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_dbg(level, fmt, ...) \
46 if (level <= NEIGH_DEBUG) \
47 pr_debug(fmt, ##__VA_ARGS__); \
50 #define PNEIGH_HASHMASK 0xF
52 static void neigh_timer_handler(unsigned long arg);
53 static void __neigh_notify(struct neighbour *n, int type, int flags);
54 static void neigh_update_notify(struct neighbour *neigh);
55 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
57 static struct neigh_table *neigh_tables;
59 static const struct file_operations neigh_stat_seq_fops;
63 Neighbour hash table buckets are protected with rwlock tbl->lock.
65 - All the scans/updates to hash buckets MUST be made under this lock.
66 - NOTHING clever should be made under this lock: no callbacks
67 to protocol backends, no attempts to send something to network.
68 It will result in deadlocks, if backend/driver wants to use neighbour
70 - If the entry requires some non-trivial actions, increase
71 its reference count and release table lock.
73 Neighbour entries are protected:
74 - with reference count.
75 - with rwlock neigh->lock
77 Reference count prevents destruction.
79 neigh->lock mainly serializes ll address data and its validity state.
80 However, the same lock is used to protect another entry fields:
84 Again, nothing clever shall be made under neigh->lock,
85 the most complicated procedure, which we allow is dev->hard_header.
86 It is supposed, that dev->hard_header is simplistic and does
87 not make callbacks to neighbour tables.
89 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
90 list of neighbour tables. This list is used only in process context,
93 static DEFINE_RWLOCK(neigh_tbl_lock);
95 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
101 static void neigh_cleanup_and_release(struct neighbour *neigh)
103 if (neigh->parms->neigh_cleanup)
104 neigh->parms->neigh_cleanup(neigh);
106 __neigh_notify(neigh, RTM_DELNEIGH, 0);
107 neigh_release(neigh);
111 * It is random distribution in the interval (1/2)*base...(3/2)*base.
112 * It corresponds to default IPv6 settings and is not overridable,
113 * because it is really reasonable choice.
116 unsigned long neigh_rand_reach_time(unsigned long base)
118 return base ? (net_random() % base) + (base >> 1) : 0;
120 EXPORT_SYMBOL(neigh_rand_reach_time);
123 static int neigh_forced_gc(struct neigh_table *tbl)
127 struct neigh_hash_table *nht;
129 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
131 write_lock_bh(&tbl->lock);
132 nht = rcu_dereference_protected(tbl->nht,
133 lockdep_is_held(&tbl->lock));
134 for (i = 0; i < (1 << nht->hash_shift); i++) {
136 struct neighbour __rcu **np;
138 np = &nht->hash_buckets[i];
139 while ((n = rcu_dereference_protected(*np,
140 lockdep_is_held(&tbl->lock))) != NULL) {
141 /* Neighbour record may be discarded if:
142 * - nobody refers to it.
143 * - it is not permanent
145 write_lock(&n->lock);
146 if (atomic_read(&n->refcnt) == 1 &&
147 !(n->nud_state & NUD_PERMANENT)) {
148 rcu_assign_pointer(*np,
149 rcu_dereference_protected(n->next,
150 lockdep_is_held(&tbl->lock)));
153 write_unlock(&n->lock);
154 neigh_cleanup_and_release(n);
157 write_unlock(&n->lock);
162 tbl->last_flush = jiffies;
164 write_unlock_bh(&tbl->lock);
169 static void neigh_add_timer(struct neighbour *n, unsigned long when)
172 if (unlikely(mod_timer(&n->timer, when))) {
173 printk("NEIGH: BUG, double timer add, state is %x\n",
179 static int neigh_del_timer(struct neighbour *n)
181 if ((n->nud_state & NUD_IN_TIMER) &&
182 del_timer(&n->timer)) {
189 static void pneigh_queue_purge(struct sk_buff_head *list)
193 while ((skb = skb_dequeue(list)) != NULL) {
199 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
202 struct neigh_hash_table *nht;
204 nht = rcu_dereference_protected(tbl->nht,
205 lockdep_is_held(&tbl->lock));
207 for (i = 0; i < (1 << nht->hash_shift); i++) {
209 struct neighbour __rcu **np = &nht->hash_buckets[i];
211 while ((n = rcu_dereference_protected(*np,
212 lockdep_is_held(&tbl->lock))) != NULL) {
213 if (dev && n->dev != dev) {
217 rcu_assign_pointer(*np,
218 rcu_dereference_protected(n->next,
219 lockdep_is_held(&tbl->lock)));
220 write_lock(&n->lock);
224 if (atomic_read(&n->refcnt) != 1) {
225 /* The most unpleasant situation.
226 We must destroy neighbour entry,
227 but someone still uses it.
229 The destroy will be delayed until
230 the last user releases us, but
231 we must kill timers etc. and move
234 skb_queue_purge(&n->arp_queue);
235 n->arp_queue_len_bytes = 0;
236 n->output = neigh_blackhole;
237 if (n->nud_state & NUD_VALID)
238 n->nud_state = NUD_NOARP;
240 n->nud_state = NUD_NONE;
241 neigh_dbg(2, "neigh %p is stray\n", n);
243 write_unlock(&n->lock);
244 neigh_cleanup_and_release(n);
249 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
251 write_lock_bh(&tbl->lock);
252 neigh_flush_dev(tbl, dev);
253 write_unlock_bh(&tbl->lock);
255 EXPORT_SYMBOL(neigh_changeaddr);
257 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
259 write_lock_bh(&tbl->lock);
260 neigh_flush_dev(tbl, dev);
261 pneigh_ifdown(tbl, dev);
262 write_unlock_bh(&tbl->lock);
264 del_timer_sync(&tbl->proxy_timer);
265 pneigh_queue_purge(&tbl->proxy_queue);
268 EXPORT_SYMBOL(neigh_ifdown);
270 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
272 struct neighbour *n = NULL;
273 unsigned long now = jiffies;
276 entries = atomic_inc_return(&tbl->entries) - 1;
277 if (entries >= tbl->gc_thresh3 ||
278 (entries >= tbl->gc_thresh2 &&
279 time_after(now, tbl->last_flush + 5 * HZ))) {
280 if (!neigh_forced_gc(tbl) &&
281 entries >= tbl->gc_thresh3)
285 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
289 skb_queue_head_init(&n->arp_queue);
290 rwlock_init(&n->lock);
291 seqlock_init(&n->ha_lock);
292 n->updated = n->used = now;
293 n->nud_state = NUD_NONE;
294 n->output = neigh_blackhole;
295 seqlock_init(&n->hh.hh_lock);
296 n->parms = neigh_parms_clone(&tbl->parms);
297 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
299 NEIGH_CACHE_STAT_INC(tbl, allocs);
301 atomic_set(&n->refcnt, 1);
307 atomic_dec(&tbl->entries);
311 static void neigh_get_hash_rnd(u32 *x)
313 get_random_bytes(x, sizeof(*x));
317 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
319 size_t size = (1 << shift) * sizeof(struct neighbour *);
320 struct neigh_hash_table *ret;
321 struct neighbour __rcu **buckets;
324 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
327 if (size <= PAGE_SIZE)
328 buckets = kzalloc(size, GFP_ATOMIC);
330 buckets = (struct neighbour __rcu **)
331 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
337 ret->hash_buckets = buckets;
338 ret->hash_shift = shift;
339 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
340 neigh_get_hash_rnd(&ret->hash_rnd[i]);
344 static void neigh_hash_free_rcu(struct rcu_head *head)
346 struct neigh_hash_table *nht = container_of(head,
347 struct neigh_hash_table,
349 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
350 struct neighbour __rcu **buckets = nht->hash_buckets;
352 if (size <= PAGE_SIZE)
355 free_pages((unsigned long)buckets, get_order(size));
359 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
360 unsigned long new_shift)
362 unsigned int i, hash;
363 struct neigh_hash_table *new_nht, *old_nht;
365 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
367 old_nht = rcu_dereference_protected(tbl->nht,
368 lockdep_is_held(&tbl->lock));
369 new_nht = neigh_hash_alloc(new_shift);
373 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
374 struct neighbour *n, *next;
376 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
377 lockdep_is_held(&tbl->lock));
380 hash = tbl->hash(n->primary_key, n->dev,
383 hash >>= (32 - new_nht->hash_shift);
384 next = rcu_dereference_protected(n->next,
385 lockdep_is_held(&tbl->lock));
387 rcu_assign_pointer(n->next,
388 rcu_dereference_protected(
389 new_nht->hash_buckets[hash],
390 lockdep_is_held(&tbl->lock)));
