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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/slab.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/socket.h>
25 #include <linux/netdevice.h>
26 #include <linux/proc_fs.h>
28 #include <linux/sysctl.h>
30 #include <linux/times.h>
31 #include <net/net_namespace.h>
32 #include <net/neighbour.h>
35 #include <net/netevent.h>
36 #include <net/netlink.h>
37 #include <linux/rtnetlink.h>
38 #include <linux/random.h>
39 #include <linux/string.h>
40 #include <linux/log2.h>
44 #define NEIGH_PRINTK(x...) printk(x)
45 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
46 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
47 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
51 #define NEIGH_PRINTK1 NEIGH_PRINTK
55 #define NEIGH_PRINTK2 NEIGH_PRINTK
58 #define PNEIGH_HASHMASK 0xF
60 static void neigh_timer_handler(unsigned long arg);
61 static void __neigh_notify(struct neighbour *n, int type, int flags);
62 static void neigh_update_notify(struct neighbour *neigh);
63 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
65 static struct neigh_table *neigh_tables;
67 static const struct file_operations neigh_stat_seq_fops;
71 Neighbour hash table buckets are protected with rwlock tbl->lock.
73 - All the scans/updates to hash buckets MUST be made under this lock.
74 - NOTHING clever should be made under this lock: no callbacks
75 to protocol backends, no attempts to send something to network.
76 It will result in deadlocks, if backend/driver wants to use neighbour
78 - If the entry requires some non-trivial actions, increase
79 its reference count and release table lock.
81 Neighbour entries are protected:
82 - with reference count.
83 - with rwlock neigh->lock
85 Reference count prevents destruction.
87 neigh->lock mainly serializes ll address data and its validity state.
88 However, the same lock is used to protect another entry fields:
92 Again, nothing clever shall be made under neigh->lock,
93 the most complicated procedure, which we allow is dev->hard_header.
94 It is supposed, that dev->hard_header is simplistic and does
95 not make callbacks to neighbour tables.
97 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
98 list of neighbour tables. This list is used only in process context,
101 static DEFINE_RWLOCK(neigh_tbl_lock);
103 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
109 static void neigh_cleanup_and_release(struct neighbour *neigh)
111 if (neigh->parms->neigh_cleanup)
112 neigh->parms->neigh_cleanup(neigh);
114 __neigh_notify(neigh, RTM_DELNEIGH, 0);
115 neigh_release(neigh);
119 * It is random distribution in the interval (1/2)*base...(3/2)*base.
120 * It corresponds to default IPv6 settings and is not overridable,
121 * because it is really reasonable choice.
124 unsigned long neigh_rand_reach_time(unsigned long base)
126 return base ? (net_random() % base) + (base >> 1) : 0;
128 EXPORT_SYMBOL(neigh_rand_reach_time);
131 static int neigh_forced_gc(struct neigh_table *tbl)
135 struct neigh_hash_table *nht;
137 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
139 write_lock_bh(&tbl->lock);
140 nht = rcu_dereference_protected(tbl->nht,
141 lockdep_is_held(&tbl->lock));
142 for (i = 0; i < (1 << nht->hash_shift); i++) {
144 struct neighbour __rcu **np;
146 np = &nht->hash_buckets[i];
147 while ((n = rcu_dereference_protected(*np,
148 lockdep_is_held(&tbl->lock))) != NULL) {
149 /* Neighbour record may be discarded if:
150 * - nobody refers to it.
151 * - it is not permanent
153 write_lock(&n->lock);
154 if (atomic_read(&n->refcnt) == 1 &&
155 !(n->nud_state & NUD_PERMANENT)) {
156 rcu_assign_pointer(*np,
157 rcu_dereference_protected(n->next,
158 lockdep_is_held(&tbl->lock)));
161 write_unlock(&n->lock);
162 neigh_cleanup_and_release(n);
165 write_unlock(&n->lock);
170 tbl->last_flush = jiffies;
172 write_unlock_bh(&tbl->lock);
177 static void neigh_add_timer(struct neighbour *n, unsigned long when)
180 if (unlikely(mod_timer(&n->timer, when))) {
181 printk("NEIGH: BUG, double timer add, state is %x\n",
187 static int neigh_del_timer(struct neighbour *n)
189 if ((n->nud_state & NUD_IN_TIMER) &&
190 del_timer(&n->timer)) {
197 static void pneigh_queue_purge(struct sk_buff_head *list)
201 while ((skb = skb_dequeue(list)) != NULL) {
207 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
210 struct neigh_hash_table *nht;
212 nht = rcu_dereference_protected(tbl->nht,
213 lockdep_is_held(&tbl->lock));
215 for (i = 0; i < (1 << nht->hash_shift); i++) {
217 struct neighbour __rcu **np = &nht->hash_buckets[i];
219 while ((n = rcu_dereference_protected(*np,
220 lockdep_is_held(&tbl->lock))) != NULL) {
221 if (dev && n->dev != dev) {
225 rcu_assign_pointer(*np,
226 rcu_dereference_protected(n->next,
227 lockdep_is_held(&tbl->lock)));
228 write_lock(&n->lock);
232 if (atomic_read(&n->refcnt) != 1) {
233 /* The most unpleasant situation.
234 We must destroy neighbour entry,
235 but someone still uses it.
237 The destroy will be delayed until
238 the last user releases us, but
239 we must kill timers etc. and move
242 skb_queue_purge(&n->arp_queue);
243 n->arp_queue_len_bytes = 0;
244 n->output = neigh_blackhole;
245 if (n->nud_state & NUD_VALID)
246 n->nud_state = NUD_NOARP;
248 n->nud_state = NUD_NONE;
249 NEIGH_PRINTK2("neigh %p is stray.\n", n);
251 write_unlock(&n->lock);
252 neigh_cleanup_and_release(n);
257 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
259 write_lock_bh(&tbl->lock);
260 neigh_flush_dev(tbl, dev);
261 write_unlock_bh(&tbl->lock);
263 EXPORT_SYMBOL(neigh_changeaddr);
265 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
267 write_lock_bh(&tbl->lock);
268 neigh_flush_dev(tbl, dev);
269 pneigh_ifdown(tbl, dev);
270 write_unlock_bh(&tbl->lock);
272 del_timer_sync(&tbl->proxy_timer);
273 pneigh_queue_purge(&tbl->proxy_queue);
276 EXPORT_SYMBOL(neigh_ifdown);
278 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
280 struct neighbour *n = NULL;
281 unsigned long now = jiffies;
284 entries = atomic_inc_return(&tbl->entries) - 1;
285 if (entries >= tbl->gc_thresh3 ||
286 (entries >= tbl->gc_thresh2 &&
287 time_after(now, tbl->last_flush + 5 * HZ))) {
288 if (!neigh_forced_gc(tbl) &&
289 entries >= tbl->gc_thresh3)
293 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
297 skb_queue_head_init(&n->arp_queue);
298 rwlock_init(&n->lock);
299 seqlock_init(&n->ha_lock);
300 n->updated = n->used = now;
301 n->nud_state = NUD_NONE;
302 n->output = neigh_blackhole;
303 seqlock_init(&n->hh.hh_lock);
304 n->parms = neigh_parms_clone(&tbl->parms);
305 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
307 NEIGH_CACHE_STAT_INC(tbl, allocs);
309 atomic_set(&n->refcnt, 1);
315 atomic_dec(&tbl->entries);
319 static void neigh_get_hash_rnd(u32 *x)
321 get_random_bytes(x, sizeof(*x));
325 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
327 size_t size = (1 << shift) * sizeof(struct neighbour *);
328 struct neigh_hash_table *ret;
329 struct neighbour __rcu **buckets;
332 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
335 if (size <= PAGE_SIZE)
336 buckets = kzalloc(size, GFP_ATOMIC);
338 buckets = (struct neighbour __rcu **)
339 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
345 ret->hash_buckets = buckets;
346 ret->hash_shift = shift;
347 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
348 neigh_get_hash_rnd(&ret->hash_rnd[i]);
352 static void neigh_hash_free_rcu(struct rcu_head *head)
354 struct neigh_hash_table *nht = container_of(head,
355 struct neigh_hash_table,
357 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
358 struct neighbour __rcu **buckets = nht->hash_buckets;
360 if (size <= PAGE_SIZE)
363 free_pages((unsigned long)buckets, get_order(size));
367 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
368 unsigned long new_shift)
370 unsigned int i, hash;
371 struct neigh_hash_table *new_nht, *old_nht;
373 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
375 old_nht = rcu_dereference_protected(tbl->nht,
376 lockdep_is_held(&tbl->lock));
377 new_nht = neigh_hash_alloc(new_shift);
381 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
382 struct neighbour *n, *next;
384 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
385 lockdep_is_held(&tbl->lock));
388 hash = tbl->hash(n->primary_key, n->dev,
