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>
41 #include <linux/inetdevice.h>
42 #include <net/addrconf.h>
46 #define neigh_dbg(level, fmt, ...) \
48 if (level <= NEIGH_DEBUG) \
49 pr_debug(fmt, ##__VA_ARGS__); \
52 #define PNEIGH_HASHMASK 0xF
54 static void neigh_timer_handler(unsigned long arg);
55 static void __neigh_notify(struct neighbour *n, int type, int flags);
56 static void neigh_update_notify(struct neighbour *neigh);
57 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
59 static struct neigh_table *neigh_tables;
61 static const struct file_operations neigh_stat_seq_fops;
65 Neighbour hash table buckets are protected with rwlock tbl->lock.
67 - All the scans/updates to hash buckets MUST be made under this lock.
68 - NOTHING clever should be made under this lock: no callbacks
69 to protocol backends, no attempts to send something to network.
70 It will result in deadlocks, if backend/driver wants to use neighbour
72 - If the entry requires some non-trivial actions, increase
73 its reference count and release table lock.
75 Neighbour entries are protected:
76 - with reference count.
77 - with rwlock neigh->lock
79 Reference count prevents destruction.
81 neigh->lock mainly serializes ll address data and its validity state.
82 However, the same lock is used to protect another entry fields:
86 Again, nothing clever shall be made under neigh->lock,
87 the most complicated procedure, which we allow is dev->hard_header.
88 It is supposed, that dev->hard_header is simplistic and does
89 not make callbacks to neighbour tables.
91 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
92 list of neighbour tables. This list is used only in process context,
95 static DEFINE_RWLOCK(neigh_tbl_lock);
97 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
103 static void neigh_cleanup_and_release(struct neighbour *neigh)
105 if (neigh->parms->neigh_cleanup)
106 neigh->parms->neigh_cleanup(neigh);
108 __neigh_notify(neigh, RTM_DELNEIGH, 0);
109 neigh_release(neigh);
113 * It is random distribution in the interval (1/2)*base...(3/2)*base.
114 * It corresponds to default IPv6 settings and is not overridable,
115 * because it is really reasonable choice.
118 unsigned long neigh_rand_reach_time(unsigned long base)
120 return base ? (net_random() % base) + (base >> 1) : 0;
122 EXPORT_SYMBOL(neigh_rand_reach_time);
125 static int neigh_forced_gc(struct neigh_table *tbl)
129 struct neigh_hash_table *nht;
131 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
133 write_lock_bh(&tbl->lock);
134 nht = rcu_dereference_protected(tbl->nht,
135 lockdep_is_held(&tbl->lock));
136 for (i = 0; i < (1 << nht->hash_shift); i++) {
138 struct neighbour __rcu **np;
140 np = &nht->hash_buckets[i];
141 while ((n = rcu_dereference_protected(*np,
142 lockdep_is_held(&tbl->lock))) != NULL) {
143 /* Neighbour record may be discarded if:
144 * - nobody refers to it.
145 * - it is not permanent
147 write_lock(&n->lock);
148 if (atomic_read(&n->refcnt) == 1 &&
149 !(n->nud_state & NUD_PERMANENT)) {
150 rcu_assign_pointer(*np,
151 rcu_dereference_protected(n->next,
152 lockdep_is_held(&tbl->lock)));
155 write_unlock(&n->lock);
156 neigh_cleanup_and_release(n);
159 write_unlock(&n->lock);
164 tbl->last_flush = jiffies;
166 write_unlock_bh(&tbl->lock);
171 static void neigh_add_timer(struct neighbour *n, unsigned long when)
174 if (unlikely(mod_timer(&n->timer, when))) {
175 printk("NEIGH: BUG, double timer add, state is %x\n",
181 static int neigh_del_timer(struct neighbour *n)
183 if ((n->nud_state & NUD_IN_TIMER) &&
184 del_timer(&n->timer)) {
191 static void pneigh_queue_purge(struct sk_buff_head *list)
195 while ((skb = skb_dequeue(list)) != NULL) {
201 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
204 struct neigh_hash_table *nht;
206 nht = rcu_dereference_protected(tbl->nht,
207 lockdep_is_held(&tbl->lock));
209 for (i = 0; i < (1 << nht->hash_shift); i++) {
211 struct neighbour __rcu **np = &nht->hash_buckets[i];
213 while ((n = rcu_dereference_protected(*np,
214 lockdep_is_held(&tbl->lock))) != NULL) {
215 if (dev && n->dev != dev) {
219 rcu_assign_pointer(*np,
220 rcu_dereference_protected(n->next,
221 lockdep_is_held(&tbl->lock)));
222 write_lock(&n->lock);
226 if (atomic_read(&n->refcnt) != 1) {
227 /* The most unpleasant situation.
228 We must destroy neighbour entry,
229 but someone still uses it.
231 The destroy will be delayed until
232 the last user releases us, but
233 we must kill timers etc. and move
236 __skb_queue_purge(&n->arp_queue);
237 n->arp_queue_len_bytes = 0;
238 n->output = neigh_blackhole;
239 if (n->nud_state & NUD_VALID)
240 n->nud_state = NUD_NOARP;
242 n->nud_state = NUD_NONE;
243 neigh_dbg(2, "neigh %p is stray\n", n);
245 write_unlock(&n->lock);
246 neigh_cleanup_and_release(n);
251 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
253 write_lock_bh(&tbl->lock);
254 neigh_flush_dev(tbl, dev);
255 write_unlock_bh(&tbl->lock);
257 EXPORT_SYMBOL(neigh_changeaddr);
259 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
261 write_lock_bh(&tbl->lock);
262 neigh_flush_dev(tbl, dev);
263 pneigh_ifdown(tbl, dev);
264 write_unlock_bh(&tbl->lock);
266 del_timer_sync(&tbl->proxy_timer);
267 pneigh_queue_purge(&tbl->proxy_queue);
270 EXPORT_SYMBOL(neigh_ifdown);
272 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
274 struct neighbour *n = NULL;
275 unsigned long now = jiffies;
278 entries = atomic_inc_return(&tbl->entries) - 1;
279 if (entries >= tbl->gc_thresh3 ||
280 (entries >= tbl->gc_thresh2 &&
281 time_after(now, tbl->last_flush + 5 * HZ))) {
282 if (!neigh_forced_gc(tbl) &&
283 entries >= tbl->gc_thresh3)
287 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
291 __skb_queue_head_init(&n->arp_queue);
292 rwlock_init(&n->lock);
293 seqlock_init(&n->ha_lock);
294 n->updated = n->used = now;
295 n->nud_state = NUD_NONE;
296 n->output = neigh_blackhole;
297 seqlock_init(&n->hh.hh_lock);
298 n->parms = neigh_parms_clone(&tbl->parms);
299 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
301 NEIGH_CACHE_STAT_INC(tbl, allocs);
303 atomic_set(&n->refcnt, 1);
309 atomic_dec(&tbl->entries);
313 static void neigh_get_hash_rnd(u32 *x)
315 get_random_bytes(x, sizeof(*x));
319 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
321 size_t size = (1 << shift) * sizeof(struct neighbour *);
322 struct neigh_hash_table *ret;
323 struct neighbour __rcu **buckets;
326 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
329 if (size <= PAGE_SIZE)
330 buckets = kzalloc(size, GFP_ATOMIC);
332 buckets = (struct neighbour __rcu **)
333 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
339 ret->hash_buckets = buckets;
340 ret->hash_shift = shift;
341 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
342 neigh_get_hash_rnd(&ret->hash_rnd[i]);
346 static void neigh_hash_free_rcu(struct rcu_head *head)
348 struct neigh_hash_table *nht = container_of(head,
349 struct neigh_hash_table,
351 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
352 struct neighbour __rcu **buckets = nht->hash_buckets;
354 if (size <= PAGE_SIZE)
357 free_pages((unsigned long)buckets, get_order(size));
361 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
362 unsigned long new_shift)
364 unsigned int i, hash;
365 struct neigh_hash_table *new_nht, *old_nht;
367 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
369 old_nht = rcu_dereference_protected(tbl->nht,
370 lockdep_is_held(&tbl->lock));
371 new_nht = neigh_hash_alloc(new_shift);
375 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
376 struct neighbour *n, *next;
378 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
379 lockdep_is_held(&tbl->lock));
382 hash = tbl->hash(n->primary_key, n->dev,
385 hash >>= (32 - new_nht->hash_shift);
386 next = rcu_dereference_protected(n->next,
387 lockdep_is_held(&tbl->lock));
389 rcu_assign_pointer(n->next,
390 rcu_dereference_protected(
391 new_nht->hash_buckets[hash],
392 lockdep_is_held(&tbl->lock)));
393 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