391 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
395 rcu_assign_pointer(tbl->nht, new_nht);
396 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
400 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
401 struct net_device *dev)
404 int key_len = tbl->key_len;
406 struct neigh_hash_table *nht;
408 NEIGH_CACHE_STAT_INC(tbl, lookups);
411 nht = rcu_dereference_bh(tbl->nht);
412 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
414 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
416 n = rcu_dereference_bh(n->next)) {
417 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
418 if (!atomic_inc_not_zero(&n->refcnt))
420 NEIGH_CACHE_STAT_INC(tbl, hits);
425 rcu_read_unlock_bh();
428 EXPORT_SYMBOL(neigh_lookup);
430 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
434 int key_len = tbl->key_len;
436 struct neigh_hash_table *nht;
438 NEIGH_CACHE_STAT_INC(tbl, lookups);
441 nht = rcu_dereference_bh(tbl->nht);
442 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
444 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
446 n = rcu_dereference_bh(n->next)) {
447 if (!memcmp(n->primary_key, pkey, key_len) &&
448 net_eq(dev_net(n->dev), net)) {
449 if (!atomic_inc_not_zero(&n->refcnt))
451 NEIGH_CACHE_STAT_INC(tbl, hits);
456 rcu_read_unlock_bh();
459 EXPORT_SYMBOL(neigh_lookup_nodev);
461 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
462 struct net_device *dev, bool want_ref)
465 int key_len = tbl->key_len;
467 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
468 struct neigh_hash_table *nht;
471 rc = ERR_PTR(-ENOBUFS);
475 memcpy(n->primary_key, pkey, key_len);
479 /* Protocol specific setup. */
480 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
482 goto out_neigh_release;
485 if (dev->netdev_ops->ndo_neigh_construct) {
486 error = dev->netdev_ops->ndo_neigh_construct(n);
489 goto out_neigh_release;
493 /* Device specific setup. */
494 if (n->parms->neigh_setup &&
495 (error = n->parms->neigh_setup(n)) < 0) {
497 goto out_neigh_release;
500 n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
502 write_lock_bh(&tbl->lock);
503 nht = rcu_dereference_protected(tbl->nht,
504 lockdep_is_held(&tbl->lock));
506 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
507 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
509 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
511 if (n->parms->dead) {
512 rc = ERR_PTR(-EINVAL);
516 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
517 lockdep_is_held(&tbl->lock));
519 n1 = rcu_dereference_protected(n1->next,
520 lockdep_is_held(&tbl->lock))) {
521 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
532 rcu_assign_pointer(n->next,
533 rcu_dereference_protected(nht->hash_buckets[hash_val],
534 lockdep_is_held(&tbl->lock)));
535 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
536 write_unlock_bh(&tbl->lock);
537 neigh_dbg(2, "neigh %p is created\n", n);
542 write_unlock_bh(&tbl->lock);
547 EXPORT_SYMBOL(__neigh_create);
549 static u32 pneigh_hash(const void *pkey, int key_len)
551 u32 hash_val = *(u32 *)(pkey + key_len - 4);
552 hash_val ^= (hash_val >> 16);
553 hash_val ^= hash_val >> 8;
554 hash_val ^= hash_val >> 4;
555 hash_val &= PNEIGH_HASHMASK;
559 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
563 struct net_device *dev)
566 if (!memcmp(n->key, pkey, key_len) &&
567 net_eq(pneigh_net(n), net) &&
568 (n->dev == dev || !n->dev))
575 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
576 struct net *net, const void *pkey, struct net_device *dev)
578 int key_len = tbl->key_len;
579 u32 hash_val = pneigh_hash(pkey, key_len);
581 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
582 net, pkey, key_len, dev);
584 EXPORT_SYMBOL_GPL(__pneigh_lookup);
586 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
587 struct net *net, const void *pkey,
588 struct net_device *dev, int creat)
590 struct pneigh_entry *n;
591 int key_len = tbl->key_len;
592 u32 hash_val = pneigh_hash(pkey, key_len);
594 read_lock_bh(&tbl->lock);
595 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
596 net, pkey, key_len, dev);
597 read_unlock_bh(&tbl->lock);
604 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
608 write_pnet(&n->net, hold_net(net));
609 memcpy(n->key, pkey, key_len);
614 if (tbl->pconstructor && tbl->pconstructor(n)) {
623 write_lock_bh(&tbl->lock);
624 n->next = tbl->phash_buckets[hash_val];
625 tbl->phash_buckets[hash_val] = n;
626 write_unlock_bh(&tbl->lock);
630 EXPORT_SYMBOL(pneigh_lookup);
633 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
634 struct net_device *dev)
636 struct pneigh_entry *n, **np;
637 int key_len = tbl->key_len;
638 u32 hash_val = pneigh_hash(pkey, key_len);
640 write_lock_bh(&tbl->lock);
641 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
643 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
644 net_eq(pneigh_net(n), net)) {
646 write_unlock_bh(&tbl->lock);
647 if (tbl->pdestructor)
651 release_net(pneigh_net(n));
656 write_unlock_bh(&tbl->lock);
660 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
662 struct pneigh_entry *n, **np;
665 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
666 np = &tbl->phash_buckets[h];
667 while ((n = *np) != NULL) {
668 if (!dev || n->dev == dev) {
670 if (tbl->pdestructor)
674 release_net(pneigh_net(n));
684 static void neigh_parms_destroy(struct neigh_parms *parms);
686 static inline void neigh_parms_put(struct neigh_parms *parms)
688 if (atomic_dec_and_test(&parms->refcnt))
689 neigh_parms_destroy(parms);
693 * neighbour must already be out of the table;
696 void neigh_destroy(struct neighbour *neigh)
698 struct net_device *dev = neigh->dev;
700 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
703 pr_warn("Destroying alive neighbour %p\n", neigh);
708 if (neigh_del_timer(neigh))
709 pr_warn("Impossible event\n");
711 skb_queue_purge(&neigh->arp_queue);
712 neigh->arp_queue_len_bytes = 0;
714 if (dev->netdev_ops->ndo_neigh_destroy)
715 dev->netdev_ops->ndo_neigh_destroy(neigh);
718 neigh_parms_put(neigh->parms);
720 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
722 atomic_dec(&neigh->tbl->entries);
723 kfree_rcu(neigh, 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)
734 neigh_dbg(2, "neigh %p is suspected\n", neigh);
736 neigh->output = neigh->ops->output;
739 /* Neighbour state is OK;
742 Called with write_locked neigh.
744 static void neigh_connect(struct neighbour *neigh)
746 neigh_dbg(2, "neigh %p is connected\n", neigh);
748 neigh->output = neigh->ops->connected_output;
751 static void neigh_periodic_work(struct work_struct *work)
753 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
755 struct neighbour __rcu **np;
757 struct neigh_hash_table *nht;
759 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
761 write_lock_bh(&tbl->lock);
762 nht = rcu_dereference_protected(tbl->nht,
763 lockdep_is_held(&tbl->lock));
765 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
769 * periodically recompute ReachableTime from random function
772 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
773 struct neigh_parms *p;
774 tbl->last_rand = jiffies;
775 for (p = &tbl->parms; p; p = p->next)
777 neigh_rand_reach_time(p->base_reachable_time);
780 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
781 np = &nht->hash_buckets[i];
783 while ((n = rcu_dereference_protected(*np,
784 lockdep_is_held(&tbl->lock))) != NULL) {
787 write_lock(&n->lock);
789 state = n->nud_state;
790 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
791 write_unlock(&n->lock);
795 if (time_before(n->used, n->confirmed))
796 n->used = n->confirmed;
798 if (atomic_read(&n->refcnt) == 1 &&
799 (state == NUD_FAILED ||
800 time_after(jiffies, n->used + n->parms->gc_staletime))) {
803 write_unlock(&n->lock);
804 neigh_cleanup_and_release(n);
807 write_unlock(&n->lock);
813 * It's fine to release lock here, even if hash table
814 * grows while we are preempted.
