2 * Generic address resolution entity
5 * Pedro Roque <roque@di.fc.ul.pt>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
15 * Harald Welte Add neighbour cache statistics like rtstat
18 #include <linux/slab.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/socket.h>
23 #include <linux/netdevice.h>
24 #include <linux/proc_fs.h>
26 #include <linux/sysctl.h>
28 #include <linux/times.h>
29 #include <net/net_namespace.h>
30 #include <net/neighbour.h>
33 #include <net/netevent.h>
34 #include <net/netlink.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/random.h>
37 #include <linux/string.h>
38 #include <linux/log2.h>
42 #define NEIGH_PRINTK(x...) printk(x)
43 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
44 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
45 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
49 #define NEIGH_PRINTK1 NEIGH_PRINTK
53 #define NEIGH_PRINTK2 NEIGH_PRINTK
56 #define PNEIGH_HASHMASK 0xF
58 static void neigh_timer_handler(unsigned long arg);
59 static void __neigh_notify(struct neighbour *n, int type, int flags);
60 static void neigh_update_notify(struct neighbour *neigh);
61 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
63 static struct neigh_table *neigh_tables;
65 static const struct file_operations neigh_stat_seq_fops;
69 Neighbour hash table buckets are protected with rwlock tbl->lock.
71 - All the scans/updates to hash buckets MUST be made under this lock.
72 - NOTHING clever should be made under this lock: no callbacks
73 to protocol backends, no attempts to send something to network.
74 It will result in deadlocks, if backend/driver wants to use neighbour
76 - If the entry requires some non-trivial actions, increase
77 its reference count and release table lock.
79 Neighbour entries are protected:
80 - with reference count.
81 - with rwlock neigh->lock
83 Reference count prevents destruction.
85 neigh->lock mainly serializes ll address data and its validity state.
86 However, the same lock is used to protect another entry fields:
90 Again, nothing clever shall be made under neigh->lock,
91 the most complicated procedure, which we allow is dev->hard_header.
92 It is supposed, that dev->hard_header is simplistic and does
93 not make callbacks to neighbour tables.
95 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
96 list of neighbour tables. This list is used only in process context,
99 static DEFINE_RWLOCK(neigh_tbl_lock);
101 static int neigh_blackhole(struct sk_buff *skb)
107 static void neigh_cleanup_and_release(struct neighbour *neigh)
109 if (neigh->parms->neigh_cleanup)
110 neigh->parms->neigh_cleanup(neigh);
112 __neigh_notify(neigh, RTM_DELNEIGH, 0);
113 neigh_release(neigh);
117 * It is random distribution in the interval (1/2)*base...(3/2)*base.
118 * It corresponds to default IPv6 settings and is not overridable,
119 * because it is really reasonable choice.
122 unsigned long neigh_rand_reach_time(unsigned long base)
124 return base ? (net_random() % base) + (base >> 1) : 0;
126 EXPORT_SYMBOL(neigh_rand_reach_time);
129 static int neigh_forced_gc(struct neigh_table *tbl)
133 struct neigh_hash_table *nht;
135 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
137 write_lock_bh(&tbl->lock);
138 nht = rcu_dereference_protected(tbl->nht,
139 lockdep_is_held(&tbl->lock));
140 for (i = 0; i <= nht->hash_mask; i++) {
142 struct neighbour __rcu **np;
144 np = &nht->hash_buckets[i];
145 while ((n = rcu_dereference_protected(*np,
146 lockdep_is_held(&tbl->lock))) != NULL) {
147 /* Neighbour record may be discarded if:
148 * - nobody refers to it.
149 * - it is not permanent
151 write_lock(&n->lock);
152 if (atomic_read(&n->refcnt) == 1 &&
153 !(n->nud_state & NUD_PERMANENT)) {
154 rcu_assign_pointer(*np,
155 rcu_dereference_protected(n->next,
156 lockdep_is_held(&tbl->lock)));
159 write_unlock(&n->lock);
160 neigh_cleanup_and_release(n);
163 write_unlock(&n->lock);
168 tbl->last_flush = jiffies;
170 write_unlock_bh(&tbl->lock);
175 static void neigh_add_timer(struct neighbour *n, unsigned long when)
178 if (unlikely(mod_timer(&n->timer, when))) {
179 printk("NEIGH: BUG, double timer add, state is %x\n",
185 static int neigh_del_timer(struct neighbour *n)
187 if ((n->nud_state & NUD_IN_TIMER) &&
188 del_timer(&n->timer)) {
195 static void pneigh_queue_purge(struct sk_buff_head *list)
199 while ((skb = skb_dequeue(list)) != NULL) {
205 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
208 struct neigh_hash_table *nht;
210 nht = rcu_dereference_protected(tbl->nht,
211 lockdep_is_held(&tbl->lock));
213 for (i = 0; i <= nht->hash_mask; i++) {
215 struct neighbour __rcu **np = &nht->hash_buckets[i];
217 while ((n = rcu_dereference_protected(*np,
218 lockdep_is_held(&tbl->lock))) != NULL) {
219 if (dev && n->dev != dev) {
223 rcu_assign_pointer(*np,
224 rcu_dereference_protected(n->next,
225 lockdep_is_held(&tbl->lock)));
226 write_lock(&n->lock);
230 if (atomic_read(&n->refcnt) != 1) {
231 /* The most unpleasant situation.
232 We must destroy neighbour entry,
233 but someone still uses it.
235 The destroy will be delayed until
236 the last user releases us, but
237 we must kill timers etc. and move
240 skb_queue_purge(&n->arp_queue);
241 n->output = neigh_blackhole;
242 if (n->nud_state & NUD_VALID)
243 n->nud_state = NUD_NOARP;
245 n->nud_state = NUD_NONE;
246 NEIGH_PRINTK2("neigh %p is stray.\n", n);
248 write_unlock(&n->lock);
249 neigh_cleanup_and_release(n);
254 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
256 write_lock_bh(&tbl->lock);
257 neigh_flush_dev(tbl, dev);
258 write_unlock_bh(&tbl->lock);
260 EXPORT_SYMBOL(neigh_changeaddr);
262 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
264 write_lock_bh(&tbl->lock);
265 neigh_flush_dev(tbl, dev);
266 pneigh_ifdown(tbl, dev);
267 write_unlock_bh(&tbl->lock);
269 del_timer_sync(&tbl->proxy_timer);
270 pneigh_queue_purge(&tbl->proxy_queue);
273 EXPORT_SYMBOL(neigh_ifdown);
275 static struct neighbour *neigh_alloc(struct neigh_table *tbl)
277 struct neighbour *n = NULL;
278 unsigned long now = jiffies;
281 entries = atomic_inc_return(&tbl->entries) - 1;
282 if (entries >= tbl->gc_thresh3 ||
283 (entries >= tbl->gc_thresh2 &&
284 time_after(now, tbl->last_flush + 5 * HZ))) {
285 if (!neigh_forced_gc(tbl) &&
286 entries >= tbl->gc_thresh3)
290 n = kmem_cache_zalloc(tbl->kmem_cachep, GFP_ATOMIC);
294 skb_queue_head_init(&n->arp_queue);
295 rwlock_init(&n->lock);
296 seqlock_init(&n->ha_lock);
297 n->updated = n->used = now;
298 n->nud_state = NUD_NONE;
299 n->output = neigh_blackhole;
300 n->parms = neigh_parms_clone(&tbl->parms);
301 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
303 NEIGH_CACHE_STAT_INC(tbl, allocs);
305 atomic_set(&n->refcnt, 1);
311 atomic_dec(&tbl->entries);
315 static struct neigh_hash_table *neigh_hash_alloc(unsigned int entries)
317 size_t size = entries * sizeof(struct neighbour *);
318 struct neigh_hash_table *ret;
319 struct neighbour __rcu **buckets;
321 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
324 if (size <= PAGE_SIZE)
325 buckets = kzalloc(size, GFP_ATOMIC);
327 buckets = (struct neighbour __rcu **)
328 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
334 ret->hash_buckets = buckets;
335 ret->hash_mask = entries - 1;
336 get_random_bytes(&ret->hash_rnd, sizeof(ret->hash_rnd));
340 static void neigh_hash_free_rcu(struct rcu_head *head)
342 struct neigh_hash_table *nht = container_of(head,
343 struct neigh_hash_table,
345 size_t size = (nht->hash_mask + 1) * sizeof(struct neighbour *);
346 struct neighbour __rcu **buckets = nht->hash_buckets;
348 if (size <= PAGE_SIZE)