397 rcu_assign_pointer(tbl->nht, new_nht);
398 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
402 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
403 struct net_device *dev)
406 int key_len = tbl->key_len;
408 struct neigh_hash_table *nht;
410 NEIGH_CACHE_STAT_INC(tbl, lookups);
413 nht = rcu_dereference_bh(tbl->nht);
414 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
416 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
418 n = rcu_dereference_bh(n->next)) {
419 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
420 if (!atomic_inc_not_zero(&n->refcnt))
422 NEIGH_CACHE_STAT_INC(tbl, hits);
427 rcu_read_unlock_bh();
430 EXPORT_SYMBOL(neigh_lookup);
432 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
436 int key_len = tbl->key_len;
438 struct neigh_hash_table *nht;
440 NEIGH_CACHE_STAT_INC(tbl, lookups);
443 nht = rcu_dereference_bh(tbl->nht);
444 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
446 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
448 n = rcu_dereference_bh(n->next)) {
449 if (!memcmp(n->primary_key, pkey, key_len) &&
450 net_eq(dev_net(n->dev), net)) {
451 if (!atomic_inc_not_zero(&n->refcnt))
453 NEIGH_CACHE_STAT_INC(tbl, hits);
458 rcu_read_unlock_bh();
461 EXPORT_SYMBOL(neigh_lookup_nodev);
463 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
464 struct net_device *dev, bool want_ref)
467 int key_len = tbl->key_len;
469 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
470 struct neigh_hash_table *nht;
473 rc = ERR_PTR(-ENOBUFS);
477 memcpy(n->primary_key, pkey, key_len);
481 /* Protocol specific setup. */
482 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
484 goto out_neigh_release;
487 if (dev->netdev_ops->ndo_neigh_construct) {
488 error = dev->netdev_ops->ndo_neigh_construct(n);
491 goto out_neigh_release;
495 /* Device specific setup. */
496 if (n->parms->neigh_setup &&
497 (error = n->parms->neigh_setup(n)) < 0) {
499 goto out_neigh_release;
502 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
504 write_lock_bh(&tbl->lock);
505 nht = rcu_dereference_protected(tbl->nht,
506 lockdep_is_held(&tbl->lock));
508 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
509 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
511 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
513 if (n->parms->dead) {
514 rc = ERR_PTR(-EINVAL);
518 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
519 lockdep_is_held(&tbl->lock));
521 n1 = rcu_dereference_protected(n1->next,
522 lockdep_is_held(&tbl->lock))) {
523 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
534 rcu_assign_pointer(n->next,
535 rcu_dereference_protected(nht->hash_buckets[hash_val],
536 lockdep_is_held(&tbl->lock)));
537 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
538 write_unlock_bh(&tbl->lock);
539 neigh_dbg(2, "neigh %p is created\n", n);
544 write_unlock_bh(&tbl->lock);
549 EXPORT_SYMBOL(__neigh_create);
551 static u32 pneigh_hash(const void *pkey, int key_len)
553 u32 hash_val = *(u32 *)(pkey + key_len - 4);
554 hash_val ^= (hash_val >> 16);
555 hash_val ^= hash_val >> 8;
556 hash_val ^= hash_val >> 4;
557 hash_val &= PNEIGH_HASHMASK;
561 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
565 struct net_device *dev)
568 if (!memcmp(n->key, pkey, key_len) &&
569 net_eq(pneigh_net(n), net) &&
570 (n->dev == dev || !n->dev))
577 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
578 struct net *net, const void *pkey, struct net_device *dev)
580 int key_len = tbl->key_len;
581 u32 hash_val = pneigh_hash(pkey, key_len);
583 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
584 net, pkey, key_len, dev);
586 EXPORT_SYMBOL_GPL(__pneigh_lookup);
588 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
589 struct net *net, const void *pkey,
590 struct net_device *dev, int creat)
592 struct pneigh_entry *n;
593 int key_len = tbl->key_len;
594 u32 hash_val = pneigh_hash(pkey, key_len);
596 read_lock_bh(&tbl->lock);
597 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
598 net, pkey, key_len, dev);
599 read_unlock_bh(&tbl->lock);
606 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
610 write_pnet(&n->net, hold_net(net));
611 memcpy(n->key, pkey, key_len);
616 if (tbl->pconstructor && tbl->pconstructor(n)) {
625 write_lock_bh(&tbl->lock);
626 n->next = tbl->phash_buckets[hash_val];
627 tbl->phash_buckets[hash_val] = n;
628 write_unlock_bh(&tbl->lock);
632 EXPORT_SYMBOL(pneigh_lookup);
635 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
636 struct net_device *dev)
638 struct pneigh_entry *n, **np;
639 int key_len = tbl->key_len;
640 u32 hash_val = pneigh_hash(pkey, key_len);
642 write_lock_bh(&tbl->lock);
643 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
645 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
646 net_eq(pneigh_net(n), net)) {
648 write_unlock_bh(&tbl->lock);
649 if (tbl->pdestructor)
653 release_net(pneigh_net(n));
658 write_unlock_bh(&tbl->lock);
662 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
664 struct pneigh_entry *n, **np;
667 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
668 np = &tbl->phash_buckets[h];
669 while ((n = *np) != NULL) {
670 if (!dev || n->dev == dev) {
672 if (tbl->pdestructor)
676 release_net(pneigh_net(n));
686 static void neigh_parms_destroy(struct neigh_parms *parms);
688 static inline void neigh_parms_put(struct neigh_parms *parms)
690 if (atomic_dec_and_test(&parms->refcnt))
691 neigh_parms_destroy(parms);
695 * neighbour must already be out of the table;
698 void neigh_destroy(struct neighbour *neigh)
700 struct net_device *dev = neigh->dev;
702 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
705 pr_warn("Destroying alive neighbour %p\n", neigh);
710 if (neigh_del_timer(neigh))
711 pr_warn("Impossible event\n");
713 write_lock_bh(&neigh->lock);
714 __skb_queue_purge(&neigh->arp_queue);
715 write_unlock_bh(&neigh->lock);
716 neigh->arp_queue_len_bytes = 0;
718 if (dev->netdev_ops->ndo_neigh_destroy)
719 dev->netdev_ops->ndo_neigh_destroy(neigh);
722 neigh_parms_put(neigh->parms);
724 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
726 atomic_dec(&neigh->tbl->entries);
727 kfree_rcu(neigh, rcu);
729 EXPORT_SYMBOL(neigh_destroy);
731 /* Neighbour state is suspicious;
734 Called with write_locked neigh.
736 static void neigh_suspect(struct neighbour *neigh)
738 neigh_dbg(2, "neigh %p is suspected\n", neigh);
740 neigh->output = neigh->ops->output;
743 /* Neighbour state is OK;
746 Called with write_locked neigh.
748 static void neigh_connect(struct neighbour *neigh)
750 neigh_dbg(2, "neigh %p is connected\n", neigh);
752 neigh->output = neigh->ops->connected_output;
755 static void neigh_periodic_work(struct work_struct *work)
757 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
759 struct neighbour __rcu **np;
761 struct neigh_hash_table *nht;
763 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
765 write_lock_bh(&tbl->lock);
766 nht = rcu_dereference_protected(tbl->nht,
767 lockdep_is_held(&tbl->lock));
769 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
773 * periodically recompute ReachableTime from random function
776 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
777 struct neigh_parms *p;
778 tbl->last_rand = jiffies;
779 for (p = &tbl->parms; p; p = p->next)
781 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
784 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
785 np = &nht->hash_buckets[i];
787 while ((n = rcu_dereference_protected(*np,
788 lockdep_is_held(&tbl->lock))) != NULL) {
791 write_lock(&n->lock);
793 state = n->nud_state;
794 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
795 write_unlock(&n->lock);
799 if (time_before(n->used, n->confirmed))
800 n->used = n->confirmed;
802 if (atomic_read(&n->refcnt) == 1 &&
803 (state == NUD_FAILED ||
804 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
807 write_unlock(&n->lock);
808 neigh_cleanup_and_release(n);
811 write_unlock(&n->lock);
817 * It's fine to release lock here, even if hash table
818 * grows while we are preempted.
