dst.c 9.68 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6 7 8 9 10 11
/*
 * 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>
12
#include <linux/workqueue.h>
Linus Torvalds's avatar
Linus Torvalds committed
13 14
#include <linux/mm.h>
#include <linux/module.h>
15
#include <linux/slab.h>
Linus Torvalds's avatar
Linus Torvalds committed
16 17 18 19
#include <linux/netdevice.h>
#include <linux/skbuff.h>
#include <linux/string.h>
#include <linux/types.h>
20
#include <net/net_namespace.h>
21
#include <linux/sched.h>
22
#include <linux/prefetch.h>
Linus Torvalds's avatar
Linus Torvalds committed
23 24 25

#include <net/dst.h>

26 27 28 29 30 31 32 33 34 35
/*
 * 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.
Linus Torvalds's avatar
Linus Torvalds committed
36 37
 */

38 39 40 41 42 43 44
/*
 * 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;
45
	struct dst_entry	*list;
46 47 48 49 50 51 52
	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);
53
static void ___dst_free(struct dst_entry *dst);
Linus Torvalds's avatar
Linus Torvalds committed
54

55
static DECLARE_DELAYED_WORK(dst_gc_work, dst_gc_task);
Linus Torvalds's avatar
Linus Torvalds committed
56

57 58 59 60 61 62 63
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)
Linus Torvalds's avatar
Linus Torvalds committed
64 65
{
	int    delayed = 0;
66 67 68 69
	int    work_performed = 0;
	unsigned long expires = ~0L;
	struct dst_entry *dst, *next, head;
	struct dst_entry *last = &head;
Linus Torvalds's avatar
Linus Torvalds committed
70

71 72
	mutex_lock(&dst_gc_mutex);
	next = dst_busy_list;
Linus Torvalds's avatar
Linus Torvalds committed
73

74 75 76 77
loop:
	while ((dst = next) != NULL) {
		next = dst->next;
		prefetch(&next->next);
78
		cond_resched();
79 80 81
		if (likely(atomic_read(&dst->__refcnt))) {
			last->next = dst;
			last = dst;
Linus Torvalds's avatar
Linus Torvalds committed
82 83 84
			delayed++;
			continue;
		}
85
		work_performed++;
Linus Torvalds's avatar
Linus Torvalds committed
86 87 88 89 90 91 92 93 94 95 96 97 98 99 100

		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);
101 102
			dst->next = next;
			next = dst;
Linus Torvalds's avatar
Linus Torvalds committed
103 104
		}
	}
105 106 107 108 109 110 111

	spin_lock_bh(&dst_garbage.lock);
	next = dst_garbage.list;
	if (next) {
		dst_garbage.list = NULL;
		spin_unlock_bh(&dst_garbage.lock);
		goto loop;
Linus Torvalds's avatar
Linus Torvalds committed
112
	}
113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132
	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;
		/*
133 134
		 * if the next desired timer is more than 4 seconds in the
		 * future then round the timer to whole seconds
135 136 137 138
		 */
		if (expires > 4*HZ)
			expires = round_jiffies_relative(expires);
		schedule_delayed_work(&dst_gc_work, expires);
139
	}
140 141 142

	spin_unlock_bh(&dst_garbage.lock);
	mutex_unlock(&dst_gc_mutex);
Linus Torvalds's avatar
Linus Torvalds committed
143 144
}

145
int dst_discard(struct sk_buff *skb)
Linus Torvalds's avatar
Linus Torvalds committed
146 147 148 149
{
	kfree_skb(skb);
	return 0;
}
150
EXPORT_SYMBOL(dst_discard);
Linus Torvalds's avatar
Linus Torvalds committed
151

152
const u32 dst_default_metrics[RTAX_MAX];
153

154 155
void *dst_alloc(struct dst_ops *ops, struct net_device *dev,
		int initial_ref, int initial_obsolete, int flags)
Linus Torvalds's avatar
Linus Torvalds committed
156
{
157
	struct dst_entry *dst;
Linus Torvalds's avatar
Linus Torvalds committed
158

