2 * Generic address resolution entity
5 * Pedro Roque <roque@di.fc.ul.pt>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
15 * Harald Welte Add neighbour cache statistics like rtstat
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/slab.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/socket.h>
25 #include <linux/netdevice.h>
26 #include <linux/proc_fs.h>
28 #include <linux/sysctl.h>
30 #include <linux/times.h>
31 #include <net/net_namespace.h>
32 #include <net/neighbour.h>
35 #include <net/netevent.h>
36 #include <net/netlink.h>
37 #include <linux/rtnetlink.h>
38 #include <linux/random.h>
39 #include <linux/string.h>
40 #include <linux/log2.h>
44 #define neigh_dbg(level, fmt, ...) \
46 if (level <= NEIGH_DEBUG) \
47 pr_debug(fmt, ##__VA_ARGS__); \
50 #define PNEIGH_HASHMASK 0xF
52 static void neigh_timer_handler(unsigned long arg);
53 static void __neigh_notify(struct neighbour *n, int type, int flags);
54 static void neigh_update_notify(struct neighbour *neigh);
55 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
57 static struct neigh_table *neigh_tables;
59 static const struct file_operations neigh_stat_seq_fops;
63 Neighbour hash table buckets are protected with rwlock tbl->lock.
65 - All the scans/updates to hash buckets MUST be made under this lock.
66 - NOTHING clever should be made under this lock: no callbacks
67 to protocol backends, no attempts to send something to network.
68 It will result in deadlocks, if backend/driver wants to use neighbour
70 - If the entry requires some non-trivial actions, increase
71 its reference count and release table lock.
73 Neighbour entries are protected:
74 - with reference count.
75 - with rwlock neigh->lock
77 Reference count prevents destruction.
79 neigh->lock mainly serializes ll address data and its validity state.
80 However, the same lock is used to protect another entry fields:
84 Again, nothing clever shall be made under neigh->lock,
85 the most complicated procedure, which we allow is dev->hard_header.
86 It is supposed, that dev->hard_header is simplistic and does
87 not make callbacks to neighbour tables.
89 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
90 list of neighbour tables. This list is used only in process context,
93 static DEFINE_RWLOCK(neigh_tbl_lock);
95 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
101 static void neigh_cleanup_and_release(struct neighbour *neigh)
103 if (neigh->parms->neigh_cleanup)
104 neigh->parms->neigh_cleanup(neigh);
106 __neigh_notify(neigh, RTM_DELNEIGH, 0);
107 neigh_release(neigh);
111 * It is random distribution in the interval (1/2)*base...(3/2)*base.
112 * It corresponds to default IPv6 settings and is not overridable,
113 * because it is really reasonable choice.
116 unsigned long neigh_rand_reach_time(unsigned long base)
118 return base ? (net_random() % base) + (base >> 1) : 0;
120 EXPORT_SYMBOL(neigh_rand_reach_time);
123 static int neigh_forced_gc(struct neigh_table *tbl)
127 struct neigh_hash_table *nht;
129 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
131 write_lock_bh(&tbl->lock);
132 nht = rcu_dereference_protected(tbl->nht,
133 lockdep_is_held(&tbl->lock));
134 for (i = 0; i < (1 << nht->hash_shift); i++) {
136 struct neighbour __rcu **np;
138 np = &nht->hash_buckets[i];
139 while ((n = rcu_dereference_protected(*np,
140 lockdep_is_held(&tbl->lock))) != NULL) {
141 /* Neighbour record may be discarded if:
142 * - nobody refers to it.
143 * - it is not permanent
145 write_lock(&n->lock);
146 if (atomic_read(&n->refcnt) == 1 &&
147 !(n->nud_state & NUD_PERMANENT)) {
148 rcu_assign_pointer(*np,
149 rcu_dereference_protected(n->next,
150 lockdep_is_held(&tbl->lock)));
153 write_unlock(&n->lock);
154 neigh_cleanup_and_release(n);
157 write_unlock(&n->lock);
162 tbl->last_flush = jiffies;
164 write_unlock_bh(&tbl->lock);
169 static void neigh_add_timer(struct neighbour *n, unsigned long when)
172 if (unlikely(mod_timer(&n->timer, when))) {
173 printk("NEIGH: BUG, double timer add, state is %x\n",
179 static int neigh_del_timer(struct neighbour *n)
181 if ((n->nud_state & NUD_IN_TIMER) &&
182 del_timer(&n->timer)) {
189 static void pneigh_queue_purge(struct sk_buff_head *list)
193 while ((skb = skb_dequeue(list)) != NULL) {
199 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
202 struct neigh_hash_table *nht;
204 nht = rcu_dereference_protected(tbl->nht,
205 lockdep_is_held(&tbl->lock));
207 for (i = 0; i < (1 << nht->hash_shift); i++) {
209 struct neighbour __rcu **np = &nht->hash_buckets[i];
211 while ((n = rcu_dereference_protected(*np,
212 lockdep_is_held(&tbl->lock))) != NULL) {
213 if (dev && n->dev != dev) {
217 rcu_assign_pointer(*np,
218 rcu_dereference_protected(n->next,
219 lockdep_is_held(&tbl->lock)));
220 write_lock(&n->lock);
224 if (atomic_read(&n->refcnt) != 1) {
225 /* The most unpleasant situation.