820 write_unlock_bh(&tbl->lock);
822 write_lock_bh(&tbl->lock);
823 nht = rcu_dereference_protected(tbl->nht,
824 lockdep_is_held(&tbl->lock));
827 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
828 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
829 * BASE_REACHABLE_TIME.
831 schedule_delayed_work(&tbl->gc_work,
832 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
833 write_unlock_bh(&tbl->lock);
836 static __inline__ int neigh_max_probes(struct neighbour *n)
838 struct neigh_parms *p = n->parms;
839 return (n->nud_state & NUD_PROBE) ?
840 NEIGH_VAR(p, UCAST_PROBES) :
841 NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
842 NEIGH_VAR(p, MCAST_PROBES);
845 static void neigh_invalidate(struct neighbour *neigh)
846 __releases(neigh->lock)
847 __acquires(neigh->lock)
851 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
852 neigh_dbg(2, "neigh %p is failed\n", neigh);
853 neigh->updated = jiffies;
855 /* It is very thin place. report_unreachable is very complicated
856 routine. Particularly, it can hit the same neighbour entry!
858 So that, we try to be accurate and avoid dead loop. --ANK
860 while (neigh->nud_state == NUD_FAILED &&
861 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
862 write_unlock(&neigh->lock);
863 neigh->ops->error_report(neigh, skb);
864 write_lock(&neigh->lock);
866 __skb_queue_purge(&neigh->arp_queue);
867 neigh->arp_queue_len_bytes = 0;
870 static void neigh_probe(struct neighbour *neigh)
871 __releases(neigh->lock)
873 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
874 /* keep skb alive even if arp_queue overflows */
876 skb = skb_copy(skb, GFP_ATOMIC);
877 write_unlock(&neigh->lock);
878 neigh->ops->solicit(neigh, skb);
879 atomic_inc(&neigh->probes);
883 /* Called when a timer expires for a neighbour entry. */
885 static void neigh_timer_handler(unsigned long arg)
887 unsigned long now, next;
888 struct neighbour *neigh = (struct neighbour *)arg;
892 write_lock(&neigh->lock);
894 state = neigh->nud_state;
898 if (!(state & NUD_IN_TIMER))
901 if (state & NUD_REACHABLE) {
902 if (time_before_eq(now,
903 neigh->confirmed + neigh->parms->reachable_time)) {
904 neigh_dbg(2, "neigh %p is still alive\n", neigh);
905 next = neigh->confirmed + neigh->parms->reachable_time;
906 } else if (time_before_eq(now,
908 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
909 neigh_dbg(2, "neigh %p is delayed\n", neigh);
910 neigh->nud_state = NUD_DELAY;
911 neigh->updated = jiffies;
912 neigh_suspect(neigh);
913 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
915 neigh_dbg(2, "neigh %p is suspected\n", neigh);
916 neigh->nud_state = NUD_STALE;
917 neigh->updated = jiffies;
918 neigh_suspect(neigh);
921 } else if (state & NUD_DELAY) {
922 if (time_before_eq(now,
924 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
925 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
926 neigh->nud_state = NUD_REACHABLE;
927 neigh->updated = jiffies;
928 neigh_connect(neigh);
930 next = neigh->confirmed + neigh->parms->reachable_time;
932 neigh_dbg(2, "neigh %p is probed\n", neigh);
933 neigh->nud_state = NUD_PROBE;
934 neigh->updated = jiffies;
935 atomic_set(&neigh->probes, 0);
936 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
939 /* NUD_PROBE|NUD_INCOMPLETE */
940 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
943 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
944 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
945 neigh->nud_state = NUD_FAILED;
947 neigh_invalidate(neigh);
950 if (neigh->nud_state & NUD_IN_TIMER) {
951 if (time_before(next, jiffies + HZ/2))
952 next = jiffies + HZ/2;
953 if (!mod_timer(&neigh->timer, next))
956 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
960 write_unlock(&neigh->lock);
964 neigh_update_notify(neigh);
966 neigh_release(neigh);
969 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
972 bool immediate_probe = false;
974 write_lock_bh(&neigh->lock);
977 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
980 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
981 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
982 NEIGH_VAR(neigh->parms, APP_PROBES)) {
983 unsigned long next, now = jiffies;
985 atomic_set(&neigh->probes,
986 NEIGH_VAR(neigh->parms, UCAST_PROBES));
987 neigh->nud_state = NUD_INCOMPLETE;
988 neigh->updated = now;
989 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
991 neigh_add_timer(neigh, next);
992 immediate_probe = true;
994 neigh->nud_state = NUD_FAILED;
995 neigh->updated = jiffies;
996 write_unlock_bh(&neigh->lock);
1001 } else if (neigh->nud_state & NUD_STALE) {
1002 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1003 neigh->nud_state = NUD_DELAY;
1004 neigh->updated = jiffies;
1005 neigh_add_timer(neigh, jiffies +
1006 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1009 if (neigh->nud_state == NUD_INCOMPLETE) {
1011 while (neigh->arp_queue_len_bytes + skb->truesize >
1012 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1013 struct sk_buff *buff;
1015 buff = __skb_dequeue(&neigh->arp_queue);
1018 neigh->arp_queue_len_bytes -= buff->truesize;
1020 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1023 __skb_queue_tail(&neigh->arp_queue, skb);
1024 neigh->arp_queue_len_bytes += skb->truesize;
1029 if (immediate_probe)
1032 write_unlock(&neigh->lock);
1036 EXPORT_SYMBOL(__neigh_event_send);
1038 static void neigh_update_hhs(struct neighbour *neigh)
1040 struct hh_cache *hh;
1041 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1044 if (neigh->dev->header_ops)
1045 update = neigh->dev->header_ops->cache_update;
1050 write_seqlock_bh(&hh->hh_lock);
1051 update(hh, neigh->dev, neigh->ha);
1052 write_sequnlock_bh(&hh->hh_lock);
1059 /* Generic update routine.
1060 -- lladdr is new lladdr or NULL, if it is not supplied.
1061 -- new is new state.
1063 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1065 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1066 lladdr instead of overriding it
1068 It also allows to retain current state
1069 if lladdr is unchanged.
1070 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1072 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1074 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1077 Caller MUST hold reference count on the entry.
1080 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1086 struct net_device *dev;
1087 int update_isrouter = 0;
1089 write_lock_bh(&neigh->lock);
1092 old = neigh->nud_state;
1095 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1096 (old & (NUD_NOARP | NUD_PERMANENT)))
1099 if (!(new & NUD_VALID)) {
1100 neigh_del_timer(neigh);
1101 if (old & NUD_CONNECTED)
1102 neigh_suspect(neigh);
1103 neigh->nud_state = new;
1105 notify = old & NUD_VALID;
1106 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1107 (new & NUD_FAILED)) {
1108 neigh_invalidate(neigh);
1114 /* Compare new lladdr with cached one */
1115 if (!dev->addr_len) {
1116 /* First case: device needs no address. */
1118 } else if (lladdr) {
1119 /* The second case: if something is already cached
1120 and a new address is proposed:
1122 - if they are different, check override flag
1124 if ((old & NUD_VALID) &&
1125 !memcmp(lladdr, neigh->ha, dev->addr_len))
1128 /* No address is supplied; if we know something,
1129 use it, otherwise discard the request.