351 free_pages((unsigned long)buckets, get_order(size));
355 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
356 unsigned long new_entries)
358 unsigned int i, hash;
359 struct neigh_hash_table *new_nht, *old_nht;
361 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
363 BUG_ON(!is_power_of_2(new_entries));
364 old_nht = rcu_dereference_protected(tbl->nht,
365 lockdep_is_held(&tbl->lock));
366 new_nht = neigh_hash_alloc(new_entries);
370 for (i = 0; i <= old_nht->hash_mask; i++) {
371 struct neighbour *n, *next;
373 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
374 lockdep_is_held(&tbl->lock));
377 hash = tbl->hash(n->primary_key, n->dev,
380 hash &= new_nht->hash_mask;
381 next = rcu_dereference_protected(n->next,
382 lockdep_is_held(&tbl->lock));
384 rcu_assign_pointer(n->next,
385 rcu_dereference_protected(
386 new_nht->hash_buckets[hash],
387 lockdep_is_held(&tbl->lock)));
388 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
392 rcu_assign_pointer(tbl->nht, new_nht);
393 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
397 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
398 struct net_device *dev)
401 int key_len = tbl->key_len;
403 struct neigh_hash_table *nht;
405 NEIGH_CACHE_STAT_INC(tbl, lookups);
408 nht = rcu_dereference_bh(tbl->nht);
409 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) & nht->hash_mask;
411 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
413 n = rcu_dereference_bh(n->next)) {
414 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
415 if (!atomic_inc_not_zero(&n->refcnt))
417 NEIGH_CACHE_STAT_INC(tbl, hits);
422 rcu_read_unlock_bh();
425 EXPORT_SYMBOL(neigh_lookup);
427 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
431 int key_len = tbl->key_len;
433 struct neigh_hash_table *nht;
435 NEIGH_CACHE_STAT_INC(tbl, lookups);
438 nht = rcu_dereference_bh(tbl->nht);
439 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) & nht->hash_mask;
441 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
443 n = rcu_dereference_bh(n->next)) {
444 if (!memcmp(n->primary_key, pkey, key_len) &&
445 net_eq(dev_net(n->dev), net)) {
446 if (!atomic_inc_not_zero(&n->refcnt))
448 NEIGH_CACHE_STAT_INC(tbl, hits);
453 rcu_read_unlock_bh();
456 EXPORT_SYMBOL(neigh_lookup_nodev);
458 struct neighbour *neigh_create(struct neigh_table *tbl, const void *pkey,
459 struct net_device *dev)
462 int key_len = tbl->key_len;
464 struct neighbour *n1, *rc, *n = neigh_alloc(tbl);
465 struct neigh_hash_table *nht;
468 rc = ERR_PTR(-ENOBUFS);
472 memcpy(n->primary_key, pkey, key_len);
476 /* Protocol specific setup. */
477 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
479 goto out_neigh_release;
482 /* Device specific setup. */
483 if (n->parms->neigh_setup &&
484 (error = n->parms->neigh_setup(n)) < 0) {
486 goto out_neigh_release;
489 n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
491 write_lock_bh(&tbl->lock);
492 nht = rcu_dereference_protected(tbl->nht,
493 lockdep_is_held(&tbl->lock));
495 if (atomic_read(&tbl->entries) > (nht->hash_mask + 1))
496 nht = neigh_hash_grow(tbl, (nht->hash_mask + 1) << 1);
498 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) & nht->hash_mask;
500 if (n->parms->dead) {
501 rc = ERR_PTR(-EINVAL);
505 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
506 lockdep_is_held(&tbl->lock));
508 n1 = rcu_dereference_protected(n1->next,
509 lockdep_is_held(&tbl->lock))) {
510 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
519 rcu_assign_pointer(n->next,
520 rcu_dereference_protected(nht->hash_buckets[hash_val],
521 lockdep_is_held(&tbl->lock)));
522 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
523 write_unlock_bh(&tbl->lock);
524 NEIGH_PRINTK2("neigh %p is created.\n", n);
529 write_unlock_bh(&tbl->lock);
534 EXPORT_SYMBOL(neigh_create);
536 static u32 pneigh_hash(const void *pkey, int key_len)
538 u32 hash_val = *(u32 *)(pkey + key_len - 4);
539 hash_val ^= (hash_val >> 16);
540 hash_val ^= hash_val >> 8;
541 hash_val ^= hash_val >> 4;
542 hash_val &= PNEIGH_HASHMASK;
546 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
550 struct net_device *dev)
553 if (!memcmp(n->key, pkey, key_len) &&
554 net_eq(pneigh_net(n), net) &&
555 (n->dev == dev || !n->dev))
562 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
563 struct net *net, const void *pkey, struct net_device *dev)
565 int key_len = tbl->key_len;
566 u32 hash_val = pneigh_hash(pkey, key_len);
568 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
569 net, pkey, key_len, dev);
571 EXPORT_SYMBOL_GPL(__pneigh_lookup);
573 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
574 struct net *net, const void *pkey,
575 struct net_device *dev, int creat)
577 struct pneigh_entry *n;
578 int key_len = tbl->key_len;
579 u32 hash_val = pneigh_hash(pkey, key_len);
581 read_lock_bh(&tbl->lock);
582 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
583 net, pkey, key_len, dev);
584 read_unlock_bh(&tbl->lock);
591 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
595 write_pnet(&n->net, hold_net(net));
596 memcpy(n->key, pkey, key_len);
601 if (tbl->pconstructor && tbl->pconstructor(n)) {
610 write_lock_bh(&tbl->lock);
611 n->next = tbl->phash_buckets[hash_val];
612 tbl->phash_buckets[hash_val] = n;
613 write_unlock_bh(&tbl->lock);
617 EXPORT_SYMBOL(pneigh_lookup);
620 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
621 struct net_device *dev)
623 struct pneigh_entry *n, **np;
624 int key_len = tbl->key_len;
625 u32 hash_val = pneigh_hash(pkey, key_len);
627 write_lock_bh(&tbl->lock);
628 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
630 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
631 net_eq(pneigh_net(n), net)) {
633 write_unlock_bh(&tbl->lock);
634 if (tbl->pdestructor)
638 release_net(pneigh_net(n));
643 write_unlock_bh(&tbl->lock);
647 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
649 struct pneigh_entry *n, **np;
652 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
653 np = &tbl->phash_buckets[h];
654 while ((n = *np) != NULL) {
655 if (!dev || n->dev == dev) {
657 if (tbl->pdestructor)
661 release_net(pneigh_net(n));
671 static void neigh_parms_destroy(struct neigh_parms *parms);
673 static inline void neigh_parms_put(struct neigh_parms *parms)
675 if (atomic_dec_and_test(&parms->refcnt))
676 neigh_parms_destroy(parms);
679 static void neigh_destroy_rcu(struct rcu_head *head)
681 struct neighbour *neigh = container_of(head, struct neighbour, rcu);
683 kmem_cache_free(neigh->tbl->kmem_cachep, neigh);
686 * neighbour must already be out of the table;
689 void neigh_destroy(struct neighbour *neigh)
693 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
697 "Destroying alive neighbour %p\n", neigh);
702 if (neigh_del_timer(neigh))
703 printk(KERN_WARNING "Impossible event.\n");
705 while ((hh = neigh->hh) != NULL) {
706 neigh->hh = hh->hh_next;
709 write_seqlock_bh(&hh->hh_lock);
710 hh->hh_output = neigh_blackhole;
711 write_sequnlock_bh(&hh->hh_lock);
715 skb_queue_purge(&neigh->arp_queue);
718 neigh_parms_put(neigh->parms);
720 NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
722 atomic_dec(&neigh->tbl->entries);
723 call_rcu(&neigh->rcu, neigh_destroy_rcu);
725 EXPORT_SYMBOL(neigh_destroy);
727 /* Neighbour state is suspicious;
730 Called with write_locked neigh.
732 static void neigh_suspect(struct neighbour *neigh)
736 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
738 neigh->output = neigh->ops->output;
740 for (hh = neigh->hh; hh; hh = hh->hh_next)
741 hh->hh_output = neigh->ops->output;
744 /* Neighbour state is OK;