816 write_unlock_bh(&tbl->lock);
818 write_lock_bh(&tbl->lock);
819 nht = rcu_dereference_protected(tbl->nht,
820 lockdep_is_held(&tbl->lock));
823 /* Cycle through all hash buckets every base_reachable_time/2 ticks.
824 * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
825 * base_reachable_time.
827 schedule_delayed_work(&tbl->gc_work,
828 tbl->parms.base_reachable_time >> 1);
829 write_unlock_bh(&tbl->lock);
832 static __inline__ int neigh_max_probes(struct neighbour *n)
834 struct neigh_parms *p = n->parms;
835 return (n->nud_state & NUD_PROBE) ?
837 p->ucast_probes + p->app_probes + p->mcast_probes;
840 static void neigh_invalidate(struct neighbour *neigh)
841 __releases(neigh->lock)
842 __acquires(neigh->lock)
846 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
847 neigh_dbg(2, "neigh %p is failed\n", neigh);
848 neigh->updated = jiffies;
850 /* It is very thin place. report_unreachable is very complicated
851 routine. Particularly, it can hit the same neighbour entry!
853 So that, we try to be accurate and avoid dead loop. --ANK
855 while (neigh->nud_state == NUD_FAILED &&
856 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
857 write_unlock(&neigh->lock);
858 neigh->ops->error_report(neigh, skb);
859 write_lock(&neigh->lock);
861 skb_queue_purge(&neigh->arp_queue);
862 neigh->arp_queue_len_bytes = 0;
865 static void neigh_probe(struct neighbour *neigh)
866 __releases(neigh->lock)
868 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
869 /* keep skb alive even if arp_queue overflows */
871 skb = skb_copy(skb, GFP_ATOMIC);
872 write_unlock(&neigh->lock);
873 neigh->ops->solicit(neigh, skb);
874 atomic_inc(&neigh->probes);
878 /* Called when a timer expires for a neighbour entry. */
880 static void neigh_timer_handler(unsigned long arg)
882 unsigned long now, next;
883 struct neighbour *neigh = (struct neighbour *)arg;
887 write_lock(&neigh->lock);
889 state = neigh->nud_state;
893 if (!(state & NUD_IN_TIMER))
896 if (state & NUD_REACHABLE) {
897 if (time_before_eq(now,
898 neigh->confirmed + neigh->parms->reachable_time)) {
899 neigh_dbg(2, "neigh %p is still alive\n", neigh);
900 next = neigh->confirmed + neigh->parms->reachable_time;
901 } else if (time_before_eq(now,
902 neigh->used + neigh->parms->delay_probe_time)) {
903 neigh_dbg(2, "neigh %p is delayed\n", neigh);
904 neigh->nud_state = NUD_DELAY;
905 neigh->updated = jiffies;
906 neigh_suspect(neigh);
907 next = now + neigh->parms->delay_probe_time;
909 neigh_dbg(2, "neigh %p is suspected\n", neigh);
910 neigh->nud_state = NUD_STALE;
911 neigh->updated = jiffies;
912 neigh_suspect(neigh);
915 } else if (state & NUD_DELAY) {
916 if (time_before_eq(now,
917 neigh->confirmed + neigh->parms->delay_probe_time)) {
918 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
919 neigh->nud_state = NUD_REACHABLE;
920 neigh->updated = jiffies;
921 neigh_connect(neigh);
923 next = neigh->confirmed + neigh->parms->reachable_time;
925 neigh_dbg(2, "neigh %p is probed\n", neigh);
926 neigh->nud_state = NUD_PROBE;
927 neigh->updated = jiffies;
928 atomic_set(&neigh->probes, 0);
929 next = now + neigh->parms->retrans_time;
932 /* NUD_PROBE|NUD_INCOMPLETE */
933 next = now + neigh->parms->retrans_time;
936 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
937 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
938 neigh->nud_state = NUD_FAILED;
940 neigh_invalidate(neigh);
943 if (neigh->nud_state & NUD_IN_TIMER) {
944 if (time_before(next, jiffies + HZ/2))
945 next = jiffies + HZ/2;
946 if (!mod_timer(&neigh->timer, next))
949 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
953 write_unlock(&neigh->lock);
957 neigh_update_notify(neigh);
959 neigh_release(neigh);
962 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
965 bool immediate_probe = false;
967 write_lock_bh(&neigh->lock);
970 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
973 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
974 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
975 unsigned long next, now = jiffies;
977 atomic_set(&neigh->probes, neigh->parms->ucast_probes);
978 neigh->nud_state = NUD_INCOMPLETE;
979 neigh->updated = now;
980 next = now + max(neigh->parms->retrans_time, HZ/2);
981 neigh_add_timer(neigh, next);
982 immediate_probe = true;
984 neigh->nud_state = NUD_FAILED;
985 neigh->updated = jiffies;
986 write_unlock_bh(&neigh->lock);
991 } else if (neigh->nud_state & NUD_STALE) {
992 neigh_dbg(2, "neigh %p is delayed\n", neigh);
993 neigh->nud_state = NUD_DELAY;
994 neigh->updated = jiffies;
995 neigh_add_timer(neigh,
996 jiffies + neigh->parms->delay_probe_time);
999 if (neigh->nud_state == NUD_INCOMPLETE) {
1001 while (neigh->arp_queue_len_bytes + skb->truesize >
1002 neigh->parms->queue_len_bytes) {
1003 struct sk_buff *buff;
1005 buff = __skb_dequeue(&neigh->arp_queue);
1008 neigh->arp_queue_len_bytes -= buff->truesize;
1010 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1013 __skb_queue_tail(&neigh->arp_queue, skb);
1014 neigh->arp_queue_len_bytes += skb->truesize;
1019 if (immediate_probe)
1022 write_unlock(&neigh->lock);
1026 EXPORT_SYMBOL(__neigh_event_send);
1028 static void neigh_update_hhs(struct neighbour *neigh)
1030 struct hh_cache *hh;
1031 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1034 if (neigh->dev->header_ops)
1035 update = neigh->dev->header_ops->cache_update;
1040 write_seqlock_bh(&hh->hh_lock);
1041 update(hh, neigh->dev, neigh->ha);
1042 write_sequnlock_bh(&hh->hh_lock);
1049 /* Generic update routine.
1050 -- lladdr is new lladdr or NULL, if it is not supplied.
1051 -- new is new state.
1053 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1055 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1056 lladdr instead of overriding it
1058 It also allows to retain current state
1059 if lladdr is unchanged.
1060 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1062 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1064 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1067 Caller MUST hold reference count on the entry.
1070 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1076 struct net_device *dev;
1077 int update_isrouter = 0;
1079 write_lock_bh(&neigh->lock);
1082 old = neigh->nud_state;
1085 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1086 (old & (NUD_NOARP | NUD_PERMANENT)))
1089 if (!(new & NUD_VALID)) {
1090 neigh_del_timer(neigh);
1091 if (old & NUD_CONNECTED)
1092 neigh_suspect(neigh);
1093 neigh->nud_state = new;
1095 notify = old & NUD_VALID;
1096 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1097 (new & NUD_FAILED)) {
1098 neigh_invalidate(neigh);
1104 /* Compare new lladdr with cached one */
1105 if (!dev->addr_len) {
1106 /* First case: device needs no address. */
1108 } else if (lladdr) {
1109 /* The second case: if something is already cached
1110 and a new address is proposed:
1112 - if they are different, check override flag
1114 if ((old & NUD_VALID) &&
1115 !memcmp(lladdr, neigh->ha, dev->addr_len))
1118 /* No address is supplied; if we know something,
1119 use it, otherwise discard the request.
1122 if (!(old & NUD_VALID))
1127 if (new & NUD_CONNECTED)
1128 neigh->confirmed = jiffies;
1129 neigh->updated = jiffies;
1131 /* If entry was valid and address is not changed,
1132 do not change entry state, if new one is STALE.
1135 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1136 if (old & NUD_VALID) {
1137 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1138 update_isrouter = 0;
1139 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1140 (old & NUD_CONNECTED)) {
1146 if (lladdr == neigh->ha && new == NUD_STALE &&
1147 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1148 (old & NUD_CONNECTED))
1155 neigh_del_timer(neigh);
1156 if (new & NUD_IN_TIMER)
1157 neigh_add_timer(neigh, (jiffies +
1158 ((new & NUD_REACHABLE) ?