226 We must destroy neighbour entry,
227 but someone still uses it.
229 The destroy will be delayed until
230 the last user releases us, but
231 we must kill timers etc. and move
234 __skb_queue_purge(&n->arp_queue);
235 n->arp_queue_len_bytes = 0;
236 n->output = neigh_blackhole;
237 if (n->nud_state & NUD_VALID)
238 n->nud_state = NUD_NOARP;
240 n->nud_state = NUD_NONE;
241 neigh_dbg(2, "neigh %p is stray\n", n);
243 write_unlock(&n->lock);
244 neigh_cleanup_and_release(n);
249 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
251 write_lock_bh(&tbl->lock);
252 neigh_flush_dev(tbl, dev);
253 write_unlock_bh(&tbl->lock);
255 EXPORT_SYMBOL(neigh_changeaddr);
257 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
259 write_lock_bh(&tbl->lock);
260 neigh_flush_dev(tbl, dev);
261 pneigh_ifdown(tbl, dev);
262 write_unlock_bh(&tbl->lock);
264 del_timer_sync(&tbl->proxy_timer);
265 pneigh_queue_purge(&tbl->proxy_queue);
268 EXPORT_SYMBOL(neigh_ifdown);
270 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
272 struct neighbour *n = NULL;
273 unsigned long now = jiffies;
276 entries = atomic_inc_return(&tbl->entries) - 1;
277 if (entries >= tbl->gc_thresh3 ||
278 (entries >= tbl->gc_thresh2 &&
279 time_after(now, tbl->last_flush + 5 * HZ))) {
280 if (!neigh_forced_gc(tbl) &&
281 entries >= tbl->gc_thresh3)
285 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
289 __skb_queue_head_init(&n->arp_queue);
290 rwlock_init(&n->lock);
291 seqlock_init(&n->ha_lock);
292 n->updated = n->used = now;
293 n->nud_state = NUD_NONE;
294 n->output = neigh_blackhole;
295 seqlock_init(&n->hh.hh_lock);
296 n->parms = neigh_parms_clone(&tbl->parms);
297 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
299 NEIGH_CACHE_STAT_INC(tbl, allocs);
301 atomic_set(&n->refcnt, 1);
307 atomic_dec(&tbl->entries);
311 static void neigh_get_hash_rnd(u32 *x)
313 get_random_bytes(x, sizeof(*x));
317 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
319 size_t size = (1 << shift) * sizeof(struct neighbour *);
320 struct neigh_hash_table *ret;
321 struct neighbour __rcu **buckets;
324 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
327 if (size <= PAGE_SIZE)
328 buckets = kzalloc(size, GFP_ATOMIC);
330 buckets = (struct neighbour __rcu **)
331 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
337 ret->hash_buckets = buckets;
338 ret->hash_shift = shift;
339 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
340 neigh_get_hash_rnd(&ret->hash_rnd[i]);
344 static void neigh_hash_free_rcu(struct rcu_head *head)
346 struct neigh_hash_table *nht = container_of(head,
347 struct neigh_hash_table,
349 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
350 struct neighbour __rcu **buckets = nht->hash_buckets;
352 if (size <= PAGE_SIZE)
355 free_pages((unsigned long)buckets, get_order(size));
359 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
360 unsigned long new_shift)
362 unsigned int i, hash;
363 struct neigh_hash_table *new_nht, *old_nht;
365 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
367 old_nht = rcu_dereference_protected(tbl->nht,
368 lockdep_is_held(&tbl->lock));
369 new_nht = neigh_hash_alloc(new_shift);
373 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
374 struct neighbour *n, *next;
376 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
377 lockdep_is_held(&tbl->lock));
380 hash = tbl->hash(n->primary_key, n->dev,
383 hash >>= (32 - new_nht->hash_shift);
384 next = rcu_dereference_protected(n->next,
385 lockdep_is_held(&tbl->lock));
387 rcu_assign_pointer(n->next,
388 rcu_dereference_protected(
389 new_nht->hash_buckets[hash],
390 lockdep_is_held(&tbl->lock)));
391 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
395 rcu_assign_pointer(tbl->nht, new_nht);
396 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
400 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
401 struct net_device *dev)
404 int key_len = tbl->key_len;
406 struct neigh_hash_table *nht;
408 NEIGH_CACHE_STAT_INC(tbl, lookups);
