[deliverable/linux.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>
#include <net/dst_metadata.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_init(struct dst_entry *dst, struct dst_ops *ops,
	      struct net_device *dev, int initial_ref, int initial_obsolete,
	      unsigned short flags)
{
	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);
}
EXPORT_SYMBOL(dst_init);

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_init(dst, ops, dev, initial_ref, initial_obsolete, flags);

	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);

	if (dst->flags & DST_METADATA)
		kfree(dst);
	else
		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);
		if (unlikely(newrefcnt < 0))
			net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
					     __func__, dst, newrefcnt);
		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);

static struct dst_ops md_dst_ops = {
	.family =		AF_UNSPEC,
};

static int dst_md_discard_sk(struct sock *sk, struct sk_buff *skb)
{
	WARN_ONCE(1, "Attempting to call output on metadata dst\n");
	kfree_skb(skb);
	return 0;
}

static int dst_md_discard(struct sk_buff *skb)
{
	WARN_ONCE(1, "Attempting to call input on metadata dst\n");
	kfree_skb(skb);
	return 0;
}

struct metadata_dst *metadata_dst_alloc(u8 optslen, gfp_t flags)
{
	struct metadata_dst *md_dst;
	struct dst_entry *dst;

	md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
	if (!md_dst)
		return ERR_PTR(-ENOMEM);

	dst = &md_dst->dst;
	dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
		 DST_METADATA | DST_NOCACHE | DST_NOCOUNT);

	dst->input = dst_md_discard;
	dst->output = dst_md_discard_sk;

	memset(dst + 1, 0, sizeof(*md_dst) + optslen - sizeof(*dst));
	md_dst->opts_len = optslen;

	return md_dst;
}
EXPORT_SYMBOL_GPL(metadata_dst_alloc);