159
	if (ops->gc && dst_entries_get_fast(ops) > ops->gc_thresh) {
160
		if (ops->gc(ops))
Linus Torvalds's avatar
Linus Torvalds committed
161 162
			return NULL;
	}
163
	dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
Linus Torvalds's avatar
Linus Torvalds committed
164 165
	if (!dst)
		return NULL;
166
	dst->child = NULL;
167 168 169
	dst->dev = dev;
	if (dev)
		dev_hold(dev);
Linus Torvalds's avatar
Linus Torvalds committed
170
	dst->ops = ops;
171
	dst_init_metrics(dst, dst_default_metrics, true);
172
	dst->expires = 0UL;
173
	dst->path = dst;
174 175 176 177
	dst->neighbour = NULL;
#ifdef CONFIG_XFRM
	dst->xfrm = NULL;
#endif
178 179
	dst->input = dst_discard;
	dst->output = dst_discard;
180
	dst->error = 0;
181
	dst->obsolete = initial_obsolete;
182 183 184 185
	dst->header_len = 0;
	dst->trailer_len = 0;
#ifdef CONFIG_IP_ROUTE_CLASSID
	dst->tclassid = 0;
Linus Torvalds's avatar
Linus Torvalds committed
186
#endif
187
	atomic_set(&dst->__refcnt, initial_ref);
188
	dst->__use = 0;
189 190
	dst->lastuse = jiffies;
	dst->flags = flags;
191
	dst->next = NULL;
192 193
	if (!(flags & DST_NOCOUNT))
		dst_entries_add(ops, 1);
Linus Torvalds's avatar
Linus Torvalds committed
194 195
	return dst;
}
196
EXPORT_SYMBOL(dst_alloc);
Linus Torvalds's avatar
Linus Torvalds committed
197

198
static void ___dst_free(struct dst_entry *dst)
Linus Torvalds's avatar
Linus Torvalds committed
199 200 201 202
{
	/* The first case (dev==NULL) is required, when
	   protocol module is unloaded.
	 */
203
	if (dst->dev == NULL || !(dst->dev->flags&IFF_UP))
204
		dst->input = dst->output = dst_discard;
Linus Torvalds's avatar
Linus Torvalds committed
205 206 207
	dst->obsolete = 2;
}

208
void __dst_free(struct dst_entry *dst)
Linus Torvalds's avatar
Linus Torvalds committed
209
{
210
	spin_lock_bh(&dst_garbage.lock);
Linus Torvalds's avatar
Linus Torvalds committed
211
	___dst_free(dst);
212 213 214 215 216
	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;
217
		cancel_delayed_work(&dst_gc_work);
218
		schedule_delayed_work(&dst_gc_work, dst_garbage.timer_expires);
Linus Torvalds's avatar
Linus Torvalds committed
219
	}
220
	spin_unlock_bh(&dst_garbage.lock);
Linus Torvalds's avatar
Linus Torvalds committed
221
}
222
EXPORT_SYMBOL(__dst_free);
Linus Torvalds's avatar
Linus Torvalds committed
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239

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

	smp_rmb();

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

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

240 241
	if (!(dst->flags & DST_NOCOUNT))
		dst_entries_add(dst->ops, -1);
Linus Torvalds's avatar
Linus Torvalds committed
242 243 244 245 246 247 248 249 250

	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) {
Herbert Xu's avatar
Herbert Xu committed
251 252
		int nohash = dst->flags & DST_NOHASH;

