2 * net/core/dst.c Protocol independent destination cache.
4 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
8 #include <linux/bitops.h>
9 #include <linux/errno.h>
10 #include <linux/init.h>
11 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/netdevice.h>
15 #include <linux/skbuff.h>
16 #include <linux/string.h>
17 #include <linux/types.h>
22 * 1) Garbage collection state of dead destination cache
23 * entries is protected by dst_lock.
24 * 2) GC is run only from BH context, and is the only remover
26 * 3) Entries are added to the garbage list from both BH
27 * and non-BH context, so local BH disabling is needed.
28 * 4) All operations modify state, so a spinlock is used.
30 static struct dst_entry *dst_garbage_list;
31 #if RT_CACHE_DEBUG >= 2
32 static atomic_t dst_total = ATOMIC_INIT(0);
34 static DEFINE_SPINLOCK(dst_lock);
36 static unsigned long dst_gc_timer_expires;
37 static unsigned long dst_gc_timer_inc = DST_GC_MAX;
38 static void dst_run_gc(unsigned long);
39 static void ___dst_free(struct dst_entry * dst);
41 static DEFINE_TIMER(dst_gc_timer, dst_run_gc, DST_GC_MIN, 0);
43 static void dst_run_gc(unsigned long dummy)
47 struct dst_entry * dst, **dstp;
49 if (!spin_trylock(&dst_lock)) {
50 mod_timer(&dst_gc_timer, jiffies + HZ/10);
54 del_timer(&dst_gc_timer);
55 dstp = &dst_garbage_list;
57 while ((dst = *dstp) != NULL) {
58 if (atomic_read(&dst->__refcnt)) {
66 dst = dst_destroy(dst);
68 /* NOHASH and still referenced. Unless it is already
69 * on gc list, invalidate it and add to gc list.
71 * Note: this is temporary. Actually, NOHASH dst's
72 * must be obsoleted when parent is obsoleted.
73 * But we do not have state "obsoleted, but
74 * referenced by parent", so it is right.
76 if (dst->obsolete > 1)
85 if (!dst_garbage_list) {
86 dst_gc_timer_inc = DST_GC_MAX;
89 if (!work_performed) {
90 if ((dst_gc_timer_expires += dst_gc_timer_inc) > DST_GC_MAX)
91 dst_gc_timer_expires = DST_GC_MAX;
92 dst_gc_timer_inc += DST_GC_INC;
94 dst_gc_timer_inc = DST_GC_INC;
95 dst_gc_timer_expires = DST_GC_MIN;
97 #if RT_CACHE_DEBUG >= 2
98 printk("dst_total: %d/%d %ld\n",
99 atomic_read(&dst_total), delayed, dst_gc_timer_expires);
101 /* if the next desired timer is more than 4 seconds in the future
102 * then round the timer to whole seconds
104 if (dst_gc_timer_expires > 4*HZ)
105 mod_timer(&dst_gc_timer,
106 round_jiffies(jiffies + dst_gc_timer_expires));
108 mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);
111 spin_unlock(&dst_lock);
114 static int dst_discard_in(struct sk_buff *skb)
120 static int dst_discard_out(struct sk_buff *skb)
126 void * dst_alloc(struct dst_ops * ops)
128 struct dst_entry * dst;
130 if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) {
134 dst = kmem_cache_zalloc(ops->kmem_cachep, GFP_ATOMIC);
137 atomic_set(&dst->__refcnt, 0);
139 dst->lastuse = jiffies;
141 dst->input = dst_discard_in;
142 dst->output = dst_discard_out;
143 #if RT_CACHE_DEBUG >= 2
144 atomic_inc(&dst_total);
146 atomic_inc(&ops->entries);
150 static void ___dst_free(struct dst_entry * dst)
152 /* The first case (dev==NULL) is required, when
153 protocol module is unloaded.
155 if (dst->dev == NULL || !(dst->dev->flags&IFF_UP)) {
156 dst->input = dst_discard_in;
157 dst->output = dst_discard_out;
162 void __dst_free(struct dst_entry * dst)
164 spin_lock_bh(&dst_lock);
166 dst->next = dst_garbage_list;
167 dst_garbage_list = dst;
168 if (dst_gc_timer_inc > DST_GC_INC) {
169 dst_gc_timer_inc = DST_GC_INC;
170 dst_gc_timer_expires = DST_GC_MIN;
171 mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);
173 spin_unlock_bh(&dst_lock);
176 struct dst_entry *dst_destroy(struct dst_entry * dst)
178 struct dst_entry *child;
179 struct neighbour *neigh;
185 neigh = dst->neighbour;
190 if (hh && atomic_dec_and_test(&hh->hh_refcnt))
194 dst->neighbour = NULL;
195 neigh_release(neigh);
198 atomic_dec(&dst->ops->entries);
200 if (dst->ops->destroy)
201 dst->ops->destroy(dst);
204 #if RT_CACHE_DEBUG >= 2
205 atomic_dec(&dst_total);
207 kmem_cache_free(dst->ops->kmem_cachep, dst);
211 int nohash = dst->flags & DST_NOHASH;
213 if (atomic_dec_and_test(&dst->__refcnt)) {
214 /* We were real parent of this dst, so kill child. */
218 /* Child is still referenced, return it for freeing. */
221 /* Child is still in his hash table */
227 /* Dirty hack. We did it in 2.2 (in __dst_free),
228 * we have _very_ good reasons not to repeat
229 * this mistake in 2.3, but we have no choice
230 * now. _It_ _is_ _explicit_ _deliberate_
231 * _race_ _condition_.
233 * Commented and originally written by Alexey.
235 static inline void dst_ifdown(struct dst_entry *dst, struct net_device *dev,
238 if (dst->ops->ifdown)
239 dst->ops->ifdown(dst, dev, unregister);
245 dst->input = dst_discard_in;
246 dst->output = dst_discard_out;
248 dst->dev = &loopback_dev;
249 dev_hold(&loopback_dev);
251 if (dst->neighbour && dst->neighbour->dev == dev) {
252 dst->neighbour->dev = &loopback_dev;
254 dev_hold(&loopback_dev);
259 static int dst_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
261 struct net_device *dev = ptr;
262 struct dst_entry *dst;
265 case NETDEV_UNREGISTER:
267 spin_lock_bh(&dst_lock);
268 for (dst = dst_garbage_list; dst; dst = dst->next) {
269 dst_ifdown(dst, dev, event != NETDEV_DOWN);
271 spin_unlock_bh(&dst_lock);
277 static struct notifier_block dst_dev_notifier = {
278 .notifier_call = dst_dev_event,
281 void __init dst_init(void)
283 register_netdevice_notifier(&dst_dev_notifier);
286 EXPORT_SYMBOL(__dst_free);
287 EXPORT_SYMBOL(dst_alloc);
288 EXPORT_SYMBOL(dst_destroy);