391 hash >>= (32 - new_nht->hash_shift);
392 next = rcu_dereference_protected(n->next,
393 lockdep_is_held(&tbl->lock));
395 rcu_assign_pointer(n->next,
396 rcu_dereference_protected(
397 new_nht->hash_buckets[hash],
398 lockdep_is_held(&tbl->lock)));
399 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
403 rcu_assign_pointer(tbl->nht, new_nht);
404 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
408 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
409 struct net_device *dev)
412 int key_len = tbl->key_len;
414 struct neigh_hash_table *nht;
416 NEIGH_CACHE_STAT_INC(tbl, lookups);
419 nht = rcu_dereference_bh(tbl->nht);
420 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
422 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
424 n = rcu_dereference_bh(n->next)) {
425 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
426 if (!atomic_inc_not_zero(&n->refcnt))
428 NEIGH_CACHE_STAT_INC(tbl, hits);
433 rcu_read_unlock_bh();
436 EXPORT_SYMBOL(neigh_lookup);
438 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
442 int key_len = tbl->key_len;
444 struct neigh_hash_table *nht;
446 NEIGH_CACHE_STAT_INC(tbl, lookups);
449 nht = rcu_dereference_bh(tbl->nht);
450 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
452 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
454 n = rcu_dereference_bh(n->next)) {
455 if (!memcmp(n->primary_key, pkey, key_len) &&
456 net_eq(dev_net(n->dev), net)) {
457 if (!atomic_inc_not_zero(&n->refcnt))
459 NEIGH_CACHE_STAT_INC(tbl, hits);
464 rcu_read_unlock_bh();
467 EXPORT_SYMBOL(neigh_lookup_nodev);
469 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
470 struct net_device *dev, bool want_ref)
473 int key_len = tbl->key_len;
475 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
476 struct neigh_hash_table *nht;
479 rc = ERR_PTR(-ENOBUFS);
483 memcpy(n->primary_key, pkey, key_len);
487 /* Protocol specific setup. */
488 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
490 goto out_neigh_release;
493 if (dev->netdev_ops->ndo_neigh_construct) {
494 error = dev->netdev_ops->ndo_neigh_construct(n);
497 goto out_neigh_release;
501 /* Device specific setup. */
502 if (n->parms->neigh_setup &&
503 (error = n->parms->neigh_setup(n)) < 0) {
505 goto out_neigh_release;
508 n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
510 write_lock_bh(&tbl->lock);
511 nht = rcu_dereference_protected(tbl->nht,
512 lockdep_is_held(&tbl->lock));
514 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
515 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
517 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
519 if (n->parms->dead) {
520 rc = ERR_PTR(-EINVAL);
524 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
525 lockdep_is_held(&tbl->lock));
527 n1 = rcu_dereference_protected(n1->next,
528 lockdep_is_held(&tbl->lock))) {
529 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
540 rcu_assign_pointer(n->next,
541 rcu_dereference_protected(nht->hash_buckets[hash_val],
542 lockdep_is_held(&tbl->lock)));
543 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
544 write_unlock_bh(&tbl->lock);
545 NEIGH_PRINTK2("neigh %p is created.\n", n);
550 write_unlock_bh(&tbl->lock);
555 EXPORT_SYMBOL(__neigh_create);
557 static u32 pneigh_hash(const void *pkey, int key_len)
559 u32 hash_val = *(u32 *)(pkey + key_len - 4);
560 hash_val ^= (hash_val >> 16);
561 hash_val ^= hash_val >> 8;
562 hash_val ^= hash_val >> 4;
563 hash_val &= PNEIGH_HASHMASK;
567 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
571 struct net_device *dev)
574 if (!memcmp(n->key, pkey, key_len) &&
575 net_eq(pneigh_net(n), net) &&
576 (n->dev == dev || !n->dev))
583 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
584 struct net *net, const void *pkey, struct net_device *dev)
586 int key_len = tbl->key_len;
587 u32 hash_val = pneigh_hash(pkey, key_len);
589 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
590 net, pkey, key_len, dev);
592 EXPORT_SYMBOL_GPL(__pneigh_lookup);
594 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
595 struct net *net, const void *pkey,
596 struct net_device *dev, int creat)
598 struct pneigh_entry *n;
599 int key_len = tbl->key_len;
600 u32 hash_val = pneigh_hash(pkey, key_len);
602 read_lock_bh(&tbl->lock);
603 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
604 net, pkey, key_len, dev);
605 read_unlock_bh(&tbl->lock);
612 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
616 write_pnet(&n->net, hold_net(net));
617 memcpy(n->key, pkey, key_len);
622 if (tbl->pconstructor && tbl->pconstructor(n)) {
631 write_lock_bh(&tbl->lock);
632 n->next = tbl->phash_buckets[hash_val];
633 tbl->phash_buckets[hash_val] = n;
634 write_unlock_bh(&tbl->lock);
638 EXPORT_SYMBOL(pneigh_lookup);
641 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
642 struct net_device *dev)
644 struct pneigh_entry *n, **np;
645 int key_len = tbl->key_len;
646 u32 hash_val = pneigh_hash(pkey, key_len);
648 write_lock_bh(&tbl->lock);
649 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
651 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
652 net_eq(pneigh_net(n), net)) {
654 write_unlock_bh(&tbl->lock);
655 if (tbl->pdestructor)
659 release_net(pneigh_net(n));
664 write_unlock_bh(&tbl->lock);
668 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
670 struct pneigh_entry *n, **np;
673 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
674 np = &tbl->phash_buckets[h];
675 while ((n = *np) != NULL) {
676 if (!dev || n->dev == dev) {
678 if (tbl->pdestructor)
682 release_net(pneigh_net(n));
692 static void neigh_parms_destroy(struct neigh_parms *parms);
694 static inline void neigh_parms_put(struct neigh_parms *parms)
696 if (atomic_dec_and_test(&parms->refcnt))
697 neigh_parms_destroy(parms);
701 * neighbour must already be out of the table;
704 void neigh_destroy(struct neighbour *neigh)
706 struct net_device *dev = neigh->dev;
708 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
711 pr_warn("Destroying alive neighbour %p\n", neigh);
716 if (neigh_del_timer(neigh))
717 pr_warn("Impossible event\n");
719 skb_queue_purge(&neigh->arp_queue);
720 neigh->arp_queue_len_bytes = 0;
722 if (dev->netdev_ops->ndo_neigh_destroy)
723 dev->netdev_ops->ndo_neigh_destroy(neigh);
726 neigh_parms_put(neigh->parms);
728 NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
730 atomic_dec(&neigh->tbl->entries);
731 kfree_rcu(neigh, rcu);
733 EXPORT_SYMBOL(neigh_destroy);
735 /* Neighbour state is suspicious;
738 Called with write_locked neigh.
740 static void neigh_suspect(struct neighbour *neigh)
742 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
744 neigh->output = neigh->ops->output;
747 /* Neighbour state is OK;
750 Called with write_locked neigh.
752 static void neigh_connect(struct neighbour *neigh)
754 NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
756 neigh->output = neigh->ops->connected_output;
759 static void neigh_periodic_work(struct work_struct *work)
761 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
763 struct neighbour __rcu **np;
765 struct neigh_hash_table *nht;
767 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
769 write_lock_bh(&tbl->lock);
770 nht = rcu_dereference_protected(tbl->nht,
771 lockdep_is_held(&tbl->lock));
773 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
777 * periodically recompute ReachableTime from random function
780 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
781 struct neigh_parms *p;
782 tbl->last_rand = jiffies;
783 for (p = &tbl->parms; p; p = p->next)
785 neigh_rand_reach_time(p->base_reachable_time);
788 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
789 np = &nht->hash_buckets[i];
791 while ((n = rcu_dereference_protected(*np,
792 lockdep_is_held(&tbl->lock))) != NULL) {
795 write_lock(&n->lock);
797 state = n->nud_state;
798 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
799 write_unlock(&n->lock);
803 if (time_before(n->used, n->confirmed))
804 n->used = n->confirmed;
806 if (atomic_read(&n->refcnt) == 1 &&
807 (state == NUD_FAILED ||
808 time_after(jiffies, n->used + n->parms->gc_staletime))) {
811 write_unlock(&n->lock);
812 neigh_cleanup_and_release(n);
815 write_unlock(&n->lock);
821 * It's fine to release lock here, even if hash table
822 * grows while we are preempted.