747 Called with write_locked neigh.
749 static void neigh_connect(struct neighbour *neigh)
753 NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
755 neigh->output = neigh->ops->connected_output;
757 for (hh = neigh->hh; hh; hh = hh->hh_next)
758 hh->hh_output = neigh->ops->hh_output;
761 static void neigh_periodic_work(struct work_struct *work)
763 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
765 struct neighbour __rcu **np;
767 struct neigh_hash_table *nht;
769 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
771 write_lock_bh(&tbl->lock);
772 nht = rcu_dereference_protected(tbl->nht,
773 lockdep_is_held(&tbl->lock));
776 * periodically recompute ReachableTime from random function
779 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
780 struct neigh_parms *p;
781 tbl->last_rand = jiffies;
782 for (p = &tbl->parms; p; p = p->next)
784 neigh_rand_reach_time(p->base_reachable_time);
787 for (i = 0 ; i <= nht->hash_mask; i++) {
788 np = &nht->hash_buckets[i];
790 while ((n = rcu_dereference_protected(*np,
791 lockdep_is_held(&tbl->lock))) != NULL) {
794 write_lock(&n->lock);
796 state = n->nud_state;
797 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
798 write_unlock(&n->lock);
802 if (time_before(n->used, n->confirmed))
803 n->used = n->confirmed;
805 if (atomic_read(&n->refcnt) == 1 &&
806 (state == NUD_FAILED ||
807 time_after(jiffies, n->used + n->parms->gc_staletime))) {
810 write_unlock(&n->lock);
811 neigh_cleanup_and_release(n);
814 write_unlock(&n->lock);
820 * It's fine to release lock here, even if hash table
821 * grows while we are preempted.
823 write_unlock_bh(&tbl->lock);
825 write_lock_bh(&tbl->lock);
826 nht = rcu_dereference_protected(tbl->nht,
827 lockdep_is_held(&tbl->lock));
829 /* Cycle through all hash buckets every base_reachable_time/2 ticks.
830 * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
831 * base_reachable_time.
833 schedule_delayed_work(&tbl->gc_work,
834 tbl->parms.base_reachable_time >> 1);
835 write_unlock_bh(&tbl->lock);
838 static __inline__ int neigh_max_probes(struct neighbour *n)
840 struct neigh_parms *p = n->parms;
841 return (n->nud_state & NUD_PROBE) ?
843 p->ucast_probes + p->app_probes + p->mcast_probes;
846 static void neigh_invalidate(struct neighbour *neigh)
847 __releases(neigh->lock)
848 __acquires(neigh->lock)
852 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
853 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
854 neigh->updated = jiffies;
856 /* It is very thin place. report_unreachable is very complicated
857 routine. Particularly, it can hit the same neighbour entry!
859 So that, we try to be accurate and avoid dead loop. --ANK
861 while (neigh->nud_state == NUD_FAILED &&
862 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
863 write_unlock(&neigh->lock);
864 neigh->ops->error_report(neigh, skb);
865 write_lock(&neigh->lock);
867 skb_queue_purge(&neigh->arp_queue);
870 /* Called when a timer expires for a neighbour entry. */
872 static void neigh_timer_handler(unsigned long arg)
874 unsigned long now, next;
875 struct neighbour *neigh = (struct neighbour *)arg;
879 write_lock(&neigh->lock);
881 state = neigh->nud_state;
885 if (!(state & NUD_IN_TIMER)) {
887 printk(KERN_WARNING "neigh: timer & !nud_in_timer\n");
892 if (state & NUD_REACHABLE) {
893 if (time_before_eq(now,
894 neigh->confirmed + neigh->parms->reachable_time)) {
895 NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
896 next = neigh->confirmed + neigh->parms->reachable_time;
897 } else if (time_before_eq(now,
898 neigh->used + neigh->parms->delay_probe_time)) {
899 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
900 neigh->nud_state = NUD_DELAY;
901 neigh->updated = jiffies;
902 neigh_suspect(neigh);
903 next = now + neigh->parms->delay_probe_time;
905 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
906 neigh->nud_state = NUD_STALE;
907 neigh->updated = jiffies;
908 neigh_suspect(neigh);
911 } else if (state & NUD_DELAY) {
912 if (time_before_eq(now,
913 neigh->confirmed + neigh->parms->delay_probe_time)) {
914 NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
915 neigh->nud_state = NUD_REACHABLE;
916 neigh->updated = jiffies;
917 neigh_connect(neigh);
919 next = neigh->confirmed + neigh->parms->reachable_time;
921 NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
922 neigh->nud_state = NUD_PROBE;
923 neigh->updated = jiffies;
924 atomic_set(&neigh->probes, 0);
925 next = now + neigh->parms->retrans_time;
928 /* NUD_PROBE|NUD_INCOMPLETE */
929 next = now + neigh->parms->retrans_time;
932 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
933 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
934 neigh->nud_state = NUD_FAILED;
936 neigh_invalidate(neigh);
939 if (neigh->nud_state & NUD_IN_TIMER) {
940 if (time_before(next, jiffies + HZ/2))
941 next = jiffies + HZ/2;
942 if (!mod_timer(&neigh->timer, next))
945 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
946 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
947 /* keep skb alive even if arp_queue overflows */
949 skb = skb_copy(skb, GFP_ATOMIC);
950 write_unlock(&neigh->lock);
951 neigh->ops->solicit(neigh, skb);
952 atomic_inc(&neigh->probes);
956 write_unlock(&neigh->lock);
960 neigh_update_notify(neigh);
962 neigh_release(neigh);
965 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
970 write_lock_bh(&neigh->lock);
973 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
978 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
979 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
980 atomic_set(&neigh->probes, neigh->parms->ucast_probes);
981 neigh->nud_state = NUD_INCOMPLETE;
982 neigh->updated = jiffies;
983 neigh_add_timer(neigh, now + 1);
985 neigh->nud_state = NUD_FAILED;
986 neigh->updated = jiffies;
987 write_unlock_bh(&neigh->lock);
992 } else if (neigh->nud_state & NUD_STALE) {
993 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
994 neigh->nud_state = NUD_DELAY;
995 neigh->updated = jiffies;
996 neigh_add_timer(neigh,
997 jiffies + neigh->parms->delay_probe_time);
1000 if (neigh->nud_state == NUD_INCOMPLETE) {
1002 if (skb_queue_len(&neigh->arp_queue) >=
1003 neigh->parms->queue_len) {
1004 struct sk_buff *buff;
1005 buff = __skb_dequeue(&neigh->arp_queue);
1007 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1010 __skb_queue_tail(&neigh->arp_queue, skb);
1015 write_unlock_bh(&neigh->lock);
1018 EXPORT_SYMBOL(__neigh_event_send);
1020 static void neigh_update_hhs(const struct neighbour *neigh)
1022 struct hh_cache *hh;
1023 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1026 if (neigh->dev->header_ops)
1027 update = neigh->dev->header_ops->cache_update;
1030 for (hh = neigh->hh; hh; hh = hh->hh_next) {
1031 write_seqlock_bh(&hh->hh_lock);
1032 update(hh, neigh->dev, neigh->ha);
1033 write_sequnlock_bh(&hh->hh_lock);
1040 /* Generic update routine.
1041 -- lladdr is new lladdr or NULL, if it is not supplied.
1042 -- new is new state.
1044 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1046 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1047 lladdr instead of overriding it
1049 It also allows to retain current state
1050 if lladdr is unchanged.
1051 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1053 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1055 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1058 Caller MUST hold reference count on the entry.
1061 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1067 struct net_device *dev;
1068 int update_isrouter = 0;
1070 write_lock_bh(&neigh->lock);
1073 old = neigh->nud_state;
1076 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1077 (old & (NUD_NOARP | NUD_PERMANENT)))
1080 if (!(new & NUD_VALID)) {
1081 neigh_del_timer(neigh);
1082 if (old & NUD_CONNECTED)
1083 neigh_suspect(neigh);
1084 neigh->nud_state = new;
1086 notify = old & NUD_VALID;
1087 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1088 (new & NUD_FAILED)) {
1089 neigh_invalidate(neigh);
1095 /* Compare new lladdr with cached one */
1096 if (!dev->addr_len) {
1097 /* First case: device needs no address. */
1099 } else if (lladdr) {
1100 /* The second case: if something is already cached
1101 and a new address is proposed:
1103 - if they are different, check override flag
1105 if ((old & NUD_VALID) &&
1106 !memcmp(lladdr, neigh->ha, dev->addr_len))
1109 /* No address is supplied; if we know something,
1110 use it, otherwise discard the request.