1132 if (!(old & NUD_VALID))
1137 if (new & NUD_CONNECTED)
1138 neigh->confirmed = jiffies;
1139 neigh->updated = jiffies;
1141 /* If entry was valid and address is not changed,
1142 do not change entry state, if new one is STALE.
1145 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1146 if (old & NUD_VALID) {
1147 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1148 update_isrouter = 0;
1149 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1150 (old & NUD_CONNECTED)) {
1156 if (lladdr == neigh->ha && new == NUD_STALE &&
1157 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1158 (old & NUD_CONNECTED))
1165 neigh_del_timer(neigh);
1166 if (new & NUD_IN_TIMER)
1167 neigh_add_timer(neigh, (jiffies +
1168 ((new & NUD_REACHABLE) ?
1169 neigh->parms->reachable_time :
1171 neigh->nud_state = new;
1174 if (lladdr != neigh->ha) {
1175 write_seqlock(&neigh->ha_lock);
1176 memcpy(&neigh->ha, lladdr, dev->addr_len);
1177 write_sequnlock(&neigh->ha_lock);
1178 neigh_update_hhs(neigh);
1179 if (!(new & NUD_CONNECTED))
1180 neigh->confirmed = jiffies -
1181 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1186 if (new & NUD_CONNECTED)
1187 neigh_connect(neigh);
1189 neigh_suspect(neigh);
1190 if (!(old & NUD_VALID)) {
1191 struct sk_buff *skb;
1193 /* Again: avoid dead loop if something went wrong */
1195 while (neigh->nud_state & NUD_VALID &&
1196 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1197 struct dst_entry *dst = skb_dst(skb);
1198 struct neighbour *n2, *n1 = neigh;
1199 write_unlock_bh(&neigh->lock);
1203 /* Why not just use 'neigh' as-is? The problem is that
1204 * things such as shaper, eql, and sch_teql can end up
1205 * using alternative, different, neigh objects to output
1206 * the packet in the output path. So what we need to do
1207 * here is re-lookup the top-level neigh in the path so
1208 * we can reinject the packet there.
1212 n2 = dst_neigh_lookup_skb(dst, skb);
1216 n1->output(n1, skb);
1221 write_lock_bh(&neigh->lock);
1223 __skb_queue_purge(&neigh->arp_queue);
1224 neigh->arp_queue_len_bytes = 0;
1227 if (update_isrouter) {
1228 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1229 (neigh->flags | NTF_ROUTER) :
1230 (neigh->flags & ~NTF_ROUTER);
1232 write_unlock_bh(&neigh->lock);
1235 neigh_update_notify(neigh);
1239 EXPORT_SYMBOL(neigh_update);
1241 /* Update the neigh to listen temporarily for probe responses, even if it is
1242 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1244 void __neigh_set_probe_once(struct neighbour *neigh)
1246 neigh->updated = jiffies;
1247 if (!(neigh->nud_state & NUD_FAILED))
1249 neigh->nud_state = NUD_PROBE;
1250 atomic_set(&neigh->probes, NEIGH_VAR(neigh->parms, UCAST_PROBES));
1251 neigh_add_timer(neigh,
1252 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1254 EXPORT_SYMBOL(__neigh_set_probe_once);
1256 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1257 u8 *lladdr, void *saddr,
1258 struct net_device *dev)
1260 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1261 lladdr || !dev->addr_len);
1263 neigh_update(neigh, lladdr, NUD_STALE,
1264 NEIGH_UPDATE_F_OVERRIDE);
1267 EXPORT_SYMBOL(neigh_event_ns);
1269 /* called with read_lock_bh(&n->lock); */
1270 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1272 struct net_device *dev = dst->dev;
1273 __be16 prot = dst->ops->protocol;
1274 struct hh_cache *hh = &n->hh;
1276 write_lock_bh(&n->lock);
1278 /* Only one thread can come in here and initialize the
1282 dev->header_ops->cache(n, hh, prot);
1284 write_unlock_bh(&n->lock);
1287 /* This function can be used in contexts, where only old dev_queue_xmit
1288 * worked, f.e. if you want to override normal output path (eql, shaper),
1289 * but resolution is not made yet.
1292 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1294 struct net_device *dev = skb->dev;
1296 __skb_pull(skb, skb_network_offset(skb));
1298 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1300 dev->header_ops->rebuild(skb))
1303 return dev_queue_xmit(skb);
1305 EXPORT_SYMBOL(neigh_compat_output);
1307 /* Slow and careful. */
1309 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1311 struct dst_entry *dst = skb_dst(skb);
1317 if (!neigh_event_send(neigh, skb)) {
1319 struct net_device *dev = neigh->dev;
1322 if (dev->header_ops->cache && !neigh->hh.hh_len)
1323 neigh_hh_init(neigh, dst);
1326 __skb_pull(skb, skb_network_offset(skb));
1327 seq = read_seqbegin(&neigh->ha_lock);
1328 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1329 neigh->ha, NULL, skb->len);
1330 } while (read_seqretry(&neigh->ha_lock, seq));
1333 rc = dev_queue_xmit(skb);
1340 neigh_dbg(1, "%s: dst=%p neigh=%p\n", __func__, dst, neigh);
1346 EXPORT_SYMBOL(neigh_resolve_output);
1348 /* As fast as possible without hh cache */
1350 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1352 struct net_device *dev = neigh->dev;
1357 __skb_pull(skb, skb_network_offset(skb));
1358 seq = read_seqbegin(&neigh->ha_lock);
1359 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1360 neigh->ha, NULL, skb->len);
1361 } while (read_seqretry(&neigh->ha_lock, seq));
1364 err = dev_queue_xmit(skb);
1371 EXPORT_SYMBOL(neigh_connected_output);
1373 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1375 return dev_queue_xmit(skb);
1377 EXPORT_SYMBOL(neigh_direct_output);
1379 static void neigh_proxy_process(unsigned long arg)
1381 struct neigh_table *tbl = (struct neigh_table *)arg;
1382 long sched_next = 0;
1383 unsigned long now = jiffies;
1384 struct sk_buff *skb, *n;
1386 spin_lock(&tbl->proxy_queue.lock);
1388 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1389 long tdif = NEIGH_CB(skb)->sched_next - now;
1392 struct net_device *dev = skb->dev;
1394 __skb_unlink(skb, &tbl->proxy_queue);
1395 if (tbl->proxy_redo && netif_running(dev)) {
1397 tbl->proxy_redo(skb);
1404 } else if (!sched_next || tdif < sched_next)
1407 del_timer(&tbl->proxy_timer);
1409 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1410 spin_unlock(&tbl->proxy_queue.lock);
1413 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1414 struct sk_buff *skb)
1416 unsigned long now = jiffies;
1417 unsigned long sched_next = now + (net_random() %
1418 NEIGH_VAR(p, PROXY_DELAY));
1420 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1425 NEIGH_CB(skb)->sched_next = sched_next;
1426 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1428 spin_lock(&tbl->proxy_queue.lock);
1429 if (del_timer(&tbl->proxy_timer)) {
1430 if (time_before(tbl->proxy_timer.expires, sched_next))
1431 sched_next = tbl->proxy_timer.expires;
1435 __skb_queue_tail(&tbl->proxy_queue, skb);
1436 mod_timer(&tbl->proxy_timer, sched_next);
1437 spin_unlock(&tbl->proxy_queue.lock);
1439 EXPORT_SYMBOL(pneigh_enqueue);
1441 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1442 struct net *net, int ifindex)
1444 struct neigh_parms *p;
1446 for (p = &tbl->parms; p; p = p->next) {
1447 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1448 (!p->dev && !ifindex && net_eq(net, &init_net)))
1455 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1456 struct neigh_table *tbl)
1458 struct neigh_parms *p;
1459 struct net *net = dev_net(dev);
1460 const struct net_device_ops *ops = dev->netdev_ops;
1462 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1465 atomic_set(&p->refcnt, 1);
1467 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1470 write_pnet(&p->net, hold_net(net));
1471 p->sysctl_table = NULL;
1473 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1480 write_lock_bh(&tbl->lock);
1481 p->next = tbl->parms.next;
1482 tbl->parms.next = p;
1483 write_unlock_bh(&tbl->lock);
1485 neigh_parms_data_state_cleanall(p);
1489 EXPORT_SYMBOL(neigh_parms_alloc);
1491 static void neigh_rcu_free_parms(struct rcu_head *head)
1493 struct neigh_parms *parms =
1494 container_of(head, struct neigh_parms, rcu_head);
1496 neigh_parms_put(parms);
1499 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1501 struct neigh_parms **p;
1503 if (!parms || parms == &tbl->parms)
1505 write_lock_bh(&tbl->lock);
1506 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1510 write_unlock_bh(&tbl->lock);
1512 dev_put(parms->dev);
1513 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1517 write_unlock_bh(&tbl->lock);
1518 neigh_dbg(1, "%s: not found\n", __func__);
1520 EXPORT_SYMBOL(neigh_parms_release);
1522 static void neigh_parms_destroy(struct neigh_parms *parms)
1524 release_net(neigh_parms_net(parms));
1528 static struct lock_class_key neigh_table_proxy_queue_class;
1530 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1532 unsigned long now = jiffies;
1533 unsigned long phsize;