411 nht = rcu_dereference_bh(tbl->nht);
412 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
414 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
416 n = rcu_dereference_bh(n->next)) {
417 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
418 if (!atomic_inc_not_zero(&n->refcnt))
420 NEIGH_CACHE_STAT_INC(tbl, hits);
425 rcu_read_unlock_bh();
428 EXPORT_SYMBOL(neigh_lookup);
430 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
434 int key_len = tbl->key_len;
436 struct neigh_hash_table *nht;
438 NEIGH_CACHE_STAT_INC(tbl, lookups);
441 nht = rcu_dereference_bh(tbl->nht);
442 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
444 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
446 n = rcu_dereference_bh(n->next)) {
447 if (!memcmp(n->primary_key, pkey, key_len) &&
448 net_eq(dev_net(n->dev), net)) {
449 if (!atomic_inc_not_zero(&n->refcnt))
451 NEIGH_CACHE_STAT_INC(tbl, hits);
456 rcu_read_unlock_bh();
459 EXPORT_SYMBOL(neigh_lookup_nodev);
461 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
462 struct net_device *dev, bool want_ref)
465 int key_len = tbl->key_len;
467 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
468 struct neigh_hash_table *nht;
471 rc = ERR_PTR(-ENOBUFS);
475 memcpy(n->primary_key, pkey, key_len);
479 /* Protocol specific setup. */
480 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
482 goto out_neigh_release;
485 if (dev->netdev_ops->ndo_neigh_construct) {
486 error = dev->netdev_ops->ndo_neigh_construct(n);
489 goto out_neigh_release;
493 /* Device specific setup. */
494 if (n->parms->neigh_setup &&
495 (error = n->parms->neigh_setup(n)) < 0) {
497 goto out_neigh_release;
500 n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
502 write_lock_bh(&tbl->lock);
503 nht = rcu_dereference_protected(tbl->nht,
504 lockdep_is_held(&tbl->lock));
506 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
507 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
509 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
511 if (n->parms->dead) {
512 rc = ERR_PTR(-EINVAL);
516 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
517 lockdep_is_held(&tbl->lock));
519 n1 = rcu_dereference_protected(n1->next,
520 lockdep_is_held(&tbl->lock))) {
521 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
532 rcu_assign_pointer(n->next,
533 rcu_dereference_protected(nht->hash_buckets[hash_val],
534 lockdep_is_held(&tbl->lock)));
535 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
536 write_unlock_bh(&tbl->lock);
537 neigh_dbg(2, "neigh %p is created\n", n);
542 write_unlock_bh(&tbl->lock);
547 EXPORT_SYMBOL(__neigh_create);
549 static u32 pneigh_hash(const void *pkey, int key_len)
551 u32 hash_val = *(u32 *)(pkey + key_len - 4);
552 hash_val ^= (hash_val >> 16);
553 hash_val ^= hash_val >> 8;
554 hash_val ^= hash_val >> 4;
555 hash_val &= PNEIGH_HASHMASK;
559 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
563 struct net_device *dev)
566 if (!memcmp(n->key, pkey, key_len) &&
567 net_eq(pneigh_net(n), net) &&
568 (n->dev == dev || !n->dev))
575 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
576 struct net *net, const void *pkey, struct net_device *dev)
578 int key_len = tbl->key_len;
579 u32 hash_val = pneigh_hash(pkey, key_len);
581 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
582 net, pkey, key_len, dev);
584 EXPORT_SYMBOL_GPL(__pneigh_lookup);
586 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
587 struct net *net, const void *pkey,
588 struct net_device *dev, int creat)
590 struct pneigh_entry *n;
591 int key_len = tbl->key_len;
592 u32 hash_val = pneigh_hash(pkey, key_len);
594 read_lock_bh(&tbl->lock);
595 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
596 net, pkey, key_len, dev);
597 read_unlock_bh(&tbl->lock);
604 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
608 write_pnet(&n->net, hold_net(net));
609 memcpy(n->key, pkey, key_len);
614 if (tbl->pconstructor && tbl->pconstructor(n)) {
623 write_lock_bh(&tbl->lock);
624 n->next = tbl->phash_buckets[hash_val];