1113 if (!(old & NUD_VALID))
1118 if (new & NUD_CONNECTED)
1119 neigh->confirmed = jiffies;
1120 neigh->updated = jiffies;
1122 /* If entry was valid and address is not changed,
1123 do not change entry state, if new one is STALE.
1126 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1127 if (old & NUD_VALID) {
1128 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1129 update_isrouter = 0;
1130 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1131 (old & NUD_CONNECTED)) {
1137 if (lladdr == neigh->ha && new == NUD_STALE &&
1138 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1139 (old & NUD_CONNECTED))
1146 neigh_del_timer(neigh);
1147 if (new & NUD_IN_TIMER)
1148 neigh_add_timer(neigh, (jiffies +
1149 ((new & NUD_REACHABLE) ?
1150 neigh->parms->reachable_time :
1152 neigh->nud_state = new;
1155 if (lladdr != neigh->ha) {
1156 write_seqlock(&neigh->ha_lock);
1157 memcpy(&neigh->ha, lladdr, dev->addr_len);
1158 write_sequnlock(&neigh->ha_lock);
1159 neigh_update_hhs(neigh);
1160 if (!(new & NUD_CONNECTED))
1161 neigh->confirmed = jiffies -
1162 (neigh->parms->base_reachable_time << 1);
1167 if (new & NUD_CONNECTED)
1168 neigh_connect(neigh);
1170 neigh_suspect(neigh);
1171 if (!(old & NUD_VALID)) {
1172 struct sk_buff *skb;
1174 /* Again: avoid dead loop if something went wrong */
1176 while (neigh->nud_state & NUD_VALID &&
1177 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1178 struct dst_entry *dst = skb_dst(skb);
1179 struct neighbour *n2, *n1 = neigh;
1180 write_unlock_bh(&neigh->lock);
1183 /* On shaper/eql skb->dst->neighbour != neigh :( */
1184 if (dst && (n2 = dst_get_neighbour(dst)) != NULL)
1189 write_lock_bh(&neigh->lock);
1191 skb_queue_purge(&neigh->arp_queue);
1194 if (update_isrouter) {
1195 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1196 (neigh->flags | NTF_ROUTER) :
1197 (neigh->flags & ~NTF_ROUTER);
1199 write_unlock_bh(&neigh->lock);
1202 neigh_update_notify(neigh);
1206 EXPORT_SYMBOL(neigh_update);
1208 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1209 u8 *lladdr, void *saddr,
1210 struct net_device *dev)
1212 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1213 lladdr || !dev->addr_len);
1215 neigh_update(neigh, lladdr, NUD_STALE,
1216 NEIGH_UPDATE_F_OVERRIDE);
1219 EXPORT_SYMBOL(neigh_event_ns);
1221 static inline bool neigh_hh_lookup(struct neighbour *n, struct dst_entry *dst,
1224 struct hh_cache *hh;
1226 smp_rmb(); /* paired with smp_wmb() in neigh_hh_init() */
1227 for (hh = n->hh; hh; hh = hh->hh_next) {
1228 if (hh->hh_type == protocol) {
1229 atomic_inc(&hh->hh_refcnt);
1230 if (unlikely(cmpxchg(&dst->hh, NULL, hh) != NULL))
1238 /* called with read_lock_bh(&n->lock); */
1239 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst,
1242 struct hh_cache *hh;
1243 struct net_device *dev = dst->dev;
1245 if (likely(neigh_hh_lookup(n, dst, protocol)))
1249 hh = kzalloc(sizeof(*hh), GFP_ATOMIC);
1253 seqlock_init(&hh->hh_lock);
1254 hh->hh_type = protocol;
1255 atomic_set(&hh->hh_refcnt, 2);
1257 if (dev->header_ops->cache(n, hh)) {
1262 write_lock_bh(&n->lock);
1264 /* must check if another thread already did the insert */
1265 if (neigh_hh_lookup(n, dst, protocol)) {
1270 if (n->nud_state & NUD_CONNECTED)
1271 hh->hh_output = n->ops->hh_output;
1273 hh->hh_output = n->ops->output;
1275 hh->hh_next = n->hh;
1276 smp_wmb(); /* paired with smp_rmb() in neigh_hh_lookup() */
1279 if (unlikely(cmpxchg(&dst->hh, NULL, hh) != NULL))
1282 write_unlock_bh(&n->lock);
1285 /* This function can be used in contexts, where only old dev_queue_xmit
1286 * worked, f.e. if you want to override normal output path (eql, shaper),
1287 * but resolution is not made yet.
1290 int neigh_compat_output(struct sk_buff *skb)
1292 struct net_device *dev = skb->dev;
1294 __skb_pull(skb, skb_network_offset(skb));
1296 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1298 dev->header_ops->rebuild(skb))
1301 return dev_queue_xmit(skb);
1303 EXPORT_SYMBOL(neigh_compat_output);
1305 /* Slow and careful. */
1307 int neigh_resolve_output(struct sk_buff *skb)
1309 struct dst_entry *dst = skb_dst(skb);
1310 struct neighbour *neigh = dst_get_neighbour(dst);
1316 __skb_pull(skb, skb_network_offset(skb));
1318 if (!neigh_event_send(neigh, skb)) {
1320 struct net_device *dev = neigh->dev;
1323 if (dev->header_ops->cache &&
1325 !(dst->flags & DST_NOCACHE))
1326 neigh_hh_init(neigh, dst, dst->ops->protocol);
1329 seq = read_seqbegin(&neigh->ha_lock);
1330 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1331 neigh->ha, NULL, skb->len);
1332 } while (read_seqretry(&neigh->ha_lock, seq));
1335 rc = neigh->ops->queue_xmit(skb);
1342 NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1349 EXPORT_SYMBOL(neigh_resolve_output);
1351 /* As fast as possible without hh cache */
1353 int neigh_connected_output(struct sk_buff *skb)
1356 struct dst_entry *dst = skb_dst(skb);
1357 struct neighbour *neigh = dst_get_neighbour(dst);
1358 struct net_device *dev = neigh->dev;
1361 __skb_pull(skb, skb_network_offset(skb));
1364 seq = read_seqbegin(&neigh->ha_lock);
1365 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1366 neigh->ha, NULL, skb->len);
1367 } while (read_seqretry(&neigh->ha_lock, seq));
1370 err = neigh->ops->queue_xmit(skb);
1377 EXPORT_SYMBOL(neigh_connected_output);
1379 static void neigh_proxy_process(unsigned long arg)
1381 struct neigh_table *tbl = (struct neigh_table *)arg;
1382 long sched_next = 0;
1383 unsigned long now = jiffies;
1384 struct sk_buff *skb, *n;
1386 spin_lock(&tbl->proxy_queue.lock);
1388 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1389 long tdif = NEIGH_CB(skb)->sched_next - now;
1392 struct net_device *dev = skb->dev;
1394 __skb_unlink(skb, &tbl->proxy_queue);
1395 if (tbl->proxy_redo && netif_running(dev)) {
1397 tbl->proxy_redo(skb);
1404 } else if (!sched_next || tdif < sched_next)
1407 del_timer(&tbl->proxy_timer);
1409 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1410 spin_unlock(&tbl->proxy_queue.lock);
1413 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1414 struct sk_buff *skb)
1416 unsigned long now = jiffies;
1417 unsigned long sched_next = now + (net_random() % p->proxy_delay);
1419 if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1424 NEIGH_CB(skb)->sched_next = sched_next;
1425 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1427 spin_lock(&tbl->proxy_queue.lock);
1428 if (del_timer(&tbl->proxy_timer)) {
1429 if (time_before(tbl->proxy_timer.expires, sched_next))
1430 sched_next = tbl->proxy_timer.expires;
1434 __skb_queue_tail(&tbl->proxy_queue, skb);
1435 mod_timer(&tbl->proxy_timer, sched_next);
1436 spin_unlock(&tbl->proxy_queue.lock);
1438 EXPORT_SYMBOL(pneigh_enqueue);
1440 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1441 struct net *net, int ifindex)
1443 struct neigh_parms *p;
1445 for (p = &tbl->parms; p; p = p->next) {
1446 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1447 (!p->dev && !ifindex))
1454 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1455 struct neigh_table *tbl)
1457 struct neigh_parms *p, *ref;
1458 struct net *net = dev_net(dev);
1459 const struct net_device_ops *ops = dev->netdev_ops;
1461 ref = lookup_neigh_parms(tbl, net, 0);
1465 p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1468 atomic_set(&p->refcnt, 1);
1470 neigh_rand_reach_time(p->base_reachable_time);
1472 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1479 write_pnet(&p->net, hold_net(net));
1480 p->sysctl_table = NULL;
1481 write_lock_bh(&tbl->lock);
1482 p->next = tbl->parms.next;
1483 tbl->parms.next = p;
1484 write_unlock_bh(&tbl->lock);