1535 write_pnet(&tbl->parms.net, &init_net);
1536 atomic_set(&tbl->parms.refcnt, 1);
1537 tbl->parms.reachable_time =
1538 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1540 tbl->stats = alloc_percpu(struct neigh_statistics);
1542 panic("cannot create neighbour cache statistics");
1544 #ifdef CONFIG_PROC_FS
1545 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1546 &neigh_stat_seq_fops, tbl))
1547 panic("cannot create neighbour proc dir entry");
1550 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1552 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1553 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1555 if (!tbl->nht || !tbl->phash_buckets)
1556 panic("cannot allocate neighbour cache hashes");
1558 if (!tbl->entry_size)
1559 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1560 tbl->key_len, NEIGH_PRIV_ALIGN);
1562 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1564 rwlock_init(&tbl->lock);
1565 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1566 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1567 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1568 skb_queue_head_init_class(&tbl->proxy_queue,
1569 &neigh_table_proxy_queue_class);
1571 tbl->last_flush = now;
1572 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1575 void neigh_table_init(struct neigh_table *tbl)
1577 struct neigh_table *tmp;
1579 neigh_table_init_no_netlink(tbl);
1580 write_lock(&neigh_tbl_lock);
1581 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1582 if (tmp->family == tbl->family)
1585 tbl->next = neigh_tables;
1587 write_unlock(&neigh_tbl_lock);
1589 if (unlikely(tmp)) {
1590 pr_err("Registering multiple tables for family %d\n",
1595 EXPORT_SYMBOL(neigh_table_init);
1597 int neigh_table_clear(struct neigh_table *tbl)
1599 struct neigh_table **tp;
1601 /* It is not clean... Fix it to unload IPv6 module safely */
1602 cancel_delayed_work_sync(&tbl->gc_work);
1603 del_timer_sync(&tbl->proxy_timer);
1604 pneigh_queue_purge(&tbl->proxy_queue);
1605 neigh_ifdown(tbl, NULL);
1606 if (atomic_read(&tbl->entries))
1607 pr_crit("neighbour leakage\n");
1608 write_lock(&neigh_tbl_lock);
1609 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1615 write_unlock(&neigh_tbl_lock);
1617 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1618 neigh_hash_free_rcu);
1621 kfree(tbl->phash_buckets);
1622 tbl->phash_buckets = NULL;
1624 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1626 free_percpu(tbl->stats);
1631 EXPORT_SYMBOL(neigh_table_clear);
1633 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1635 struct net *net = sock_net(skb->sk);
1637 struct nlattr *dst_attr;
1638 struct neigh_table *tbl;
1639 struct net_device *dev = NULL;
1643 if (nlmsg_len(nlh) < sizeof(*ndm))
1646 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1647 if (dst_attr == NULL)
1650 ndm = nlmsg_data(nlh);
1651 if (ndm->ndm_ifindex) {
1652 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1659 read_lock(&neigh_tbl_lock);
1660 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1661 struct neighbour *neigh;
1663 if (tbl->family != ndm->ndm_family)
1665 read_unlock(&neigh_tbl_lock);
1667 if (nla_len(dst_attr) < tbl->key_len)
1670 if (ndm->ndm_flags & NTF_PROXY) {
1671 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1678 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1679 if (neigh == NULL) {
1684 err = neigh_update(neigh, NULL, NUD_FAILED,
1685 NEIGH_UPDATE_F_OVERRIDE |
1686 NEIGH_UPDATE_F_ADMIN);
1687 neigh_release(neigh);
1690 read_unlock(&neigh_tbl_lock);
1691 err = -EAFNOSUPPORT;
1697 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1699 struct net *net = sock_net(skb->sk);
1701 struct nlattr *tb[NDA_MAX+1];
1702 struct neigh_table *tbl;
1703 struct net_device *dev = NULL;
1707 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1712 if (tb[NDA_DST] == NULL)
1715 ndm = nlmsg_data(nlh);
1716 if (ndm->ndm_ifindex) {
1717 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1723 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1727 read_lock(&neigh_tbl_lock);
1728 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1729 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1730 struct neighbour *neigh;
1733 if (tbl->family != ndm->ndm_family)
1735 read_unlock(&neigh_tbl_lock);
1737 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1739 dst = nla_data(tb[NDA_DST]);
1740 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1742 if (ndm->ndm_flags & NTF_PROXY) {
1743 struct pneigh_entry *pn;
1746 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1748 pn->flags = ndm->ndm_flags;
1757 neigh = neigh_lookup(tbl, dst, dev);
1758 if (neigh == NULL) {
1759 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1764 neigh = __neigh_lookup_errno(tbl, dst, dev);
1765 if (IS_ERR(neigh)) {
1766 err = PTR_ERR(neigh);
1770 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1772 neigh_release(neigh);
1776 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1777 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1780 if (ndm->ndm_flags & NTF_USE) {
1781 neigh_event_send(neigh, NULL);
1784 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1785 neigh_release(neigh);
1789 read_unlock(&neigh_tbl_lock);
1790 err = -EAFNOSUPPORT;
1795 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1797 struct nlattr *nest;
1799 nest = nla_nest_start(skb, NDTA_PARMS);
1804 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1805 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1806 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1807 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1808 /* approximative value for deprecated QUEUE_LEN (in packets) */
1809 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1810 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1811 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1812 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1813 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1814 NEIGH_VAR(parms, UCAST_PROBES)) ||
1815 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1816 NEIGH_VAR(parms, MCAST_PROBES)) ||
1817 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1818 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1819 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1820 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1821 NEIGH_VAR(parms, GC_STALETIME)) ||
1822 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1823 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1824 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1825 NEIGH_VAR(parms, RETRANS_TIME)) ||
1826 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1827 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1828 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1829 NEIGH_VAR(parms, PROXY_DELAY)) ||
1830 nla_put_msecs(skb, NDTPA_LOCKTIME,
1831 NEIGH_VAR(parms, LOCKTIME)))
1832 goto nla_put_failure;
1833 return nla_nest_end(skb, nest);
1836 nla_nest_cancel(skb, nest);
1840 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1841 u32 pid, u32 seq, int type, int flags)
1843 struct nlmsghdr *nlh;
1844 struct ndtmsg *ndtmsg;
1846 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1850 ndtmsg = nlmsg_data(nlh);
1852 read_lock_bh(&tbl->lock);
1853 ndtmsg->ndtm_family = tbl->family;
1854 ndtmsg->ndtm_pad1 = 0;
1855 ndtmsg->ndtm_pad2 = 0;
1857 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1858 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1859 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1860 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1861 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1862 goto nla_put_failure;
1864 unsigned long now = jiffies;
1865 unsigned int flush_delta = now - tbl->last_flush;
1866 unsigned int rand_delta = now - tbl->last_rand;
1867 struct neigh_hash_table *nht;
1868 struct ndt_config ndc = {
1869 .ndtc_key_len = tbl->key_len,
1870 .ndtc_entry_size = tbl->entry_size,
1871 .ndtc_entries = atomic_read(&tbl->entries),
1872 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1873 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1874 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1878 nht = rcu_dereference_bh(tbl->nht);
1879 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1880 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1881 rcu_read_unlock_bh();
1883 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1884 goto nla_put_failure;
1889 struct ndt_stats ndst;
1891 memset(&ndst, 0, sizeof(ndst));
1893 for_each_possible_cpu(cpu) {
1894 struct neigh_statistics *st;
1896 st = per_cpu_ptr(tbl->stats, cpu);
1897 ndst.ndts_allocs += st->allocs;