824 write_unlock_bh(&tbl->lock);
826 write_lock_bh(&tbl->lock);
827 nht = rcu_dereference_protected(tbl->nht,
828 lockdep_is_held(&tbl->lock));
831 /* Cycle through all hash buckets every base_reachable_time/2 ticks.
832 * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
833 * base_reachable_time.
835 schedule_delayed_work(&tbl->gc_work,
836 tbl->parms.base_reachable_time >> 1);
837 write_unlock_bh(&tbl->lock);
840 static __inline__ int neigh_max_probes(struct neighbour *n)
842 struct neigh_parms *p = n->parms;
843 return (n->nud_state & NUD_PROBE) ?
845 p->ucast_probes + p->app_probes + p->mcast_probes;
848 static void neigh_invalidate(struct neighbour *neigh)
849 __releases(neigh->lock)
850 __acquires(neigh->lock)
854 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
855 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
856 neigh->updated = jiffies;
858 /* It is very thin place. report_unreachable is very complicated
859 routine. Particularly, it can hit the same neighbour entry!
861 So that, we try to be accurate and avoid dead loop. --ANK
863 while (neigh->nud_state == NUD_FAILED &&
864 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
865 write_unlock(&neigh->lock);
866 neigh->ops->error_report(neigh, skb);
867 write_lock(&neigh->lock);
869 skb_queue_purge(&neigh->arp_queue);
870 neigh->arp_queue_len_bytes = 0;
873 static void neigh_probe(struct neighbour *neigh)
874 __releases(neigh->lock)
876 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
877 /* keep skb alive even if arp_queue overflows */
879 skb = skb_copy(skb, GFP_ATOMIC);
880 write_unlock(&neigh->lock);
881 neigh->ops->solicit(neigh, skb);
882 atomic_inc(&neigh->probes);
886 /* Called when a timer expires for a neighbour entry. */
888 static void neigh_timer_handler(unsigned long arg)
890 unsigned long now, next;
891 struct neighbour *neigh = (struct neighbour *)arg;
895 write_lock(&neigh->lock);
897 state = neigh->nud_state;
901 if (!(state & NUD_IN_TIMER))
904 if (state & NUD_REACHABLE) {
905 if (time_before_eq(now,
906 neigh->confirmed + neigh->parms->reachable_time)) {
907 NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
908 next = neigh->confirmed + neigh->parms->reachable_time;
909 } else if (time_before_eq(now,
910 neigh->used + neigh->parms->delay_probe_time)) {
911 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
912 neigh->nud_state = NUD_DELAY;
913 neigh->updated = jiffies;
914 neigh_suspect(neigh);
915 next = now + neigh->parms->delay_probe_time;
917 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
918 neigh->nud_state = NUD_STALE;
919 neigh->updated = jiffies;
920 neigh_suspect(neigh);
923 } else if (state & NUD_DELAY) {
924 if (time_before_eq(now,
925 neigh->confirmed + neigh->parms->delay_probe_time)) {
926 NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
927 neigh->nud_state = NUD_REACHABLE;
928 neigh->updated = jiffies;
929 neigh_connect(neigh);
931 next = neigh->confirmed + neigh->parms->reachable_time;
933 NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
934 neigh->nud_state = NUD_PROBE;
935 neigh->updated = jiffies;
936 atomic_set(&neigh->probes, 0);
937 next = now + neigh->parms->retrans_time;
940 /* NUD_PROBE|NUD_INCOMPLETE */
941 next = now + neigh->parms->retrans_time;
944 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
945 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
946 neigh->nud_state = NUD_FAILED;
948 neigh_invalidate(neigh);
951 if (neigh->nud_state & NUD_IN_TIMER) {
952 if (time_before(next, jiffies + HZ/2))
953 next = jiffies + HZ/2;
954 if (!mod_timer(&neigh->timer, next))
957 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
961 write_unlock(&neigh->lock);
965 neigh_update_notify(neigh);
967 neigh_release(neigh);
970 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
973 bool immediate_probe = false;
975 write_lock_bh(&neigh->lock);
978 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
981 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
982 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
983 unsigned long next, now = jiffies;
985 atomic_set(&neigh->probes, neigh->parms->ucast_probes);
986 neigh->nud_state = NUD_INCOMPLETE;
987 neigh->updated = now;
988 next = now + max(neigh->parms->retrans_time, HZ/2);
989 neigh_add_timer(neigh, next);
990 immediate_probe = true;
992 neigh->nud_state = NUD_FAILED;
993 neigh->updated = jiffies;
994 write_unlock_bh(&neigh->lock);
999 } else if (neigh->nud_state & NUD_STALE) {
1000 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
1001 neigh->nud_state = NUD_DELAY;
1002 neigh->updated = jiffies;
1003 neigh_add_timer(neigh,
1004 jiffies + neigh->parms->delay_probe_time);
1007 if (neigh->nud_state == NUD_INCOMPLETE) {
1009 while (neigh->arp_queue_len_bytes + skb->truesize >
1010 neigh->parms->queue_len_bytes) {
1011 struct sk_buff *buff;
1013 buff = __skb_dequeue(&neigh->arp_queue);
1016 neigh->arp_queue_len_bytes -= buff->truesize;
1018 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1021 __skb_queue_tail(&neigh->arp_queue, skb);
1022 neigh->arp_queue_len_bytes += skb->truesize;
1027 if (immediate_probe)
1030 write_unlock(&neigh->lock);
1034 EXPORT_SYMBOL(__neigh_event_send);
1036 static void neigh_update_hhs(struct neighbour *neigh)
1038 struct hh_cache *hh;
1039 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1042 if (neigh->dev->header_ops)
1043 update = neigh->dev->header_ops->cache_update;
1048 write_seqlock_bh(&hh->hh_lock);
1049 update(hh, neigh->dev, neigh->ha);
1050 write_sequnlock_bh(&hh->hh_lock);
1057 /* Generic update routine.
1058 -- lladdr is new lladdr or NULL, if it is not supplied.
1059 -- new is new state.
1061 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1063 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1064 lladdr instead of overriding it
1066 It also allows to retain current state
1067 if lladdr is unchanged.
1068 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1070 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1072 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1075 Caller MUST hold reference count on the entry.
1078 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1084 struct net_device *dev;
1085 int update_isrouter = 0;
1087 write_lock_bh(&neigh->lock);
1090 old = neigh->nud_state;
1093 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1094 (old & (NUD_NOARP | NUD_PERMANENT)))
1097 if (!(new & NUD_VALID)) {
1098 neigh_del_timer(neigh);
1099 if (old & NUD_CONNECTED)
1100 neigh_suspect(neigh);
1101 neigh->nud_state = new;
1103 notify = old & NUD_VALID;
1104 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1105 (new & NUD_FAILED)) {
1106 neigh_invalidate(neigh);
1112 /* Compare new lladdr with cached one */
1113 if (!dev->addr_len) {
1114 /* First case: device needs no address. */
1116 } else if (lladdr) {
1117 /* The second case: if something is already cached
1118 and a new address is proposed:
1120 - if they are different, check override flag
1122 if ((old & NUD_VALID) &&
1123 !memcmp(lladdr, neigh->ha, dev->addr_len))
1126 /* No address is supplied; if we know something,
1127 use it, otherwise discard the request.
1130 if (!(old & NUD_VALID))
1135 if (new & NUD_CONNECTED)
1136 neigh->confirmed = jiffies;
1137 neigh->updated = jiffies;
1139 /* If entry was valid and address is not changed,
1140 do not change entry state, if new one is STALE.
1143 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1144 if (old & NUD_VALID) {
1145 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1146 update_isrouter = 0;
1147 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1148 (old & NUD_CONNECTED)) {
1154 if (lladdr == neigh->ha && new == NUD_STALE &&
1155 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1156 (old & NUD_CONNECTED))
1163 neigh_del_timer(neigh);
1164 if (new & NUD_IN_TIMER)
1165 neigh_add_timer(neigh, (jiffies +
1166 ((new & NUD_REACHABLE) ?
