;
	221	child = dst->child;
	222
	223	dst->hh = NULL;
	224	if (hh && atomic_dec_and_test(&hh->hh_refcnt))
	225	kfree(hh);
	226
	227	if (neigh) {
	228	dst->neighbour = NULL;
	229	neigh_release(neigh);
	230	}
	231
	232	atomic_dec(&dst->ops->entries);
	233
	234	if (dst->ops->destroy)
	235	dst->ops->destroy(dst);
	236	if (dst->dev)
	237	dev_put(dst->dev);
4ec93edb	238	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	239	atomic_dec(&dst_total);
	240	#endif
	241	kmem_cache_free(dst->ops->kmem_cachep, dst);
	242
	243	dst = child;
	244	if (dst) {
6775cab9 HX	245	int nohash = dst->flags & DST_NOHASH;
6775cab9 HX	246
1da177e4 LT	247	if (atomic_dec_and_test(&dst->__refcnt)) {
1da177e4 LT	248	/* We were real parent of this dst, so kill child. */
6775cab9	249	if (nohash)
1da177e4 LT	250	goto again;
	251	} else {
	252	/* Child is still referenced, return it for freeing. */
6775cab9	253	if (nohash)
1da177e4 LT	254	return dst;
	255	/* Child is still in his hash table */
	256	}
	257	}
	258	return NULL;
	259	}
	260
	261	/* Dirty hack. We did it in 2.2 (in __dst_free),
	262	* we have _very_ good reasons not to repeat
	263	* this mistake in 2.3, but we have no choice
	264	* now. _It_ _is_ _explicit_ _deliberate_
	265	* _race_ _condition_.
	266	*
	267	* Commented and originally written by Alexey.
	268	*/
	269	static inline void dst_ifdown(struct dst_entry dst, struct net_device dev,
	270	int unregister)
	271	{
	272	if (dst->ops->ifdown)
	273	dst->ops->ifdown(dst, dev, unregister);
	274
	275	if (dev != dst->dev)
	276	return;
	277
	278	if (!unregister) {
c4b1010f	279	dst->input = dst->output = dst_discard;
1da177e4	280	} else {
2774c7ab	281	dst->dev = init_net.loopback_dev;
de3cb747	282	dev_hold(dst->dev);
1da177e4 LT	283	dev_put(dev);
1da177e4 LT	284	if (dst->neighbour && dst->neighbour->dev == dev) {
2774c7ab	285	dst->neighbour->dev = init_net.loopback_dev;
1da177e4	286	dev_put(dev);
de3cb747	287	dev_hold(dst->neighbour->dev);
1da177e4 LT	288	}
	289	}
	290	}
	291
	292	static int dst_dev_event(struct notifier_block this, unsigned long event, void ptr)
	293	{
	294	struct net_device *dev = ptr;
86bba269	295	struct dst_entry dst, last = NULL;
1da177e4	296
e9dc8653 EB	297	if (dev->nd_net != &init_net)
	298	return NOTIFY_DONE;
	299
1da177e4 LT	300	switch (event) {
	301	case NETDEV_UNREGISTER:
	302	case NETDEV_DOWN:
86bba269 ED	303	mutex_lock(&dst_gc_mutex);
	304	for (dst = dst_busy_list; dst; dst = dst->next) {
	305	last = dst;
	306	dst_ifdown(dst, dev, event != NETDEV_DOWN);
	307	}
	308
	309	spin_lock_bh(&dst_garbage.lock);
	310	dst = dst_garbage.list;
	311	dst_garbage.list = NULL;
	312	spin_unlock_bh(&dst_garbage.lock);
	313
	314	if (last)
	315	last->next = dst;
	316	else
	317	dst_busy_list = dst;
	318	for (; dst; dst = dst->next) {
1da177e4 LT	319	dst_ifdown(dst, dev, event != NETDEV_DOWN);
1da177e4 LT	320	}
86bba269	321	mutex_unlock(&dst_gc_mutex);
1da177e4 LT	322	break;
	323	}
	324	return NOTIFY_DONE;
	325	}
	326
	327	static struct notifier_block dst_dev_notifier = {
	328	.notifier_call = dst_dev_event,
	329	};
	330
	331	void __init dst_init(void)
	332	{
	333	register_netdevice_notifier(&dst_dev_notifier);
	334	}
	335
	336	EXPORT_SYMBOL(__dst_free);
	337	EXPORT_SYMBOL(dst_alloc);
	338	EXPORT_SYMBOL(dst_destroy);