1898 ndst.ndts_destroys += st->destroys;
1899 ndst.ndts_hash_grows += st->hash_grows;
1900 ndst.ndts_res_failed += st->res_failed;
1901 ndst.ndts_lookups += st->lookups;
1902 ndst.ndts_hits += st->hits;
1903 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1904 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1905 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1906 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1909 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1910 goto nla_put_failure;
1913 BUG_ON(tbl->parms.dev);
1914 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1915 goto nla_put_failure;
1917 read_unlock_bh(&tbl->lock);
1918 return nlmsg_end(skb, nlh);
1921 read_unlock_bh(&tbl->lock);
1922 nlmsg_cancel(skb, nlh);
1926 static int neightbl_fill_param_info(struct sk_buff *skb,
1927 struct neigh_table *tbl,
1928 struct neigh_parms *parms,
1929 u32 pid, u32 seq, int type,
1932 struct ndtmsg *ndtmsg;
1933 struct nlmsghdr *nlh;
1935 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1939 ndtmsg = nlmsg_data(nlh);
1941 read_lock_bh(&tbl->lock);
1942 ndtmsg->ndtm_family = tbl->family;
1943 ndtmsg->ndtm_pad1 = 0;
1944 ndtmsg->ndtm_pad2 = 0;
1946 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1947 neightbl_fill_parms(skb, parms) < 0)
1950 read_unlock_bh(&tbl->lock);
1951 return nlmsg_end(skb, nlh);
1953 read_unlock_bh(&tbl->lock);
1954 nlmsg_cancel(skb, nlh);
1958 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1959 [NDTA_NAME] = { .type = NLA_STRING },
1960 [NDTA_THRESH1] = { .type = NLA_U32 },
1961 [NDTA_THRESH2] = { .type = NLA_U32 },
1962 [NDTA_THRESH3] = { .type = NLA_U32 },
1963 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1964 [NDTA_PARMS] = { .type = NLA_NESTED },
1967 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1968 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1969 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1970 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1971 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1972 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1973 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1974 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1975 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1976 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1977 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1978 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1979 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1980 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1983 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1985 struct net *net = sock_net(skb->sk);
1986 struct neigh_table *tbl;
1987 struct ndtmsg *ndtmsg;
1988 struct nlattr *tb[NDTA_MAX+1];
1991 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1992 nl_neightbl_policy);
1996 if (tb[NDTA_NAME] == NULL) {
2001 ndtmsg = nlmsg_data(nlh);
2002 read_lock(&neigh_tbl_lock);
2003 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
2004 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2007 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
2017 * We acquire tbl->lock to be nice to the periodic timers and
2018 * make sure they always see a consistent set of values.
2020 write_lock_bh(&tbl->lock);
2022 if (tb[NDTA_PARMS]) {
2023 struct nlattr *tbp[NDTPA_MAX+1];
2024 struct neigh_parms *p;
2027 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2028 nl_ntbl_parm_policy);
2030 goto errout_tbl_lock;
2032 if (tbp[NDTPA_IFINDEX])
2033 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2035 p = lookup_neigh_parms(tbl, net, ifindex);
2038 goto errout_tbl_lock;
2041 for (i = 1; i <= NDTPA_MAX; i++) {
2046 case NDTPA_QUEUE_LEN:
2047 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2048 nla_get_u32(tbp[i]) *
2049 SKB_TRUESIZE(ETH_FRAME_LEN));
2051 case NDTPA_QUEUE_LENBYTES:
2052 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2053 nla_get_u32(tbp[i]));
2055 case NDTPA_PROXY_QLEN:
2056 NEIGH_VAR_SET(p, PROXY_QLEN,
2057 nla_get_u32(tbp[i]));
2059 case NDTPA_APP_PROBES:
2060 NEIGH_VAR_SET(p, APP_PROBES,
2061 nla_get_u32(tbp[i]));
2063 case NDTPA_UCAST_PROBES:
2064 NEIGH_VAR_SET(p, UCAST_PROBES,
2065 nla_get_u32(tbp[i]));
2067 case NDTPA_MCAST_PROBES:
2068 NEIGH_VAR_SET(p, MCAST_PROBES,
2069 nla_get_u32(tbp[i]));
2071 case NDTPA_BASE_REACHABLE_TIME:
2072 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2073 nla_get_msecs(tbp[i]));
2075 case NDTPA_GC_STALETIME:
2076 NEIGH_VAR_SET(p, GC_STALETIME,
2077 nla_get_msecs(tbp[i]));
2079 case NDTPA_DELAY_PROBE_TIME:
2080 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2081 nla_get_msecs(tbp[i]));
2083 case NDTPA_RETRANS_TIME:
2084 NEIGH_VAR_SET(p, RETRANS_TIME,
2085 nla_get_msecs(tbp[i]));
2087 case NDTPA_ANYCAST_DELAY:
2088 NEIGH_VAR_SET(p, ANYCAST_DELAY, nla_get_msecs(tbp[i]));
2090 case NDTPA_PROXY_DELAY:
2091 NEIGH_VAR_SET(p, PROXY_DELAY, nla_get_msecs(tbp[i]));
2093 case NDTPA_LOCKTIME:
2094 NEIGH_VAR_SET(p, LOCKTIME, nla_get_msecs(tbp[i]));
2101 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2102 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2103 !net_eq(net, &init_net))
2104 goto errout_tbl_lock;
2106 if (tb[NDTA_THRESH1])
2107 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2109 if (tb[NDTA_THRESH2])
2110 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2112 if (tb[NDTA_THRESH3])
2113 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2115 if (tb[NDTA_GC_INTERVAL])
2116 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2121 write_unlock_bh(&tbl->lock);
2123 read_unlock(&neigh_tbl_lock);
2128 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2130 struct net *net = sock_net(skb->sk);
2131 int family, tidx, nidx = 0;
2132 int tbl_skip = cb->args[0];
2133 int neigh_skip = cb->args[1];
2134 struct neigh_table *tbl;
2136 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2138 read_lock(&neigh_tbl_lock);
2139 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2140 struct neigh_parms *p;
2142 if (tidx < tbl_skip || (family && tbl->family != family))
2145 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2146 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2150 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2151 if (!net_eq(neigh_parms_net(p), net))
2154 if (nidx < neigh_skip)
2157 if (neightbl_fill_param_info(skb, tbl, p,
2158 NETLINK_CB(cb->skb).portid,
2170 read_unlock(&neigh_tbl_lock);
2177 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2178 u32 pid, u32 seq, int type, unsigned int flags)
2180 unsigned long now = jiffies;
2181 struct nda_cacheinfo ci;
2182 struct nlmsghdr *nlh;
2185 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2189 ndm = nlmsg_data(nlh);
2190 ndm->ndm_family = neigh->ops->family;
2193 ndm->ndm_flags = neigh->flags;
2194 ndm->ndm_type = neigh->type;
2195 ndm->ndm_ifindex = neigh->dev->ifindex;
2197 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2198 goto nla_put_failure;
2200 read_lock_bh(&neigh->lock);
2201 ndm->ndm_state = neigh->nud_state;
2202 if (neigh->nud_state & NUD_VALID) {
2203 char haddr[MAX_ADDR_LEN];
2205 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2206 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2207 read_unlock_bh(&neigh->lock);
2208 goto nla_put_failure;
2212 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2213 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2214 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2215 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2216 read_unlock_bh(&neigh->lock);
2218 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2219 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2220 goto nla_put_failure;
2222 return nlmsg_end(skb, nlh);
2225 nlmsg_cancel(skb, nlh);
2229 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2230 u32 pid, u32 seq, int type, unsigned int flags,
2231 struct neigh_table *tbl)
2233 struct nlmsghdr *nlh;
2236 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2240 ndm = nlmsg_data(nlh);
2241 ndm->ndm_family = tbl->family;
2244 ndm->ndm_flags = pn->flags | NTF_PROXY;
2245 ndm->ndm_type = NDA_DST;
2246 ndm->ndm_ifindex = pn->dev->ifindex;
2247 ndm->ndm_state = NUD_NONE;
2249 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2250 goto nla_put_failure;
2252 return nlmsg_end(skb, nlh);
2255 nlmsg_cancel(skb, nlh);
2259 static void neigh_update_notify(struct neighbour *neigh)
2261 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2262 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2265 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2266 struct netlink_callback *cb)
2268 struct net *net = sock_net(skb->sk);