625 tbl->phash_buckets[hash_val] = n;
626 write_unlock_bh(&tbl->lock);
630 EXPORT_SYMBOL(pneigh_lookup);
633 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
634 struct net_device *dev)
636 struct pneigh_entry *n, **np;
637 int key_len = tbl->key_len;
638 u32 hash_val = pneigh_hash(pkey, key_len);
640 write_lock_bh(&tbl->lock);
641 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
643 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
644 net_eq(pneigh_net(n), net)) {
646 write_unlock_bh(&tbl->lock);
647 if (tbl->pdestructor)
651 release_net(pneigh_net(n));
656 write_unlock_bh(&tbl->lock);
660 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
662 struct pneigh_entry *n, **np;
665 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
666 np = &tbl->phash_buckets[h];
667 while ((n = *np) != NULL) {
668 if (!dev || n->dev == dev) {
670 if (tbl->pdestructor)
674 release_net(pneigh_net(n));
684 static void neigh_parms_destroy(struct neigh_parms *parms);
686 static inline void neigh_parms_put(struct neigh_parms *parms)
688 if (atomic_dec_and_test(&parms->refcnt))
689 neigh_parms_destroy(parms);
693 * neighbour must already be out of the table;
696 void neigh_destroy(struct neighbour *neigh)
698 struct net_device *dev = neigh->dev;
700 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
703 pr_warn("Destroying alive neighbour %p\n", neigh);
708 if (neigh_del_timer(neigh))
709 pr_warn("Impossible event\n");
711 write_lock_bh(&neigh->lock);
712 __skb_queue_purge(&neigh->arp_queue);
713 write_unlock_bh(&neigh->lock);
714 neigh->arp_queue_len_bytes = 0;
716 if (dev->netdev_ops->ndo_neigh_destroy)
717 dev->netdev_ops->ndo_neigh_destroy(neigh);
720 neigh_parms_put(neigh->parms);
722 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
724 atomic_dec(&neigh->tbl->entries);
725 kfree_rcu(neigh, rcu);
727 EXPORT_SYMBOL(neigh_destroy);
729 /* Neighbour state is suspicious;
732 Called with write_locked neigh.
734 static void neigh_suspect(struct neighbour *neigh)
736 neigh_dbg(2, "neigh %p is suspected\n", neigh);
738 neigh->output = neigh->ops->output;
741 /* Neighbour state is OK;
744 Called with write_locked neigh.
746 static void neigh_connect(struct neighbour *neigh)
748 neigh_dbg(2, "neigh %p is connected\n", neigh);
750 neigh->output = neigh->ops->connected_output;
753 static void neigh_periodic_work(struct work_struct *work)
755 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
757 struct neighbour __rcu **np;
759 struct neigh_hash_table *nht;
761 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
763 write_lock_bh(&tbl->lock);
764 nht = rcu_dereference_protected(tbl->nht,
765 lockdep_is_held(&tbl->lock));
768 * periodically recompute ReachableTime from random function
771 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
772 struct neigh_parms *p;
773 tbl->last_rand = jiffies;
774 for (p = &tbl->parms; p; p = p->next)
776 neigh_rand_reach_time(p->base_reachable_time);
779 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
782 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
783 np = &nht->hash_buckets[i];
785 while ((n = rcu_dereference_protected(*np,
786 lockdep_is_held(&tbl->lock))) != NULL) {
789 write_lock(&n->lock);
791 state = n->nud_state;
792 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
793 write_unlock(&n->lock);
797 if (time_before(n->used, n->confirmed))
798 n->used = n->confirmed;
800 if (atomic_read(&n->refcnt) == 1 &&
801 (state == NUD_FAILED ||
802 time_after(jiffies, n->used + n->parms->gc_staletime))) {
805 write_unlock(&n->lock);
806 neigh_cleanup_and_release(n);
809 write_unlock(&n->lock);
815 * It's fine to release lock here, even if hash table
816 * grows while we are preempted.
818 write_unlock_bh(&tbl->lock);
820 write_lock_bh(&tbl->lock);
821 nht = rcu_dereference_protected(tbl->nht,
822 lockdep_is_held(&tbl->lock));
825 /* Cycle through all hash buckets every base_reachable_time/2 ticks.
826 * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
827 * base_reachable_time.
