[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/slab.h>
#include <linux/netdevice.h>
#include <linux/skbuff.h>
#include <linux/string.h>
#include <linux/types.h>
#include <net/net_namespace.h>
#include <linux/sched.h>
#include <linux/prefetch.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.
 */

/*
 * 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;

	mutex_lock(&dst_gc_mutex);
	next = dst_busy_list;

loop:
	while ((dst = next) != NULL) {
		next = dst->next;
		prefetch(&next->next);
		cond_resched();
		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 > 0)
				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);
}

int dst_discard_sk(struct sock *sk, struct sk_buff *skb)
{
	kfree_skb(skb);
	return 0;
}
EXPORT_SYMBOL(dst_discard_sk);

const u32 dst_default_metrics[RTAX_MAX + 1] = {
	/* This initializer is needed to force linker to place this variable
	 * into const section. Otherwise it might end into bss section.
	 * We really want to avoid false sharing on this variable, and catch
	 * any writes on it.
	 */
	[RTAX_MAX] = 0xdeadbeef,
};


void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
		int initial_ref, int initial_obsolete, unsigned short flags)
{
	struct dst_entry *dst;

	if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
		if (ops->gc(ops))
			return NULL;
	}
	dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
	if (!dst)
		return NULL;
	dst->child = NULL;
	dst->dev = dev;
	if (dev)
		dev_hold(dev);
	dst->ops = ops;
	dst_init_metrics(dst, dst_default_metrics, true);
	dst->expires = 0UL;
	dst->path = dst;
	dst->from = NULL;
#ifdef CONFIG_XFRM
	dst->xfrm = NULL;
#endif
	dst->input = dst_discard;
	dst->output = dst_discard_sk;
	dst->error = 0;
	dst->obsolete = initial_obsolete;
	dst->header_len = 0;
	dst->trailer_len = 0;
#ifdef CONFIG_IP_ROUTE_CLASSID
	dst->tclassid = 0;
#endif
	atomic_set(&dst->__refcnt, initial_ref);
	dst->__use = 0;
	dst->lastuse = jiffies;
	dst->flags = flags;
	dst->pending_confirm = 0;
	dst->next = NULL;
	if (!(flags & DST_NOCOUNT))
		dst_entries_add(ops, 1);
	return dst;
}
EXPORT_SYMBOL(dst_alloc);

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_discard;
		dst->output = dst_discard_sk;
	}
	dst->obsolete = DST_OBSOLETE_DEAD;
}

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;
		mod_delayed_work(system_wq, &dst_gc_work,
				 dst_garbage.timer_expires);
	}
	spin_unlock_bh(&dst_garbage.lock);
}
EXPORT_SYMBOL(__dst_free);

struct dst_entry *dst_destroy(struct dst_entry * dst)
{
	struct dst_entry *child;

	smp_rmb();

again:
	child = dst->child;

	if (!(dst->flags & DST_NOCOUNT))
		dst_entries_add(dst->ops, -1);

	if (dst->ops->destroy)
		dst->ops->destroy(dst);
	if (dst->dev)
		dev_put(dst->dev);
	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;
}
EXPORT_SYMBOL(dst_destroy);

static void dst_destroy_rcu(struct rcu_head *head)
{
	struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);

	dst = dst_destroy(dst);
	if (dst)
		__dst_free(dst);
}

void dst_release(struct dst_entry *dst)
{
	if (dst) {
		int newrefcnt;

		newrefcnt = atomic_dec_return(&dst->__refcnt);
		WARN_ON(newrefcnt < 0);
		if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt)
			call_rcu(&dst->rcu_head, dst_destroy_rcu);
	}
}
EXPORT_SYMBOL(dst_release);

u32 *dst_cow_metrics_generic(struct dst_entry *dst, unsigned long old)
{
	u32 *p = kmalloc(sizeof(u32) * RTAX_MAX, GFP_ATOMIC);

	if (p) {
		u32 *old_p = __DST_METRICS_PTR(old);
		unsigned long prev, new;

		memcpy(p, old_p, sizeof(u32) * RTAX_MAX);

		new = (unsigned long) p;
		prev = cmpxchg(&dst->_metrics, old, new);

		if (prev != old) {
			kfree(p);
			p = __DST_METRICS_PTR(prev);
			if (prev & DST_METRICS_READ_ONLY)
				p = NULL;
		}
	}
	return p;
}
EXPORT_SYMBOL(dst_cow_metrics_generic);