1488 EXPORT_SYMBOL(neigh_parms_alloc);
1490 static void neigh_rcu_free_parms(struct rcu_head *head)
1492 struct neigh_parms *parms =
1493 container_of(head, struct neigh_parms, rcu_head);
1495 neigh_parms_put(parms);
1498 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1500 struct neigh_parms **p;
1502 if (!parms || parms == &tbl->parms)
1504 write_lock_bh(&tbl->lock);
1505 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1509 write_unlock_bh(&tbl->lock);
1511 dev_put(parms->dev);
1512 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1516 write_unlock_bh(&tbl->lock);
1517 NEIGH_PRINTK1("neigh_parms_release: not found\n");
1519 EXPORT_SYMBOL(neigh_parms_release);
1521 static void neigh_parms_destroy(struct neigh_parms *parms)
1523 release_net(neigh_parms_net(parms));
1527 static struct lock_class_key neigh_table_proxy_queue_class;
1529 void neigh_table_init_no_netlink(struct neigh_table *tbl)
1531 unsigned long now = jiffies;
1532 unsigned long phsize;
1534 write_pnet(&tbl->parms.net, &init_net);
1535 atomic_set(&tbl->parms.refcnt, 1);
1536 tbl->parms.reachable_time =
1537 neigh_rand_reach_time(tbl->parms.base_reachable_time);
1539 if (!tbl->kmem_cachep)
1541 kmem_cache_create(tbl->id, tbl->entry_size, 0,
1542 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1544 tbl->stats = alloc_percpu(struct neigh_statistics);
1546 panic("cannot create neighbour cache statistics");
1548 #ifdef CONFIG_PROC_FS
1549 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1550 &neigh_stat_seq_fops, tbl))
1551 panic("cannot create neighbour proc dir entry");
1554 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(8));
1556 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1557 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1559 if (!tbl->nht || !tbl->phash_buckets)
1560 panic("cannot allocate neighbour cache hashes");
1562 rwlock_init(&tbl->lock);
1563 INIT_DELAYED_WORK_DEFERRABLE(&tbl->gc_work, neigh_periodic_work);
1564 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1565 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1566 skb_queue_head_init_class(&tbl->proxy_queue,
1567 &neigh_table_proxy_queue_class);
1569 tbl->last_flush = now;
1570 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1572 EXPORT_SYMBOL(neigh_table_init_no_netlink);
1574 void neigh_table_init(struct neigh_table *tbl)
1576 struct neigh_table *tmp;
1578 neigh_table_init_no_netlink(tbl);
1579 write_lock(&neigh_tbl_lock);
1580 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1581 if (tmp->family == tbl->family)
1584 tbl->next = neigh_tables;
1586 write_unlock(&neigh_tbl_lock);
1588 if (unlikely(tmp)) {
1589 printk(KERN_ERR "NEIGH: Registering multiple tables for "
1590 "family %d\n", tbl->family);
1594 EXPORT_SYMBOL(neigh_table_init);
1596 int neigh_table_clear(struct neigh_table *tbl)
1598 struct neigh_table **tp;
1600 /* It is not clean... Fix it to unload IPv6 module safely */
1601 cancel_delayed_work_sync(&tbl->gc_work);
1602 del_timer_sync(&tbl->proxy_timer);
1603 pneigh_queue_purge(&tbl->proxy_queue);
1604 neigh_ifdown(tbl, NULL);
1605 if (atomic_read(&tbl->entries))
1606 printk(KERN_CRIT "neighbour leakage\n");
1607 write_lock(&neigh_tbl_lock);
1608 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1614 write_unlock(&neigh_tbl_lock);
1616 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1617 neigh_hash_free_rcu);
1620 kfree(tbl->phash_buckets);
1621 tbl->phash_buckets = NULL;
1623 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1625 free_percpu(tbl->stats);
1628 kmem_cache_destroy(tbl->kmem_cachep);
1629 tbl->kmem_cachep = NULL;
1633 EXPORT_SYMBOL(neigh_table_clear);
1635 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1637 struct net *net = sock_net(skb->sk);
1639 struct nlattr *dst_attr;
1640 struct neigh_table *tbl;
1641 struct net_device *dev = NULL;
1645 if (nlmsg_len(nlh) < sizeof(*ndm))
1648 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1649 if (dst_attr == NULL)
1652 ndm = nlmsg_data(nlh);
1653 if (ndm->ndm_ifindex) {
1654 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1661 read_lock(&neigh_tbl_lock);
1662 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1663 struct neighbour *neigh;
1665 if (tbl->family != ndm->ndm_family)
1667 read_unlock(&neigh_tbl_lock);
1669 if (nla_len(dst_attr) < tbl->key_len)
1672 if (ndm->ndm_flags & NTF_PROXY) {
1673 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1680 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1681 if (neigh == NULL) {
1686 err = neigh_update(neigh, NULL, NUD_FAILED,
1687 NEIGH_UPDATE_F_OVERRIDE |
1688 NEIGH_UPDATE_F_ADMIN);
1689 neigh_release(neigh);
1692 read_unlock(&neigh_tbl_lock);
1693 err = -EAFNOSUPPORT;
1699 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1701 struct net *net = sock_net(skb->sk);
1703 struct nlattr *tb[NDA_MAX+1];
1704 struct neigh_table *tbl;
1705 struct net_device *dev = NULL;
1709 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1714 if (tb[NDA_DST] == NULL)
1717 ndm = nlmsg_data(nlh);
1718 if (ndm->ndm_ifindex) {
1719 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1725 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1729 read_lock(&neigh_tbl_lock);
1730 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1731 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1732 struct neighbour *neigh;
1735 if (tbl->family != ndm->ndm_family)
1737 read_unlock(&neigh_tbl_lock);
1739 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1741 dst = nla_data(tb[NDA_DST]);
1742 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1744 if (ndm->ndm_flags & NTF_PROXY) {
1745 struct pneigh_entry *pn;
1748 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1750 pn->flags = ndm->ndm_flags;
1759 neigh = neigh_lookup(tbl, dst, dev);
1760 if (neigh == NULL) {
1761 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1766 neigh = __neigh_lookup_errno(tbl, dst, dev);
1767 if (IS_ERR(neigh)) {
1768 err = PTR_ERR(neigh);
1772 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1774 neigh_release(neigh);
1778 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1779 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1782 if (ndm->ndm_flags & NTF_USE) {
1783 neigh_event_send(neigh, NULL);
1786 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1787 neigh_release(neigh);
1791 read_unlock(&neigh_tbl_lock);
1792 err = -EAFNOSUPPORT;
1797 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1799 struct nlattr *nest;
1801 nest = nla_nest_start(skb, NDTA_PARMS);
1806 NLA_PUT_U32(skb, NDTPA_IFINDEX, parms->dev->ifindex);
1808 NLA_PUT_U32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt));
1809 NLA_PUT_U32(skb, NDTPA_QUEUE_LEN, parms->queue_len);
1810 NLA_PUT_U32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen);
1811 NLA_PUT_U32(skb, NDTPA_APP_PROBES, parms->app_probes);
1812 NLA_PUT_U32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes);
1813 NLA_PUT_U32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes);
1814 NLA_PUT_MSECS(skb, NDTPA_REACHABLE_TIME, parms->reachable_time);
1815 NLA_PUT_MSECS(skb, NDTPA_BASE_REACHABLE_TIME,
1816 parms->base_reachable_time);
1817 NLA_PUT_MSECS(skb, NDTPA_GC_STALETIME, parms->gc_staletime);
1818 NLA_PUT_MSECS(skb, NDTPA_DELAY_PROBE_TIME, parms->delay_probe_time);
1819 NLA_PUT_MSECS(skb, NDTPA_RETRANS_TIME, parms->retrans_time);
1820 NLA_PUT_MSECS(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay);
1821 NLA_PUT_MSECS(skb, NDTPA_PROXY_DELAY, parms->proxy_delay);
1822 NLA_PUT_MSECS(skb, NDTPA_LOCKTIME, parms->locktime);
1824 return nla_nest_end(skb, nest);
1827 nla_nest_cancel(skb, nest);
