[mirror_ubuntu-zesty-kernel.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 <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);
		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 > 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
}

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

static const u32 dst_default_metrics[RTAX_MAX];

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

	if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
		if (ops->gc(ops))
			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;
	dst_init_metrics(dst, dst_default_metrics, true);
#if RT_CACHE_DEBUG >= 2
	atomic_inc(&dst_total);
#endif
	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->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;
		cancel_delayed_work(&dst_gc_work);
		schedule_delayed_work(&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;
	struct neighbour *neigh;
	struct hh_cache *hh;

	smp_rmb();

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

	dst->hh = NULL;
	if (hh)
		hh_cache_put(hh);

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

	dst_entries_add(dst->ops, -1);

	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;
}
EXPORT_SYMBOL(dst_destroy);

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) {
			dst = dst_destroy(dst);
			if (dst)
				__dst_free(dst);
		}
	}
}
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;
	prev = cmpxchg(&dst->_metrics, old, new);
	if (prev == old)
		kfree(__DST_METRICS_PTR(old));
}
EXPORT_SYMBOL(__dst_destroy_metrics_generic);

/**
 * skb_dst_set_noref - sets skb dst, without a reference
 * @skb: buffer
 * @dst: dst entry
 *
 * Sets skb dst, assuming a reference was not taken on dst
 * skb_dst_drop() should not dst_release() this dst
 */
void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
{
	WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
	/* If dst not in cache, we must take a reference, because
	 * dst_release() will destroy dst as soon as its refcount becomes zero
	 */
	if (unlikely(dst->flags & DST_NOCACHE)) {
		dst_hold(dst);
		skb_dst_set(skb, dst);
	} else {
		skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
	}
}
EXPORT_SYMBOL(skb_dst_set_noref);

/* 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->output = dst_discard;
	} else {
		dst->dev = dev_net(dst->dev)->loopback_dev;
		dev_hold(dst->dev);
		dev_put(dev);
		if (dst->neighbour && dst->neighbour->dev == dev) {
			dst->neighbour->dev = dst->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 = ptr;
	struct dst_entry *dst, *last = NULL;