829 schedule_delayed_work(&tbl->gc_work,
830 tbl->parms.base_reachable_time >> 1);
831 write_unlock_bh(&tbl->lock);
834 static __inline__ int neigh_max_probes(struct neighbour *n)
836 struct neigh_parms *p = n->parms;
837 return (n->nud_state & NUD_PROBE) ?
839 p->ucast_probes + p->app_probes + p->mcast_probes;
842 static void neigh_invalidate(struct neighbour *neigh)
843 __releases(neigh->lock)
844 __acquires(neigh->lock)
848 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
849 neigh_dbg(2, "neigh %p is failed\n", neigh);
850 neigh->updated = jiffies;
852 /* It is very thin place. report_unreachable is very complicated
853 routine. Particularly, it can hit the same neighbour entry!
855 So that, we try to be accurate and avoid dead loop. --ANK
857 while (neigh->nud_state == NUD_FAILED &&
858 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
859 write_unlock(&neigh->lock);
860 neigh->ops->error_report(neigh, skb);
861 write_lock(&neigh->lock);
863 __skb_queue_purge(&neigh->arp_queue);
864 neigh->arp_queue_len_bytes = 0;
867 static void neigh_probe(struct neighbour *neigh)
868 __releases(neigh->lock)
870 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
871 /* keep skb alive even if arp_queue overflows */
873 skb = skb_copy(skb, GFP_ATOMIC);
874 write_unlock(&neigh->lock);
875 neigh->ops->solicit(neigh, skb);
876 atomic_inc(&neigh->probes);
880 /* Called when a timer expires for a neighbour entry. */
882 static void neigh_timer_handler(unsigned long arg)
884 unsigned long now, next;
885 struct neighbour *neigh = (struct neighbour *)arg;
889 write_lock(&neigh->lock);
891 state = neigh->nud_state;
895 if (!(state & NUD_IN_TIMER))
898 if (state & NUD_REACHABLE) {
899 if (time_before_eq(now,
900 neigh->confirmed + neigh->parms->reachable_time)) {
901 neigh_dbg(2, "neigh %p is still alive\n", neigh);
902 next = neigh->confirmed + neigh->parms->reachable_time;
903 } else if (time_before_eq(now,
904 neigh->used + neigh->parms->delay_probe_time)) {
905 neigh_dbg(2, "neigh %p is delayed\n", neigh);
906 neigh->nud_state = NUD_DELAY;
907 neigh->updated = jiffies;
908 neigh_suspect(neigh);
909 next = now + neigh->parms->delay_probe_time;
911 neigh_dbg(2, "neigh %p is suspected\n", neigh);
912 neigh->nud_state = NUD_STALE;
913 neigh->updated = jiffies;
914 neigh_suspect(neigh);
917 } else if (state & NUD_DELAY) {
918 if (time_before_eq(now,
919 neigh->confirmed + neigh->parms->delay_probe_time)) {
920 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
921 neigh->nud_state = NUD_REACHABLE;
922 neigh->updated = jiffies;
923 neigh_connect(neigh);
925 next = neigh->confirmed + neigh->parms->reachable_time;
927 neigh_dbg(2, "neigh %p is probed\n", neigh);
928 neigh->nud_state = NUD_PROBE;
929 neigh->updated = jiffies;
930 atomic_set(&neigh->probes, 0);
932 next = now + neigh->parms->retrans_time;
935 /* NUD_PROBE|NUD_INCOMPLETE */
936 next = now + neigh->parms->retrans_time;
939 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
940 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
941 neigh->nud_state = NUD_FAILED;
943 neigh_invalidate(neigh);
946 if (neigh->nud_state & NUD_IN_TIMER) {
947 if (time_before(next, jiffies + HZ/2))
948 next = jiffies + HZ/2;
949 if (!mod_timer(&neigh->timer, next))
952 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
956 write_unlock(&neigh->lock);
960 neigh_update_notify(neigh);
962 neigh_release(neigh);
965 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
968 bool immediate_probe = false;
970 write_lock_bh(&neigh->lock);
973 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
976 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
977 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
978 unsigned long next, now = jiffies;
980 atomic_set(&neigh->probes, neigh->parms->ucast_probes);
981 neigh->nud_state = NUD_INCOMPLETE;
982 neigh->updated = now;
983 next = now + max(neigh->parms->retrans_time, HZ/2);
984 neigh_add_timer(neigh, next);
985 immediate_probe = true;
987 neigh->nud_state = NUD_FAILED;
988 neigh->updated = jiffies;
989 write_unlock_bh(&neigh->lock);
994 } else if (neigh->nud_state & NUD_STALE) {
995 neigh_dbg(2, "neigh %p is delayed\n", neigh);
996 neigh->nud_state = NUD_DELAY;
997 neigh->updated = jiffies;
998 neigh_add_timer(neigh,
999 jiffies + neigh->parms->delay_probe_time);
1002 if (neigh->nud_state == NUD_INCOMPLETE) {
1004 while (neigh->arp_queue_len_bytes + skb->truesize >
1005 neigh->parms->queue_len_bytes) {
1006 struct sk_buff *buff;
1008 buff = __skb_dequeue(&neigh->arp_queue);
1011 neigh->arp_queue_len_bytes -= buff->truesize;
1013 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1016 __skb_queue_tail(&neigh->arp_queue, skb);
1017 neigh->arp_queue_len_bytes += skb->truesize;
1022 if (immediate_probe)
1025 write_unlock(&neigh->lock);
1029 EXPORT_SYMBOL(__neigh_event_send);
1031 static void neigh_update_hhs(struct neighbour *neigh)
1033 struct hh_cache *hh;
1034 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1037 if (neigh->dev->header_ops)
1038 update = neigh->dev->header_ops->cache_update;
1043 write_seqlock_bh(&hh->hh_lock);
1044 update(hh, neigh->dev, neigh->ha);
1045 write_sequnlock_bh(&hh->hh_lock);
1052 /* Generic update routine.