/* 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_subsys_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
1da177e4 LT	24	#include <net/dst.h>
f38a9eb1	25	#include <net/dst_metadata.h>
1da177e4	26
86bba269 ED	27	/*
	28	* Theory of operations:
	29	* 1) We use a list, protected by a spinlock, to add
	30	* new entries from both BH and non-BH context.
	31	* 2) In order to keep spinlock held for a small delay,
	32	* we use a second list where are stored long lived
	33	* entries, that are handled by the garbage collect thread
	34	* fired by a workqueue.
	35	* 3) This list is guarded by a mutex,
	36	* so that the gc_task and dst_dev_event() can be synchronized.
1da177e4	37	*/
1da177e4	38
86bba269 ED	39	/*
	40	* We want to keep lock & list close together
	41	* to dirty as few cache lines as possible in __dst_free().
	42	* As this is not a very strong hint, we dont force an alignment on SMP.
	43	*/
	44	static struct {
	45	spinlock_t lock;
598ed936	46	struct dst_entry *list;
86bba269 ED	47	unsigned long timer_inc;
	48	unsigned long timer_expires;
	49	} dst_garbage = {
	50	.lock = __SPIN_LOCK_UNLOCKED(dst_garbage.lock),
	51	.timer_inc = DST_GC_MAX,
	52	};
	53	static void dst_gc_task(struct work_struct *work);
598ed936	54	static void ___dst_free(struct dst_entry *dst);
1da177e4	55
86bba269	56	static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task);
1da177e4	57
86bba269 ED	58	static DEFINE_MUTEX(dst_gc_mutex);
	59	/*
	60	* long lived entries are maintained in this list, guarded by dst_gc_mutex
	61	*/
	62	static struct dst_entry *dst_busy_list;
	63
	64	static void dst_gc_task(struct work_struct *work)
1da177e4 LT	65	{
1da177e4 LT	66	int delayed = 0;
86bba269 ED	67	int work_performed = 0;
	68	unsigned long expires = ~0L;
	69	struct dst_entry dst, next, head;
	70	struct dst_entry *last = &head;
1da177e4	71
86bba269 ED	72	mutex_lock(&dst_gc_mutex);
86bba269 ED	73	next = dst_busy_list;
1da177e4	74
86bba269 ED	75	loop:
	76	while ((dst = next) != NULL) {
	77	next = dst->next;
	78	prefetch(&next->next);
2fc1b5dd	79	cond_resched();
86bba269 ED	80	if (likely(atomic_read(&dst->__refcnt))) {
	81	last->next = dst;
	82	last = dst;
1da177e4 LT	83	delayed++;
	84	continue;
	85	}
86bba269	86	work_performed++;
1da177e4 LT	87
	88	dst = dst_destroy(dst);
	89	if (dst) {
	90	/* NOHASH and still referenced. Unless it is already
	91	* on gc list, invalidate it and add to gc list.
	92	*
	93	* Note: this is temporary. Actually, NOHASH dst's
	94	* must be obsoleted when parent is obsoleted.
	95	* But we do not have state "obsoleted, but
	96	* referenced by parent", so it is right.
	97	*/
f5b0a874	98	if (dst->obsolete > 0)
1da177e4 LT	99	continue;
	100
	101	___dst_free(dst);
86bba269 ED	102	dst->next = next;
86bba269 ED	103	next = dst;
1da177e4 LT	104	}
1da177e4 LT	105	}
86bba269 ED	106
	107	spin_lock_bh(&dst_garbage.lock);
	108	next = dst_garbage.list;
	109	if (next) {
	110	dst_garbage.list = NULL;
	111	spin_unlock_bh(&dst_garbage.lock);
	112	goto loop;
1da177e4	113	}
86bba269 ED	114	last->next = NULL;
	115	dst_busy_list = head.next;
	116	if (!dst_busy_list)
	117	dst_garbage.timer_inc = DST_GC_MAX;
	118	else {
	119	/*
	120	* if we freed less than 1/10 of delayed entries,
	121	* we can sleep longer.
	122	*/
	123	if (work_performed <= delayed/10) {
	124	dst_garbage.timer_expires += dst_garbage.timer_inc;
