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 ? (prandom_u32() % 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));
770 * periodically recompute ReachableTime from random function
773 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
774 struct neigh_parms *p;
775 tbl->last_rand = jiffies;
776 list_for_each_entry(p, &tbl->parms_list, list)
778 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
781 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
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 queue_delayed_work(system_power_efficient_wq, &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 int max_probes = NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES);
840 if (!(n->nud_state & NUD_PROBE))
841 max_probes += 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);
951 if (neigh->nud_state & NUD_IN_TIMER) {
952 if (time_before(next, jiffies + HZ/2))
953 next = jiffies + HZ/2;
954 if (!mod_timer(&neigh->timer, next))
957 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
961 write_unlock(&neigh->lock);
965 neigh_update_notify(neigh);
967 neigh_release(neigh);
970 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
973 bool immediate_probe = false;
975 write_lock_bh(&neigh->lock);
978 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
981 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
982 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
983 NEIGH_VAR(neigh->parms, APP_PROBES)) {
984 unsigned long next, now = jiffies;
986 atomic_set(&neigh->probes,
987 NEIGH_VAR(neigh->parms, UCAST_PROBES));
988 neigh->nud_state = NUD_INCOMPLETE;
989 neigh->updated = now;
990 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
992 neigh_add_timer(neigh, next);
993 immediate_probe = true;
995 neigh->nud_state = NUD_FAILED;
996 neigh->updated = jiffies;
997 write_unlock_bh(&neigh->lock);
1002 } else if (neigh->nud_state & NUD_STALE) {
1003 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1004 neigh->nud_state = NUD_DELAY;
1005 neigh->updated = jiffies;
1006 neigh_add_timer(neigh, jiffies +
1007 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1010 if (neigh->nud_state == NUD_INCOMPLETE) {
1012 while (neigh->arp_queue_len_bytes + skb->truesize >
1013 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1014 struct sk_buff *buff;
1016 buff = __skb_dequeue(&neigh->arp_queue);
1019 neigh->arp_queue_len_bytes -= buff->truesize;
1021 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1024 __skb_queue_tail(&neigh->arp_queue, skb);
1025 neigh->arp_queue_len_bytes += skb->truesize;
1030 if (immediate_probe)
1033 write_unlock(&neigh->lock);
1037 EXPORT_SYMBOL(__neigh_event_send);
1039 static void neigh_update_hhs(struct neighbour *neigh)
1041 struct hh_cache *hh;
1042 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1045 if (neigh->dev->header_ops)
1046 update = neigh->dev->header_ops->cache_update;
1051 write_seqlock_bh(&hh->hh_lock);
1052 update(hh, neigh->dev, neigh->ha);
1053 write_sequnlock_bh(&hh->hh_lock);
1060 /* Generic update routine.
1061 -- lladdr is new lladdr or NULL, if it is not supplied.
1062 -- new is new state.
1064 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1066 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1067 lladdr instead of overriding it
1069 It also allows to retain current state
1070 if lladdr is unchanged.
1071 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1073 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1075 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1078 Caller MUST hold reference count on the entry.
1081 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1087 struct net_device *dev;
1088 int update_isrouter = 0;
1090 write_lock_bh(&neigh->lock);
1093 old = neigh->nud_state;
1096 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1097 (old & (NUD_NOARP | NUD_PERMANENT)))
1100 if (!(new & NUD_VALID)) {
1101 neigh_del_timer(neigh);
1102 if (old & NUD_CONNECTED)
1103 neigh_suspect(neigh);
1104 neigh->nud_state = new;
1106 notify = old & NUD_VALID;
1107 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1108 (new & NUD_FAILED)) {
1109 neigh_invalidate(neigh);
1115 /* Compare new lladdr with cached one */
1116 if (!dev->addr_len) {
1117 /* First case: device needs no address. */
1119 } else if (lladdr) {
1120 /* The second case: if something is already cached
1121 and a new address is proposed:
1123 - if they are different, check override flag
1125 if ((old & NUD_VALID) &&
1126 !memcmp(lladdr, neigh->ha, dev->addr_len))
1129 /* No address is supplied; if we know something,
1130 use it, otherwise discard the request.
1133 if (!(old & NUD_VALID))
1138 if (new & NUD_CONNECTED)
1139 neigh->confirmed = jiffies;
1140 neigh->updated = jiffies;
1142 /* If entry was valid and address is not changed,
1143 do not change entry state, if new one is STALE.
1146 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1147 if (old & NUD_VALID) {
1148 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1149 update_isrouter = 0;
1150 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1151 (old & NUD_CONNECTED)) {
1157 if (lladdr == neigh->ha && new == NUD_STALE &&
1158 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1159 (old & NUD_CONNECTED))
1166 neigh_del_timer(neigh);
1167 if (new & NUD_IN_TIMER)
1168 neigh_add_timer(neigh, (jiffies +
1169 ((new & NUD_REACHABLE) ?
1170 neigh->parms->reachable_time :
1172 neigh->nud_state = new;
1176 if (lladdr != neigh->ha) {
1177 write_seqlock(&neigh->ha_lock);
1178 memcpy(&neigh->ha, lladdr, dev->addr_len);
1179 write_sequnlock(&neigh->ha_lock);
1180 neigh_update_hhs(neigh);
1181 if (!(new & NUD_CONNECTED))
1182 neigh->confirmed = jiffies -
1183 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1188 if (new & NUD_CONNECTED)
1189 neigh_connect(neigh);
1191 neigh_suspect(neigh);
1192 if (!(old & NUD_VALID)) {
1193 struct sk_buff *skb;
1195 /* Again: avoid dead loop if something went wrong */
1197 while (neigh->nud_state & NUD_VALID &&
1198 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1199 struct dst_entry *dst = skb_dst(skb);
1200 struct neighbour *n2, *n1 = neigh;
1201 write_unlock_bh(&neigh->lock);
1205 /* Why not just use 'neigh' as-is? The problem is that
1206 * things such as shaper, eql, and sch_teql can end up
1207 * using alternative, different, neigh objects to output
1208 * the packet in the output path. So what we need to do
1209 * here is re-lookup the top-level neigh in the path so
1210 * we can reinject the packet there.