1053 -- lladdr is new lladdr or NULL, if it is not supplied.
1054 -- new is new state.
1056 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1058 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1059 lladdr instead of overriding it
1061 It also allows to retain current state
1062 if lladdr is unchanged.
1063 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1065 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1067 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1070 Caller MUST hold reference count on the entry.
1073 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1079 struct net_device *dev;
1080 int update_isrouter = 0;
1082 write_lock_bh(&neigh->lock);
1085 old = neigh->nud_state;
1088 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1089 (old & (NUD_NOARP | NUD_PERMANENT)))
1092 if (!(new & NUD_VALID)) {
1093 neigh_del_timer(neigh);
1094 if (old & NUD_CONNECTED)
1095 neigh_suspect(neigh);
1096 neigh->nud_state = new;
1098 notify = old & NUD_VALID;
1099 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1100 (new & NUD_FAILED)) {
1101 neigh_invalidate(neigh);
1107 /* Compare new lladdr with cached one */
1108 if (!dev->addr_len) {
1109 /* First case: device needs no address. */
1111 } else if (lladdr) {
1112 /* The second case: if something is already cached
1113 and a new address is proposed:
1115 - if they are different, check override flag
1117 if ((old & NUD_VALID) &&
1118 !memcmp(lladdr, neigh->ha, dev->addr_len))
1121 /* No address is supplied; if we know something,
1122 use it, otherwise discard the request.
1125 if (!(old & NUD_VALID))
1130 if (new & NUD_CONNECTED)
1131 neigh->confirmed = jiffies;
1132 neigh->updated = jiffies;
1134 /* If entry was valid and address is not changed,
1135 do not change entry state, if new one is STALE.
1138 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1139 if (old & NUD_VALID) {
1140 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1141 update_isrouter = 0;
1142 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1143 (old & NUD_CONNECTED)) {
1149 if (lladdr == neigh->ha && new == NUD_STALE &&
1150 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1151 (old & NUD_CONNECTED))
1158 neigh_del_timer(neigh);
1159 if (new & NUD_PROBE)
1160 atomic_set(&neigh->probes, 0);
1161 if (new & NUD_IN_TIMER)
1162 neigh_add_timer(neigh, (jiffies +
1163 ((new & NUD_REACHABLE) ?
1164 neigh->parms->reachable_time :
1166 neigh->nud_state = new;
1169 if (lladdr != neigh->ha) {
1170 write_seqlock(&neigh->ha_lock);
1171 memcpy(&neigh->ha, lladdr, dev->addr_len);
1172 write_sequnlock(&neigh->ha_lock);
1173 neigh_update_hhs(neigh);
1174 if (!(new & NUD_CONNECTED))
1175 neigh->confirmed = jiffies -
1176 (neigh->parms->base_reachable_time << 1);
1181 if (new & NUD_CONNECTED)
1182 neigh_connect(neigh);
1184 neigh_suspect(neigh);
1185 if (!(old & NUD_VALID)) {
1186 struct sk_buff *skb;
1188 /* Again: avoid dead loop if something went wrong */
1190 while (neigh->nud_state & NUD_VALID &&
1191 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1192 struct dst_entry *dst = skb_dst(skb);
1193 struct neighbour *n2, *n1 = neigh;
1194 write_unlock_bh(&neigh->lock);
1198 /* Why not just use 'neigh' as-is? The problem is that
1199 * things such as shaper, eql, and sch_teql can end up
1200 * using alternative, different, neigh objects to output
1201 * the packet in the output path. So what we need to do
1202 * here is re-lookup the top-level neigh in the path so
1203 * we can reinject the packet there.