/* Caller asserts that dst_metrics_read_only(dst) is false.  */
void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
{
	unsigned long prev, new;

	new = ((unsigned long) dst_default_metrics) | DST_METRICS_READ_ONLY;
	prev = cmpxchg(&dst->_metrics, old, new);
	if (prev == old)
		kfree(__DST_METRICS_PTR(old));
}
EXPORT_SYMBOL(__dst_destroy_metrics_generic);

/* 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 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_discard;
		dst->output = dst_discard_sk;
	} else {
		dst->dev = dev_net(dst->dev)->loopback_dev;
		dev_hold(dst->dev);
		dev_put(dev);
	}
}

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

	switch (event) {
	case NETDEV_UNREGISTER_FINAL:
	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,
	.priority = -10, /* must be called after other network notifiers */
};

void __init dst_init(void)
{
	register_netdevice_notifier(&dst_dev_notifier);
}
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>
1da177e4 LT	14	#include <linux/module.h>
5a0e3ad6	15	#include <linux/slab.h>
1da177e4	16	#include <linux/netdevice.h>
1da177e4 LT	17	#include <linux/skbuff.h>
	18	#include <linux/string.h>
	19	#include <linux/types.h>
e9dc8653	20	#include <net/net_namespace.h>
2fc1b5dd	21	#include <linux/sched.h>
268bb0ce	22	#include <linux/prefetch.h>
1da177e4 LT	23
	24	#include <net/dst.h>
	25
86bba269 ED	26	/*
	27	* Theory of operations:
	28	* 1) We use a list, protected by a spinlock, to add
	29	* new entries from both BH and non-BH context.
	30	* 2) In order to keep spinlock held for a small delay,
	31	* we use a second list where are stored long lived
	32	* entries, that are handled by the garbage collect thread
	33	* fired by a workqueue.
	34	* 3) This list is guarded by a mutex,
	35	* so that the gc_task and dst_dev_event() can be synchronized.
1da177e4	36	*/
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;
598ed936	45	struct dst_entry *list;
86bba269 ED	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);
598ed936	53	static void ___dst_free(struct dst_entry *dst);
1da177e4	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;
1da177e4	70
86bba269 ED	71	mutex_lock(&dst_gc_mutex);
86bba269 ED	72	next = dst_busy_list;
1da177e4	73
86bba269 ED	74	loop:
	75	while ((dst = next) != NULL) {
	76	next = dst->next;
	77	prefetch(&next->next);
2fc1b5dd	78	cond_resched();
86bba269 ED	79	if (likely(atomic_read(&dst->__refcnt))) {
	80	last->next = dst;
	81	last = dst;
1da177e4 LT	82	delayed++;
	83	continue;
	84	}
86bba269	85	work_performed++;
1da177e4 LT	86
	87	dst = dst_destroy(dst);
	88	if (dst) {
	89	/* NOHASH and still referenced. Unless it is already
	90	* on gc list, invalidate it and add to gc list.
	91	*
	92	* Note: this is temporary. Actually, NOHASH dst's
	93	* must be obsoleted when parent is obsoleted.
	94	* But we do not have state "obsoleted, but
	95	* referenced by parent", so it is right.
	96	*/
f5b0a874	97	if (dst->obsolete > 0)
1da177e4 LT	98	continue;
	99
	100	___dst_free(dst);
86bba269 ED	101	dst->next = next;
86bba269 ED	102	next = dst;