1159 neigh->parms->reachable_time :
1161 neigh->nud_state = new;
1164 if (lladdr != neigh->ha) {
1165 write_seqlock(&neigh->ha_lock);
1166 memcpy(&neigh->ha, lladdr, dev->addr_len);
1167 write_sequnlock(&neigh->ha_lock);
1168 neigh_update_hhs(neigh);
1169 if (!(new & NUD_CONNECTED))
1170 neigh->confirmed = jiffies -
1171 (neigh->parms->base_reachable_time << 1);
1176 if (new & NUD_CONNECTED)
1177 neigh_connect(neigh);
1179 neigh_suspect(neigh);
1180 if (!(old & NUD_VALID)) {
1181 struct sk_buff *skb;
1183 /* Again: avoid dead loop if something went wrong */
1185 while (neigh->nud_state & NUD_VALID &&
1186 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1187 struct dst_entry *dst = skb_dst(skb);
1188 struct neighbour *n2, *n1 = neigh;
1189 write_unlock_bh(&neigh->lock);
1193 /* Why not just use 'neigh' as-is? The problem is that
1194 * things such as shaper, eql, and sch_teql can end up
1195 * using alternative, different, neigh objects to output
1196 * the packet in the output path. So what we need to do
1197 * here is re-lookup the top-level neigh in the path so
1198 * we can reinject the packet there.
1202 n2 = dst_neigh_lookup_skb(dst, skb);
1206 n1->output(n1, skb);
1211 write_lock_bh(&neigh->lock);
1213 skb_queue_purge(&neigh->arp_queue);
1214 neigh->arp_queue_len_bytes = 0;
1217 if (update_isrouter) {
1218 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1219 (neigh->flags | NTF_ROUTER) :
1220 (neigh->flags & ~NTF_ROUTER);
1222 write_unlock_bh(&neigh->lock);
1225 neigh_update_notify(neigh);
1229 EXPORT_SYMBOL(neigh_update);
1231 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1232 u8 *lladdr, void *saddr,
1233 struct net_device *dev)
1235 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1236 lladdr || !dev->addr_len);
1238 neigh_update(neigh, lladdr, NUD_STALE,
1239 NEIGH_UPDATE_F_OVERRIDE);
1242 EXPORT_SYMBOL(neigh_event_ns);
1244 /* called with read_lock_bh(&n->lock); */
1245 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1247 struct net_device *dev = dst->dev;
1248 __be16 prot = dst->ops->protocol;
1249 struct hh_cache *hh = &n->hh;
1251 write_lock_bh(&n->lock);
1253 /* Only one thread can come in here and initialize the
1257 dev->header_ops->cache(n, hh, prot);
1259 write_unlock_bh(&n->lock);
1262 /* This function can be used in contexts, where only old dev_queue_xmit
1263 * worked, f.e. if you want to override normal output path (eql, shaper),
1264 * but resolution is not made yet.
1267 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1269 struct net_device *dev = skb->dev;
1271 __skb_pull(skb, skb_network_offset(skb));
1273 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1275 dev->header_ops->rebuild(skb))
1278 return dev_queue_xmit(skb);
1280 EXPORT_SYMBOL(neigh_compat_output);
1282 /* Slow and careful. */
1284 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1286 struct dst_entry *dst = skb_dst(skb);
1292 if (!neigh_event_send(neigh, skb)) {
1294 struct net_device *dev = neigh->dev;
1297 if (dev->header_ops->cache && !neigh->hh.hh_len)
1298 neigh_hh_init(neigh, dst);
1301 __skb_pull(skb, skb_network_offset(skb));
1302 seq = read_seqbegin(&neigh->ha_lock);
1303 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1304 neigh->ha, NULL, skb->len);
1305 } while (read_seqretry(&neigh->ha_lock, seq));
1308 rc = dev_queue_xmit(skb);
1315 neigh_dbg(1, "%s: dst=%p neigh=%p\n", __func__, dst, neigh);
1321 EXPORT_SYMBOL(neigh_resolve_output);
1323 /* As fast as possible without hh cache */
1325 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1327 struct net_device *dev = neigh->dev;
1332 __skb_pull(skb, skb_network_offset(skb));
1333 seq = read_seqbegin(&neigh->ha_lock);
1334 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1335 neigh->ha, NULL, skb->len);
1336 } while (read_seqretry(&neigh->ha_lock, seq));
1339 err = dev_queue_xmit(skb);
1346 EXPORT_SYMBOL(neigh_connected_output);
1348 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1350 return dev_queue_xmit(skb);
1352 EXPORT_SYMBOL(neigh_direct_output);
1354 static void neigh_proxy_process(unsigned long arg)
1356 struct neigh_table *tbl = (struct neigh_table *)arg;
1357 long sched_next = 0;
1358 unsigned long now = jiffies;
1359 struct sk_buff *skb, *n;
1361 spin_lock(&tbl->proxy_queue.lock);
1363 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1364 long tdif = NEIGH_CB(skb)->sched_next - now;
1367 struct net_device *dev = skb->dev;
1369 __skb_unlink(skb, &tbl->proxy_queue);
1370 if (tbl->proxy_redo && netif_running(dev)) {
1372 tbl->proxy_redo(skb);
1379 } else if (!sched_next || tdif < sched_next)
1382 del_timer(&tbl->proxy_timer);
1384 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1385 spin_unlock(&tbl->proxy_queue.lock);
1388 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1389 struct sk_buff *skb)
1391 unsigned long now = jiffies;
1392 unsigned long sched_next = now + (net_random() % p->proxy_delay);
1394 if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1399 NEIGH_CB(skb)->sched_next = sched_next;
1400 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1402 spin_lock(&tbl->proxy_queue.lock);
1403 if (del_timer(&tbl->proxy_timer)) {
1404 if (time_before(tbl->proxy_timer.expires, sched_next))
1405 sched_next = tbl->proxy_timer.expires;
1409 __skb_queue_tail(&tbl->proxy_queue, skb);
1410 mod_timer(&tbl->proxy_timer, sched_next);
1411 spin_unlock(&tbl->proxy_queue.lock);
1413 EXPORT_SYMBOL(pneigh_enqueue);
1415 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1416 struct net *net, int ifindex)
1418 struct neigh_parms *p;
1420 for (p = &tbl->parms; p; p = p->next) {
1421 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1422 (!p->dev && !ifindex && net_eq(net, &init_net)))
1429 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1430 struct neigh_table *tbl)
1432 struct neigh_parms *p;
1433 struct net *net = dev_net(dev);
1434 const struct net_device_ops *ops = dev->netdev_ops;
1436 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1439 atomic_set(&p->refcnt, 1);
1441 neigh_rand_reach_time(p->base_reachable_time);
1443 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1450 write_pnet(&p->net, hold_net(net));
1451 p->sysctl_table = NULL;
1452 write_lock_bh(&tbl->lock);
1453 p->next = tbl->parms.next;
1454 tbl->parms.next = p;
1455 write_unlock_bh(&tbl->lock);
1459 EXPORT_SYMBOL(neigh_parms_alloc);
1461 static void neigh_rcu_free_parms(struct rcu_head *head)
1463 struct neigh_parms *parms =
1464 container_of(head, struct neigh_parms, rcu_head);
1466 neigh_parms_put(parms);
1469 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1471 struct neigh_parms **p;
1473 if (!parms || parms == &tbl->parms)
1475 write_lock_bh(&tbl->lock);
1476 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1480 write_unlock_bh(&tbl->lock);
1482 dev_put(parms->dev);
1483 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1487 write_unlock_bh(&tbl->lock);
1488 neigh_dbg(1, "%s: not found\n", __func__);
1490 EXPORT_SYMBOL(neigh_parms_release);
1492 static void neigh_parms_destroy(struct neigh_parms *parms)
1494 release_net(neigh_parms_net(parms));
1498 static struct lock_class_key neigh_table_proxy_queue_class;
1500 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1502 unsigned long now = jiffies;
1503 unsigned long phsize;
1505 write_pnet(&tbl->parms.net, &init_net);
1506 atomic_set(&tbl->parms.refcnt, 1);
1507 tbl->parms.reachable_time =
1508 neigh_rand_reach_time(tbl->parms.base_reachable_time);
1510 tbl->stats = alloc_percpu(struct neigh_statistics);
1512 panic("cannot create neighbour cache statistics");
1514 #ifdef CONFIG_PROC_FS
1515 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1516 &neigh_stat_seq_fops, tbl))
1517 panic("cannot create neighbour proc dir entry");
1520 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1522 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1523 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1525 if (!tbl->nht || !tbl->phash_buckets)
1526 panic("cannot allocate neighbour cache hashes");
1528 if (!tbl->entry_size)
1529 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1530 tbl->key_len, NEIGH_PRIV_ALIGN);
1532 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1534 rwlock_init(&tbl->lock);
1535 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1536 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1537 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1538 skb_queue_head_init_class(&tbl->proxy_queue,
1539 &neigh_table_proxy_queue_class);
1541 tbl->last_flush = now;
1542 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1545 void neigh_table_init(struct neigh_table *tbl)
1547 struct neigh_table *tmp;
1549 neigh_table_init_no_netlink(tbl);
1550 write_lock(&neigh_tbl_lock);
1551 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1552 if (tmp->family == tbl->family)
1555 tbl->next = neigh_tables;
1557 write_unlock(&neigh_tbl_lock);
1559 if (unlikely(tmp)) {
1560 pr_err("Registering multiple tables for family %d\n",
1565 EXPORT_SYMBOL(neigh_table_init);
1567 int neigh_table_clear(struct neigh_table *tbl)
1569 struct neigh_table **tp;
1571 /* It is not clean... Fix it to unload IPv6 module safely */
1572 cancel_delayed_work_sync(&tbl->gc_work);
1573 del_timer_sync(&tbl->proxy_timer);
1574 pneigh_queue_purge(&tbl->proxy_queue);
1575 neigh_ifdown(tbl, NULL);
1576 if (atomic_read(&tbl->entries))
1577 pr_crit("neighbour leakage\n");
1578 write_lock(&neigh_tbl_lock);