1207 n2 = dst_neigh_lookup_skb(dst, skb);
1211 n1->output(n1, skb);
1216 write_lock_bh(&neigh->lock);
1218 __skb_queue_purge(&neigh->arp_queue);
1219 neigh->arp_queue_len_bytes = 0;
1222 if (update_isrouter) {
1223 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1224 (neigh->flags | NTF_ROUTER) :
1225 (neigh->flags & ~NTF_ROUTER);
1227 write_unlock_bh(&neigh->lock);
1230 neigh_update_notify(neigh);
1234 EXPORT_SYMBOL(neigh_update);
1236 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1237 u8 *lladdr, void *saddr,
1238 struct net_device *dev)
1240 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1241 lladdr || !dev->addr_len);
1243 neigh_update(neigh, lladdr, NUD_STALE,
1244 NEIGH_UPDATE_F_OVERRIDE);
1247 EXPORT_SYMBOL(neigh_event_ns);
1249 /* called with read_lock_bh(&n->lock); */
1250 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1252 struct net_device *dev = dst->dev;
1253 __be16 prot = dst->ops->protocol;
1254 struct hh_cache *hh = &n->hh;
1256 write_lock_bh(&n->lock);
1258 /* Only one thread can come in here and initialize the
1262 dev->header_ops->cache(n, hh, prot);
1264 write_unlock_bh(&n->lock);
1267 /* This function can be used in contexts, where only old dev_queue_xmit
1268 * worked, f.e. if you want to override normal output path (eql, shaper),
1269 * but resolution is not made yet.
1272 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1274 struct net_device *dev = skb->dev;
1276 __skb_pull(skb, skb_network_offset(skb));
1278 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1280 dev_rebuild_header(skb))
1283 return dev_queue_xmit(skb);
1285 EXPORT_SYMBOL(neigh_compat_output);
1287 /* Slow and careful. */
1289 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1291 struct dst_entry *dst = skb_dst(skb);
1297 if (!neigh_event_send(neigh, skb)) {
1299 struct net_device *dev = neigh->dev;
1302 if (dev->header_ops->cache && !neigh->hh.hh_len)
1303 neigh_hh_init(neigh, dst);
1306 __skb_pull(skb, skb_network_offset(skb));
1307 seq = read_seqbegin(&neigh->ha_lock);
1308 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1309 neigh->ha, NULL, skb->len);
1310 } while (read_seqretry(&neigh->ha_lock, seq));
1313 rc = dev_queue_xmit(skb);
1320 neigh_dbg(1, "%s: dst=%p neigh=%p\n", __func__, dst, neigh);
1326 EXPORT_SYMBOL(neigh_resolve_output);
1328 /* As fast as possible without hh cache */
1330 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1332 struct net_device *dev = neigh->dev;
1337 __skb_pull(skb, skb_network_offset(skb));
1338 seq = read_seqbegin(&neigh->ha_lock);
1339 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1340 neigh->ha, NULL, skb->len);
1341 } while (read_seqretry(&neigh->ha_lock, seq));
1344 err = dev_queue_xmit(skb);
1351 EXPORT_SYMBOL(neigh_connected_output);
1353 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1355 return dev_queue_xmit(skb);
1357 EXPORT_SYMBOL(neigh_direct_output);
1359 static void neigh_proxy_process(unsigned long arg)
1361 struct neigh_table *tbl = (struct neigh_table *)arg;
1362 long sched_next = 0;
1363 unsigned long now = jiffies;
1364 struct sk_buff *skb, *n;
1366 spin_lock(&tbl->proxy_queue.lock);
1368 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1369 long tdif = NEIGH_CB(skb)->sched_next - now;
1372 struct net_device *dev = skb->dev;
1374 __skb_unlink(skb, &tbl->proxy_queue);
1375 if (tbl->proxy_redo && netif_running(dev)) {
1377 tbl->proxy_redo(skb);
1384 } else if (!sched_next || tdif < sched_next)
1387 del_timer(&tbl->proxy_timer);
1389 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1390 spin_unlock(&tbl->proxy_queue.lock);
1393 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1394 struct sk_buff *skb)
1396 unsigned long now = jiffies;
1397 unsigned long sched_next = now + (net_random() % p->proxy_delay);
1399 if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1404 NEIGH_CB(skb)->sched_next = sched_next;
1405 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1407 spin_lock(&tbl->proxy_queue.lock);
1408 if (del_timer(&tbl->proxy_timer)) {
1409 if (time_before(tbl->proxy_timer.expires, sched_next))