1831 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1832 u32 pid, u32 seq, int type, int flags)
1834 struct nlmsghdr *nlh;
1835 struct ndtmsg *ndtmsg;
1837 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1841 ndtmsg = nlmsg_data(nlh);
1843 read_lock_bh(&tbl->lock);
1844 ndtmsg->ndtm_family = tbl->family;
1845 ndtmsg->ndtm_pad1 = 0;
1846 ndtmsg->ndtm_pad2 = 0;
1848 NLA_PUT_STRING(skb, NDTA_NAME, tbl->id);
1849 NLA_PUT_MSECS(skb, NDTA_GC_INTERVAL, tbl->gc_interval);
1850 NLA_PUT_U32(skb, NDTA_THRESH1, tbl->gc_thresh1);
1851 NLA_PUT_U32(skb, NDTA_THRESH2, tbl->gc_thresh2);
1852 NLA_PUT_U32(skb, NDTA_THRESH3, tbl->gc_thresh3);
1855 unsigned long now = jiffies;
1856 unsigned int flush_delta = now - tbl->last_flush;
1857 unsigned int rand_delta = now - tbl->last_rand;
1858 struct neigh_hash_table *nht;
1859 struct ndt_config ndc = {
1860 .ndtc_key_len = tbl->key_len,
1861 .ndtc_entry_size = tbl->entry_size,
1862 .ndtc_entries = atomic_read(&tbl->entries),
1863 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1864 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1865 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1869 nht = rcu_dereference_bh(tbl->nht);
1870 ndc.ndtc_hash_rnd = nht->hash_rnd;
1871 ndc.ndtc_hash_mask = nht->hash_mask;
1872 rcu_read_unlock_bh();
1874 NLA_PUT(skb, NDTA_CONFIG, sizeof(ndc), &ndc);
1879 struct ndt_stats ndst;
1881 memset(&ndst, 0, sizeof(ndst));
1883 for_each_possible_cpu(cpu) {
1884 struct neigh_statistics *st;
1886 st = per_cpu_ptr(tbl->stats, cpu);
1887 ndst.ndts_allocs += st->allocs;
1888 ndst.ndts_destroys += st->destroys;
1889 ndst.ndts_hash_grows += st->hash_grows;
1890 ndst.ndts_res_failed += st->res_failed;
1891 ndst.ndts_lookups += st->lookups;
1892 ndst.ndts_hits += st->hits;
1893 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1894 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1895 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1896 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1899 NLA_PUT(skb, NDTA_STATS, sizeof(ndst), &ndst);
1902 BUG_ON(tbl->parms.dev);
1903 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1904 goto nla_put_failure;
1906 read_unlock_bh(&tbl->lock);
1907 return nlmsg_end(skb, nlh);
1910 read_unlock_bh(&tbl->lock);
1911 nlmsg_cancel(skb, nlh);
1915 static int neightbl_fill_param_info(struct sk_buff *skb,
1916 struct neigh_table *tbl,
1917 struct neigh_parms *parms,
1918 u32 pid, u32 seq, int type,
1921 struct ndtmsg *ndtmsg;
1922 struct nlmsghdr *nlh;
1924 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1928 ndtmsg = nlmsg_data(nlh);
1930 read_lock_bh(&tbl->lock);
1931 ndtmsg->ndtm_family = tbl->family;
1932 ndtmsg->ndtm_pad1 = 0;
1933 ndtmsg->ndtm_pad2 = 0;
1935 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1936 neightbl_fill_parms(skb, parms) < 0)
1939 read_unlock_bh(&tbl->lock);
1940 return nlmsg_end(skb, nlh);
1942 read_unlock_bh(&tbl->lock);
1943 nlmsg_cancel(skb, nlh);
1947 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1948 [NDTA_NAME] = { .type = NLA_STRING },
1949 [NDTA_THRESH1] = { .type = NLA_U32 },
1950 [NDTA_THRESH2] = { .type = NLA_U32 },
1951 [NDTA_THRESH3] = { .type = NLA_U32 },
1952 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1953 [NDTA_PARMS] = { .type = NLA_NESTED },
1956 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1957 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1958 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1959 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1960 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1961 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1962 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1963 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1964 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1965 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1966 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1967 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1968 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1969 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1972 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1974 struct net *net = sock_net(skb->sk);
1975 struct neigh_table *tbl;
1976 struct ndtmsg *ndtmsg;
1977 struct nlattr *tb[NDTA_MAX+1];
1980 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1981 nl_neightbl_policy);
1985 if (tb[NDTA_NAME] == NULL) {
1990 ndtmsg = nlmsg_data(nlh);
1991 read_lock(&neigh_tbl_lock);
1992 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1993 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1996 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
2006 * We acquire tbl->lock to be nice to the periodic timers and
2007 * make sure they always see a consistent set of values.
2009 write_lock_bh(&tbl->lock);
2011 if (tb[NDTA_PARMS]) {
2012 struct nlattr *tbp[NDTPA_MAX+1];
2013 struct neigh_parms *p;
2016 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2017 nl_ntbl_parm_policy);
2019 goto errout_tbl_lock;
2021 if (tbp[NDTPA_IFINDEX])
2022 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2024 p = lookup_neigh_parms(tbl, net, ifindex);
2027 goto errout_tbl_lock;
2030 for (i = 1; i <= NDTPA_MAX; i++) {
2035 case NDTPA_QUEUE_LEN:
2036 p->queue_len = nla_get_u32(tbp[i]);
2038 case NDTPA_PROXY_QLEN:
2039 p->proxy_qlen = nla_get_u32(tbp[i]);
2041 case NDTPA_APP_PROBES:
2042 p->app_probes = nla_get_u32(tbp[i]);
2044 case NDTPA_UCAST_PROBES:
2045 p->ucast_probes = nla_get_u32(tbp[i]);
2047 case NDTPA_MCAST_PROBES:
2048 p->mcast_probes = nla_get_u32(tbp[i]);
2050 case NDTPA_BASE_REACHABLE_TIME:
2051 p->base_reachable_time = nla_get_msecs(tbp[i]);
2053 case NDTPA_GC_STALETIME:
2054 p->gc_staletime = nla_get_msecs(tbp[i]);
2056 case NDTPA_DELAY_PROBE_TIME:
2057 p->delay_probe_time = nla_get_msecs(tbp[i]);
2059 case NDTPA_RETRANS_TIME:
2060 p->retrans_time = nla_get_msecs(tbp[i]);
2062 case NDTPA_ANYCAST_DELAY:
2063 p->anycast_delay = nla_get_msecs(tbp[i]);
2065 case NDTPA_PROXY_DELAY:
2066 p->proxy_delay = nla_get_msecs(tbp[i]);
2068 case NDTPA_LOCKTIME:
2069 p->locktime = nla_get_msecs(tbp[i]);
2075 if (tb[NDTA_THRESH1])
2076 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2078 if (tb[NDTA_THRESH2])
2079 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2081 if (tb[NDTA_THRESH3])
2082 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2084 if (tb[NDTA_GC_INTERVAL])
2085 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2090 write_unlock_bh(&tbl->lock);
2092 read_unlock(&neigh_tbl_lock);
2097 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2099 struct net *net = sock_net(skb->sk);
2100 int family, tidx, nidx = 0;
2101 int tbl_skip = cb->args[0];
2102 int neigh_skip = cb->args[1];
2103 struct neigh_table *tbl;
2105 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2107 read_lock(&neigh_tbl_lock);
2108 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2109 struct neigh_parms *p;
2111 if (tidx < tbl_skip || (family && tbl->family != family))
2114 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).pid,
2115 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2119 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2120 if (!net_eq(neigh_parms_net(p), net))
2123 if (nidx < neigh_skip)
2126 if (neightbl_fill_param_info(skb, tbl, p,
2127 NETLINK_CB(cb->skb).pid,
2139 read_unlock(&neigh_tbl_lock);
2146 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2147 u32 pid, u32 seq, int type, unsigned int flags)
2149 unsigned long now = jiffies;
2150 struct nda_cacheinfo ci;
2151 struct nlmsghdr *nlh;
2154 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2158 ndm = nlmsg_data(nlh);