1579 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1585 write_unlock(&neigh_tbl_lock);
1587 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1588 neigh_hash_free_rcu);
1591 kfree(tbl->phash_buckets);
1592 tbl->phash_buckets = NULL;
1594 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1596 free_percpu(tbl->stats);
1601 EXPORT_SYMBOL(neigh_table_clear);
1603 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1605 struct net *net = sock_net(skb->sk);
1607 struct nlattr *dst_attr;
1608 struct neigh_table *tbl;
1609 struct net_device *dev = NULL;
1613 if (nlmsg_len(nlh) < sizeof(*ndm))
1616 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1617 if (dst_attr == NULL)
1620 ndm = nlmsg_data(nlh);
1621 if (ndm->ndm_ifindex) {
1622 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1629 read_lock(&neigh_tbl_lock);
1630 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1631 struct neighbour *neigh;
1633 if (tbl->family != ndm->ndm_family)
1635 read_unlock(&neigh_tbl_lock);
1637 if (nla_len(dst_attr) < tbl->key_len)
1640 if (ndm->ndm_flags & NTF_PROXY) {
1641 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1648 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1649 if (neigh == NULL) {
1654 err = neigh_update(neigh, NULL, NUD_FAILED,
1655 NEIGH_UPDATE_F_OVERRIDE |
1656 NEIGH_UPDATE_F_ADMIN);
1657 neigh_release(neigh);
1660 read_unlock(&neigh_tbl_lock);
1661 err = -EAFNOSUPPORT;
1667 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1669 struct net *net = sock_net(skb->sk);
1671 struct nlattr *tb[NDA_MAX+1];
1672 struct neigh_table *tbl;
1673 struct net_device *dev = NULL;
1677 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1682 if (tb[NDA_DST] == NULL)
1685 ndm = nlmsg_data(nlh);
1686 if (ndm->ndm_ifindex) {
1687 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1693 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1697 read_lock(&neigh_tbl_lock);
1698 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1699 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1700 struct neighbour *neigh;
1703 if (tbl->family != ndm->ndm_family)
1705 read_unlock(&neigh_tbl_lock);
1707 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1709 dst = nla_data(tb[NDA_DST]);
1710 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1712 if (ndm->ndm_flags & NTF_PROXY) {
1713 struct pneigh_entry *pn;
1716 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1718 pn->flags = ndm->ndm_flags;
1727 neigh = neigh_lookup(tbl, dst, dev);
1728 if (neigh == NULL) {
1729 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1734 neigh = __neigh_lookup_errno(tbl, dst, dev);
1735 if (IS_ERR(neigh)) {
1736 err = PTR_ERR(neigh);
1740 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1742 neigh_release(neigh);
1746 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1747 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1750 if (ndm->ndm_flags & NTF_USE) {
1751 neigh_event_send(neigh, NULL);
1754 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1755 neigh_release(neigh);
1759 read_unlock(&neigh_tbl_lock);
1760 err = -EAFNOSUPPORT;
1765 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1767 struct nlattr *nest;
1769 nest = nla_nest_start(skb, NDTA_PARMS);
1774 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1775 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1776 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES, parms->queue_len_bytes) ||
1777 /* approximative value for deprecated QUEUE_LEN (in packets) */
1778 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1779 parms->queue_len_bytes / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1780 nla_put_u32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen) ||
1781 nla_put_u32(skb, NDTPA_APP_PROBES, parms->app_probes) ||
1782 nla_put_u32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes) ||
1783 nla_put_u32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes) ||
1784 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1785 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1786 parms->base_reachable_time) ||
1787 nla_put_msecs(skb, NDTPA_GC_STALETIME, parms->gc_staletime) ||
1788 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1789 parms->delay_probe_time) ||
1790 nla_put_msecs(skb, NDTPA_RETRANS_TIME, parms->retrans_time) ||
1791 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay) ||
1792 nla_put_msecs(skb, NDTPA_PROXY_DELAY, parms->proxy_delay) ||
1793 nla_put_msecs(skb, NDTPA_LOCKTIME, parms->locktime))
1794 goto nla_put_failure;
1795 return nla_nest_end(skb, nest);
1798 nla_nest_cancel(skb, nest);
1802 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1803 u32 pid, u32 seq, int type, int flags)
1805 struct nlmsghdr *nlh;
1806 struct ndtmsg *ndtmsg;
1808 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1812 ndtmsg = nlmsg_data(nlh);
1814 read_lock_bh(&tbl->lock);
1815 ndtmsg->ndtm_family = tbl->family;
1816 ndtmsg->ndtm_pad1 = 0;
1817 ndtmsg->ndtm_pad2 = 0;
1819 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1820 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1821 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1822 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1823 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1824 goto nla_put_failure;
1826 unsigned long now = jiffies;
1827 unsigned int flush_delta = now - tbl->last_flush;
1828 unsigned int rand_delta = now - tbl->last_rand;
1829 struct neigh_hash_table *nht;
1830 struct ndt_config ndc = {
1831 .ndtc_key_len = tbl->key_len,
1832 .ndtc_entry_size = tbl->entry_size,
1833 .ndtc_entries = atomic_read(&tbl->entries),
1834 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1835 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1836 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1840 nht = rcu_dereference_bh(tbl->nht);
1841 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1842 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1843 rcu_read_unlock_bh();
1845 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1846 goto nla_put_failure;
1851 struct ndt_stats ndst;
1853 memset(&ndst, 0, sizeof(ndst));
1855 for_each_possible_cpu(cpu) {
1856 struct neigh_statistics *st;
1858 st = per_cpu_ptr(tbl->stats, cpu);
1859 ndst.ndts_allocs += st->allocs;
1860 ndst.ndts_destroys += st->destroys;
1861 ndst.ndts_hash_grows += st->hash_grows;
1862 ndst.ndts_res_failed += st->res_failed;
1863 ndst.ndts_lookups += st->lookups;
1864 ndst.ndts_hits += st->hits;
1865 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1866 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1867 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1868 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1871 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1872 goto nla_put_failure;
1875 BUG_ON(tbl->parms.dev);
1876 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1877 goto nla_put_failure;
1879 read_unlock_bh(&tbl->lock);
1880 return nlmsg_end(skb, nlh);
1883 read_unlock_bh(&tbl->lock);
1884 nlmsg_cancel(skb, nlh);
1888 static int neightbl_fill_param_info(struct sk_buff *skb,
1889 struct neigh_table *tbl,
1890 struct neigh_parms *parms,
1891 u32 pid, u32 seq, int type,
1894 struct ndtmsg *ndtmsg;
1895 struct nlmsghdr *nlh;
1897 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1901 ndtmsg = nlmsg_data(nlh);
1903 read_lock_bh(&tbl->lock);
1904 ndtmsg->ndtm_family = tbl->family;
1905 ndtmsg->ndtm_pad1 = 0;
1906 ndtmsg->ndtm_pad2 = 0;
1908 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1909 neightbl_fill_parms(skb, parms) < 0)
1912 read_unlock_bh(&tbl->lock);
1913 return nlmsg_end(skb, nlh);
1915 read_unlock_bh(&tbl->lock);
1916 nlmsg_cancel(skb, nlh);
1920 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1921 [NDTA_NAME] = { .type = NLA_STRING },
1922 [NDTA_THRESH1] = { .type = NLA_U32 },
1923 [NDTA_THRESH2] = { .type = NLA_U32 },
1924 [NDTA_THRESH3] = { .type = NLA_U32 },
1925 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1926 [NDTA_PARMS] = { .type = NLA_NESTED },
1929 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1930 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1931 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1932 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1933 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1934 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1935 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1936 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1937 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1938 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1939 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1940 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1941 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1942 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1945 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1947 struct net *net = sock_net(skb->sk);
1948 struct neigh_table *tbl;
1949 struct ndtmsg *ndtmsg;
1950 struct nlattr *tb[NDTA_MAX+1];
1953 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1954 nl_neightbl_policy);
1958 if (tb[NDTA_NAME] == NULL) {
1963 ndtmsg = nlmsg_data(nlh);
1964 read_lock(&neigh_tbl_lock);
1965 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1966 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1969 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1979 * We acquire tbl->lock to be nice to the periodic timers and
1980 * make sure they always see a consistent set of values.