	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,
	.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>
1da177e4 LT	22
	23	#include <net/dst.h>
	24
86bba269 ED	25	/*
	26	* Theory of operations:
	27	* 1) We use a list, protected by a spinlock, to add
	28	* new entries from both BH and non-BH context.
	29	* 2) In order to keep spinlock held for a small delay,
	30	* we use a second list where are stored long lived
	31	* entries, that are handled by the garbage collect thread
	32	* fired by a workqueue.
	33	* 3) This list is guarded by a mutex,
	34	* so that the gc_task and dst_dev_event() can be synchronized.
1da177e4	35	*/
4ec93edb	36	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	37	static atomic_t dst_total = ATOMIC_INIT(0);
1da177e4 LT	38	#endif
1da177e4	39
86bba269 ED	40	/*
	41	* We want to keep lock & list close together
	42	* to dirty as few cache lines as possible in __dst_free().
	43	* As this is not a very strong hint, we dont force an alignment on SMP.
	44	*/
	45	static struct {
	46	spinlock_t lock;
598ed936	47	struct dst_entry *list;
86bba269 ED	48	unsigned long timer_inc;
	49	unsigned long timer_expires;
	50	} dst_garbage = {
	51	.lock = __SPIN_LOCK_UNLOCKED(dst_garbage.lock),
	52	.timer_inc = DST_GC_MAX,
	53	};
	54	static void dst_gc_task(struct work_struct *work);
598ed936	55	static void ___dst_free(struct dst_entry *dst);
1da177e4	56
86bba269	57	static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task);
1da177e4	58
86bba269 ED	59	static DEFINE_MUTEX(dst_gc_mutex);
	60	/*
	61	* long lived entries are maintained in this list, guarded by dst_gc_mutex
	62	*/
	63	static struct dst_entry *dst_busy_list;
	64
	65	static void dst_gc_task(struct work_struct *work)
1da177e4 LT	66	{
1da177e4 LT	67	int delayed = 0;
86bba269 ED	68	int work_performed = 0;
	69	unsigned long expires = ~0L;
	70	struct dst_entry dst, next, head;
	71	struct dst_entry *last = &head;
	72	#if RT_CACHE_DEBUG >= 2
	73	ktime_t time_start = ktime_get();
	74	struct timespec elapsed;
	75	#endif
1da177e4	76
86bba269 ED	77	mutex_lock(&dst_gc_mutex);
86bba269 ED	78	next = dst_busy_list;
1da177e4	79
86bba269 ED	80	loop:
	81	while ((dst = next) != NULL) {
	82	next = dst->next;
	83	prefetch(&next->next);
2fc1b5dd	84	cond_resched();
86bba269 ED	85	if (likely(atomic_read(&dst->__refcnt))) {
	86	last->next = dst;
	87	last = dst;
1da177e4 LT	88	delayed++;
	89	continue;
	90	}
86bba269	91	work_performed++;
1da177e4 LT	92
	93	dst = dst_destroy(dst);
	94	if (dst) {
	95	/* NOHASH and still referenced. Unless it is already
	96	* on gc list, invalidate it and add to gc list.
	97	*
	98	* Note: this is temporary. Actually, NOHASH dst's
	99	* must be obsoleted when parent is obsoleted.
	100	* But we do not have state "obsoleted, but
	101	* referenced by parent", so it is right.
	102	*/
	103	if (dst->obsolete > 1)
	104	continue;
	105
	106	___dst_free(dst);
86bba269 ED	107	dst->next = next;
86bba269 ED	108	next = dst;
1da177e4 LT	109	}
1da177e4 LT	110	}
86bba269 ED	111
	112	spin_lock_bh(&dst_garbage.lock);
	113	next = dst_garbage.list;
	114	if (next) {
	115	dst_garbage.list = NULL;
	116	spin_unlock_bh(&dst_garbage.lock);
	117	goto loop;
1da177e4	118	}
86bba269 ED	119	last->next = NULL;
	120	dst_busy_list = head.next;
	121	if (!dst_busy_list)
	122	dst_garbage.timer_inc = DST_GC_MAX;
	123	else {
	124	/*
	125	* if we freed less than 1/10 of delayed entries,
	126	* we can sleep longer.
	127	*/
	128	if (work_performed <= delayed/10) {
	129	dst_garbage.timer_expires += dst_garbage.timer_inc;
	130	if (dst_garbage.timer_expires > DST_GC_MAX)
	131	dst_garbage.timer_expires = DST_GC_MAX;
	132	dst_garbage.timer_inc += DST_GC_INC;
	133	} else {
	134	dst_garbage.timer_inc = DST_GC_INC;
	135	dst_garbage.timer_expires = DST_GC_MIN;
	136	}
	137	expires = dst_garbage.timer_expires;
	138	/*
598ed936	139	* if the next desired timer is more than 4 seconds in the
598ed936	140	* future then round the timer to whole seconds