1167 neigh->parms->reachable_time :
1169 neigh->nud_state = new;
1172 if (lladdr != neigh->ha) {
1173 write_seqlock(&neigh->ha_lock);
1174 memcpy(&neigh->ha, lladdr, dev->addr_len);
1175 write_sequnlock(&neigh->ha_lock);
1176 neigh_update_hhs(neigh);
1177 if (!(new & NUD_CONNECTED))
1178 neigh->confirmed = jiffies -
1179 (neigh->parms->base_reachable_time << 1);
1184 if (new & NUD_CONNECTED)
1185 neigh_connect(neigh);
1187 neigh_suspect(neigh);
1188 if (!(old & NUD_VALID)) {
1189 struct sk_buff *skb;
1191 /* Again: avoid dead loop if something went wrong */
1193 while (neigh->nud_state & NUD_VALID &&
1194 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1195 struct dst_entry *dst = skb_dst(skb);
1196 struct neighbour *n2, *n1 = neigh;
1197 write_unlock_bh(&neigh->lock);
1201 /* Why not just use 'neigh' as-is? The problem is that
1202 * things such as shaper, eql, and sch_teql can end up
1203 * using alternative, different, neigh objects to output
1204 * the packet in the output path. So what we need to do
1205 * here is re-lookup the top-level neigh in the path so
1206 * we can reinject the packet there.
1210 n2 = dst_neigh_lookup_skb(dst, skb);
1214 n1->output(n1, skb);
1219 write_lock_bh(&neigh->lock);
1221 skb_queue_purge(&neigh->arp_queue);
1222 neigh->arp_queue_len_bytes = 0;
1225 if (update_isrouter) {
1226 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1227 (neigh->flags | NTF_ROUTER) :
1228 (neigh->flags & ~NTF_ROUTER);
1230 write_unlock_bh(&neigh->lock);
1233 neigh_update_notify(neigh);
1237 EXPORT_SYMBOL(neigh_update);
1239 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1240 u8 *lladdr, void *saddr,
1241 struct net_device *dev)
1243 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1244 lladdr || !dev->addr_len);
1246 neigh_update(neigh, lladdr, NUD_STALE,
1247 NEIGH_UPDATE_F_OVERRIDE);
1250 EXPORT_SYMBOL(neigh_event_ns);
1252 /* called with read_lock_bh(&n->lock); */
1253 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1255 struct net_device *dev = dst->dev;
1256 __be16 prot = dst->ops->protocol;
1257 struct hh_cache *hh = &n->hh;
1259 write_lock_bh(&n->lock);
1261 /* Only one thread can come in here and initialize the
1265 dev->header_ops->cache(n, hh, prot);
1267 write_unlock_bh(&n->lock);
1270 /* This function can be used in contexts, where only old dev_queue_xmit
1271 * worked, f.e. if you want to override normal output path (eql, shaper),
1272 * but resolution is not made yet.
1275 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1277 struct net_device *dev = skb->dev;
1279 __skb_pull(skb, skb_network_offset(skb));
1281 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1283 dev->header_ops->rebuild(skb))
1286 return dev_queue_xmit(skb);
1288 EXPORT_SYMBOL(neigh_compat_output);
1290 /* Slow and careful. */
1292 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1294 struct dst_entry *dst = skb_dst(skb);
1300 if (!neigh_event_send(neigh, skb)) {
1302 struct net_device *dev = neigh->dev;
1305 if (dev->header_ops->cache && !neigh->hh.hh_len)
1306 neigh_hh_init(neigh, dst);
1309 __skb_pull(skb, skb_network_offset(skb));
1310 seq = read_seqbegin(&neigh->ha_lock);
1311 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1312 neigh->ha, NULL, skb->len);
1313 } while (read_seqretry(&neigh->ha_lock, seq));
1316 rc = dev_queue_xmit(skb);
1323 NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1330 EXPORT_SYMBOL(neigh_resolve_output);
1332 /* As fast as possible without hh cache */
1334 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1336 struct net_device *dev = neigh->dev;
1341 __skb_pull(skb, skb_network_offset(skb));
1342 seq = read_seqbegin(&neigh->ha_lock);
1343 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1344 neigh->ha, NULL, skb->len);
1345 } while (read_seqretry(&neigh->ha_lock, seq));
1348 err = dev_queue_xmit(skb);
1355 EXPORT_SYMBOL(neigh_connected_output);
1357 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1359 return dev_queue_xmit(skb);
1361 EXPORT_SYMBOL(neigh_direct_output);
1363 static void neigh_proxy_process(unsigned long arg)
1365 struct neigh_table *tbl = (struct neigh_table *)arg;
1366 long sched_next = 0;
1367 unsigned long now = jiffies;
1368 struct sk_buff *skb, *n;
1370 spin_lock(&tbl->proxy_queue.lock);
1372 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1373 long tdif = NEIGH_CB(skb)->sched_next - now;
1376 struct net_device *dev = skb->dev;
1378 __skb_unlink(skb, &tbl->proxy_queue);
1379 if (tbl->proxy_redo && netif_running(dev)) {
1381 tbl->proxy_redo(skb);
1388 } else if (!sched_next || tdif < sched_next)
1391 del_timer(&tbl->proxy_timer);
1393 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1394 spin_unlock(&tbl->proxy_queue.lock);
1397 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1398 struct sk_buff *skb)
1400 unsigned long now = jiffies;
1401 unsigned long sched_next = now + (net_random() % p->proxy_delay);
1403 if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1408 NEIGH_CB(skb)->sched_next = sched_next;
1409 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1411 spin_lock(&tbl->proxy_queue.lock);
1412 if (del_timer(&tbl->proxy_timer)) {
1413 if (time_before(tbl->proxy_timer.expires, sched_next))
1414 sched_next = tbl->proxy_timer.expires;
1418 __skb_queue_tail(&tbl->proxy_queue, skb);
1419 mod_timer(&tbl->proxy_timer, sched_next);
1420 spin_unlock(&tbl->proxy_queue.lock);
1422 EXPORT_SYMBOL(pneigh_enqueue);
1424 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1425 struct net *net, int ifindex)
1427 struct neigh_parms *p;
1429 for (p = &tbl->parms; p; p = p->next) {
1430 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1431 (!p->dev && !ifindex))
1438 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1439 struct neigh_table *tbl)
1441 struct neigh_parms *p, *ref;
1442 struct net *net = dev_net(dev);
1443 const struct net_device_ops *ops = dev->netdev_ops;
1445 ref = lookup_neigh_parms(tbl, net, 0);
1449 p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1452 atomic_set(&p->refcnt, 1);
1454 neigh_rand_reach_time(p->base_reachable_time);
1456 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1463 write_pnet(&p->net, hold_net(net));
1464 p->sysctl_table = NULL;
1465 write_lock_bh(&tbl->lock);
1466 p->next = tbl->parms.next;
1467 tbl->parms.next = p;
1468 write_unlock_bh(&tbl->lock);
1472 EXPORT_SYMBOL(neigh_parms_alloc);
1474 static void neigh_rcu_free_parms(struct rcu_head *head)
1476 struct neigh_parms *parms =
1477 container_of(head, struct neigh_parms, rcu_head);
1479 neigh_parms_put(parms);
1482 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1484 struct neigh_parms **p;
1486 if (!parms || parms == &tbl->parms)
1488 write_lock_bh(&tbl->lock);
1489 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1493 write_unlock_bh(&tbl->lock);
1495 dev_put(parms->dev);
1496 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1500 write_unlock_bh(&tbl->lock);
1501 NEIGH_PRINTK1("neigh_parms_release: not found\n");
1503 EXPORT_SYMBOL(neigh_parms_release);
1505 static void neigh_parms_destroy(struct neigh_parms *parms)
1507 release_net(neigh_parms_net(parms));
1511 static struct lock_class_key neigh_table_proxy_queue_class;
1513 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1515 unsigned long now = jiffies;
1516 unsigned long phsize;
1518 write_pnet(&tbl->parms.net, &init_net);
1519 atomic_set(&tbl->parms.refcnt, 1);
1520 tbl->parms.reachable_time =
1521 neigh_rand_reach_time(tbl->parms.base_reachable_time);
1523 tbl->stats = alloc_percpu(struct neigh_statistics);
1525 panic("cannot create neighbour cache statistics");
1527 #ifdef CONFIG_PROC_FS
1528 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1529 &neigh_stat_seq_fops, tbl))
1530 panic("cannot create neighbour proc dir entry");
1533 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1535 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1536 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1538 if (!tbl->nht || !tbl->phash_buckets)
1539 panic("cannot allocate neighbour cache hashes");
1541 if (!tbl->entry_size)
1542 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1543 tbl->key_len, NEIGH_PRIV_ALIGN);
1545 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1547 rwlock_init(&tbl->lock);
1548 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1549 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1550 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1551 skb_queue_head_init_class(&tbl->proxy_queue,
1552 &neigh_table_proxy_queue_class);
1554 tbl->last_flush = now;
1555 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1558 void neigh_table_init(struct neigh_table *tbl)
1560 struct neigh_table *tmp;
1562 neigh_table_init_no_netlink(tbl);
1563 write_lock(&neigh_tbl_lock);
1564 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1565 if (tmp->family == tbl->family)
1568 tbl->next = neigh_tables;
1570 write_unlock(&neigh_tbl_lock);
1572 if (unlikely(tmp)) {
1573 pr_err("Registering multiple tables for family %d\n",
1578 EXPORT_SYMBOL(neigh_table_init);
1580 int neigh_table_clear(struct neigh_table *tbl)
1582 struct neigh_table **tp;
1584 /* It is not clean... Fix it to unload IPv6 module safely */
1585 cancel_delayed_work_sync(&tbl->gc_work);
1586 del_timer_sync(&tbl->proxy_timer);
1587 pneigh_queue_purge(&tbl->proxy_queue);
1588 neigh_ifdown(tbl, NULL);
1589 if (atomic_read(&tbl->entries))
1590 pr_crit("neighbour leakage\n");
1591 write_lock(&neigh_tbl_lock);