2269 struct neighbour *n;
2270 int rc, h, s_h = cb->args[1];
2271 int idx, s_idx = idx = cb->args[2];
2272 struct neigh_hash_table *nht;
2275 nht = rcu_dereference_bh(tbl->nht);
2277 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2280 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2282 n = rcu_dereference_bh(n->next)) {
2283 if (!net_eq(dev_net(n->dev), net))
2287 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2290 NLM_F_MULTI) <= 0) {
2300 rcu_read_unlock_bh();
2306 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2307 struct netlink_callback *cb)
2309 struct pneigh_entry *n;
2310 struct net *net = sock_net(skb->sk);
2311 int rc, h, s_h = cb->args[3];
2312 int idx, s_idx = idx = cb->args[4];
2314 read_lock_bh(&tbl->lock);
2316 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2319 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2320 if (dev_net(n->dev) != net)
2324 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2327 NLM_F_MULTI, tbl) <= 0) {
2328 read_unlock_bh(&tbl->lock);
2337 read_unlock_bh(&tbl->lock);
2346 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2348 struct neigh_table *tbl;
2353 read_lock(&neigh_tbl_lock);
2354 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2356 /* check for full ndmsg structure presence, family member is
2357 * the same for both structures
2359 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2360 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2365 for (tbl = neigh_tables, t = 0; tbl;
2366 tbl = tbl->next, t++) {
2367 if (t < s_t || (family && tbl->family != family))
2370 memset(&cb->args[1], 0, sizeof(cb->args) -
2371 sizeof(cb->args[0]));
2373 err = pneigh_dump_table(tbl, skb, cb);
2375 err = neigh_dump_table(tbl, skb, cb);
2379 read_unlock(&neigh_tbl_lock);
2385 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2388 struct neigh_hash_table *nht;
2391 nht = rcu_dereference_bh(tbl->nht);
2393 read_lock(&tbl->lock); /* avoid resizes */
2394 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2395 struct neighbour *n;
2397 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2399 n = rcu_dereference_bh(n->next))
2402 read_unlock(&tbl->lock);
2403 rcu_read_unlock_bh();
2405 EXPORT_SYMBOL(neigh_for_each);
2407 /* The tbl->lock must be held as a writer and BH disabled. */
2408 void __neigh_for_each_release(struct neigh_table *tbl,
2409 int (*cb)(struct neighbour *))
2412 struct neigh_hash_table *nht;
2414 nht = rcu_dereference_protected(tbl->nht,
2415 lockdep_is_held(&tbl->lock));
2416 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2417 struct neighbour *n;
2418 struct neighbour __rcu **np;
2420 np = &nht->hash_buckets[chain];
2421 while ((n = rcu_dereference_protected(*np,
2422 lockdep_is_held(&tbl->lock))) != NULL) {
2425 write_lock(&n->lock);
2428 rcu_assign_pointer(*np,
2429 rcu_dereference_protected(n->next,
2430 lockdep_is_held(&tbl->lock)));
2434 write_unlock(&n->lock);
2436 neigh_cleanup_and_release(n);
2440 EXPORT_SYMBOL(__neigh_for_each_release);
2442 #ifdef CONFIG_PROC_FS
2444 static struct neighbour *neigh_get_first(struct seq_file *seq)
2446 struct neigh_seq_state *state = seq->private;
2447 struct net *net = seq_file_net(seq);
2448 struct neigh_hash_table *nht = state->nht;
2449 struct neighbour *n = NULL;
2450 int bucket = state->bucket;
2452 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2453 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2454 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2457 if (!net_eq(dev_net(n->dev), net))
2459 if (state->neigh_sub_iter) {
2463 v = state->neigh_sub_iter(state, n, &fakep);
2467 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2469 if (n->nud_state & ~NUD_NOARP)
2472 n = rcu_dereference_bh(n->next);
2478 state->bucket = bucket;
2483 static struct neighbour *neigh_get_next(struct seq_file *seq,
2484 struct neighbour *n,
2487 struct neigh_seq_state *state = seq->private;
2488 struct net *net = seq_file_net(seq);
2489 struct neigh_hash_table *nht = state->nht;
2491 if (state->neigh_sub_iter) {
2492 void *v = state->neigh_sub_iter(state, n, pos);
2496 n = rcu_dereference_bh(n->next);
2500 if (!net_eq(dev_net(n->dev), net))
2502 if (state->neigh_sub_iter) {
2503 void *v = state->neigh_sub_iter(state, n, pos);
2508 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2511 if (n->nud_state & ~NUD_NOARP)
2514 n = rcu_dereference_bh(n->next);
2520 if (++state->bucket >= (1 << nht->hash_shift))
2523 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2531 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2533 struct neighbour *n = neigh_get_first(seq);
2538 n = neigh_get_next(seq, n, pos);
2543 return *pos ? NULL : n;
2546 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2548 struct neigh_seq_state *state = seq->private;
2549 struct net *net = seq_file_net(seq);
2550 struct neigh_table *tbl = state->tbl;
2551 struct pneigh_entry *pn = NULL;
2552 int bucket = state->bucket;
2554 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2555 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2556 pn = tbl->phash_buckets[bucket];
2557 while (pn && !net_eq(pneigh_net(pn), net))
2562 state->bucket = bucket;
2567 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2568 struct pneigh_entry *pn,
2571 struct neigh_seq_state *state = seq->private;
2572 struct net *net = seq_file_net(seq);
2573 struct neigh_table *tbl = state->tbl;
2577 } while (pn && !net_eq(pneigh_net(pn), net));
2580 if (++state->bucket > PNEIGH_HASHMASK)
2582 pn = tbl->phash_buckets[state->bucket];
2583 while (pn && !net_eq(pneigh_net(pn), net))
2595 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2597 struct pneigh_entry *pn = pneigh_get_first(seq);
2602 pn = pneigh_get_next(seq, pn, pos);
2607 return *pos ? NULL : pn;
2610 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2612 struct neigh_seq_state *state = seq->private;
2614 loff_t idxpos = *pos;
2616 rc = neigh_get_idx(seq, &idxpos);
2617 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2618 rc = pneigh_get_idx(seq, &idxpos);
2623 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2626 struct neigh_seq_state *state = seq->private;
2630 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2633 state->nht = rcu_dereference_bh(tbl->nht);
2635 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2637 EXPORT_SYMBOL(neigh_seq_start);
2639 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2641 struct neigh_seq_state *state;
2644 if (v == SEQ_START_TOKEN) {
2645 rc = neigh_get_first(seq);
2649 state = seq->private;
2650 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2651 rc = neigh_get_next(seq, v, NULL);
2654 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2655 rc = pneigh_get_first(seq);
2657 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2658 rc = pneigh_get_next(seq, v, NULL);
2664 EXPORT_SYMBOL(neigh_seq_next);
2666 void neigh_seq_stop(struct seq_file *seq, void *v)
2669 rcu_read_unlock_bh();
2671 EXPORT_SYMBOL(neigh_seq_stop);
2673 /* statistics via seq_file */
2675 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2677 struct neigh_table *tbl = seq->private;
2681 return SEQ_START_TOKEN;
2683 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2684 if (!cpu_possible(cpu))
2687 return per_cpu_ptr(tbl->stats, cpu);
2692 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2694 struct neigh_table *tbl = seq->private;
2697 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2698 if (!cpu_possible(cpu))
2701 return per_cpu_ptr(tbl->stats, cpu);
2706 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2711 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2713 struct neigh_table *tbl = seq->private;
2714 struct neigh_statistics *st = v;
2716 if (v == SEQ_START_TOKEN) {
2717 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");
2721 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2722 "%08lx %08lx %08lx %08lx %08lx\n",
2723 atomic_read(&tbl->entries),
2734 st->rcv_probes_mcast,
2735 st->rcv_probes_ucast,
2737 st->periodic_gc_runs,
2745 static const struct seq_operations neigh_stat_seq_ops = {
2746 .start = neigh_stat_seq_start,
2747 .next = neigh_stat_seq_next,
2748 .stop = neigh_stat_seq_stop,
2749 .show = neigh_stat_seq_show,
2752 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2754 int ret = seq_open(file, &neigh_stat_seq_ops);
2757 struct seq_file *sf = file->private_data;
2758 sf->private = PDE_DATA(inode);
2763 static const struct file_operations neigh_stat_seq_fops = {
2764 .owner = THIS_MODULE,
2765 .open = neigh_stat_seq_open,
2767 .llseek = seq_lseek,
2768 .release = seq_release,
2771 #endif /* CONFIG_PROC_FS */
