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/sched.h>
16 #include <linux/skbuff.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
22 const char dst_underflow_bug_msg[] = KERN_DEBUG "BUG: dst underflow %d: %p at %p\n";
25 * 1) Garbage collection state of dead destination cache
26 * entries is protected by dst_lock.
27 * 2) GC is run only from BH context, and is the only remover
29 * 3) Entries are added to the garbage list from both BH
30 * and non-BH context, so local BH disabling is needed.
31 * 4) All operations modify state, so a spinlock is used.
33 static struct dst_entry *dst_garbage_list;
34 #if RT_CACHE_DEBUG >= 2
35 static atomic_t dst_total = ATOMIC_INIT(0);
37 static spinlock_t dst_lock = SPIN_LOCK_UNLOCKED;
39 static unsigned long dst_gc_timer_expires;
40 static unsigned long dst_gc_timer_inc = DST_GC_MAX;
41 static void dst_run_gc(unsigned long);
42 static void ___dst_free(struct dst_entry * dst);
44 static struct timer_list dst_gc_timer =
45 TIMER_INITIALIZER(dst_run_gc, DST_GC_MIN, 0);
47 static void dst_run_gc(unsigned long dummy)
50 struct dst_entry * dst, **dstp;
52 if (!spin_trylock(&dst_lock)) {
53 mod_timer(&dst_gc_timer, jiffies + HZ/10);
58 del_timer(&dst_gc_timer);
59 dstp = &dst_garbage_list;
60 while ((dst = *dstp) != NULL) {
61 if (atomic_read(&dst->__refcnt)) {
68 dst = dst_destroy(dst);
70 /* NOHASH and still referenced. Unless it is already
71 * on gc list, invalidate it and add to gc list.
73 * Note: this is temporary. Actually, NOHASH dst's
74 * must be obsoleted when parent is obsoleted.
75 * But we do not have state "obsoleted, but
76 * referenced by parent", so it is right.
78 if (dst->obsolete > 1)
87 if (!dst_garbage_list) {
88 dst_gc_timer_inc = DST_GC_MAX;
91 if ((dst_gc_timer_expires += dst_gc_timer_inc) > DST_GC_MAX)
92 dst_gc_timer_expires = DST_GC_MAX;
93 dst_gc_timer_inc += DST_GC_INC;
94 dst_gc_timer.expires = jiffies + dst_gc_timer_expires;
95 #if RT_CACHE_DEBUG >= 2
96 printk("dst_total: %d/%d %ld\n",
97 atomic_read(&dst_total), delayed, dst_gc_timer_expires);
99 add_timer(&dst_gc_timer);
102 spin_unlock(&dst_lock);
105 static int dst_discard_in(struct sk_buff *skb)
111 static int dst_discard_out(struct sk_buff **pskb)
117 void * dst_alloc(struct dst_ops * ops)
119 struct dst_entry * dst;
121 if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) {
125 dst = kmem_cache_alloc(ops->kmem_cachep, SLAB_ATOMIC);
128 memset(dst, 0, ops->entry_size);
129 atomic_set(&dst->__refcnt, 0);
131 dst->lastuse = jiffies;
133 dst->input = dst_discard_in;
134 dst->output = dst_discard_out;
135 #if RT_CACHE_DEBUG >= 2
136 atomic_inc(&dst_total);
138 atomic_inc(&ops->entries);
142 static void ___dst_free(struct dst_entry * dst)
144 /* The first case (dev==NULL) is required, when
145 protocol module is unloaded.
147 if (dst->dev == NULL || !(dst->dev->flags&IFF_UP)) {
148 dst->input = dst_discard_in;
149 dst->output = dst_discard_out;
154 void __dst_free(struct dst_entry * dst)
156 spin_lock_bh(&dst_lock);
158 dst->next = dst_garbage_list;
159 dst_garbage_list = dst;
160 if (dst_gc_timer_inc > DST_GC_INC) {
161 dst_gc_timer_inc = DST_GC_INC;
162 dst_gc_timer_expires = DST_GC_MIN;
163 mod_timer(&dst_gc_timer, jiffies + dst_gc_timer_expires);
165 spin_unlock_bh(&dst_lock);
168 struct dst_entry *dst_destroy(struct dst_entry * dst)
170 struct dst_entry *child;
171 struct neighbour *neigh;
175 neigh = dst->neighbour;
180 if (hh && atomic_dec_and_test(&hh->hh_refcnt))
184 dst->neighbour = NULL;
185 neigh_release(neigh);
188 atomic_dec(&dst->ops->entries);
190 if (dst->ops->destroy)
191 dst->ops->destroy(dst);
194 #if RT_CACHE_DEBUG >= 2
195 atomic_dec(&dst_total);
197 kmem_cache_free(dst->ops->kmem_cachep, dst);
201 if (atomic_dec_and_test(&dst->__refcnt)) {
202 /* We were real parent of this dst, so kill child. */
203 if (dst->flags&DST_NOHASH)
206 /* Child is still referenced, return it for freeing. */
207 if (dst->flags&DST_NOHASH)
209 /* Child is still in his hash table */
215 /* Dirty hack. We did it in 2.2 (in __dst_free),
216 * we have _very_ good reasons not to repeat
217 * this mistake in 2.3, but we have no choice
218 * now. _It_ _is_ _explicit_ _deliberate_
219 * _race_ _condition_.
221 * Commented and originally written by Alexey.
223 static void dst_ifdown(struct dst_entry *dst, int unregister)
225 struct net_device *dev = dst->dev;
228 dst->input = dst_discard_in;
229 dst->output = dst_discard_out;
234 dst->dev = &loopback_dev;
235 dev_hold(&loopback_dev);
237 if (dst->neighbour && dst->neighbour->dev == dev) {
238 dst->neighbour->dev = &loopback_dev;
240 dev_hold(&loopback_dev);
244 if (dst->ops->ifdown)
245 dst->ops->ifdown(dst, unregister);
246 } while ((dst = dst->child) && dst->flags & DST_NOHASH &&
250 static int dst_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
252 struct net_device *dev = ptr;
253 struct dst_entry *dst;
256 case NETDEV_UNREGISTER:
258 spin_lock_bh(&dst_lock);
259 for (dst = dst_garbage_list; dst; dst = dst->next) {
261 dst_ifdown(dst, event != NETDEV_DOWN);
263 spin_unlock_bh(&dst_lock);
269 struct notifier_block dst_dev_notifier = {
270 .notifier_call = dst_dev_event,
273 void __init dst_init(void)
275 register_netdevice_notifier(&dst_dev_notifier);
278 EXPORT_SYMBOL(dst_underflow_bug_msg);
279 EXPORT_SYMBOL(__dst_free);
280 EXPORT_SYMBOL(dst_alloc);
281 EXPORT_SYMBOL(dst_destroy);