2159 ndm->ndm_family = neigh->ops->family;
2162 ndm->ndm_flags = neigh->flags;
2163 ndm->ndm_type = neigh->type;
2164 ndm->ndm_ifindex = neigh->dev->ifindex;
2166 NLA_PUT(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key);
2168 read_lock_bh(&neigh->lock);
2169 ndm->ndm_state = neigh->nud_state;
2170 if (neigh->nud_state & NUD_VALID) {
2171 char haddr[MAX_ADDR_LEN];
2173 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2174 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2175 read_unlock_bh(&neigh->lock);
2176 goto nla_put_failure;
2180 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2181 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2182 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2183 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2184 read_unlock_bh(&neigh->lock);
2186 NLA_PUT_U32(skb, NDA_PROBES, atomic_read(&neigh->probes));
2187 NLA_PUT(skb, NDA_CACHEINFO, sizeof(ci), &ci);
2189 return nlmsg_end(skb, nlh);
2192 nlmsg_cancel(skb, nlh);
2196 static void neigh_update_notify(struct neighbour *neigh)
2198 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2199 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2202 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2203 struct netlink_callback *cb)
2205 struct net *net = sock_net(skb->sk);
2206 struct neighbour *n;
2207 int rc, h, s_h = cb->args[1];
2208 int idx, s_idx = idx = cb->args[2];
2209 struct neigh_hash_table *nht;
2212 nht = rcu_dereference_bh(tbl->nht);
2214 for (h = 0; h <= nht->hash_mask; h++) {
2219 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2221 n = rcu_dereference_bh(n->next)) {
2222 if (!net_eq(dev_net(n->dev), net))
2226 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).pid,
2229 NLM_F_MULTI) <= 0) {
2239 rcu_read_unlock_bh();
2245 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2247 struct neigh_table *tbl;
2250 read_lock(&neigh_tbl_lock);
2251 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2254 for (tbl = neigh_tables, t = 0; tbl; tbl = tbl->next, t++) {
2255 if (t < s_t || (family && tbl->family != family))
2258 memset(&cb->args[1], 0, sizeof(cb->args) -
2259 sizeof(cb->args[0]));
2260 if (neigh_dump_table(tbl, skb, cb) < 0)
2263 read_unlock(&neigh_tbl_lock);
2269 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2272 struct neigh_hash_table *nht;
2275 nht = rcu_dereference_bh(tbl->nht);
2277 read_lock(&tbl->lock); /* avoid resizes */
2278 for (chain = 0; chain <= nht->hash_mask; chain++) {
2279 struct neighbour *n;
2281 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2283 n = rcu_dereference_bh(n->next))
2286 read_unlock(&tbl->lock);
2287 rcu_read_unlock_bh();
2289 EXPORT_SYMBOL(neigh_for_each);
2291 /* The tbl->lock must be held as a writer and BH disabled. */
2292 void __neigh_for_each_release(struct neigh_table *tbl,
2293 int (*cb)(struct neighbour *))
2296 struct neigh_hash_table *nht;
2298 nht = rcu_dereference_protected(tbl->nht,
2299 lockdep_is_held(&tbl->lock));
2300 for (chain = 0; chain <= nht->hash_mask; chain++) {
2301 struct neighbour *n;
2302 struct neighbour __rcu **np;
2304 np = &nht->hash_buckets[chain];
2305 while ((n = rcu_dereference_protected(*np,
2306 lockdep_is_held(&tbl->lock))) != NULL) {
2309 write_lock(&n->lock);
2312 rcu_assign_pointer(*np,
2313 rcu_dereference_protected(n->next,
2314 lockdep_is_held(&tbl->lock)));
2318 write_unlock(&n->lock);
2320 neigh_cleanup_and_release(n);
2324 EXPORT_SYMBOL(__neigh_for_each_release);
2326 #ifdef CONFIG_PROC_FS
2328 static struct neighbour *neigh_get_first(struct seq_file *seq)
2330 struct neigh_seq_state *state = seq->private;
2331 struct net *net = seq_file_net(seq);
2332 struct neigh_hash_table *nht = state->nht;
2333 struct neighbour *n = NULL;
2334 int bucket = state->bucket;
2336 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2337 for (bucket = 0; bucket <= nht->hash_mask; bucket++) {
2338 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2341 if (!net_eq(dev_net(n->dev), net))
2343 if (state->neigh_sub_iter) {
2347 v = state->neigh_sub_iter(state, n, &fakep);
2351 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2353 if (n->nud_state & ~NUD_NOARP)
2356 n = rcu_dereference_bh(n->next);
2362 state->bucket = bucket;
2367 static struct neighbour *neigh_get_next(struct seq_file *seq,
2368 struct neighbour *n,
2371 struct neigh_seq_state *state = seq->private;
2372 struct net *net = seq_file_net(seq);
2373 struct neigh_hash_table *nht = state->nht;
2375 if (state->neigh_sub_iter) {
2376 void *v = state->neigh_sub_iter(state, n, pos);
2380 n = rcu_dereference_bh(n->next);
2384 if (!net_eq(dev_net(n->dev), net))
2386 if (state->neigh_sub_iter) {
2387 void *v = state->neigh_sub_iter(state, n, pos);
2392 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2395 if (n->nud_state & ~NUD_NOARP)
2398 n = rcu_dereference_bh(n->next);
2404 if (++state->bucket > nht->hash_mask)
2407 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2415 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2417 struct neighbour *n = neigh_get_first(seq);
2422 n = neigh_get_next(seq, n, pos);
2427 return *pos ? NULL : n;
2430 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2432 struct neigh_seq_state *state = seq->private;
2433 struct net *net = seq_file_net(seq);
2434 struct neigh_table *tbl = state->tbl;
2435 struct pneigh_entry *pn = NULL;
2436 int bucket = state->bucket;
2438 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2439 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2440 pn = tbl->phash_buckets[bucket];
2441 while (pn && !net_eq(pneigh_net(pn), net))
2446 state->bucket = bucket;
2451 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2452 struct pneigh_entry *pn,
2455 struct neigh_seq_state *state = seq->private;
2456 struct net *net = seq_file_net(seq);
2457 struct neigh_table *tbl = state->tbl;
2461 if (++state->bucket > PNEIGH_HASHMASK)
2463 pn = tbl->phash_buckets[state->bucket];
2464 while (pn && !net_eq(pneigh_net(pn), net))
2476 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2478 struct pneigh_entry *pn = pneigh_get_first(seq);
2483 pn = pneigh_get_next(seq, pn, pos);
2488 return *pos ? NULL : pn;
2491 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2493 struct neigh_seq_state *state = seq->private;
2495 loff_t idxpos = *pos;
2497 rc = neigh_get_idx(seq, &idxpos);
2498 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2499 rc = pneigh_get_idx(seq, &idxpos);
2504 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2507 struct neigh_seq_state *state = seq->private;
2511 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2514 state->nht = rcu_dereference_bh(tbl->nht);
2516 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2518 EXPORT_SYMBOL(neigh_seq_start);
2520 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2522 struct neigh_seq_state *state;
2525 if (v == SEQ_START_TOKEN) {
2526 rc = neigh_get_first(seq);
2530 state = seq->private;
2531 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2532 rc = neigh_get_next(seq, v, NULL);
2535 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2536 rc = pneigh_get_first(seq);
2538 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2539 rc = pneigh_get_next(seq, v, NULL);
2545 EXPORT_SYMBOL(neigh_seq_next);
2547 void neigh_seq_stop(struct seq_file *seq, void *v)
2550 rcu_read_unlock_bh();
2552 EXPORT_SYMBOL(neigh_seq_stop);
2554 /* statistics via seq_file */
2556 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2558 struct neigh_table *tbl = seq->private;
2562 return SEQ_START_TOKEN;