1214 n2 = dst_neigh_lookup_skb(dst, skb);
1218 n1->output(n1, skb);
1223 write_lock_bh(&neigh->lock);
1225 __skb_queue_purge(&neigh->arp_queue);
1226 neigh->arp_queue_len_bytes = 0;
1229 if (update_isrouter) {
1230 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1231 (neigh->flags | NTF_ROUTER) :
1232 (neigh->flags & ~NTF_ROUTER);
1234 write_unlock_bh(&neigh->lock);
1237 neigh_update_notify(neigh);
1241 EXPORT_SYMBOL(neigh_update);
1243 /* Update the neigh to listen temporarily for probe responses, even if it is
1244 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1246 void __neigh_set_probe_once(struct neighbour *neigh)
1248 neigh->updated = jiffies;
1249 if (!(neigh->nud_state & NUD_FAILED))
1251 neigh->nud_state = NUD_INCOMPLETE;
1252 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1253 neigh_add_timer(neigh,
1254 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1256 EXPORT_SYMBOL(__neigh_set_probe_once);
1258 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1259 u8 *lladdr, void *saddr,
1260 struct net_device *dev)
1262 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1263 lladdr || !dev->addr_len);
1265 neigh_update(neigh, lladdr, NUD_STALE,
1266 NEIGH_UPDATE_F_OVERRIDE);
1269 EXPORT_SYMBOL(neigh_event_ns);
1271 /* called with read_lock_bh(&n->lock); */
1272 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1274 struct net_device *dev = dst->dev;
1275 __be16 prot = dst->ops->protocol;
1276 struct hh_cache *hh = &n->hh;
1278 write_lock_bh(&n->lock);
1280 /* Only one thread can come in here and initialize the
1284 dev->header_ops->cache(n, hh, prot);
1286 write_unlock_bh(&n->lock);
1289 /* This function can be used in contexts, where only old dev_queue_xmit
1290 * worked, f.e. if you want to override normal output path (eql, shaper),
1291 * but resolution is not made yet.
1294 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1296 struct net_device *dev = skb->dev;
1298 __skb_pull(skb, skb_network_offset(skb));
1300 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1302 dev_rebuild_header(skb))
1305 return dev_queue_xmit(skb);
1307 EXPORT_SYMBOL(neigh_compat_output);
1309 /* Slow and careful. */
1311 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1313 struct dst_entry *dst = skb_dst(skb);
1319 if (!neigh_event_send(neigh, skb)) {
1321 struct net_device *dev = neigh->dev;
1324 if (dev->header_ops->cache && !neigh->hh.hh_len)
1325 neigh_hh_init(neigh, dst);
1328 __skb_pull(skb, skb_network_offset(skb));
1329 seq = read_seqbegin(&neigh->ha_lock);
1330 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1331 neigh->ha, NULL, skb->len);
1332 } while (read_seqretry(&neigh->ha_lock, seq));
1335 rc = dev_queue_xmit(skb);
1342 neigh_dbg(1, "%s: dst=%p neigh=%p\n", __func__, dst, neigh);
1348 EXPORT_SYMBOL(neigh_resolve_output);
1350 /* As fast as possible without hh cache */
1352 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1354 struct net_device *dev = neigh->dev;
1359 __skb_pull(skb, skb_network_offset(skb));
1360 seq = read_seqbegin(&neigh->ha_lock);
1361 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1362 neigh->ha, NULL, skb->len);
1363 } while (read_seqretry(&neigh->ha_lock, seq));
1366 err = dev_queue_xmit(skb);
1373 EXPORT_SYMBOL(neigh_connected_output);
1375 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1377 return dev_queue_xmit(skb);
1379 EXPORT_SYMBOL(neigh_direct_output);
1381 static void neigh_proxy_process(unsigned long arg)
1383 struct neigh_table *tbl = (struct neigh_table *)arg;
1384 long sched_next = 0;
1385 unsigned long now = jiffies;
1386 struct sk_buff *skb, *n;
1388 spin_lock(&tbl->proxy_queue.lock);
1390 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1391 long tdif = NEIGH_CB(skb)->sched_next - now;
1394 struct net_device *dev = skb->dev;
1396 __skb_unlink(skb, &tbl->proxy_queue);
1397 if (tbl->proxy_redo && netif_running(dev)) {
1399 tbl->proxy_redo(skb);
1406 } else if (!sched_next || tdif < sched_next)
1409 del_timer(&tbl->proxy_timer);
1411 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1412 spin_unlock(&tbl->proxy_queue.lock);
1415 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1416 struct sk_buff *skb)
1418 unsigned long now = jiffies;
1420 unsigned long sched_next = now + (prandom_u32() %
1421 NEIGH_VAR(p, PROXY_DELAY));
1423 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1428 NEIGH_CB(skb)->sched_next = sched_next;
1429 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1431 spin_lock(&tbl->proxy_queue.lock);
1432 if (del_timer(&tbl->proxy_timer)) {
1433 if (time_before(tbl->proxy_timer.expires, sched_next))
1434 sched_next = tbl->proxy_timer.expires;
1438 __skb_queue_tail(&tbl->proxy_queue, skb);
1439 mod_timer(&tbl->proxy_timer, sched_next);
1440 spin_unlock(&tbl->proxy_queue.lock);
1442 EXPORT_SYMBOL(pneigh_enqueue);
1444 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1445 struct net *net, int ifindex)
1447 struct neigh_parms *p;
1449 list_for_each_entry(p, &tbl->parms_list, list) {
1450 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1451 (!p->dev && !ifindex && net_eq(net, &init_net)))
1458 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1459 struct neigh_table *tbl)
1461 struct neigh_parms *p;
1462 struct net *net = dev_net(dev);
1463 const struct net_device_ops *ops = dev->netdev_ops;
1465 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1468 atomic_set(&p->refcnt, 1);
1470 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1473 write_pnet(&p->net, hold_net(net));
1474 p->sysctl_table = NULL;
1476 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1483 write_lock_bh(&tbl->lock);
1484 list_add(&p->list, &tbl->parms.list);
1485 write_unlock_bh(&tbl->lock);
1487 neigh_parms_data_state_cleanall(p);
1491 EXPORT_SYMBOL(neigh_parms_alloc);
1493 static void neigh_rcu_free_parms(struct rcu_head *head)
1495 struct neigh_parms *parms =
1496 container_of(head, struct neigh_parms, rcu_head);
1498 neigh_parms_put(parms);
1501 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1503 if (!parms || parms == &tbl->parms)
1505 write_lock_bh(&tbl->lock);
1506 list_del(&parms->list);
1508 write_unlock_bh(&tbl->lock);
1510 dev_put(parms->dev);