2773 static inline size_t neigh_nlmsg_size(void)
2775 return NLMSG_ALIGN(sizeof(struct ndmsg))
2776 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2777 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2778 + nla_total_size(sizeof(struct nda_cacheinfo))
2779 + nla_total_size(4); /* NDA_PROBES */
2782 static void __neigh_notify(struct neighbour *n, int type, int flags)
2784 struct net *net = dev_net(n->dev);
2785 struct sk_buff *skb;
2788 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2792 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2794 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2795 WARN_ON(err == -EMSGSIZE);
2799 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2803 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2806 void neigh_app_ns(struct neighbour *n)
2808 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2810 EXPORT_SYMBOL(neigh_app_ns);
2812 #ifdef CONFIG_SYSCTL
2814 static int int_max = INT_MAX;
2815 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2817 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2818 void __user *buffer, size_t *lenp, loff_t *ppos)
2821 struct ctl_table tmp = *ctl;
2824 tmp.extra2 = &unres_qlen_max;
2827 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2828 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2831 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2835 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2840 return __in_dev_arp_parms_get_rcu(dev);
2842 return __in6_dev_nd_parms_get_rcu(dev);
2847 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2850 struct net_device *dev;
2851 int family = neigh_parms_family(p);
2854 for_each_netdev_rcu(net, dev) {
2855 struct neigh_parms *dst_p =
2856 neigh_get_dev_parms_rcu(dev, family);
2858 if (dst_p && !test_bit(index, dst_p->data_state))
2859 dst_p->data[index] = p->data[index];
2864 static void neigh_proc_update(struct ctl_table *ctl, int write)
2866 struct net_device *dev = ctl->extra1;
2867 struct neigh_parms *p = ctl->extra2;
2868 struct net *net = neigh_parms_net(p);
2869 int index = (int *) ctl->data - p->data;
2874 set_bit(index, p->data_state);
2875 if (!dev) /* NULL dev means this is default value */
2876 neigh_copy_dflt_parms(net, p, index);
2879 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2880 void __user *buffer,
2881 size_t *lenp, loff_t *ppos)
2883 struct ctl_table tmp = *ctl;
2887 tmp.extra2 = &int_max;
2889 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2890 neigh_proc_update(ctl, write);
2894 int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2895 void __user *buffer, size_t *lenp, loff_t *ppos)
2897 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2899 neigh_proc_update(ctl, write);
2902 EXPORT_SYMBOL(neigh_proc_dointvec);
2904 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2905 void __user *buffer,
2906 size_t *lenp, loff_t *ppos)
2908 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2910 neigh_proc_update(ctl, write);
2913 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2915 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2916 void __user *buffer,
2917 size_t *lenp, loff_t *ppos)
2919 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2921 neigh_proc_update(ctl, write);
2925 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2926 void __user *buffer,
2927 size_t *lenp, loff_t *ppos)
2929 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2931 neigh_proc_update(ctl, write);
2934 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2936 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2937 void __user *buffer,
2938 size_t *lenp, loff_t *ppos)
2940 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2942 neigh_proc_update(ctl, write);
2946 #define NEIGH_PARMS_DATA_OFFSET(index) \
2947 (&((struct neigh_parms *) 0)->data[index])
2949 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
2950 [NEIGH_VAR_ ## attr] = { \
2952 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
2953 .maxlen = sizeof(int), \
2955 .proc_handler = proc, \
2958 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
2959 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
2961 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
2962 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
2964 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
2965 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
2967 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
2968 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2970 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
2971 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2973 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
2974 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
2976 static struct neigh_sysctl_table {
2977 struct ctl_table_header *sysctl_header;
2978 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2979 } neigh_sysctl_template __read_mostly = {
2981 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
2982 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
2983 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
2984 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
2985 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
2986 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
2987 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
2988 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
2989 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
2990 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
2991 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
2992 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
2993 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
2994 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
2995 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
2996 [NEIGH_VAR_GC_INTERVAL] = {
2997 .procname = "gc_interval",
2998 .maxlen = sizeof(int),
3000 .proc_handler = proc_dointvec_jiffies,
3002 [NEIGH_VAR_GC_THRESH1] = {
3003 .procname = "gc_thresh1",
3004 .maxlen = sizeof(int),
3008 .proc_handler = proc_dointvec_minmax,
3010 [NEIGH_VAR_GC_THRESH2] = {
3011 .procname = "gc_thresh2",
3012 .maxlen = sizeof(int),
3016 .proc_handler = proc_dointvec_minmax,
3018 [NEIGH_VAR_GC_THRESH3] = {
3019 .procname = "gc_thresh3",
3020 .maxlen = sizeof(int),
3024 .proc_handler = proc_dointvec_minmax,
3030 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3031 proc_handler *handler)
3034 struct neigh_sysctl_table *t;
3035 const char *dev_name_source;
3036 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3039 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3043 for (i = 0; i < ARRAY_SIZE(t->neigh_vars); i++) {
3044 t->neigh_vars[i].data += (long) p;
3045 t->neigh_vars[i].extra1 = dev;
3046 t->neigh_vars[i].extra2 = p;
3050 dev_name_source = dev->name;
3051 /* Terminate the table early */
3052 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3053 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3055 dev_name_source = "default";
3056 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1);
3057 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1;
3058 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2;
3059 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3;
3064 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3066 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3067 /* RetransTime (in milliseconds)*/
3068 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3069 /* ReachableTime (in milliseconds) */
3070 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3073 /* Don't export sysctls to unprivileged users */
3074 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3075 t->neigh_vars[0].procname = NULL;
3077 switch (neigh_parms_family(p)) {
3088 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3089 p_name, dev_name_source);
3091 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3092 if (!t->sysctl_header)
3095 p->sysctl_table = t;
3103 EXPORT_SYMBOL(neigh_sysctl_register);
3105 void neigh_sysctl_unregister(struct neigh_parms *p)
3107 if (p->sysctl_table) {
3108 struct neigh_sysctl_table *t = p->sysctl_table;
3109 p->sysctl_table = NULL;
3110 unregister_net_sysctl_table(t->sysctl_header);
3114 EXPORT_SYMBOL(neigh_sysctl_unregister);
3116 #endif /* CONFIG_SYSCTL */
3118 static int __init neigh_init(void)
3120 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3121 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3122 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3124 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3126 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3131 subsys_initcall(neigh_init);