86bba269 ED	141	*/
	142	if (expires > 4*HZ)
	143	expires = round_jiffies_relative(expires);
	144	schedule_delayed_work(&dst_gc_work, expires);
f0098f78	145	}
86bba269 ED	146
	147	spin_unlock_bh(&dst_garbage.lock);
	148	mutex_unlock(&dst_gc_mutex);
1da177e4	149	#if RT_CACHE_DEBUG >= 2
86bba269 ED	150	elapsed = ktime_to_timespec(ktime_sub(ktime_get(), time_start));
	151	printk(KERN_DEBUG "dst_total: %d delayed: %d work_perf: %d"
	152	" expires: %lu elapsed: %lu us\n",
	153	atomic_read(&dst_total), delayed, work_performed,
	154	expires,
598ed936	155	elapsed.tv_sec * USEC_PER_SEC +
598ed936	156	elapsed.tv_nsec / NSEC_PER_USEC);
1da177e4	157	#endif
1da177e4 LT	158	}
1da177e4 LT	159
352e512c	160	int dst_discard(struct sk_buff *skb)
1da177e4 LT	161	{
	162	kfree_skb(skb);
	163	return 0;
	164	}
352e512c	165	EXPORT_SYMBOL(dst_discard);
1da177e4	166
62fa8a84 DM	167	static const u32 dst_default_metrics[RTAX_MAX];
62fa8a84 DM	168
598ed936	169	void dst_alloc(struct dst_ops ops)
1da177e4	170	{
598ed936	171	struct dst_entry *dst;
1da177e4	172
fc66f95c	173	if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
569d3645	174	if (ops->gc(ops))
1da177e4 LT	175	return NULL;
1da177e4 LT	176	}
c3762229	177	dst = kmem_cache_zalloc(ops->kmem_cachep, GFP_ATOMIC);
1da177e4 LT	178	if (!dst)
1da177e4 LT	179	return NULL;
1da177e4 LT	180	atomic_set(&dst->__refcnt, 0);
	181	dst->ops = ops;
	182	dst->lastuse = jiffies;
	183	dst->path = dst;
c4b1010f	184	dst->input = dst->output = dst_discard;
62fa8a84	185	dst_init_metrics(dst, dst_default_metrics, true);
4ec93edb	186	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	187	atomic_inc(&dst_total);
1da177e4 LT	188	#endif
fc66f95c	189	dst_entries_add(ops, 1);
1da177e4 LT	190	return dst;
1da177e4 LT	191	}
598ed936	192	EXPORT_SYMBOL(dst_alloc);
1da177e4	193
598ed936	194	static void ___dst_free(struct dst_entry *dst)
1da177e4 LT	195	{
	196	/* The first case (dev==NULL) is required, when
	197	protocol module is unloaded.
	198	*/
598ed936	199	if (dst->dev == NULL \|\| !(dst->dev->flags&IFF_UP))
c4b1010f	200	dst->input = dst->output = dst_discard;
1da177e4 LT	201	dst->obsolete = 2;
	202	}
	203
598ed936	204	void __dst_free(struct dst_entry *dst)
1da177e4	205	{
86bba269	206	spin_lock_bh(&dst_garbage.lock);
1da177e4	207	___dst_free(dst);
86bba269 ED	208	dst->next = dst_garbage.list;
	209	dst_garbage.list = dst;
	210	if (dst_garbage.timer_inc > DST_GC_INC) {
	211	dst_garbage.timer_inc = DST_GC_INC;
	212	dst_garbage.timer_expires = DST_GC_MIN;
f262b59b	213	cancel_delayed_work(&dst_gc_work);
86bba269	214	schedule_delayed_work(&dst_gc_work, dst_garbage.timer_expires);
1da177e4	215	}
86bba269	216	spin_unlock_bh(&dst_garbage.lock);
1da177e4	217	}
d79d9913	218	EXPORT_SYMBOL(__dst_free);
1da177e4 LT	219
	220	struct dst_entry dst_destroy(struct dst_entry dst)
	221	{
	222	struct dst_entry *child;
	223	struct neighbour *neigh;
	224	struct hh_cache *hh;
	225
	226	smp_rmb();
	227
	228	again:
	229	neigh = dst->neighbour;
	230	hh = dst->hh;
	231	child = dst->child;
	232
	233	dst->hh = NULL;
34d101dd ED	234	if (hh)
34d101dd ED	235	hh_cache_put(hh);
1da177e4 LT	236
	237	if (neigh) {
	238	dst->neighbour = NULL;
	239	neigh_release(neigh);
	240	}
	241
fc66f95c	242	dst_entries_add(dst->ops, -1);
1da177e4 LT	243
	244	if (dst->ops->destroy)
	245	dst->ops->destroy(dst);
	246	if (dst->dev)
	247	dev_put(dst->dev);
4ec93edb	248	#if RT_CACHE_DEBUG >= 2
1da177e4 LT	249	atomic_dec(&dst_total);
	250	#endif
	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
8d330868 IJ	272	void dst_release(struct dst_entry *dst)
	273	{
	274	if (dst) {
598ed936	275	int newrefcnt;
ef711cf1	276
598ed936	277	newrefcnt = atomic_dec_return(&dst->__refcnt);
598ed936	278	WARN_ON(newrefcnt < 0);
27b75c95 ED	279	if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt) {
	280	dst = dst_destroy(dst);
	281	if (dst)
	282	__dst_free(dst);
	283	}
8d330868 IJ	284	}
	285	}
	286	EXPORT_SYMBOL(dst_release);
	287