1592 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1598 write_unlock(&neigh_tbl_lock);
1600 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1601 neigh_hash_free_rcu);
1604 kfree(tbl->phash_buckets);
1605 tbl->phash_buckets = NULL;
1607 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1609 free_percpu(tbl->stats);
1614 EXPORT_SYMBOL(neigh_table_clear);
1616 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1618 struct net *net = sock_net(skb->sk);
1620 struct nlattr *dst_attr;
1621 struct neigh_table *tbl;
1622 struct net_device *dev = NULL;
1626 if (nlmsg_len(nlh) < sizeof(*ndm))
1629 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1630 if (dst_attr == NULL)
1633 ndm = nlmsg_data(nlh);
1634 if (ndm->ndm_ifindex) {
1635 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1642 read_lock(&neigh_tbl_lock);
1643 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1644 struct neighbour *neigh;
1646 if (tbl->family != ndm->ndm_family)
1648 read_unlock(&neigh_tbl_lock);
1650 if (nla_len(dst_attr) < tbl->key_len)
1653 if (ndm->ndm_flags & NTF_PROXY) {
1654 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1661 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1662 if (neigh == NULL) {
1667 err = neigh_update(neigh, NULL, NUD_FAILED,
1668 NEIGH_UPDATE_F_OVERRIDE |
1669 NEIGH_UPDATE_F_ADMIN);
1670 neigh_release(neigh);
1673 read_unlock(&neigh_tbl_lock);
1674 err = -EAFNOSUPPORT;
1680 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1682 struct net *net = sock_net(skb->sk);
1684 struct nlattr *tb[NDA_MAX+1];
1685 struct neigh_table *tbl;
1686 struct net_device *dev = NULL;
1690 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1695 if (tb[NDA_DST] == NULL)
1698 ndm = nlmsg_data(nlh);
1699 if (ndm->ndm_ifindex) {
1700 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1706 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1710 read_lock(&neigh_tbl_lock);
1711 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1712 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1713 struct neighbour *neigh;
1716 if (tbl->family != ndm->ndm_family)
1718 read_unlock(&neigh_tbl_lock);
1720 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1722 dst = nla_data(tb[NDA_DST]);
1723 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1725 if (ndm->ndm_flags & NTF_PROXY) {
1726 struct pneigh_entry *pn;
1729 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1731 pn->flags = ndm->ndm_flags;
1740 neigh = neigh_lookup(tbl, dst, dev);
1741 if (neigh == NULL) {
1742 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1747 neigh = __neigh_lookup_errno(tbl, dst, dev);
1748 if (IS_ERR(neigh)) {
1749 err = PTR_ERR(neigh);
1753 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1755 neigh_release(neigh);
1759 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1760 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1763 if (ndm->ndm_flags & NTF_USE) {
1764 neigh_event_send(neigh, NULL);
1767 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1768 neigh_release(neigh);
1772 read_unlock(&neigh_tbl_lock);
1773 err = -EAFNOSUPPORT;
1778 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1780 struct nlattr *nest;
1782 nest = nla_nest_start(skb, NDTA_PARMS);
1787 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1788 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1789 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES, parms->queue_len_bytes) ||
1790 /* approximative value for deprecated QUEUE_LEN (in packets) */
1791 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1792 parms->queue_len_bytes / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1793 nla_put_u32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen) ||
1794 nla_put_u32(skb, NDTPA_APP_PROBES, parms->app_probes) ||
1795 nla_put_u32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes) ||
1796 nla_put_u32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes) ||
1797 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1798 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1799 parms->base_reachable_time) ||
1800 nla_put_msecs(skb, NDTPA_GC_STALETIME, parms->gc_staletime) ||
1801 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1802 parms->delay_probe_time) ||
1803 nla_put_msecs(skb, NDTPA_RETRANS_TIME, parms->retrans_time) ||
1804 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay) ||
1805 nla_put_msecs(skb, NDTPA_PROXY_DELAY, parms->proxy_delay) ||
1806 nla_put_msecs(skb, NDTPA_LOCKTIME, parms->locktime))
1807 goto nla_put_failure;
1808 return nla_nest_end(skb, nest);
1811 nla_nest_cancel(skb, nest);
1815 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1816 u32 pid, u32 seq, int type, int flags)
1818 struct nlmsghdr *nlh;
1819 struct ndtmsg *ndtmsg;
1821 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1825 ndtmsg = nlmsg_data(nlh);
1827 read_lock_bh(&tbl->lock);
1828 ndtmsg->ndtm_family = tbl->family;
1829 ndtmsg->ndtm_pad1 = 0;
1830 ndtmsg->ndtm_pad2 = 0;
1832 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1833 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1834 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1835 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1836 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1837 goto nla_put_failure;
1839 unsigned long now = jiffies;
1840 unsigned int flush_delta = now - tbl->last_flush;
1841 unsigned int rand_delta = now - tbl->last_rand;
1842 struct neigh_hash_table *nht;
1843 struct ndt_config ndc = {
1844 .ndtc_key_len = tbl->key_len,
1845 .ndtc_entry_size = tbl->entry_size,
1846 .ndtc_entries = atomic_read(&tbl->entries),
1847 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1848 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1849 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1853 nht = rcu_dereference_bh(tbl->nht);
1854 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1855 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1856 rcu_read_unlock_bh();
1858 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1859 goto nla_put_failure;
1864 struct ndt_stats ndst;
1866 memset(&ndst, 0, sizeof(ndst));
1868 for_each_possible_cpu(cpu) {
1869 struct neigh_statistics *st;
1871 st = per_cpu_ptr(tbl->stats, cpu);
1872 ndst.ndts_allocs += st->allocs;
1873 ndst.ndts_destroys += st->destroys;
1874 ndst.ndts_hash_grows += st->hash_grows;
1875 ndst.ndts_res_failed += st->res_failed;
1876 ndst.ndts_lookups += st->lookups;
1877 ndst.ndts_hits += st->hits;
1878 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1879 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1880 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1881 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1884 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1885 goto nla_put_failure;
1888 BUG_ON(tbl->parms.dev);
1889 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1890 goto nla_put_failure;
1892 read_unlock_bh(&tbl->lock);
1893 return nlmsg_end(skb, nlh);
1896 read_unlock_bh(&tbl->lock);
1897 nlmsg_cancel(skb, nlh);
1901 static int neightbl_fill_param_info(struct sk_buff *skb,
1902 struct neigh_table *tbl,
1903 struct neigh_parms *parms,
1904 u32 pid, u32 seq, int type,
1907 struct ndtmsg *ndtmsg;
1908 struct nlmsghdr *nlh;
1910 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1914 ndtmsg = nlmsg_data(nlh);
1916 read_lock_bh(&tbl->lock);
1917 ndtmsg->ndtm_family = tbl->family;
1918 ndtmsg->ndtm_pad1 = 0;
1919 ndtmsg->ndtm_pad2 = 0;
1921 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1922 neightbl_fill_parms(skb, parms) < 0)
1925 read_unlock_bh(&tbl->lock);
1926 return nlmsg_end(skb, nlh);
1928 read_unlock_bh(&tbl->lock);
1929 nlmsg_cancel(skb, nlh);
1933 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1934 [NDTA_NAME] = { .type = NLA_STRING },
1935 [NDTA_THRESH1] = { .type = NLA_U32 },
1936 [NDTA_THRESH2] = { .type = NLA_U32 },
1937 [NDTA_THRESH3] = { .type = NLA_U32 },
1938 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1939 [NDTA_PARMS] = { .type = NLA_NESTED },
1942 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1943 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1944 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1945 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1946 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1947 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1948 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1949 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1950 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1951 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1952 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1953 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1954 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1955 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1958 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1960 struct net *net = sock_net(skb->sk);
1961 struct neigh_table *tbl;
1962 struct ndtmsg *ndtmsg;
1963 struct nlattr *tb[NDTA_MAX+1];
1966 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1967 nl_neightbl_policy);
1971 if (tb[NDTA_NAME] == NULL) {
1976 ndtmsg = nlmsg_data(nlh);
1977 read_lock(&neigh_tbl_lock);
1978 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1979 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1982 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1992 * We acquire tbl->lock to be nice to the periodic timers and
1993 * make sure they always see a consistent set of values.