	125	if (dst_garbage.timer_expires > DST_GC_MAX)
	126	dst_garbage.timer_expires = DST_GC_MAX;
	127	dst_garbage.timer_inc += DST_GC_INC;
	128	} else {
	129	dst_garbage.timer_inc = DST_GC_INC;
	130	dst_garbage.timer_expires = DST_GC_MIN;
	131	}
	132	expires = dst_garbage.timer_expires;
	133	/*
598ed936	134	* if the next desired timer is more than 4 seconds in the
598ed936	135	* future then round the timer to whole seconds
86bba269 ED	136	*/
	137	if (expires > 4*HZ)
	138	expires = round_jiffies_relative(expires);
	139	schedule_delayed_work(&dst_gc_work, expires);
f0098f78	140	}
86bba269 ED	141
	142	spin_unlock_bh(&dst_garbage.lock);
	143	mutex_unlock(&dst_gc_mutex);
1da177e4 LT	144	}
1da177e4 LT	145
aad88724	146	int dst_discard_sk(struct sock sk, struct sk_buff skb)
1da177e4 LT	147	{
	148	kfree_skb(skb);
	149	return 0;
	150	}
aad88724	151	EXPORT_SYMBOL(dst_discard_sk);
1da177e4	152
a37e6e34 ED	153	const u32 dst_default_metrics[RTAX_MAX + 1] = {
	154	/* This initializer is needed to force linker to place this variable
	155	* into const section. Otherwise it might end into bss section.
	156	* We really want to avoid false sharing on this variable, and catch
	157	* any writes on it.
	158	*/
	159	[RTAX_MAX] = 0xdeadbeef,
	160	};
	161
f38a9eb1 TG	162	void dst_init(struct dst_entry dst, struct dst_ops ops,
	163	struct net_device *dev, int initial_ref, int initial_obsolete,
	164	unsigned short flags)
1da177e4	165	{
cf911662	166	dst->child = NULL;
5c1e6aa3 DM	167	dst->dev = dev;
	168	if (dev)
	169	dev_hold(dev);
1da177e4	170	dst->ops = ops;
62fa8a84	171	dst_init_metrics(dst, dst_default_metrics, true);
cf911662	172	dst->expires = 0UL;
5c1e6aa3	173	dst->path = dst;
ecd98837	174	dst->from = NULL;
cf911662 DM	175	#ifdef CONFIG_XFRM
	176	dst->xfrm = NULL;
	177	#endif
5c1e6aa3	178	dst->input = dst_discard;
aad88724	179	dst->output = dst_discard_sk;
cf911662	180	dst->error = 0;
5c1e6aa3	181	dst->obsolete = initial_obsolete;
cf911662 DM	182	dst->header_len = 0;
	183	dst->trailer_len = 0;
	184	#ifdef CONFIG_IP_ROUTE_CLASSID
	185	dst->tclassid = 0;
1da177e4	186	#endif
5c1e6aa3	187	atomic_set(&dst->__refcnt, initial_ref);
cf911662	188	dst->__use = 0;
5c1e6aa3 DM	189	dst->lastuse = jiffies;
5c1e6aa3 DM	190	dst->flags = flags;
5110effe	191	dst->pending_confirm = 0;
cf911662	192	dst->next = NULL;
957c665f DM	193	if (!(flags & DST_NOCOUNT))
957c665f DM	194	dst_entries_add(ops, 1);
f38a9eb1 TG	195	}
	196	EXPORT_SYMBOL(dst_init);
	197
	198	void dst_alloc(struct dst_ops ops, struct net_device *dev,
	199	int initial_ref, int initial_obsolete, unsigned short flags)
	200	{
	201	struct dst_entry *dst;
	202
	203	if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
	204	if (ops->gc(ops))
	205	return NULL;
	206	}
	207
	208	dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
	209	if (!dst)
	210	return NULL;
	211
	212	dst_init(dst, ops, dev, initial_ref, initial_obsolete, flags);
	213
1da177e4 LT	214	return dst;
1da177e4 LT	215	}
598ed936	216	EXPORT_SYMBOL(dst_alloc);
1da177e4	217
598ed936	218	static void ___dst_free(struct dst_entry *dst)
1da177e4 LT	219	{
	220	/* The first case (dev==NULL) is required, when
	221	protocol module is unloaded.
	222	*/
aad88724 ED	223	if (dst->dev == NULL \|\| !(dst->dev->flags&IFF_UP)) {
	224	dst->input = dst_discard;
	225	dst->output = dst_discard_sk;
	226	}
f5b0a874	227	dst->obsolete = DST_OBSOLETE_DEAD;
1da177e4 LT	228	}