1982 write_lock_bh(&tbl->lock);
1984 if (tb[NDTA_PARMS]) {
1985 struct nlattr *tbp[NDTPA_MAX+1];
1986 struct neigh_parms *p;
1989 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1990 nl_ntbl_parm_policy);
1992 goto errout_tbl_lock;
1994 if (tbp[NDTPA_IFINDEX])
1995 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1997 p = lookup_neigh_parms(tbl, net, ifindex);
2000 goto errout_tbl_lock;
2003 for (i = 1; i <= NDTPA_MAX; i++) {
2008 case NDTPA_QUEUE_LEN:
2009 p->queue_len_bytes = nla_get_u32(tbp[i]) *
2010 SKB_TRUESIZE(ETH_FRAME_LEN);
2012 case NDTPA_QUEUE_LENBYTES:
2013 p->queue_len_bytes = nla_get_u32(tbp[i]);
2015 case NDTPA_PROXY_QLEN:
2016 p->proxy_qlen = nla_get_u32(tbp[i]);
2018 case NDTPA_APP_PROBES:
2019 p->app_probes = nla_get_u32(tbp[i]);
2021 case NDTPA_UCAST_PROBES:
2022 p->ucast_probes = nla_get_u32(tbp[i]);
2024 case NDTPA_MCAST_PROBES:
2025 p->mcast_probes = nla_get_u32(tbp[i]);
2027 case NDTPA_BASE_REACHABLE_TIME:
2028 p->base_reachable_time = nla_get_msecs(tbp[i]);
2030 case NDTPA_GC_STALETIME:
2031 p->gc_staletime = nla_get_msecs(tbp[i]);
2033 case NDTPA_DELAY_PROBE_TIME:
2034 p->delay_probe_time = nla_get_msecs(tbp[i]);
2036 case NDTPA_RETRANS_TIME:
2037 p->retrans_time = nla_get_msecs(tbp[i]);
2039 case NDTPA_ANYCAST_DELAY:
2040 p->anycast_delay = nla_get_msecs(tbp[i]);
2042 case NDTPA_PROXY_DELAY:
2043 p->proxy_delay = nla_get_msecs(tbp[i]);
2045 case NDTPA_LOCKTIME:
2046 p->locktime = nla_get_msecs(tbp[i]);
2052 if (tb[NDTA_THRESH1])
2053 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2055 if (tb[NDTA_THRESH2])
2056 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2058 if (tb[NDTA_THRESH3])
2059 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2061 if (tb[NDTA_GC_INTERVAL])
2062 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2067 write_unlock_bh(&tbl->lock);
2069 read_unlock(&neigh_tbl_lock);
2074 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2076 struct net *net = sock_net(skb->sk);
2077 int family, tidx, nidx = 0;
2078 int tbl_skip = cb->args[0];
2079 int neigh_skip = cb->args[1];
2080 struct neigh_table *tbl;
2082 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2084 read_lock(&neigh_tbl_lock);
2085 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2086 struct neigh_parms *p;
2088 if (tidx < tbl_skip || (family && tbl->family != family))
2091 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2092 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2096 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2097 if (!net_eq(neigh_parms_net(p), net))
2100 if (nidx < neigh_skip)
2103 if (neightbl_fill_param_info(skb, tbl, p,
2104 NETLINK_CB(cb->skb).portid,
2116 read_unlock(&neigh_tbl_lock);
2123 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2124 u32 pid, u32 seq, int type, unsigned int flags)
2126 unsigned long now = jiffies;
2127 struct nda_cacheinfo ci;
2128 struct nlmsghdr *nlh;
2131 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2135 ndm = nlmsg_data(nlh);
2136 ndm->ndm_family = neigh->ops->family;
2139 ndm->ndm_flags = neigh->flags;
2140 ndm->ndm_type = neigh->type;
2141 ndm->ndm_ifindex = neigh->dev->ifindex;
2143 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2144 goto nla_put_failure;
2146 read_lock_bh(&neigh->lock);
2147 ndm->ndm_state = neigh->nud_state;
2148 if (neigh->nud_state & NUD_VALID) {
2149 char haddr[MAX_ADDR_LEN];
2151 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2152 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2153 read_unlock_bh(&neigh->lock);
2154 goto nla_put_failure;
2158 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2159 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2160 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2161 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2162 read_unlock_bh(&neigh->lock);
2164 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2165 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2166 goto nla_put_failure;
2168 return nlmsg_end(skb, nlh);
2171 nlmsg_cancel(skb, nlh);
2175 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2176 u32 pid, u32 seq, int type, unsigned int flags,
2177 struct neigh_table *tbl)
2179 struct nlmsghdr *nlh;
2182 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2186 ndm = nlmsg_data(nlh);
2187 ndm->ndm_family = tbl->family;
2190 ndm->ndm_flags = pn->flags | NTF_PROXY;
2191 ndm->ndm_type = NDA_DST;
2192 ndm->ndm_ifindex = pn->dev->ifindex;
2193 ndm->ndm_state = NUD_NONE;
2195 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2196 goto nla_put_failure;
2198 return nlmsg_end(skb, nlh);
2201 nlmsg_cancel(skb, nlh);
2205 static void neigh_update_notify(struct neighbour *neigh)
2207 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2208 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2211 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2212 struct netlink_callback *cb)
2214 struct net *net = sock_net(skb->sk);
2215 struct neighbour *n;
2216 int rc, h, s_h = cb->args[1];
2217 int idx, s_idx = idx = cb->args[2];
2218 struct neigh_hash_table *nht;
2221 nht = rcu_dereference_bh(tbl->nht);
2223 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2226 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2228 n = rcu_dereference_bh(n->next)) {
2229 if (!net_eq(dev_net(n->dev), net))
2233 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2236 NLM_F_MULTI) <= 0) {
2246 rcu_read_unlock_bh();
2252 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2253 struct netlink_callback *cb)
2255 struct pneigh_entry *n;
2256 struct net *net = sock_net(skb->sk);
2257 int rc, h, s_h = cb->args[3];
2258 int idx, s_idx = idx = cb->args[4];
2260 read_lock_bh(&tbl->lock);
2262 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2265 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2266 if (dev_net(n->dev) != net)
2270 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2273 NLM_F_MULTI, tbl) <= 0) {
2274 read_unlock_bh(&tbl->lock);
2283 read_unlock_bh(&tbl->lock);
2292 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2294 struct neigh_table *tbl;
2299 read_lock(&neigh_tbl_lock);
2300 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2302 /* check for full ndmsg structure presence, family member is
2303 * the same for both structures
2305 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2306 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2311 for (tbl = neigh_tables, t = 0; tbl;
2312 tbl = tbl->next, t++) {
2313 if (t < s_t || (family && tbl->family != family))
2316 memset(&cb->args[1], 0, sizeof(cb->args) -
2317 sizeof(cb->args[0]));
2319 err = pneigh_dump_table(tbl, skb, cb);
2321 err = neigh_dump_table(tbl, skb, cb);
2325 read_unlock(&neigh_tbl_lock);
2331 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2334 struct neigh_hash_table *nht;
2337 nht = rcu_dereference_bh(tbl->nht);
2339 read_lock(&tbl->lock); /* avoid resizes */
2340 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2341 struct neighbour *n;
2343 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2345 n = rcu_dereference_bh(n->next))
2348 read_unlock(&tbl->lock);
2349 rcu_read_unlock_bh();
2351 EXPORT_SYMBOL(neigh_for_each);
2353 /* The tbl->lock must be held as a writer and BH disabled. */
2354 void __neigh_for_each_release(struct neigh_table *tbl,