1511 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1513 EXPORT_SYMBOL(neigh_parms_release);
1515 static void neigh_parms_destroy(struct neigh_parms *parms)
1517 release_net(neigh_parms_net(parms));
1521 static struct lock_class_key neigh_table_proxy_queue_class;
1523 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1525 unsigned long now = jiffies;
1526 unsigned long phsize;
1528 INIT_LIST_HEAD(&tbl->parms_list);
1529 list_add(&tbl->parms.list, &tbl->parms_list);
1530 write_pnet(&tbl->parms.net, &init_net);
1531 atomic_set(&tbl->parms.refcnt, 1);
1532 tbl->parms.reachable_time =
1533 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1535 tbl->stats = alloc_percpu(struct neigh_statistics);
1537 panic("cannot create neighbour cache statistics");
1539 #ifdef CONFIG_PROC_FS
1540 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1541 &neigh_stat_seq_fops, tbl))
1542 panic("cannot create neighbour proc dir entry");
1545 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1547 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1548 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1550 if (!tbl->nht || !tbl->phash_buckets)
1551 panic("cannot allocate neighbour cache hashes");
1553 if (!tbl->entry_size)
1554 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1555 tbl->key_len, NEIGH_PRIV_ALIGN);
1557 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1559 rwlock_init(&tbl->lock);
1560 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1561 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1562 tbl->parms.reachable_time);
1563 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1564 skb_queue_head_init_class(&tbl->proxy_queue,
1565 &neigh_table_proxy_queue_class);
1567 tbl->last_flush = now;
1568 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1571 void neigh_table_init(struct neigh_table *tbl)
1573 struct neigh_table *tmp;
1575 neigh_table_init_no_netlink(tbl);
1576 write_lock(&neigh_tbl_lock);
1577 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1578 if (tmp->family == tbl->family)
1581 tbl->next = neigh_tables;
1583 write_unlock(&neigh_tbl_lock);
1585 if (unlikely(tmp)) {
1586 pr_err("Registering multiple tables for family %d\n",
1591 EXPORT_SYMBOL(neigh_table_init);
1593 int neigh_table_clear(struct neigh_table *tbl)
1595 struct neigh_table **tp;
1597 /* It is not clean... Fix it to unload IPv6 module safely */
1598 cancel_delayed_work_sync(&tbl->gc_work);
1599 del_timer_sync(&tbl->proxy_timer);
1600 pneigh_queue_purge(&tbl->proxy_queue);
1601 neigh_ifdown(tbl, NULL);
1602 if (atomic_read(&tbl->entries))
1603 pr_crit("neighbour leakage\n");
1604 write_lock(&neigh_tbl_lock);
1605 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1611 write_unlock(&neigh_tbl_lock);
1613 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1614 neigh_hash_free_rcu);
1617 kfree(tbl->phash_buckets);
1618 tbl->phash_buckets = NULL;
1620 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1622 free_percpu(tbl->stats);
1627 EXPORT_SYMBOL(neigh_table_clear);
1629 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1631 struct net *net = sock_net(skb->sk);
1633 struct nlattr *dst_attr;
1634 struct neigh_table *tbl;
1635 struct net_device *dev = NULL;
1639 if (nlmsg_len(nlh) < sizeof(*ndm))
1642 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1643 if (dst_attr == NULL)
1646 ndm = nlmsg_data(nlh);
1647 if (ndm->ndm_ifindex) {
1648 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1655 read_lock(&neigh_tbl_lock);
1656 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1657 struct neighbour *neigh;
1659 if (tbl->family != ndm->ndm_family)
1661 read_unlock(&neigh_tbl_lock);
1663 if (nla_len(dst_attr) < tbl->key_len)
1666 if (ndm->ndm_flags & NTF_PROXY) {
1667 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1674 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1675 if (neigh == NULL) {
1680 err = neigh_update(neigh, NULL, NUD_FAILED,
1681 NEIGH_UPDATE_F_OVERRIDE |
1682 NEIGH_UPDATE_F_ADMIN);
1683 neigh_release(neigh);
1686 read_unlock(&neigh_tbl_lock);
1687 err = -EAFNOSUPPORT;
1693 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1695 struct net *net = sock_net(skb->sk);
1697 struct nlattr *tb[NDA_MAX+1];
1698 struct neigh_table *tbl;
1699 struct net_device *dev = NULL;
1703 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1708 if (tb[NDA_DST] == NULL)
1711 ndm = nlmsg_data(nlh);
1712 if (ndm->ndm_ifindex) {
1713 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1719 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1723 read_lock(&neigh_tbl_lock);
1724 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1725 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1726 struct neighbour *neigh;
1729 if (tbl->family != ndm->ndm_family)
1731 read_unlock(&neigh_tbl_lock);
1733 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1735 dst = nla_data(tb[NDA_DST]);
1736 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1738 if (ndm->ndm_flags & NTF_PROXY) {
1739 struct pneigh_entry *pn;
1742 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1744 pn->flags = ndm->ndm_flags;
1753 neigh = neigh_lookup(tbl, dst, dev);
1754 if (neigh == NULL) {
1755 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1760 neigh = __neigh_lookup_errno(tbl, dst, dev);
1761 if (IS_ERR(neigh)) {
1762 err = PTR_ERR(neigh);
1766 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1768 neigh_release(neigh);
1772 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1773 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1776 if (ndm->ndm_flags & NTF_USE) {
1777 neigh_event_send(neigh, NULL);
1780 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1781 neigh_release(neigh);
1785 read_unlock(&neigh_tbl_lock);
1786 err = -EAFNOSUPPORT;
1791 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1793 struct nlattr *nest;
1795 nest = nla_nest_start(skb, NDTA_PARMS);
1800 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1801 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1802 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1803 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1804 /* approximative value for deprecated QUEUE_LEN (in packets) */