1da177e4 LT	103	}
1da177e4 LT	104	}
86bba269 ED	105
	106	spin_lock_bh(&dst_garbage.lock);
	107	next = dst_garbage.list;
	108	if (next) {
	109	dst_garbage.list = NULL;
	110	spin_unlock_bh(&dst_garbage.lock);
	111	goto loop;
1da177e4	112	}
86bba269 ED	113	last->next = NULL;
	114	dst_busy_list = head.next;
	115	if (!dst_busy_list)
	116	dst_garbage.timer_inc = DST_GC_MAX;
	117	else {
	118	/*
	119	* if we freed less than 1/10 of delayed entries,
	120	* we can sleep longer.
	121	*/
	122	if (work_performed <= delayed/10) {
	123	dst_garbage.timer_expires += dst_garbage.timer_inc;
	124	if (dst_garbage.timer_expires > DST_GC_MAX)
	125	dst_garbage.timer_expires = DST_GC_MAX;
	126	dst_garbage.timer_inc += DST_GC_INC;
	127	} else {
	128	dst_garbage.timer_inc = DST_GC_INC;
	129	dst_garbage.timer_expires = DST_GC_MIN;
	130	}
	131	expires = dst_garbage.timer_expires;
	132	/*
598ed936	133	* if the next desired timer is more than 4 seconds in the
598ed936	134	* future then round the timer to whole seconds
86bba269 ED	135	*/
	136	if (expires > 4*HZ)
	137	expires = round_jiffies_relative(expires);
	138	schedule_delayed_work(&dst_gc_work, expires);
f0098f78	139	}
86bba269 ED	140
	141	spin_unlock_bh(&dst_garbage.lock);
	142	mutex_unlock(&dst_gc_mutex);
1da177e4 LT	143	}
1da177e4 LT	144
aad88724	145	int dst_discard_sk(struct sock sk, struct sk_buff skb)
1da177e4 LT	146	{
	147	kfree_skb(skb);
	148	return 0;
	149	}
aad88724	150	EXPORT_SYMBOL(dst_discard_sk);
1da177e4	151
a37e6e34 ED	152	const u32 dst_default_metrics[RTAX_MAX + 1] = {
	153	/* This initializer is needed to force linker to place this variable
	154	* into const section. Otherwise it might end into bss section.
	155	* We really want to avoid false sharing on this variable, and catch
	156	* any writes on it.
	157	*/
	158	[RTAX_MAX] = 0xdeadbeef,
	159	};
	160
62fa8a84	161
5c1e6aa3	162	void dst_alloc(struct dst_ops ops, struct net_device *dev,
5110effe	163	int initial_ref, int initial_obsolete, unsigned short flags)
1da177e4	164	{
598ed936	165	struct dst_entry *dst;
1da177e4	166
fc66f95c	167	if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
569d3645	168	if (ops->gc(ops))
1da177e4 LT	169	return NULL;
1da177e4 LT	170	}
cf911662	171	dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
1da177e4 LT	172	if (!dst)
1da177e4 LT	173	return NULL;
cf911662	174	dst->child = NULL;
5c1e6aa3 DM	175	dst->dev = dev;
	176	if (dev)
	177	dev_hold(dev);
1da177e4	178	dst->ops = ops;
62fa8a84	179	dst_init_metrics(dst, dst_default_metrics, true);
cf911662	180	dst->expires = 0UL;
5c1e6aa3	181	dst->path = dst;
ecd98837	182	dst->from = NULL;
cf911662 DM	183	#ifdef CONFIG_XFRM
	184	dst->xfrm = NULL;
	185	#endif
5c1e6aa3	186	dst->input = dst_discard;
aad88724	187	dst->output = dst_discard_sk;
cf911662	188	dst->error = 0;
5c1e6aa3	189	dst->obsolete = initial_obsolete;
cf911662 DM	190	dst->header_len = 0;
	191	dst->trailer_len = 0;
	192	#ifdef CONFIG_IP_ROUTE_CLASSID
	193	dst->tclassid = 0;
1da177e4	194	#endif
5c1e6aa3	195	atomic_set(&dst->__refcnt, initial_ref);
cf911662	196	dst->__use = 0;