1da177e4 LT	229
598ed936	230	void __dst_free(struct dst_entry *dst)
1da177e4	231	{
86bba269	232	spin_lock_bh(&dst_garbage.lock);
1da177e4	233	___dst_free(dst);
86bba269 ED	234	dst->next = dst_garbage.list;
	235	dst_garbage.list = dst;
	236	if (dst_garbage.timer_inc > DST_GC_INC) {
	237	dst_garbage.timer_inc = DST_GC_INC;
	238	dst_garbage.timer_expires = DST_GC_MIN;
41f63c53 TH	239	mod_delayed_work(system_wq, &dst_gc_work,
41f63c53 TH	240	dst_garbage.timer_expires);
1da177e4	241	}
86bba269	242	spin_unlock_bh(&dst_garbage.lock);
1da177e4	243	}
d79d9913	244	EXPORT_SYMBOL(__dst_free);
1da177e4 LT	245
	246	struct dst_entry dst_destroy(struct dst_entry dst)
	247	{
	248	struct dst_entry *child;
1da177e4 LT	249
	250	smp_rmb();
	251
	252	again:
1da177e4 LT	253	child = dst->child;
1da177e4 LT	254
957c665f DM	255	if (!(dst->flags & DST_NOCOUNT))
957c665f DM	256	dst_entries_add(dst->ops, -1);
1da177e4 LT	257
	258	if (dst->ops->destroy)
	259	dst->ops->destroy(dst);
	260	if (dst->dev)
	261	dev_put(dst->dev);
f38a9eb1 TG	262
	263	if (dst->flags & DST_METADATA)
	264	kfree(dst);
	265	else
	266	kmem_cache_free(dst->ops->kmem_cachep, dst);
1da177e4 LT	267
	268	dst = child;
	269	if (dst) {
6775cab9 HX	270	int nohash = dst->flags & DST_NOHASH;
6775cab9 HX	271
1da177e4 LT	272	if (atomic_dec_and_test(&dst->__refcnt)) {
1da177e4 LT	273	/* We were real parent of this dst, so kill child. */
6775cab9	274	if (nohash)
1da177e4 LT	275	goto again;
	276	} else {
	277	/* Child is still referenced, return it for freeing. */
6775cab9	278	if (nohash)
1da177e4 LT	279	return dst;
	280	/* Child is still in his hash table */
	281	}
	282	}
	283	return NULL;
	284	}
598ed936	285	EXPORT_SYMBOL(dst_destroy);
1da177e4	286
f8864972 ED	287	static void dst_destroy_rcu(struct rcu_head *head)
	288	{
	289	struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
	290
	291	dst = dst_destroy(dst);
	292	if (dst)
	293	__dst_free(dst);
	294	}
	295
8d330868 IJ	296	void dst_release(struct dst_entry *dst)
	297	{
	298	if (dst) {
598ed936	299	int newrefcnt;
ef711cf1	300
598ed936	301	newrefcnt = atomic_dec_return(&dst->__refcnt);
8bf4ada2 KK	302	if (unlikely(newrefcnt < 0))
	303	net_warn_ratelimited("%s: dst:%p refcnt:%d\n",
	304	__func__, dst, newrefcnt);
f8864972 ED	305	if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt)
f8864972 ED	306	call_rcu(&dst->rcu_head, dst_destroy_rcu);
8d330868 IJ	307	}
	308	}
	309	EXPORT_SYMBOL(dst_release);
	310
62fa8a84 DM	311	u32 dst_cow_metrics_generic(struct dst_entry dst, unsigned long old)
	312	{
	313	u32 p = kmalloc(sizeof(u32) RTAX_MAX, GFP_ATOMIC);
	314
	315	if (p) {
	316	u32 *old_p = __DST_METRICS_PTR(old);
	317	unsigned long prev, new;
	318
	319	memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
	320
	321	new = (unsigned long) p;
	322	prev = cmpxchg(&dst->_metrics, old, new);
	323
	324	if (prev != old) {
	325	kfree(p);
	326	p = __DST_METRICS_PTR(prev);
	327	if (prev & DST_METRICS_READ_ONLY)
	328	p = NULL;
	329	}
	330	}
	331	return p;
	332	}
	333	EXPORT_SYMBOL(dst_cow_metrics_generic);
	334
	335	/* Caller asserts that dst_metrics_read_only(dst) is false. */
	336	void __dst_destroy_metrics_generic(struct dst_entry *dst, unsigned long old)
	337	{
	338	unsigned long prev, new;
	339
b30c516f	340	new = ((unsigned long) dst_default_metrics) \| DST_METRICS_READ_ONLY;
62fa8a84 DM	341	prev = cmpxchg(&dst->_metrics, old, new);