1805 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1806 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1807 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1808 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1809 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1810 NEIGH_VAR(parms, UCAST_PROBES)) ||
1811 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1812 NEIGH_VAR(parms, MCAST_PROBES)) ||
1813 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1814 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1815 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1816 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1817 NEIGH_VAR(parms, GC_STALETIME)) ||
1818 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1819 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1820 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1821 NEIGH_VAR(parms, RETRANS_TIME)) ||
1822 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1823 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1824 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1825 NEIGH_VAR(parms, PROXY_DELAY)) ||
1826 nla_put_msecs(skb, NDTPA_LOCKTIME,
1827 NEIGH_VAR(parms, LOCKTIME)))
1828 goto nla_put_failure;
1829 return nla_nest_end(skb, nest);
1832 nla_nest_cancel(skb, nest);
1836 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1837 u32 pid, u32 seq, int type, int flags)
1839 struct nlmsghdr *nlh;
1840 struct ndtmsg *ndtmsg;
1842 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1846 ndtmsg = nlmsg_data(nlh);
1848 read_lock_bh(&tbl->lock);
1849 ndtmsg->ndtm_family = tbl->family;
1850 ndtmsg->ndtm_pad1 = 0;
1851 ndtmsg->ndtm_pad2 = 0;
1853 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1854 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1855 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1856 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1857 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1858 goto nla_put_failure;
1860 unsigned long now = jiffies;
1861 unsigned int flush_delta = now - tbl->last_flush;
1862 unsigned int rand_delta = now - tbl->last_rand;
1863 struct neigh_hash_table *nht;
1864 struct ndt_config ndc = {
1865 .ndtc_key_len = tbl->key_len,
1866 .ndtc_entry_size = tbl->entry_size,
1867 .ndtc_entries = atomic_read(&tbl->entries),
1868 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1869 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1870 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1874 nht = rcu_dereference_bh(tbl->nht);
1875 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1876 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1877 rcu_read_unlock_bh();
1879 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1880 goto nla_put_failure;
1885 struct ndt_stats ndst;
1887 memset(&ndst, 0, sizeof(ndst));
1889 for_each_possible_cpu(cpu) {
1890 struct neigh_statistics *st;
1892 st = per_cpu_ptr(tbl->stats, cpu);
1893 ndst.ndts_allocs += st->allocs;
1894 ndst.ndts_destroys += st->destroys;
1895 ndst.ndts_hash_grows += st->hash_grows;
1896 ndst.ndts_res_failed += st->res_failed;
1897 ndst.ndts_lookups += st->lookups;
1898 ndst.ndts_hits += st->hits;
1899 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1900 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1901 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1902 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1905 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1906 goto nla_put_failure;
1909 BUG_ON(tbl->parms.dev);
1910 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1911 goto nla_put_failure;
1913 read_unlock_bh(&tbl->lock);
1914 return nlmsg_end(skb, nlh);
1917 read_unlock_bh(&tbl->lock);
1918 nlmsg_cancel(skb, nlh);
1922 static int neightbl_fill_param_info(struct sk_buff *skb,
1923 struct neigh_table *tbl,
1924 struct neigh_parms *parms,
1925 u32 pid, u32 seq, int type,
1928 struct ndtmsg *ndtmsg;
1929 struct nlmsghdr *nlh;
1931 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1935 ndtmsg = nlmsg_data(nlh);
1937 read_lock_bh(&tbl->lock);
1938 ndtmsg->ndtm_family = tbl->family;
1939 ndtmsg->ndtm_pad1 = 0;
1940 ndtmsg->ndtm_pad2 = 0;
1942 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1943 neightbl_fill_parms(skb, parms) < 0)
1946 read_unlock_bh(&tbl->lock);
1947 return nlmsg_end(skb, nlh);
1949 read_unlock_bh(&tbl->lock);
1950 nlmsg_cancel(skb, nlh);
1954 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1955 [NDTA_NAME] = { .type = NLA_STRING },
1956 [NDTA_THRESH1] = { .type = NLA_U32 },
1957 [NDTA_THRESH2] = { .type = NLA_U32 },
1958 [NDTA_THRESH3] = { .type = NLA_U32 },
1959 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1960 [NDTA_PARMS] = { .type = NLA_NESTED },
1963 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1964 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1965 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1966 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1967 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1968 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1969 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1970 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1971 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1972 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1973 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1974 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1975 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1976 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1979 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1981 struct net *net = sock_net(skb->sk);
1982 struct neigh_table *tbl;
1983 struct ndtmsg *ndtmsg;
1984 struct nlattr *tb[NDTA_MAX+1];
1987 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1988 nl_neightbl_policy);
1992 if (tb[NDTA_NAME] == NULL) {
1997 ndtmsg = nlmsg_data(nlh);
1998 read_lock(&neigh_tbl_lock);
1999 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
2000 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2003 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
2013 * We acquire tbl->lock to be nice to the periodic timers and
2014 * make sure they always see a consistent set of values.