5c1e6aa3 DM	197	dst->lastuse = jiffies;
5c1e6aa3 DM	198	dst->flags = flags;
5110effe	199	dst->pending_confirm = 0;
cf911662	200	dst->next = NULL;
957c665f DM	201	if (!(flags & DST_NOCOUNT))
957c665f DM	202	dst_entries_add(ops, 1);
1da177e4 LT	203	return dst;
1da177e4 LT	204	}
598ed936	205	EXPORT_SYMBOL(dst_alloc);
1da177e4	206
598ed936	207	static void ___dst_free(struct dst_entry *dst)
1da177e4 LT	208	{
	209	/* The first case (dev==NULL) is required, when
	210	protocol module is unloaded.
	211	*/
aad88724 ED	212	if (dst->dev == NULL \|\| !(dst->dev->flags&IFF_UP)) {
	213	dst->input = dst_discard;
	214	dst->output = dst_discard_sk;
	215	}
f5b0a874	216	dst->obsolete = DST_OBSOLETE_DEAD;
1da177e4 LT	217	}
1da177e4 LT	218
598ed936	219	void __dst_free(struct dst_entry *dst)
1da177e4	220	{
86bba269	221	spin_lock_bh(&dst_garbage.lock);
1da177e4	222	___dst_free(dst);
86bba269 ED	223	dst->next = dst_garbage.list;
	224	dst_garbage.list = dst;
	225	if (dst_garbage.timer_inc > DST_GC_INC) {
	226	dst_garbage.timer_inc = DST_GC_INC;
	227	dst_garbage.timer_expires = DST_GC_MIN;
41f63c53 TH	228	mod_delayed_work(system_wq, &dst_gc_work,
41f63c53 TH	229	dst_garbage.timer_expires);
1da177e4	230	}
86bba269	231	spin_unlock_bh(&dst_garbage.lock);
1da177e4	232	}
d79d9913	233	EXPORT_SYMBOL(__dst_free);
1da177e4 LT	234
	235	struct dst_entry dst_destroy(struct dst_entry dst)
	236	{
	237	struct dst_entry *child;
1da177e4 LT	238
	239	smp_rmb();
	240
	241	again:
1da177e4 LT	242	child = dst->child;
1da177e4 LT	243
957c665f DM	244	if (!(dst->flags & DST_NOCOUNT))
957c665f DM	245	dst_entries_add(dst->ops, -1);
1da177e4 LT	246
	247	if (dst->ops->destroy)
	248	dst->ops->destroy(dst);
	249	if (dst->dev)
	250	dev_put(dst->dev);
1da177e4 LT	251	kmem_cache_free(dst->ops->kmem_cachep, dst);
	252
	253	dst = child;
	254	if (dst) {
6775cab9 HX	255	int nohash = dst->flags & DST_NOHASH;
6775cab9 HX	256
1da177e4 LT	257	if (atomic_dec_and_test(&dst->__refcnt)) {
1da177e4 LT	258	/* We were real parent of this dst, so kill child. */
6775cab9	259	if (nohash)
1da177e4 LT	260	goto again;
	261	} else {
	262	/* Child is still referenced, return it for freeing. */
6775cab9	263	if (nohash)
1da177e4 LT	264	return dst;
	265	/* Child is still in his hash table */
	266	}
	267	}
	268	return NULL;
	269	}
598ed936	270	EXPORT_SYMBOL(dst_destroy);
1da177e4	271
f8864972 ED	272	static void dst_destroy_rcu(struct rcu_head *head)
	273	{
	274	struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
	275
	276	dst = dst_destroy(dst);
	277	if (dst)
	278	__dst_free(dst);
	279	}
	280
8d330868 IJ	281	void dst_release(struct dst_entry *dst)
	282	{
	283	if (dst) {
598ed936	284	int newrefcnt;
ef711cf1	285
598ed936	286	newrefcnt = atomic_dec_return(&dst->__refcnt);
598ed936	287	WARN_ON(newrefcnt < 0);
f8864972 ED	288	if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt)
f8864972 ED	289	call_rcu(&dst->rcu_head, dst_destroy_rcu);
8d330868 IJ	290	}
	291	}
	292	EXPORT_SYMBOL(dst_release);
	293
62fa8a84 DM	294	u32 dst_cow_metrics_generic(struct dst_entry dst, unsigned long old)