1995 write_lock_bh(&tbl->lock);
1997 if (tb[NDTA_PARMS]) {
1998 struct nlattr *tbp[NDTPA_MAX+1];
1999 struct neigh_parms *p;
2002 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2003 nl_ntbl_parm_policy);
2005 goto errout_tbl_lock;
2007 if (tbp[NDTPA_IFINDEX])
2008 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2010 p = lookup_neigh_parms(tbl, net, ifindex);
2013 goto errout_tbl_lock;
2016 for (i = 1; i <= NDTPA_MAX; i++) {
2021 case NDTPA_QUEUE_LEN:
2022 p->queue_len_bytes = nla_get_u32(tbp[i]) *
2023 SKB_TRUESIZE(ETH_FRAME_LEN);
2025 case NDTPA_QUEUE_LENBYTES:
2026 p->queue_len_bytes = nla_get_u32(tbp[i]);
2028 case NDTPA_PROXY_QLEN:
2029 p->proxy_qlen = nla_get_u32(tbp[i]);
2031 case NDTPA_APP_PROBES:
2032 p->app_probes = nla_get_u32(tbp[i]);
2034 case NDTPA_UCAST_PROBES:
2035 p->ucast_probes = nla_get_u32(tbp[i]);
2037 case NDTPA_MCAST_PROBES:
2038 p->mcast_probes = nla_get_u32(tbp[i]);
2040 case NDTPA_BASE_REACHABLE_TIME:
2041 p->base_reachable_time = nla_get_msecs(tbp[i]);
2043 case NDTPA_GC_STALETIME:
2044 p->gc_staletime = nla_get_msecs(tbp[i]);
2046 case NDTPA_DELAY_PROBE_TIME:
2047 p->delay_probe_time = nla_get_msecs(tbp[i]);
2049 case NDTPA_RETRANS_TIME:
2050 p->retrans_time = nla_get_msecs(tbp[i]);
2052 case NDTPA_ANYCAST_DELAY:
2053 p->anycast_delay = nla_get_msecs(tbp[i]);
2055 case NDTPA_PROXY_DELAY:
2056 p->proxy_delay = nla_get_msecs(tbp[i]);
2058 case NDTPA_LOCKTIME:
2059 p->locktime = nla_get_msecs(tbp[i]);
2065 if (tb[NDTA_THRESH1])
2066 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2068 if (tb[NDTA_THRESH2])
2069 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2071 if (tb[NDTA_THRESH3])
2072 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2074 if (tb[NDTA_GC_INTERVAL])
2075 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2080 write_unlock_bh(&tbl->lock);
2082 read_unlock(&neigh_tbl_lock);
2087 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2089 struct net *net = sock_net(skb->sk);
2090 int family, tidx, nidx = 0;
2091 int tbl_skip = cb->args[0];
2092 int neigh_skip = cb->args[1];
2093 struct neigh_table *tbl;
2095 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2097 read_lock(&neigh_tbl_lock);
2098 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2099 struct neigh_parms *p;
2101 if (tidx < tbl_skip || (family && tbl->family != family))
2104 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2105 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2109 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2110 if (!net_eq(neigh_parms_net(p), net))
2113 if (nidx < neigh_skip)
2116 if (neightbl_fill_param_info(skb, tbl, p,
2117 NETLINK_CB(cb->skb).portid,
2129 read_unlock(&neigh_tbl_lock);
2136 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2137 u32 pid, u32 seq, int type, unsigned int flags)
2139 unsigned long now = jiffies;
2140 struct nda_cacheinfo ci;
2141 struct nlmsghdr *nlh;
2144 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2148 ndm = nlmsg_data(nlh);
2149 ndm->ndm_family = neigh->ops->family;
2152 ndm->ndm_flags = neigh->flags;
2153 ndm->ndm_type = neigh->type;
2154 ndm->ndm_ifindex = neigh->dev->ifindex;
2156 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2157 goto nla_put_failure;
2159 read_lock_bh(&neigh->lock);
2160 ndm->ndm_state = neigh->nud_state;
2161 if (neigh->nud_state & NUD_VALID) {
2162 char haddr[MAX_ADDR_LEN];
2164 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2165 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2166 read_unlock_bh(&neigh->lock);
2167 goto nla_put_failure;
2171 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2172 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2173 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2174 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2175 read_unlock_bh(&neigh->lock);
2177 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2178 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2179 goto nla_put_failure;
2181 return nlmsg_end(skb, nlh);
2184 nlmsg_cancel(skb, nlh);
2188 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2189 u32 pid, u32 seq, int type, unsigned int flags,
2190 struct neigh_table *tbl)
2192 struct nlmsghdr *nlh;
2195 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2199 ndm = nlmsg_data(nlh);
2200 ndm->ndm_family = tbl->family;
2203 ndm->ndm_flags = pn->flags | NTF_PROXY;
2204 ndm->ndm_type = NDA_DST;
2205 ndm->ndm_ifindex = pn->dev->ifindex;
2206 ndm->ndm_state = NUD_NONE;
2208 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2209 goto nla_put_failure;
2211 return nlmsg_end(skb, nlh);
2214 nlmsg_cancel(skb, nlh);
2218 static void neigh_update_notify(struct neighbour *neigh)
2220 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2221 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2224 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2225 struct netlink_callback *cb)
2227 struct net *net = sock_net(skb->sk);
2228 struct neighbour *n;
2229 int rc, h, s_h = cb->args[1];
2230 int idx, s_idx = idx = cb->args[2];
2231 struct neigh_hash_table *nht;
2234 nht = rcu_dereference_bh(tbl->nht);
2236 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2239 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2241 n = rcu_dereference_bh(n->next)) {
2242 if (!net_eq(dev_net(n->dev), net))
2246 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2249 NLM_F_MULTI) <= 0) {
2259 rcu_read_unlock_bh();
2265 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2266 struct netlink_callback *cb)
2268 struct pneigh_entry *n;
2269 struct net *net = sock_net(skb->sk);
2270 int rc, h, s_h = cb->args[3];
2271 int idx, s_idx = idx = cb->args[4];
2273 read_lock_bh(&tbl->lock);
2275 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2278 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2279 if (dev_net(n->dev) != net)
2283 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2286 NLM_F_MULTI, tbl) <= 0) {
2287 read_unlock_bh(&tbl->lock);
2296 read_unlock_bh(&tbl->lock);
2305 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2307 struct neigh_table *tbl;
2312 read_lock(&neigh_tbl_lock);
2313 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2315 /* check for full ndmsg structure presence, family member is
2316 * the same for both structures
2318 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2319 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2324 for (tbl = neigh_tables, t = 0; tbl;
2325 tbl = tbl->next, t++) {
2326 if (t < s_t || (family && tbl->family != family))
2329 memset(&cb->args[1], 0, sizeof(cb->args) -
2330 sizeof(cb->args[0]));
2332 err = pneigh_dump_table(tbl, skb, cb);
2334 err = neigh_dump_table(tbl, skb, cb);
2338 read_unlock(&neigh_tbl_lock);
2344 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2347 struct neigh_hash_table *nht;
2350 nht = rcu_dereference_bh(tbl->nht);
2352 read_lock(&tbl->lock); /* avoid resizes */
2353 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2354 struct neighbour *n;
2356 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2358 n = rcu_dereference_bh(n->next))
2361 read_unlock(&tbl->lock);
2362 rcu_read_unlock_bh();
2364 EXPORT_SYMBOL(neigh_for_each);
2366 /* The tbl->lock must be held as a writer and BH disabled. */
2367 void __neigh_for_each_release(struct neigh_table *tbl,