2016 write_lock_bh(&tbl->lock);
2018 if (tb[NDTA_PARMS]) {
2019 struct nlattr *tbp[NDTPA_MAX+1];
2020 struct neigh_parms *p;
2023 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2024 nl_ntbl_parm_policy);
2026 goto errout_tbl_lock;
2028 if (tbp[NDTPA_IFINDEX])
2029 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2031 p = lookup_neigh_parms(tbl, net, ifindex);
2034 goto errout_tbl_lock;
2037 for (i = 1; i <= NDTPA_MAX; i++) {
2042 case NDTPA_QUEUE_LEN:
2043 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2044 nla_get_u32(tbp[i]) *
2045 SKB_TRUESIZE(ETH_FRAME_LEN));
2047 case NDTPA_QUEUE_LENBYTES:
2048 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2049 nla_get_u32(tbp[i]));
2051 case NDTPA_PROXY_QLEN:
2052 NEIGH_VAR_SET(p, PROXY_QLEN,
2053 nla_get_u32(tbp[i]));
2055 case NDTPA_APP_PROBES:
2056 NEIGH_VAR_SET(p, APP_PROBES,
2057 nla_get_u32(tbp[i]));
2059 case NDTPA_UCAST_PROBES:
2060 NEIGH_VAR_SET(p, UCAST_PROBES,
2061 nla_get_u32(tbp[i]));
2063 case NDTPA_MCAST_PROBES:
2064 NEIGH_VAR_SET(p, MCAST_PROBES,
2065 nla_get_u32(tbp[i]));
2067 case NDTPA_BASE_REACHABLE_TIME:
2068 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2069 nla_get_msecs(tbp[i]));
2071 case NDTPA_GC_STALETIME:
2072 NEIGH_VAR_SET(p, GC_STALETIME,
2073 nla_get_msecs(tbp[i]));
2075 case NDTPA_DELAY_PROBE_TIME:
2076 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2077 nla_get_msecs(tbp[i]));
2079 case NDTPA_RETRANS_TIME:
2080 NEIGH_VAR_SET(p, RETRANS_TIME,
2081 nla_get_msecs(tbp[i]));
2083 case NDTPA_ANYCAST_DELAY:
2084 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2085 nla_get_msecs(tbp[i]));
2087 case NDTPA_PROXY_DELAY:
2088 NEIGH_VAR_SET(p, PROXY_DELAY,
2089 nla_get_msecs(tbp[i]));
2091 case NDTPA_LOCKTIME:
2092 NEIGH_VAR_SET(p, LOCKTIME,
2093 nla_get_msecs(tbp[i]));
2100 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2101 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2102 !net_eq(net, &init_net))
2103 goto errout_tbl_lock;
2105 if (tb[NDTA_THRESH1])
2106 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2108 if (tb[NDTA_THRESH2])
2109 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2111 if (tb[NDTA_THRESH3])
2112 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2114 if (tb[NDTA_GC_INTERVAL])
2115 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2120 write_unlock_bh(&tbl->lock);
2122 read_unlock(&neigh_tbl_lock);
2127 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2129 struct net *net = sock_net(skb->sk);
2130 int family, tidx, nidx = 0;
2131 int tbl_skip = cb->args[0];
2132 int neigh_skip = cb->args[1];
2133 struct neigh_table *tbl;
2135 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2137 read_lock(&neigh_tbl_lock);
2138 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2139 struct neigh_parms *p;
2141 if (tidx < tbl_skip || (family && tbl->family != family))
2144 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2145 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2150 p = list_next_entry(&tbl->parms, list);
2151 list_for_each_entry_from(p, &tbl->parms_list, list) {
2152 if (!net_eq(neigh_parms_net(p), net))
2155 if (nidx < neigh_skip)
2158 if (neightbl_fill_param_info(skb, tbl, p,
2159 NETLINK_CB(cb->skb).portid,
2171 read_unlock(&neigh_tbl_lock);
2178 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2179 u32 pid, u32 seq, int type, unsigned int flags)
2181 unsigned long now = jiffies;
2182 struct nda_cacheinfo ci;
2183 struct nlmsghdr *nlh;
2186 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2190 ndm = nlmsg_data(nlh);
2191 ndm->ndm_family = neigh->ops->family;
2194 ndm->ndm_flags = neigh->flags;
2195 ndm->ndm_type = neigh->type;
2196 ndm->ndm_ifindex = neigh->dev->ifindex;
2198 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2199 goto nla_put_failure;
2201 read_lock_bh(&neigh->lock);
2202 ndm->ndm_state = neigh->nud_state;
2203 if (neigh->nud_state & NUD_VALID) {
2204 char haddr[MAX_ADDR_LEN];
2206 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2207 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2208 read_unlock_bh(&neigh->lock);
2209 goto nla_put_failure;
2213 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2214 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2215 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2216 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2217 read_unlock_bh(&neigh->lock);
2219 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2220 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2221 goto nla_put_failure;
2223 return nlmsg_end(skb, nlh);
2226 nlmsg_cancel(skb, nlh);
2230 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2231 u32 pid, u32 seq, int type, unsigned int flags,
2232 struct neigh_table *tbl)
2234 struct nlmsghdr *nlh;
2237 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2241 ndm = nlmsg_data(nlh);
2242 ndm->ndm_family = tbl->family;
2245 ndm->ndm_flags = pn->flags | NTF_PROXY;
2246 ndm->ndm_type = RTN_UNICAST;
2247 ndm->ndm_ifindex = pn->dev->ifindex;
2248 ndm->ndm_state = NUD_NONE;
2250 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2251 goto nla_put_failure;
2253 return nlmsg_end(skb, nlh);
2256 nlmsg_cancel(skb, nlh);
2260 static void neigh_update_notify(struct neighbour *neigh)
2262 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2263 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2266 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2267 struct netlink_callback *cb)
2269 struct net *net = sock_net(skb->sk);
2270 struct neighbour *n;
2271 int rc, h, s_h = cb->args[1];
2272 int idx, s_idx = idx = cb->args[2];
2273 struct neigh_hash_table *nht;
2276 nht = rcu_dereference_bh(tbl->nht);