2355 int (*cb)(struct neighbour *))
2358 struct neigh_hash_table *nht;
2360 nht = rcu_dereference_protected(tbl->nht,
2361 lockdep_is_held(&tbl->lock));
2362 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2363 struct neighbour *n;
2364 struct neighbour __rcu **np;
2366 np = &nht->hash_buckets[chain];
2367 while ((n = rcu_dereference_protected(*np,
2368 lockdep_is_held(&tbl->lock))) != NULL) {
2371 write_lock(&n->lock);
2374 rcu_assign_pointer(*np,
2375 rcu_dereference_protected(n->next,
2376 lockdep_is_held(&tbl->lock)));
2380 write_unlock(&n->lock);
2382 neigh_cleanup_and_release(n);
2386 EXPORT_SYMBOL(__neigh_for_each_release);
2388 #ifdef CONFIG_PROC_FS
2390 static struct neighbour *neigh_get_first(struct seq_file *seq)
2392 struct neigh_seq_state *state = seq->private;
2393 struct net *net = seq_file_net(seq);
2394 struct neigh_hash_table *nht = state->nht;
2395 struct neighbour *n = NULL;
2396 int bucket = state->bucket;
2398 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2399 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2400 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2403 if (!net_eq(dev_net(n->dev), net))
2405 if (state->neigh_sub_iter) {
2409 v = state->neigh_sub_iter(state, n, &fakep);
2413 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2415 if (n->nud_state & ~NUD_NOARP)
2418 n = rcu_dereference_bh(n->next);
2424 state->bucket = bucket;
2429 static struct neighbour *neigh_get_next(struct seq_file *seq,
2430 struct neighbour *n,
2433 struct neigh_seq_state *state = seq->private;
2434 struct net *net = seq_file_net(seq);
2435 struct neigh_hash_table *nht = state->nht;
2437 if (state->neigh_sub_iter) {
2438 void *v = state->neigh_sub_iter(state, n, pos);
2442 n = rcu_dereference_bh(n->next);
2446 if (!net_eq(dev_net(n->dev), net))
2448 if (state->neigh_sub_iter) {
2449 void *v = state->neigh_sub_iter(state, n, pos);
2454 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2457 if (n->nud_state & ~NUD_NOARP)
2460 n = rcu_dereference_bh(n->next);
2466 if (++state->bucket >= (1 << nht->hash_shift))
2469 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2477 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2479 struct neighbour *n = neigh_get_first(seq);
2484 n = neigh_get_next(seq, n, pos);
2489 return *pos ? NULL : n;
2492 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2494 struct neigh_seq_state *state = seq->private;
2495 struct net *net = seq_file_net(seq);
2496 struct neigh_table *tbl = state->tbl;
2497 struct pneigh_entry *pn = NULL;
2498 int bucket = state->bucket;
2500 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2501 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2502 pn = tbl->phash_buckets[bucket];
2503 while (pn && !net_eq(pneigh_net(pn), net))
2508 state->bucket = bucket;
2513 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2514 struct pneigh_entry *pn,
2517 struct neigh_seq_state *state = seq->private;
2518 struct net *net = seq_file_net(seq);
2519 struct neigh_table *tbl = state->tbl;
2523 } while (pn && !net_eq(pneigh_net(pn), net));
2526 if (++state->bucket > PNEIGH_HASHMASK)
2528 pn = tbl->phash_buckets[state->bucket];
2529 while (pn && !net_eq(pneigh_net(pn), net))
2541 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2543 struct pneigh_entry *pn = pneigh_get_first(seq);
2548 pn = pneigh_get_next(seq, pn, pos);
2553 return *pos ? NULL : pn;
2556 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2558 struct neigh_seq_state *state = seq->private;
2560 loff_t idxpos = *pos;
2562 rc = neigh_get_idx(seq, &idxpos);
2563 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2564 rc = pneigh_get_idx(seq, &idxpos);
2569 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2572 struct neigh_seq_state *state = seq->private;
2576 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2579 state->nht = rcu_dereference_bh(tbl->nht);
2581 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2583 EXPORT_SYMBOL(neigh_seq_start);
2585 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2587 struct neigh_seq_state *state;
2590 if (v == SEQ_START_TOKEN) {
2591 rc = neigh_get_first(seq);
2595 state = seq->private;
2596 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2597 rc = neigh_get_next(seq, v, NULL);
2600 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2601 rc = pneigh_get_first(seq);
2603 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2604 rc = pneigh_get_next(seq, v, NULL);
2610 EXPORT_SYMBOL(neigh_seq_next);
2612 void neigh_seq_stop(struct seq_file *seq, void *v)
2615 rcu_read_unlock_bh();
2617 EXPORT_SYMBOL(neigh_seq_stop);
2619 /* statistics via seq_file */
2621 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2623 struct neigh_table *tbl = seq->private;
2627 return SEQ_START_TOKEN;
2629 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2630 if (!cpu_possible(cpu))
2633 return per_cpu_ptr(tbl->stats, cpu);
2638 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2640 struct neigh_table *tbl = seq->private;
2643 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2644 if (!cpu_possible(cpu))
2647 return per_cpu_ptr(tbl->stats, cpu);
2652 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2657 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2659 struct neigh_table *tbl = seq->private;
2660 struct neigh_statistics *st = v;
2662 if (v == SEQ_START_TOKEN) {
2663 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");
2667 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2668 "%08lx %08lx %08lx %08lx %08lx\n",
2669 atomic_read(&tbl->entries),
2680 st->rcv_probes_mcast,
2681 st->rcv_probes_ucast,
2683 st->periodic_gc_runs,
2691 static const struct seq_operations neigh_stat_seq_ops = {
2692 .start = neigh_stat_seq_start,
2693 .next = neigh_stat_seq_next,
2694 .stop = neigh_stat_seq_stop,
2695 .show = neigh_stat_seq_show,
2698 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2700 int ret = seq_open(file, &neigh_stat_seq_ops);
2703 struct seq_file *sf = file->private_data;
2704 sf->private = PDE_DATA(inode);
2709 static const struct file_operations neigh_stat_seq_fops = {
2710 .owner = THIS_MODULE,
2711 .open = neigh_stat_seq_open,
2713 .llseek = seq_lseek,
2714 .release = seq_release,
2717 #endif /* CONFIG_PROC_FS */
2719 static inline size_t neigh_nlmsg_size(void)
2721 return NLMSG_ALIGN(sizeof(struct ndmsg))
2722 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2723 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2724 + nla_total_size(sizeof(struct nda_cacheinfo))
2725 + nla_total_size(4); /* NDA_PROBES */
2728 static void __neigh_notify(struct neighbour *n, int type, int flags)
2730 struct net *net = dev_net(n->dev);
2731 struct sk_buff *skb;
2734 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2738 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2740 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2741 WARN_ON(err == -EMSGSIZE);
2745 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2749 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2753 void neigh_app_ns(struct neighbour *n)
2755 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2757 EXPORT_SYMBOL(neigh_app_ns);
2758 #endif /* CONFIG_ARPD */
2760 #ifdef CONFIG_SYSCTL