2564 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2565 if (!cpu_possible(cpu))
2568 return per_cpu_ptr(tbl->stats, cpu);
2573 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2575 struct neigh_table *tbl = seq->private;
2578 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2579 if (!cpu_possible(cpu))
2582 return per_cpu_ptr(tbl->stats, cpu);
2587 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2592 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2594 struct neigh_table *tbl = seq->private;
2595 struct neigh_statistics *st = v;
2597 if (v == SEQ_START_TOKEN) {
2598 seq_printf(seq, "entries allocs destroys hash_grows lookups hits res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards\n");
2602 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2603 "%08lx %08lx %08lx %08lx %08lx\n",
2604 atomic_read(&tbl->entries),
2615 st->rcv_probes_mcast,
2616 st->rcv_probes_ucast,
2618 st->periodic_gc_runs,
2626 static const struct seq_operations neigh_stat_seq_ops = {
2627 .start = neigh_stat_seq_start,
2628 .next = neigh_stat_seq_next,
2629 .stop = neigh_stat_seq_stop,
2630 .show = neigh_stat_seq_show,
2633 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2635 int ret = seq_open(file, &neigh_stat_seq_ops);
2638 struct seq_file *sf = file->private_data;
2639 sf->private = PDE(inode)->data;
2644 static const struct file_operations neigh_stat_seq_fops = {
2645 .owner = THIS_MODULE,
2646 .open = neigh_stat_seq_open,
2648 .llseek = seq_lseek,
2649 .release = seq_release,
2652 #endif /* CONFIG_PROC_FS */
2654 static inline size_t neigh_nlmsg_size(void)
2656 return NLMSG_ALIGN(sizeof(struct ndmsg))
2657 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2658 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2659 + nla_total_size(sizeof(struct nda_cacheinfo))
2660 + nla_total_size(4); /* NDA_PROBES */
2663 static void __neigh_notify(struct neighbour *n, int type, int flags)
2665 struct net *net = dev_net(n->dev);
2666 struct sk_buff *skb;
2669 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2673 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2675 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2676 WARN_ON(err == -EMSGSIZE);
2680 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2684 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2688 void neigh_app_ns(struct neighbour *n)
2690 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2692 EXPORT_SYMBOL(neigh_app_ns);
2693 #endif /* CONFIG_ARPD */
2695 #ifdef CONFIG_SYSCTL
2697 #define NEIGH_VARS_MAX 19
2699 static struct neigh_sysctl_table {
2700 struct ctl_table_header *sysctl_header;
2701 struct ctl_table neigh_vars[NEIGH_VARS_MAX];
2703 } neigh_sysctl_template __read_mostly = {
2706 .procname = "mcast_solicit",
2707 .maxlen = sizeof(int),
2709 .proc_handler = proc_dointvec,
2712 .procname = "ucast_solicit",
2713 .maxlen = sizeof(int),
2715 .proc_handler = proc_dointvec,
2718 .procname = "app_solicit",
2719 .maxlen = sizeof(int),
2721 .proc_handler = proc_dointvec,
2724 .procname = "retrans_time",
2725 .maxlen = sizeof(int),
2727 .proc_handler = proc_dointvec_userhz_jiffies,
2730 .procname = "base_reachable_time",
2731 .maxlen = sizeof(int),
2733 .proc_handler = proc_dointvec_jiffies,
2736 .procname = "delay_first_probe_time",
2737 .maxlen = sizeof(int),
2739 .proc_handler = proc_dointvec_jiffies,
2742 .procname = "gc_stale_time",
2743 .maxlen = sizeof(int),
2745 .proc_handler = proc_dointvec_jiffies,
2748 .procname = "unres_qlen",
2749 .maxlen = sizeof(int),
2751 .proc_handler = proc_dointvec,
2754 .procname = "proxy_qlen",
2755 .maxlen = sizeof(int),
2757 .proc_handler = proc_dointvec,
2760 .procname = "anycast_delay",
2761 .maxlen = sizeof(int),
2763 .proc_handler = proc_dointvec_userhz_jiffies,
2766 .procname = "proxy_delay",
2767 .maxlen = sizeof(int),
2769 .proc_handler = proc_dointvec_userhz_jiffies,
2772 .procname = "locktime",
2773 .maxlen = sizeof(int),
2775 .proc_handler = proc_dointvec_userhz_jiffies,
2778 .procname = "retrans_time_ms",
2779 .maxlen = sizeof(int),
2781 .proc_handler = proc_dointvec_ms_jiffies,
2784 .procname = "base_reachable_time_ms",
2785 .maxlen = sizeof(int),
2787 .proc_handler = proc_dointvec_ms_jiffies,
2790 .procname = "gc_interval",
2791 .maxlen = sizeof(int),
2793 .proc_handler = proc_dointvec_jiffies,
2796 .procname = "gc_thresh1",
2797 .maxlen = sizeof(int),
2799 .proc_handler = proc_dointvec,
2802 .procname = "gc_thresh2",
2803 .maxlen = sizeof(int),
2805 .proc_handler = proc_dointvec,
2808 .procname = "gc_thresh3",
2809 .maxlen = sizeof(int),
2811 .proc_handler = proc_dointvec,
2817 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2818 char *p_name, proc_handler *handler)
2820 struct neigh_sysctl_table *t;
2821 const char *dev_name_source = NULL;
2823 #define NEIGH_CTL_PATH_ROOT 0
2824 #define NEIGH_CTL_PATH_PROTO 1
2825 #define NEIGH_CTL_PATH_NEIGH 2
2826 #define NEIGH_CTL_PATH_DEV 3
2828 struct ctl_path neigh_path[] = {
2829 { .procname = "net", },
2830 { .procname = "proto", },
2831 { .procname = "neigh", },
2832 { .procname = "default", },
2836 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2840 t->neigh_vars[0].data = &p->mcast_probes;
2841 t->neigh_vars[1].data = &p->ucast_probes;
2842 t->neigh_vars[2].data = &p->app_probes;
2843 t->neigh_vars[3].data = &p->retrans_time;
2844 t->neigh_vars[4].data = &p->base_reachable_time;
2845 t->neigh_vars[5].data = &p->delay_probe_time;
2846 t->neigh_vars[6].data = &p->gc_staletime;
2847 t->neigh_vars[7].data = &p->queue_len;
2848 t->neigh_vars[8].data = &p->proxy_qlen;
2849 t->neigh_vars[9].data = &p->anycast_delay;
2850 t->neigh_vars[10].data = &p->proxy_delay;
2851 t->neigh_vars[11].data = &p->locktime;
2852 t->neigh_vars[12].data = &p->retrans_time;
2853 t->neigh_vars[13].data = &p->base_reachable_time;
2856 dev_name_source = dev->name;
2857 /* Terminate the table early */
2858 memset(&t->neigh_vars[14], 0, sizeof(t->neigh_vars[14]));
2860 dev_name_source = neigh_path[NEIGH_CTL_PATH_DEV].procname;
2861 t->neigh_vars[14].data = (int *)(p + 1);
2862 t->neigh_vars[15].data = (int *)(p + 1) + 1;
2863 t->neigh_vars[16].data = (int *)(p + 1) + 2;
2864 t->neigh_vars[17].data = (int *)(p + 1) + 3;
2870 t->neigh_vars[3].proc_handler = handler;
2871 t->neigh_vars[3].extra1 = dev;
2873 t->neigh_vars[4].proc_handler = handler;
2874 t->neigh_vars[4].extra1 = dev;
2875 /* RetransTime (in milliseconds)*/
2876 t->neigh_vars[12].proc_handler = handler;
2877 t->neigh_vars[12].extra1 = dev;
2878 /* ReachableTime (in milliseconds) */
2879 t->neigh_vars[13].proc_handler = handler;
2880 t->neigh_vars[13].extra1 = dev;
2883 t->dev_name = kstrdup(dev_name_source, GFP_KERNEL);
2887 neigh_path[NEIGH_CTL_PATH_DEV].procname = t->dev_name;
2888 neigh_path[NEIGH_CTL_PATH_PROTO].procname = p_name;
2891 register_net_sysctl_table(neigh_parms_net(p), neigh_path, t->neigh_vars);
2892 if (!t->sysctl_header)
2895 p->sysctl_table = t;
2905 EXPORT_SYMBOL(neigh_sysctl_register);
2907 void neigh_sysctl_unregister(struct neigh_parms *p)
2909 if (p->sysctl_table) {
2910 struct neigh_sysctl_table *t = p->sysctl_table;
2911 p->sysctl_table = NULL;
2912 unregister_sysctl_table(t->sysctl_header);
2917 EXPORT_SYMBOL(neigh_sysctl_unregister);
2919 #endif /* CONFIG_SYSCTL */
2921 static int __init neigh_init(void)
2923 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL);
2924 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL);
2925 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info);
2927 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info);
2928 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL);
2933 subsys_initcall(neigh_init);