2278 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2281 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2283 n = rcu_dereference_bh(n->next)) {
2284 if (!net_eq(dev_net(n->dev), net))
2288 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2291 NLM_F_MULTI) <= 0) {
2301 rcu_read_unlock_bh();
2307 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2308 struct netlink_callback *cb)
2310 struct pneigh_entry *n;
2311 struct net *net = sock_net(skb->sk);
2312 int rc, h, s_h = cb->args[3];
2313 int idx, s_idx = idx = cb->args[4];
2315 read_lock_bh(&tbl->lock);
2317 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2320 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2321 if (dev_net(n->dev) != net)
2325 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2328 NLM_F_MULTI, tbl) <= 0) {
2329 read_unlock_bh(&tbl->lock);
2338 read_unlock_bh(&tbl->lock);
2347 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2349 struct neigh_table *tbl;
2354 read_lock(&neigh_tbl_lock);
2355 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2357 /* check for full ndmsg structure presence, family member is
2358 * the same for both structures
2360 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2361 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2366 for (tbl = neigh_tables, t = 0; tbl;
2367 tbl = tbl->next, t++) {
2368 if (t < s_t || (family && tbl->family != family))
2371 memset(&cb->args[1], 0, sizeof(cb->args) -
2372 sizeof(cb->args[0]));
2374 err = pneigh_dump_table(tbl, skb, cb);
2376 err = neigh_dump_table(tbl, skb, cb);
2380 read_unlock(&neigh_tbl_lock);
2386 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2389 struct neigh_hash_table *nht;
2392 nht = rcu_dereference_bh(tbl->nht);
2394 read_lock(&tbl->lock); /* avoid resizes */
2395 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2396 struct neighbour *n;
2398 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2400 n = rcu_dereference_bh(n->next))
2403 read_unlock(&tbl->lock);
2404 rcu_read_unlock_bh();
2406 EXPORT_SYMBOL(neigh_for_each);
2408 /* The tbl->lock must be held as a writer and BH disabled. */
2409 void __neigh_for_each_release(struct neigh_table *tbl,
2410 int (*cb)(struct neighbour *))
2413 struct neigh_hash_table *nht;
2415 nht = rcu_dereference_protected(tbl->nht,
2416 lockdep_is_held(&tbl->lock));
2417 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2418 struct neighbour *n;
2419 struct neighbour __rcu **np;
2421 np = &nht->hash_buckets[chain];
2422 while ((n = rcu_dereference_protected(*np,
2423 lockdep_is_held(&tbl->lock))) != NULL) {
2426 write_lock(&n->lock);
2429 rcu_assign_pointer(*np,
2430 rcu_dereference_protected(n->next,
2431 lockdep_is_held(&tbl->lock)));
2435 write_unlock(&n->lock);
2437 neigh_cleanup_and_release(n);
2441 EXPORT_SYMBOL(__neigh_for_each_release);
2443 #ifdef CONFIG_PROC_FS
2445 static struct neighbour *neigh_get_first(struct seq_file *seq)
2447 struct neigh_seq_state *state = seq->private;
2448 struct net *net = seq_file_net(seq);
2449 struct neigh_hash_table *nht = state->nht;
2450 struct neighbour *n = NULL;
2451 int bucket = state->bucket;
2453 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2454 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2455 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2458 if (!net_eq(dev_net(n->dev), net))
2460 if (state->neigh_sub_iter) {
2464 v = state->neigh_sub_iter(state, n, &fakep);
2468 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2470 if (n->nud_state & ~NUD_NOARP)
2473 n = rcu_dereference_bh(n->next);
2479 state->bucket = bucket;
2484 static struct neighbour *neigh_get_next(struct seq_file *seq,
2485 struct neighbour *n,
2488 struct neigh_seq_state *state = seq->private;
2489 struct net *net = seq_file_net(seq);
2490 struct neigh_hash_table *nht = state->nht;
2492 if (state->neigh_sub_iter) {
2493 void *v = state->neigh_sub_iter(state, n, pos);
2497 n = rcu_dereference_bh(n->next);
2501 if (!net_eq(dev_net(n->dev), net))
2503 if (state->neigh_sub_iter) {
2504 void *v = state->neigh_sub_iter(state, n, pos);
2509 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2512 if (n->nud_state & ~NUD_NOARP)
2515 n = rcu_dereference_bh(n->next);
2521 if (++state->bucket >= (1 << nht->hash_shift))
2524 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2532 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2534 struct neighbour *n = neigh_get_first(seq);
2539 n = neigh_get_next(seq, n, pos);
2544 return *pos ? NULL : n;
2547 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2549 struct neigh_seq_state *state = seq->private;
2550 struct net *net = seq_file_net(seq);
2551 struct neigh_table *tbl = state->tbl;
2552 struct pneigh_entry *pn = NULL;
2553 int bucket = state->bucket;
2555 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2556 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2557 pn = tbl->phash_buckets[bucket];
2558 while (pn && !net_eq(pneigh_net(pn), net))
2563 state->bucket = bucket;
2568 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2569 struct pneigh_entry *pn,
2572 struct neigh_seq_state *state = seq->private;
2573 struct net *net = seq_file_net(seq);
2574 struct neigh_table *tbl = state->tbl;
2578 } while (pn && !net_eq(pneigh_net(pn), net));
2581 if (++state->bucket > PNEIGH_HASHMASK)
2583 pn = tbl->phash_buckets[state->bucket];
2584 while (pn && !net_eq(pneigh_net(pn), net))
2596 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2598 struct pneigh_entry *pn = pneigh_get_first(seq);
2603 pn = pneigh_get_next(seq, pn, pos);
2608 return *pos ? NULL : pn;
2611 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2613 struct neigh_seq_state *state = seq->private;
2615 loff_t idxpos = *pos;