62fa8a84 DM	288	u32 dst_cow_metrics_generic(struct dst_entry dst, unsigned long old)
	289	{
	290	u32 p = kmalloc(sizeof(u32) RTAX_MAX, GFP_ATOMIC);
	291
	292	if (p) {
	293	u32 *old_p = __DST_METRICS_PTR(old);
	294	unsigned long prev, new;
	295
	296	memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
	297
	298	new = (unsigned long) p;
	299	prev = cmpxchg(&dst->_metrics, old, new);
	300
	301	if (prev != old) {
	302	kfree(p);
	303	p = __DST_METRICS_PTR(prev);
	304	if (prev & DST_METRICS_READ_ONLY)
	305	p = NULL;
	306	}
	307	}
	308	return p;
	309	}
	310	EXPORT_SYMBOL(dst_cow_metrics_generic);
	311
	312	/* Caller asserts that dst_metrics_read_only(dst) is false. */
	313	void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
	314	{
	315	unsigned long prev, new;
	316
	317	new = (unsigned long) dst_default_metrics;
	318	prev = cmpxchg(&dst->_metrics, old, new);
	319	if (prev == old)
	320	kfree(__DST_METRICS_PTR(old));
	321	}
	322	EXPORT_SYMBOL(__dst_destroy_metrics_generic);
	323
27b75c95 ED	324	/**
	325	* skb_dst_set_noref - sets skb dst, without a reference
	326	* @skb: buffer
	327	* @dst: dst entry
	328	*
	329	* Sets skb dst, assuming a reference was not taken on dst
	330	* skb_dst_drop() should not dst_release() this dst
	331	*/
	332	void skb_dst_set_noref(struct sk_buff skb, struct dst_entry dst)
	333	{
	334	WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
	335	/* If dst not in cache, we must take a reference, because
	336	* dst_release() will destroy dst as soon as its refcount becomes zero
	337	*/
	338	if (unlikely(dst->flags & DST_NOCACHE)) {
	339	dst_hold(dst);
	340	skb_dst_set(skb, dst);
	341	} else {
	342	skb->_skb_refdst = (unsigned long)dst \| SKB_DST_NOREF;
	343	}
	344	}
	345	EXPORT_SYMBOL(skb_dst_set_noref);
	346
1da177e4 LT	347	/* Dirty hack. We did it in 2.2 (in __dst_free),
	348	* we have _very_ good reasons not to repeat
	349	* this mistake in 2.3, but we have no choice
	350	* now. _It_ _is_ _explicit_ _deliberate_
	351	* _race_ _condition_.
	352	*
	353	* Commented and originally written by Alexey.
	354	*/
56115511	355	static void dst_ifdown(struct dst_entry dst, struct net_device dev,
56115511	356	int unregister)
1da177e4 LT	357	{
	358	if (dst->ops->ifdown)
	359	dst->ops->ifdown(dst, dev, unregister);
	360
	361	if (dev != dst->dev)
	362	return;
	363
	364	if (!unregister) {
c4b1010f	365	dst->input = dst->output = dst_discard;
1da177e4	366	} else {
c346dca1	367	dst->dev = dev_net(dst->dev)->loopback_dev;
de3cb747	368	dev_hold(dst->dev);
1da177e4 LT	369	dev_put(dev);
1da177e4 LT	370	if (dst->neighbour && dst->neighbour->dev == dev) {
5a3e55d6	371	dst->neighbour->dev = dst->dev;
64b7d961	372	dev_hold(dst->dev);
1da177e4	373	dev_put(dev);
1da177e4 LT	374	}
	375	}
	376	}
	377
598ed936	378	static int dst_dev_event(struct notifier_block *this, unsigned long event,
598ed936	379	void *ptr)
1da177e4 LT	380	{
1da177e4 LT	381	struct net_device *dev = ptr;
86bba269	382	struct dst_entry dst, last = NULL;
1da177e4 LT	383
	384	switch (event) {
	385	case NETDEV_UNREGISTER:
	386	case NETDEV_DOWN:
86bba269 ED	387	mutex_lock(&dst_gc_mutex);
	388	for (dst = dst_busy_list; dst; dst = dst->next) {
	389	last = dst;
	390	dst_ifdown(dst, dev, event != NETDEV_DOWN);
	391	}
	392
	393	spin_lock_bh(&dst_garbage.lock);
	394	dst = dst_garbage.list;
	395	dst_garbage.list = NULL;
	396	spin_unlock_bh(&dst_garbage.lock);
	397
	398	if (last)
	399	last->next = dst;
	400	else
	401	dst_busy_list = dst;
598ed936	402	for (; dst; dst = dst->next)
1da177e4	403	dst_ifdown(dst, dev, event != NETDEV_DOWN);
86bba269	404	mutex_unlock(&dst_gc_mutex);
1da177e4 LT	405	break;
	406	}
	407	return NOTIFY_DONE;
	408	}
	409
	410	static struct notifier_block dst_dev_notifier = {
	411	.notifier_call = dst_dev_event,
332dd96f	412	.priority = -10, /* must be called after other network notifiers */
1da177e4 LT	413	};
	414
	415	void __init dst_init(void)
	416	{
	417	register_netdevice_notifier(&dst_dev_notifier);
	418	}