2368 int (*cb)(struct neighbour *))
2371 struct neigh_hash_table *nht;
2373 nht = rcu_dereference_protected(tbl->nht,
2374 lockdep_is_held(&tbl->lock));
2375 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2376 struct neighbour *n;
2377 struct neighbour __rcu **np;
2379 np = &nht->hash_buckets[chain];
2380 while ((n = rcu_dereference_protected(*np,
2381 lockdep_is_held(&tbl->lock))) != NULL) {
2384 write_lock(&n->lock);
2387 rcu_assign_pointer(*np,
2388 rcu_dereference_protected(n->next,
2389 lockdep_is_held(&tbl->lock)));
2393 write_unlock(&n->lock);
2395 neigh_cleanup_and_release(n);
2399 EXPORT_SYMBOL(__neigh_for_each_release);
2401 #ifdef CONFIG_PROC_FS
2403 static struct neighbour *neigh_get_first(struct seq_file *seq)
2405 struct neigh_seq_state *state = seq->private;
2406 struct net *net = seq_file_net(seq);
2407 struct neigh_hash_table *nht = state->nht;
2408 struct neighbour *n = NULL;
2409 int bucket = state->bucket;
2411 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2412 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2413 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2416 if (!net_eq(dev_net(n->dev), net))
2418 if (state->neigh_sub_iter) {
2422 v = state->neigh_sub_iter(state, n, &fakep);
2426 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2428 if (n->nud_state & ~NUD_NOARP)
2431 n = rcu_dereference_bh(n->next);
2437 state->bucket = bucket;
2442 static struct neighbour *neigh_get_next(struct seq_file *seq,
2443 struct neighbour *n,
2446 struct neigh_seq_state *state = seq->private;
2447 struct net *net = seq_file_net(seq);
2448 struct neigh_hash_table *nht = state->nht;
2450 if (state->neigh_sub_iter) {
2451 void *v = state->neigh_sub_iter(state, n, pos);
2455 n = rcu_dereference_bh(n->next);
2459 if (!net_eq(dev_net(n->dev), net))
2461 if (state->neigh_sub_iter) {
2462 void *v = state->neigh_sub_iter(state, n, pos);
2467 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2470 if (n->nud_state & ~NUD_NOARP)
2473 n = rcu_dereference_bh(n->next);
2479 if (++state->bucket >= (1 << nht->hash_shift))
2482 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2490 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2492 struct neighbour *n = neigh_get_first(seq);
2497 n = neigh_get_next(seq, n, pos);
2502 return *pos ? NULL : n;
2505 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2507 struct neigh_seq_state *state = seq->private;
2508 struct net *net = seq_file_net(seq);
2509 struct neigh_table *tbl = state->tbl;
2510 struct pneigh_entry *pn = NULL;
2511 int bucket = state->bucket;
2513 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2514 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2515 pn = tbl->phash_buckets[bucket];
2516 while (pn && !net_eq(pneigh_net(pn), net))
2521 state->bucket = bucket;
2526 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2527 struct pneigh_entry *pn,
2530 struct neigh_seq_state *state = seq->private;
2531 struct net *net = seq_file_net(seq);
2532 struct neigh_table *tbl = state->tbl;
2536 } while (pn && !net_eq(pneigh_net(pn), net));
2539 if (++state->bucket > PNEIGH_HASHMASK)
2541 pn = tbl->phash_buckets[state->bucket];
2542 while (pn && !net_eq(pneigh_net(pn), net))
2554 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2556 struct pneigh_entry *pn = pneigh_get_first(seq);
2561 pn = pneigh_get_next(seq, pn, pos);
2566 return *pos ? NULL : pn;
2569 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2571 struct neigh_seq_state *state = seq->private;
2573 loff_t idxpos = *pos;
2575 rc = neigh_get_idx(seq, &idxpos);
2576 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2577 rc = pneigh_get_idx(seq, &idxpos);
2582 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2585 struct neigh_seq_state *state = seq->private;
2589 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2592 state->nht = rcu_dereference_bh(tbl->nht);
2594 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2596 EXPORT_SYMBOL(neigh_seq_start);
2598 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2600 struct neigh_seq_state *state;
2603 if (v == SEQ_START_TOKEN) {
2604 rc = neigh_get_first(seq);
2608 state = seq->private;
2609 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2610 rc = neigh_get_next(seq, v, NULL);
2613 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2614 rc = pneigh_get_first(seq);
2616 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2617 rc = pneigh_get_next(seq, v, NULL);
2623 EXPORT_SYMBOL(neigh_seq_next);
2625 void neigh_seq_stop(struct seq_file *seq, void *v)
2628 rcu_read_unlock_bh();
2630 EXPORT_SYMBOL(neigh_seq_stop);
2632 /* statistics via seq_file */
2634 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2636 struct neigh_table *tbl = seq->private;
2640 return SEQ_START_TOKEN;
2642 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2643 if (!cpu_possible(cpu))
2646 return per_cpu_ptr(tbl->stats, cpu);
2651 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2653 struct neigh_table *tbl = seq->private;
2656 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2657 if (!cpu_possible(cpu))
2660 return per_cpu_ptr(tbl->stats, cpu);
2665 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2670 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2672 struct neigh_table *tbl = seq->private;
2673 struct neigh_statistics *st = v;
2675 if (v == SEQ_START_TOKEN) {
2676 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");
2680 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2681 "%08lx %08lx %08lx %08lx %08lx\n",
2682 atomic_read(&tbl->entries),
2693 st->rcv_probes_mcast,
2694 st->rcv_probes_ucast,
2696 st->periodic_gc_runs,
2704 static const struct seq_operations neigh_stat_seq_ops = {
2705 .start = neigh_stat_seq_start,
2706 .next = neigh_stat_seq_next,
2707 .stop = neigh_stat_seq_stop,
2708 .show = neigh_stat_seq_show,
2711 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2713 int ret = seq_open(file, &neigh_stat_seq_ops);
2716 struct seq_file *sf = file->private_data;
2717 sf->private = PDE(inode)->data;
2722 static const struct file_operations neigh_stat_seq_fops = {
2723 .owner = THIS_MODULE,
2724 .open = neigh_stat_seq_open,
2726 .llseek = seq_lseek,
2727 .release = seq_release,
2730 #endif /* CONFIG_PROC_FS */
2732 static inline size_t neigh_nlmsg_size(void)
2734 return NLMSG_ALIGN(sizeof(struct ndmsg))
2735 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2736 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2737 + nla_total_size(sizeof(struct nda_cacheinfo))
2738 + nla_total_size(4); /* NDA_PROBES */
2741 static void __neigh_notify(struct neighbour *n, int type, int flags)
2743 struct net *net = dev_net(n->dev);
2744 struct sk_buff *skb;
2747 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2751 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2753 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2754 WARN_ON(err == -EMSGSIZE);
2758 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2762 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2766 void neigh_app_ns(struct neighbour *n)
2768 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2770 EXPORT_SYMBOL(neigh_app_ns);
2771 #endif /* CONFIG_ARPD */
2773 #ifdef CONFIG_SYSCTL