2762 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2764 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2765 void __user *buffer, size_t *lenp, loff_t *ppos)
2768 struct ctl_table tmp = *ctl;
2771 tmp.extra2 = &unres_qlen_max;
2774 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2775 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2778 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2783 NEIGH_VAR_MCAST_PROBE,
2784 NEIGH_VAR_UCAST_PROBE,
2785 NEIGH_VAR_APP_PROBE,
2786 NEIGH_VAR_RETRANS_TIME,
2787 NEIGH_VAR_BASE_REACHABLE_TIME,
2788 NEIGH_VAR_DELAY_PROBE_TIME,
2789 NEIGH_VAR_GC_STALETIME,
2790 NEIGH_VAR_QUEUE_LEN,
2791 NEIGH_VAR_QUEUE_LEN_BYTES,
2792 NEIGH_VAR_PROXY_QLEN,
2793 NEIGH_VAR_ANYCAST_DELAY,
2794 NEIGH_VAR_PROXY_DELAY,
2796 NEIGH_VAR_RETRANS_TIME_MS,
2797 NEIGH_VAR_BASE_REACHABLE_TIME_MS,
2798 NEIGH_VAR_GC_INTERVAL,
2799 NEIGH_VAR_GC_THRESH1,
2800 NEIGH_VAR_GC_THRESH2,
2801 NEIGH_VAR_GC_THRESH3,
2805 static struct neigh_sysctl_table {
2806 struct ctl_table_header *sysctl_header;
2807 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2808 } neigh_sysctl_template __read_mostly = {
2810 [NEIGH_VAR_MCAST_PROBE] = {
2811 .procname = "mcast_solicit",
2812 .maxlen = sizeof(int),
2814 .proc_handler = proc_dointvec,
2816 [NEIGH_VAR_UCAST_PROBE] = {
2817 .procname = "ucast_solicit",
2818 .maxlen = sizeof(int),
2820 .proc_handler = proc_dointvec,
2822 [NEIGH_VAR_APP_PROBE] = {
2823 .procname = "app_solicit",
2824 .maxlen = sizeof(int),
2826 .proc_handler = proc_dointvec,
2828 [NEIGH_VAR_RETRANS_TIME] = {
2829 .procname = "retrans_time",
2830 .maxlen = sizeof(int),
2832 .proc_handler = proc_dointvec_userhz_jiffies,
2834 [NEIGH_VAR_BASE_REACHABLE_TIME] = {
2835 .procname = "base_reachable_time",
2836 .maxlen = sizeof(int),
2838 .proc_handler = proc_dointvec_jiffies,
2840 [NEIGH_VAR_DELAY_PROBE_TIME] = {
2841 .procname = "delay_first_probe_time",
2842 .maxlen = sizeof(int),
2844 .proc_handler = proc_dointvec_jiffies,
2846 [NEIGH_VAR_GC_STALETIME] = {
2847 .procname = "gc_stale_time",
2848 .maxlen = sizeof(int),
2850 .proc_handler = proc_dointvec_jiffies,
2852 [NEIGH_VAR_QUEUE_LEN] = {
2853 .procname = "unres_qlen",
2854 .maxlen = sizeof(int),
2856 .proc_handler = proc_unres_qlen,
2858 [NEIGH_VAR_QUEUE_LEN_BYTES] = {
2859 .procname = "unres_qlen_bytes",
2860 .maxlen = sizeof(int),
2863 .proc_handler = proc_dointvec_minmax,
2865 [NEIGH_VAR_PROXY_QLEN] = {
2866 .procname = "proxy_qlen",
2867 .maxlen = sizeof(int),
2869 .proc_handler = proc_dointvec,
2871 [NEIGH_VAR_ANYCAST_DELAY] = {
2872 .procname = "anycast_delay",
2873 .maxlen = sizeof(int),
2875 .proc_handler = proc_dointvec_userhz_jiffies,
2877 [NEIGH_VAR_PROXY_DELAY] = {
2878 .procname = "proxy_delay",
2879 .maxlen = sizeof(int),
2881 .proc_handler = proc_dointvec_userhz_jiffies,
2883 [NEIGH_VAR_LOCKTIME] = {
2884 .procname = "locktime",
2885 .maxlen = sizeof(int),
2887 .proc_handler = proc_dointvec_userhz_jiffies,
2889 [NEIGH_VAR_RETRANS_TIME_MS] = {
2890 .procname = "retrans_time_ms",
2891 .maxlen = sizeof(int),
2893 .proc_handler = proc_dointvec_ms_jiffies,
2895 [NEIGH_VAR_BASE_REACHABLE_TIME_MS] = {
2896 .procname = "base_reachable_time_ms",
2897 .maxlen = sizeof(int),
2899 .proc_handler = proc_dointvec_ms_jiffies,
2901 [NEIGH_VAR_GC_INTERVAL] = {
2902 .procname = "gc_interval",
2903 .maxlen = sizeof(int),
2905 .proc_handler = proc_dointvec_jiffies,
2907 [NEIGH_VAR_GC_THRESH1] = {
2908 .procname = "gc_thresh1",
2909 .maxlen = sizeof(int),
2911 .proc_handler = proc_dointvec,
2913 [NEIGH_VAR_GC_THRESH2] = {
2914 .procname = "gc_thresh2",
2915 .maxlen = sizeof(int),
2917 .proc_handler = proc_dointvec,
2919 [NEIGH_VAR_GC_THRESH3] = {
2920 .procname = "gc_thresh3",
2921 .maxlen = sizeof(int),
2923 .proc_handler = proc_dointvec,
2929 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2930 char *p_name, proc_handler *handler)
2932 struct neigh_sysctl_table *t;
2933 const char *dev_name_source = NULL;
2934 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
2936 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2940 t->neigh_vars[NEIGH_VAR_MCAST_PROBE].data = &p->mcast_probes;
2941 t->neigh_vars[NEIGH_VAR_UCAST_PROBE].data = &p->ucast_probes;
2942 t->neigh_vars[NEIGH_VAR_APP_PROBE].data = &p->app_probes;
2943 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].data = &p->retrans_time;
2944 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].data = &p->base_reachable_time;
2945 t->neigh_vars[NEIGH_VAR_DELAY_PROBE_TIME].data = &p->delay_probe_time;
2946 t->neigh_vars[NEIGH_VAR_GC_STALETIME].data = &p->gc_staletime;
2947 t->neigh_vars[NEIGH_VAR_QUEUE_LEN].data = &p->queue_len_bytes;
2948 t->neigh_vars[NEIGH_VAR_QUEUE_LEN_BYTES].data = &p->queue_len_bytes;
2949 t->neigh_vars[NEIGH_VAR_PROXY_QLEN].data = &p->proxy_qlen;
2950 t->neigh_vars[NEIGH_VAR_ANYCAST_DELAY].data = &p->anycast_delay;
2951 t->neigh_vars[NEIGH_VAR_PROXY_DELAY].data = &p->proxy_delay;
2952 t->neigh_vars[NEIGH_VAR_LOCKTIME].data = &p->locktime;
2953 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].data = &p->retrans_time;
2954 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].data = &p->base_reachable_time;
2957 dev_name_source = dev->name;
2958 /* Terminate the table early */
2959 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
2960 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
2962 dev_name_source = "default";
2963 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1);
2964 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1;
2965 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2;
2966 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3;
2972 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
2973 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].extra1 = dev;
2975 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
2976 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].extra1 = dev;
2977 /* RetransTime (in milliseconds)*/
2978 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
2979 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].extra1 = dev;
2980 /* ReachableTime (in milliseconds) */
2981 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
2982 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].extra1 = dev;
2985 /* Don't export sysctls to unprivileged users */
2986 if (neigh_parms_net(p)->user_ns != &init_user_ns)
2987 t->neigh_vars[0].procname = NULL;
2989 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
2990 p_name, dev_name_source);
2992 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
2993 if (!t->sysctl_header)
2996 p->sysctl_table = t;
3004 EXPORT_SYMBOL(neigh_sysctl_register);
3006 void neigh_sysctl_unregister(struct neigh_parms *p)
3008 if (p->sysctl_table) {
3009 struct neigh_sysctl_table *t = p->sysctl_table;
3010 p->sysctl_table = NULL;
3011 unregister_net_sysctl_table(t->sysctl_header);
3015 EXPORT_SYMBOL(neigh_sysctl_unregister);
3017 #endif /* CONFIG_SYSCTL */
3019 static int __init neigh_init(void)
3021 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3022 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3023 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3025 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3027 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3032 subsys_initcall(neigh_init);