	295	{
	296	u32 p = kmalloc(sizeof(u32) RTAX_MAX, GFP_ATOMIC);
	297
	298	if (p) {
	299	u32 *old_p = __DST_METRICS_PTR(old);
	300	unsigned long prev, new;
	301
	302	memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
	303
	304	new = (unsigned long) p;
	305	prev = cmpxchg(&dst->_metrics, old, new);
	306
	307	if (prev != old) {
	308	kfree(p);
	309	p = __DST_METRICS_PTR(prev);
	310	if (prev & DST_METRICS_READ_ONLY)
	311	p = NULL;
	312	}
	313	}
	314	return p;
	315	}
	316	EXPORT_SYMBOL(dst_cow_metrics_generic);
	317
	318	/* Caller asserts that dst_metrics_read_only(dst) is false. */
	319	void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
	320	{
	321	unsigned long prev, new;
	322
b30c516f	323	new = ((unsigned long) dst_default_metrics) \| DST_METRICS_READ_ONLY;
62fa8a84 DM	324	prev = cmpxchg(&dst->_metrics, old, new);
	325	if (prev == old)
	326	kfree(__DST_METRICS_PTR(old));
	327	}
	328	EXPORT_SYMBOL(__dst_destroy_metrics_generic);
	329
1da177e4 LT	330	/* Dirty hack. We did it in 2.2 (in __dst_free),
	331	* we have _very_ good reasons not to repeat
	332	* this mistake in 2.3, but we have no choice
	333	* now. _It_ _is_ _explicit_ _deliberate_
	334	* _race_ _condition_.
	335	*
	336	* Commented and originally written by Alexey.
	337	*/
56115511	338	static void dst_ifdown(struct dst_entry dst, struct net_device dev,
56115511	339	int unregister)
1da177e4 LT	340	{
	341	if (dst->ops->ifdown)
	342	dst->ops->ifdown(dst, dev, unregister);
	343
	344	if (dev != dst->dev)
	345	return;
	346
	347	if (!unregister) {
aad88724 ED	348	dst->input = dst_discard;
aad88724 ED	349	dst->output = dst_discard_sk;
1da177e4	350	} else {
c346dca1	351	dst->dev = dev_net(dst->dev)->loopback_dev;
de3cb747	352	dev_hold(dst->dev);
1da177e4	353	dev_put(dev);
1da177e4 LT	354	}
	355	}
	356
598ed936	357	static int dst_dev_event(struct notifier_block *this, unsigned long event,
598ed936	358	void *ptr)
1da177e4	359	{
351638e7	360	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
86bba269	361	struct dst_entry dst, last = NULL;
1da177e4 LT	362
1da177e4 LT	363	switch (event) {
0115e8e3	364	case NETDEV_UNREGISTER_FINAL:
1da177e4	365	case NETDEV_DOWN:
86bba269 ED	366	mutex_lock(&dst_gc_mutex);
	367	for (dst = dst_busy_list; dst; dst = dst->next) {
	368	last = dst;
	369	dst_ifdown(dst, dev, event != NETDEV_DOWN);
	370	}
	371
	372	spin_lock_bh(&dst_garbage.lock);
	373	dst = dst_garbage.list;
	374	dst_garbage.list = NULL;
	375	spin_unlock_bh(&dst_garbage.lock);
	376
	377	if (last)
	378	last->next = dst;
	379	else
	380	dst_busy_list = dst;
598ed936	381	for (; dst; dst = dst->next)
1da177e4	382	dst_ifdown(dst, dev, event != NETDEV_DOWN);
86bba269	383	mutex_unlock(&dst_gc_mutex);
1da177e4 LT	384	break;
	385	}
	386	return NOTIFY_DONE;
	387	}
	388
	389	static struct notifier_block dst_dev_notifier = {
	390	.notifier_call = dst_dev_event,
332dd96f	391	.priority = -10, /* must be called after other network notifiers */
1da177e4 LT	392	};
	393
	394	void __init dst_init(void)
	395	{
	396	register_netdevice_notifier(&dst_dev_notifier);
	397	}