2775 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2777 static int proc_unres_qlen(ctl_table *ctl, int write, void __user *buffer,
2778 size_t *lenp, loff_t *ppos)
2781 ctl_table tmp = *ctl;
2784 tmp.extra2 = &unres_qlen_max;
2787 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2788 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2791 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2796 NEIGH_VAR_MCAST_PROBE,
2797 NEIGH_VAR_UCAST_PROBE,
2798 NEIGH_VAR_APP_PROBE,
2799 NEIGH_VAR_RETRANS_TIME,
2800 NEIGH_VAR_BASE_REACHABLE_TIME,
2801 NEIGH_VAR_DELAY_PROBE_TIME,
2802 NEIGH_VAR_GC_STALETIME,
2803 NEIGH_VAR_QUEUE_LEN,
2804 NEIGH_VAR_QUEUE_LEN_BYTES,
2805 NEIGH_VAR_PROXY_QLEN,
2806 NEIGH_VAR_ANYCAST_DELAY,
2807 NEIGH_VAR_PROXY_DELAY,
2809 NEIGH_VAR_RETRANS_TIME_MS,
2810 NEIGH_VAR_BASE_REACHABLE_TIME_MS,
2811 NEIGH_VAR_GC_INTERVAL,
2812 NEIGH_VAR_GC_THRESH1,
2813 NEIGH_VAR_GC_THRESH2,
2814 NEIGH_VAR_GC_THRESH3,
2818 static struct neigh_sysctl_table {
2819 struct ctl_table_header *sysctl_header;
2820 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2821 } neigh_sysctl_template __read_mostly = {
2823 [NEIGH_VAR_MCAST_PROBE] = {
2824 .procname = "mcast_solicit",
2825 .maxlen = sizeof(int),
2827 .proc_handler = proc_dointvec,
2829 [NEIGH_VAR_UCAST_PROBE] = {
2830 .procname = "ucast_solicit",
2831 .maxlen = sizeof(int),
2833 .proc_handler = proc_dointvec,
2835 [NEIGH_VAR_APP_PROBE] = {
2836 .procname = "app_solicit",
2837 .maxlen = sizeof(int),
2839 .proc_handler = proc_dointvec,
2841 [NEIGH_VAR_RETRANS_TIME] = {
2842 .procname = "retrans_time",
2843 .maxlen = sizeof(int),
2845 .proc_handler = proc_dointvec_userhz_jiffies,
2847 [NEIGH_VAR_BASE_REACHABLE_TIME] = {
2848 .procname = "base_reachable_time",
2849 .maxlen = sizeof(int),
2851 .proc_handler = proc_dointvec_jiffies,
2853 [NEIGH_VAR_DELAY_PROBE_TIME] = {
2854 .procname = "delay_first_probe_time",
2855 .maxlen = sizeof(int),
2857 .proc_handler = proc_dointvec_jiffies,
2859 [NEIGH_VAR_GC_STALETIME] = {
2860 .procname = "gc_stale_time",
2861 .maxlen = sizeof(int),
2863 .proc_handler = proc_dointvec_jiffies,
2865 [NEIGH_VAR_QUEUE_LEN] = {
2866 .procname = "unres_qlen",
2867 .maxlen = sizeof(int),
2869 .proc_handler = proc_unres_qlen,
2871 [NEIGH_VAR_QUEUE_LEN_BYTES] = {
2872 .procname = "unres_qlen_bytes",
2873 .maxlen = sizeof(int),
2876 .proc_handler = proc_dointvec_minmax,
2878 [NEIGH_VAR_PROXY_QLEN] = {
2879 .procname = "proxy_qlen",
2880 .maxlen = sizeof(int),
2882 .proc_handler = proc_dointvec,
2884 [NEIGH_VAR_ANYCAST_DELAY] = {
2885 .procname = "anycast_delay",
2886 .maxlen = sizeof(int),
2888 .proc_handler = proc_dointvec_userhz_jiffies,
2890 [NEIGH_VAR_PROXY_DELAY] = {
2891 .procname = "proxy_delay",
2892 .maxlen = sizeof(int),
2894 .proc_handler = proc_dointvec_userhz_jiffies,
2896 [NEIGH_VAR_LOCKTIME] = {
2897 .procname = "locktime",
2898 .maxlen = sizeof(int),
2900 .proc_handler = proc_dointvec_userhz_jiffies,
2902 [NEIGH_VAR_RETRANS_TIME_MS] = {
2903 .procname = "retrans_time_ms",
2904 .maxlen = sizeof(int),
2906 .proc_handler = proc_dointvec_ms_jiffies,
2908 [NEIGH_VAR_BASE_REACHABLE_TIME_MS] = {
2909 .procname = "base_reachable_time_ms",
2910 .maxlen = sizeof(int),
2912 .proc_handler = proc_dointvec_ms_jiffies,
2914 [NEIGH_VAR_GC_INTERVAL] = {
2915 .procname = "gc_interval",
2916 .maxlen = sizeof(int),
2918 .proc_handler = proc_dointvec_jiffies,
2920 [NEIGH_VAR_GC_THRESH1] = {
2921 .procname = "gc_thresh1",
2922 .maxlen = sizeof(int),
2924 .proc_handler = proc_dointvec,
2926 [NEIGH_VAR_GC_THRESH2] = {
2927 .procname = "gc_thresh2",
2928 .maxlen = sizeof(int),
2930 .proc_handler = proc_dointvec,
2932 [NEIGH_VAR_GC_THRESH3] = {
2933 .procname = "gc_thresh3",
2934 .maxlen = sizeof(int),
2936 .proc_handler = proc_dointvec,
2942 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2943 char *p_name, proc_handler *handler)
2945 struct neigh_sysctl_table *t;
2946 const char *dev_name_source = NULL;
2947 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
2949 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2953 t->neigh_vars[NEIGH_VAR_MCAST_PROBE].data = &p->mcast_probes;
2954 t->neigh_vars[NEIGH_VAR_UCAST_PROBE].data = &p->ucast_probes;
2955 t->neigh_vars[NEIGH_VAR_APP_PROBE].data = &p->app_probes;
2956 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].data = &p->retrans_time;
2957 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].data = &p->base_reachable_time;
2958 t->neigh_vars[NEIGH_VAR_DELAY_PROBE_TIME].data = &p->delay_probe_time;
2959 t->neigh_vars[NEIGH_VAR_GC_STALETIME].data = &p->gc_staletime;
2960 t->neigh_vars[NEIGH_VAR_QUEUE_LEN].data = &p->queue_len_bytes;
2961 t->neigh_vars[NEIGH_VAR_QUEUE_LEN_BYTES].data = &p->queue_len_bytes;
2962 t->neigh_vars[NEIGH_VAR_PROXY_QLEN].data = &p->proxy_qlen;
2963 t->neigh_vars[NEIGH_VAR_ANYCAST_DELAY].data = &p->anycast_delay;
2964 t->neigh_vars[NEIGH_VAR_PROXY_DELAY].data = &p->proxy_delay;
2965 t->neigh_vars[NEIGH_VAR_LOCKTIME].data = &p->locktime;
2966 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].data = &p->retrans_time;
2967 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].data = &p->base_reachable_time;
2970 dev_name_source = dev->name;
2971 /* Terminate the table early */
2972 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
2973 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
2975 dev_name_source = "default";
2976 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1);
2977 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1;
2978 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2;
2979 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3;
2985 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
2986 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].extra1 = dev;
2988 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
2989 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].extra1 = dev;
2990 /* RetransTime (in milliseconds)*/
2991 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
2992 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].extra1 = dev;
2993 /* ReachableTime (in milliseconds) */
2994 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
2995 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].extra1 = dev;
2998 /* Don't export sysctls to unprivileged users */
2999 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3000 t->neigh_vars[0].procname = NULL;
3002 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3003 p_name, dev_name_source);
3005 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3006 if (!t->sysctl_header)
3009 p->sysctl_table = t;
3017 EXPORT_SYMBOL(neigh_sysctl_register);
3019 void neigh_sysctl_unregister(struct neigh_parms *p)
3021 if (p->sysctl_table) {
3022 struct neigh_sysctl_table *t = p->sysctl_table;
3023 p->sysctl_table = NULL;
3024 unregister_net_sysctl_table(t->sysctl_header);
3028 EXPORT_SYMBOL(neigh_sysctl_unregister);
3030 #endif /* CONFIG_SYSCTL */
3032 static int __init neigh_init(void)
3034 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3035 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3036 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3038 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3040 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3045 subsys_initcall(neigh_init);