2617 rc = neigh_get_idx(seq, &idxpos);
2618 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2619 rc = pneigh_get_idx(seq, &idxpos);
2624 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2627 struct neigh_seq_state *state = seq->private;
2631 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2634 state->nht = rcu_dereference_bh(tbl->nht);
2636 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2638 EXPORT_SYMBOL(neigh_seq_start);
2640 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2642 struct neigh_seq_state *state;
2645 if (v == SEQ_START_TOKEN) {
2646 rc = neigh_get_first(seq);
2650 state = seq->private;
2651 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2652 rc = neigh_get_next(seq, v, NULL);
2655 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2656 rc = pneigh_get_first(seq);
2658 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2659 rc = pneigh_get_next(seq, v, NULL);
2665 EXPORT_SYMBOL(neigh_seq_next);
2667 void neigh_seq_stop(struct seq_file *seq, void *v)
2670 rcu_read_unlock_bh();
2672 EXPORT_SYMBOL(neigh_seq_stop);
2674 /* statistics via seq_file */
2676 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2678 struct neigh_table *tbl = seq->private;
2682 return SEQ_START_TOKEN;
2684 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2685 if (!cpu_possible(cpu))
2688 return per_cpu_ptr(tbl->stats, cpu);
2693 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2695 struct neigh_table *tbl = seq->private;
2698 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2699 if (!cpu_possible(cpu))
2702 return per_cpu_ptr(tbl->stats, cpu);
2707 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2712 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2714 struct neigh_table *tbl = seq->private;
2715 struct neigh_statistics *st = v;
2717 if (v == SEQ_START_TOKEN) {
2718 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");
2722 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2723 "%08lx %08lx %08lx %08lx %08lx\n",
2724 atomic_read(&tbl->entries),
2735 st->rcv_probes_mcast,
2736 st->rcv_probes_ucast,
2738 st->periodic_gc_runs,
2746 static const struct seq_operations neigh_stat_seq_ops = {
2747 .start = neigh_stat_seq_start,
2748 .next = neigh_stat_seq_next,
2749 .stop = neigh_stat_seq_stop,
2750 .show = neigh_stat_seq_show,
2753 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2755 int ret = seq_open(file, &neigh_stat_seq_ops);
2758 struct seq_file *sf = file->private_data;
2759 sf->private = PDE_DATA(inode);
2764 static const struct file_operations neigh_stat_seq_fops = {
2765 .owner = THIS_MODULE,
2766 .open = neigh_stat_seq_open,
2768 .llseek = seq_lseek,
2769 .release = seq_release,
2772 #endif /* CONFIG_PROC_FS */
2774 static inline size_t neigh_nlmsg_size(void)
2776 return NLMSG_ALIGN(sizeof(struct ndmsg))
2777 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2778 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2779 + nla_total_size(sizeof(struct nda_cacheinfo))
2780 + nla_total_size(4); /* NDA_PROBES */
2783 static void __neigh_notify(struct neighbour *n, int type, int flags)
2785 struct net *net = dev_net(n->dev);
2786 struct sk_buff *skb;
2789 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2793 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2795 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2796 WARN_ON(err == -EMSGSIZE);
2800 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2804 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2807 void neigh_app_ns(struct neighbour *n)
2809 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2811 EXPORT_SYMBOL(neigh_app_ns);
2813 #ifdef CONFIG_SYSCTL
2815 static int int_max = INT_MAX;
2816 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2818 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2819 void __user *buffer, size_t *lenp, loff_t *ppos)
2822 struct ctl_table tmp = *ctl;
2825 tmp.extra2 = &unres_qlen_max;
2828 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2829 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2832 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2836 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2841 return __in_dev_arp_parms_get_rcu(dev);
2843 return __in6_dev_nd_parms_get_rcu(dev);
2848 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2851 struct net_device *dev;
2852 int family = neigh_parms_family(p);
2855 for_each_netdev_rcu(net, dev) {
2856 struct neigh_parms *dst_p =
2857 neigh_get_dev_parms_rcu(dev, family);
2859 if (dst_p && !test_bit(index, dst_p->data_state))
2860 dst_p->data[index] = p->data[index];
2865 static void neigh_proc_update(struct ctl_table *ctl, int write)
2867 struct net_device *dev = ctl->extra1;
2868 struct neigh_parms *p = ctl->extra2;
2869 struct net *net = neigh_parms_net(p);
2870 int index = (int *) ctl->data - p->data;
2875 set_bit(index, p->data_state);
2876 if (!dev) /* NULL dev means this is default value */
2877 neigh_copy_dflt_parms(net, p, index);
2880 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2881 void __user *buffer,
2882 size_t *lenp, loff_t *ppos)
2884 struct ctl_table tmp = *ctl;
2888 tmp.extra2 = &int_max;
2890 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2891 neigh_proc_update(ctl, write);
2895 int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2896 void __user *buffer, size_t *lenp, loff_t *ppos)
2898 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2900 neigh_proc_update(ctl, write);
2903 EXPORT_SYMBOL(neigh_proc_dointvec);
2905 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2906 void __user *buffer,
2907 size_t *lenp, loff_t *ppos)
2909 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2911 neigh_proc_update(ctl, write);