	342	if (prev == old)
	343	kfree(__DST_METRICS_PTR(old));
	344	}
	345	EXPORT_SYMBOL(__dst_destroy_metrics_generic);
	346
f38a9eb1 TG	347	static struct dst_ops md_dst_ops = {
	348	.family = AF_UNSPEC,
	349	};
	350
	351	static int dst_md_discard_sk(struct sock sk, struct sk_buff skb)
	352	{
	353	WARN_ONCE(1, "Attempting to call output on metadata dst\n");
	354	kfree_skb(skb);
	355	return 0;
	356	}
	357
	358	static int dst_md_discard(struct sk_buff *skb)
	359	{
	360	WARN_ONCE(1, "Attempting to call input on metadata dst\n");
	361	kfree_skb(skb);
	362	return 0;
	363	}
	364
	365	struct metadata_dst *metadata_dst_alloc(u8 optslen, gfp_t flags)
	366	{
	367	struct metadata_dst *md_dst;
	368	struct dst_entry *dst;
	369
	370	md_dst = kmalloc(sizeof(*md_dst) + optslen, flags);
	371	if (!md_dst)
	372	return ERR_PTR(-ENOMEM);
	373
	374	dst = &md_dst->dst;
	375	dst_init(dst, &md_dst_ops, NULL, 1, DST_OBSOLETE_NONE,
	376	DST_METADATA \| DST_NOCACHE \| DST_NOCOUNT);
	377
	378	dst->input = dst_md_discard;
	379	dst->output = dst_md_discard_sk;
	380
	381	memset(dst + 1, 0, sizeof(md_dst) + optslen - sizeof(dst));
	382	md_dst->opts_len = optslen;
	383
	384	return md_dst;
	385	}
	386	EXPORT_SYMBOL_GPL(metadata_dst_alloc);
	387
1da177e4 LT	388	/* Dirty hack. We did it in 2.2 (in __dst_free),
	389	* we have _very_ good reasons not to repeat
	390	* this mistake in 2.3, but we have no choice
	391	* now. _It_ _is_ _explicit_ _deliberate_
	392	* _race_ _condition_.
	393	*
	394	* Commented and originally written by Alexey.
	395	*/
56115511	396	static void dst_ifdown(struct dst_entry dst, struct net_device dev,
56115511	397	int unregister)
1da177e4 LT	398	{
	399	if (dst->ops->ifdown)
	400	dst->ops->ifdown(dst, dev, unregister);
	401
	402	if (dev != dst->dev)
	403	return;
	404
	405	if (!unregister) {
aad88724 ED	406	dst->input = dst_discard;
aad88724 ED	407	dst->output = dst_discard_sk;
1da177e4	408	} else {
c346dca1	409	dst->dev = dev_net(dst->dev)->loopback_dev;
de3cb747	410	dev_hold(dst->dev);
1da177e4	411	dev_put(dev);
1da177e4 LT	412	}
	413	}
	414
598ed936	415	static int dst_dev_event(struct notifier_block *this, unsigned long event,
598ed936	416	void *ptr)
1da177e4	417	{
351638e7	418	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
86bba269	419	struct dst_entry dst, last = NULL;
1da177e4 LT	420
1da177e4 LT	421	switch (event) {
0115e8e3	422	case NETDEV_UNREGISTER_FINAL:
1da177e4	423	case NETDEV_DOWN:
86bba269 ED	424	mutex_lock(&dst_gc_mutex);
	425	for (dst = dst_busy_list; dst; dst = dst->next) {
	426	last = dst;
	427	dst_ifdown(dst, dev, event != NETDEV_DOWN);
	428	}
	429
	430	spin_lock_bh(&dst_garbage.lock);
	431	dst = dst_garbage.list;
	432	dst_garbage.list = NULL;
	433	spin_unlock_bh(&dst_garbage.lock);
	434
	435	if (last)
	436	last->next = dst;
	437	else
	438	dst_busy_list = dst;
598ed936	439	for (; dst; dst = dst->next)
1da177e4	440	dst_ifdown(dst, dev, event != NETDEV_DOWN);
86bba269	441	mutex_unlock(&dst_gc_mutex);
1da177e4 LT	442	break;
	443	}
	444	return NOTIFY_DONE;
	445	}
	446
	447	static struct notifier_block dst_dev_notifier = {
	448	.notifier_call = dst_dev_event,
332dd96f	449	.priority = -10, /* must be called after other network notifiers */
1da177e4 LT	450	};
1da177e4 LT	451
f38a9eb1	452	void __init dst_subsys_init(void)
1da177e4 LT	453	{
	454	register_netdevice_notifier(&dst_dev_notifier);
	455	}