Linus Torvalds's avatar
Linus Torvalds committed
253 254
		if (atomic_dec_and_test(&dst->__refcnt)) {
			/* We were real parent of this dst, so kill child. */
Herbert Xu's avatar
Herbert Xu committed
255
			if (nohash)
Linus Torvalds's avatar
Linus Torvalds committed
256 257 258
				goto again;
		} else {
			/* Child is still referenced, return it for freeing. */
Herbert Xu's avatar
Herbert Xu committed
259
			if (nohash)
Linus Torvalds's avatar
Linus Torvalds committed
260 261 262 263 264 265
				return dst;
			/* Child is still in his hash table */
		}
	}
	return NULL;
}
266
EXPORT_SYMBOL(dst_destroy);
Linus Torvalds's avatar
Linus Torvalds committed
267

Ilpo Järvinen's avatar
Ilpo Järvinen committed
268 269 270
void dst_release(struct dst_entry *dst)
{
	if (dst) {
271
		int newrefcnt;
Eric Dumazet's avatar
Eric Dumazet committed
272

273 274
		newrefcnt = atomic_dec_return(&dst->__refcnt);
		WARN_ON(newrefcnt < 0);
275 276 277 278 279
		if (unlikely(dst->flags & DST_NOCACHE) && !newrefcnt) {
			dst = dst_destroy(dst);
			if (dst)
				__dst_free(dst);
		}
Ilpo Järvinen's avatar
Ilpo Järvinen committed
280 281 282 283
	}
}
EXPORT_SYMBOL(dst_release);

284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312
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;

313
	new = ((unsigned long) dst_default_metrics) | DST_METRICS_READ_ONLY;
314 315 316 317 318 319
	prev = cmpxchg(&dst->_metrics, old, new);
	if (prev == old)
		kfree(__DST_METRICS_PTR(old));
}
EXPORT_SYMBOL(__dst_destroy_metrics_generic);

320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342
/**
 * 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);

Linus Torvalds's avatar
Linus Torvalds committed
343 344 345 346 347 348 349 350
/* 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.
 */
351 352
static void dst_ifdown(struct dst_entry *dst, struct net_device *dev,
		       int unregister)
Linus Torvalds's avatar
Linus Torvalds committed
353 354 355 356 357 358 359 360
{
	if (dst->ops->ifdown)
		dst->ops->ifdown(dst, dev, unregister);

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

	if (!unregister) {
361
		dst->input = dst->output = dst_discard;
Linus Torvalds's avatar
Linus Torvalds committed
362
	} else {
363
		dst->dev = dev_net(dst->dev)->loopback_dev;
364
		dev_hold(dst->dev);
Linus Torvalds's avatar
Linus Torvalds committed
365 366
		dev_put(dev);
		if (dst->neighbour && dst->neighbour->dev == dev) {
367
			dst->neighbour->dev = dst->dev;
Eric Dumazet's avatar
Eric Dumazet committed
368
			dev_hold(dst->dev);
Linus Torvalds's avatar
Linus Torvalds committed
369 370 371 372 373
			dev_put(dev);
		}
	}
}

374 375
static int dst_dev_event(struct notifier_block *this, unsigned long event,
			 void *ptr)
Linus Torvalds's avatar
Linus Torvalds committed
376 377
{
	struct net_device *dev = ptr;
378
	struct dst_entry *dst, *last = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
379 380 381 382

	switch (event) {
	case NETDEV_UNREGISTER:
	case NETDEV_DOWN:
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
		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;
398
		for (; dst; dst = dst->next)
Linus Torvalds's avatar
Linus Torvalds committed
399
			dst_ifdown(dst, dev, event != NETDEV_DOWN);
400
		mutex_unlock(&dst_gc_mutex);
Linus Torvalds's avatar
Linus Torvalds committed
401 402 403 404 405 406 407
		break;
	}
	return NOTIFY_DONE;
}

static struct notifier_block dst_dev_notifier = {
	.notifier_call	= dst_dev_event,
408
	.priority = -10, /* must be called after other network notifiers */
Linus Torvalds's avatar
Linus Torvalds committed
409 410 411 412 413 414
};

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