2914 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2916 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2917 void __user *buffer,
2918 size_t *lenp, loff_t *ppos)
2920 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2922 neigh_proc_update(ctl, write);
2926 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2927 void __user *buffer,
2928 size_t *lenp, loff_t *ppos)
2930 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2932 neigh_proc_update(ctl, write);
2935 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2937 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2938 void __user *buffer,
2939 size_t *lenp, loff_t *ppos)
2941 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2943 neigh_proc_update(ctl, write);
2947 #define NEIGH_PARMS_DATA_OFFSET(index) \
2948 (&((struct neigh_parms *) 0)->data[index])
2950 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
2951 [NEIGH_VAR_ ## attr] = { \
2953 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
2954 .maxlen = sizeof(int), \
2956 .proc_handler = proc, \
2959 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
2960 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
2962 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
2963 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
2965 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
2966 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
2968 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
2969 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2971 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
2972 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2974 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
2975 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
2977 static struct neigh_sysctl_table {
2978 struct ctl_table_header *sysctl_header;
2979 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2980 } neigh_sysctl_template __read_mostly = {
2982 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
2983 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
2984 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
2985 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
2986 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
2987 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
2988 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
2989 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
2990 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
2991 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
2992 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
2993 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
2994 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
2995 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
2996 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
2997 [NEIGH_VAR_GC_INTERVAL] = {
2998 .procname = "gc_interval",
2999 .maxlen = sizeof(int),
3001 .proc_handler = proc_dointvec_jiffies,
3003 [NEIGH_VAR_GC_THRESH1] = {
3004 .procname = "gc_thresh1",
3005 .maxlen = sizeof(int),
3009 .proc_handler = proc_dointvec_minmax,
3011 [NEIGH_VAR_GC_THRESH2] = {
3012 .procname = "gc_thresh2",
3013 .maxlen = sizeof(int),
3017 .proc_handler = proc_dointvec_minmax,
3019 [NEIGH_VAR_GC_THRESH3] = {
3020 .procname = "gc_thresh3",
3021 .maxlen = sizeof(int),
3025 .proc_handler = proc_dointvec_minmax,
3031 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3032 proc_handler *handler)
3035 struct neigh_sysctl_table *t;
3036 const char *dev_name_source;
3037 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3040 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3044 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3045 t->neigh_vars[i].data += (long) p;
3046 t->neigh_vars[i].extra1 = dev;
3047 t->neigh_vars[i].extra2 = p;
3051 dev_name_source = dev->name;
3052 /* Terminate the table early */
3053 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3054 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3056 struct neigh_table *tbl = p->tbl;
3057 dev_name_source = "default";
3058 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3059 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3060 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3061 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3066 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3068 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3069 /* RetransTime (in milliseconds)*/
3070 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3071 /* ReachableTime (in milliseconds) */
3072 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3075 /* Don't export sysctls to unprivileged users */
3076 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3077 t->neigh_vars[0].procname = NULL;
3079 switch (neigh_parms_family(p)) {
3090 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3091 p_name, dev_name_source);
3093 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3094 if (!t->sysctl_header)
3097 p->sysctl_table = t;
3105 EXPORT_SYMBOL(neigh_sysctl_register);
3107 void neigh_sysctl_unregister(struct neigh_parms *p)
3109 if (p->sysctl_table) {
3110 struct neigh_sysctl_table *t = p->sysctl_table;
3111 p->sysctl_table = NULL;
3112 unregister_net_sysctl_table(t->sysctl_header);
3116 EXPORT_SYMBOL(neigh_sysctl_unregister);
3118 #endif /* CONFIG_SYSCTL */
3120 static int __init neigh_init(void)
3122 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3123 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3124 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3126 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3128 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3133 subsys_initcall(neigh_init);