1410 sched_next = tbl->proxy_timer.expires;
1414 __skb_queue_tail(&tbl->proxy_queue, skb);
1415 mod_timer(&tbl->proxy_timer, sched_next);
1416 spin_unlock(&tbl->proxy_queue.lock);
1418 EXPORT_SYMBOL(pneigh_enqueue);
1420 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1421 struct net *net, int ifindex)
1423 struct neigh_parms *p;
1425 for (p = &tbl->parms; p; p = p->next) {
1426 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1427 (!p->dev && !ifindex))
1434 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1435 struct neigh_table *tbl)
1437 struct neigh_parms *p, *ref;
1438 struct net *net = dev_net(dev);
1439 const struct net_device_ops *ops = dev->netdev_ops;
1441 ref = lookup_neigh_parms(tbl, net, 0);
1445 p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1448 atomic_set(&p->refcnt, 1);
1450 neigh_rand_reach_time(p->base_reachable_time);
1453 write_pnet(&p->net, hold_net(net));
1454 p->sysctl_table = NULL;
1456 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1463 write_lock_bh(&tbl->lock);
1464 p->next = tbl->parms.next;
1465 tbl->parms.next = p;
1466 write_unlock_bh(&tbl->lock);
1470 EXPORT_SYMBOL(neigh_parms_alloc);
1472 static void neigh_rcu_free_parms(struct rcu_head *head)
1474 struct neigh_parms *parms =
1475 container_of(head, struct neigh_parms, rcu_head);
1477 neigh_parms_put(parms);
1480 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1482 struct neigh_parms **p;
1484 if (!parms || parms == &tbl->parms)
1486 write_lock_bh(&tbl->lock);
1487 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1491 write_unlock_bh(&tbl->lock);
1493 dev_put(parms->dev);
1494 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1498 write_unlock_bh(&tbl->lock);
1499 neigh_dbg(1, "%s: not found\n", __func__);
1501 EXPORT_SYMBOL(neigh_parms_release);
1503 static void neigh_parms_destroy(struct neigh_parms *parms)
1505 release_net(neigh_parms_net(parms));
1509 static struct lock_class_key neigh_table_proxy_queue_class;
1511 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1513 unsigned long now = jiffies;
1514 unsigned long phsize;
1516 write_pnet(&tbl->parms.net, &init_net);
1517 atomic_set(&tbl->parms.refcnt, 1);
1518 tbl->parms.reachable_time =
1519 neigh_rand_reach_time(tbl->parms.base_reachable_time);
1521 tbl->stats = alloc_percpu(struct neigh_statistics);
1523 panic("cannot create neighbour cache statistics");
1525 #ifdef CONFIG_PROC_FS
1526 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1527 &neigh_stat_seq_fops, tbl))
1528 panic("cannot create neighbour proc dir entry");
1531 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1533 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1534 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1536 if (!tbl->nht || !tbl->phash_buckets)
1537 panic("cannot allocate neighbour cache hashes");
1539 if (!tbl->entry_size)
1540 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1541 tbl->key_len, NEIGH_PRIV_ALIGN);
1543 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1545 rwlock_init(&tbl->lock);
1546 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1547 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1548 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1549 skb_queue_head_init_class(&tbl->proxy_queue,
1550 &neigh_table_proxy_queue_class);
1552 tbl->last_flush = now;
1553 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1556 void neigh_table_init(struct neigh_table *tbl)
1558 struct neigh_table *tmp;
1560 neigh_table_init_no_netlink(tbl);
1561 write_lock(&neigh_tbl_lock);
1562 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1563 if (tmp->family == tbl->family)
1566 tbl->next = neigh_tables;
1568 write_unlock(&neigh_tbl_lock);
1570 if (unlikely(tmp)) {
1571 pr_err("Registering multiple tables for family %d\n",
1576 EXPORT_SYMBOL(neigh_table_init);
1578 int neigh_table_clear(struct neigh_table *tbl)
1580 struct neigh_table **tp;
1582 /* It is not clean... Fix it to unload IPv6 module safely */
1583 cancel_delayed_work_sync(&tbl->gc_work);
1584 del_timer_sync(&tbl->proxy_timer);
1585 pneigh_queue_purge(&tbl->proxy_queue);
1586 neigh_ifdown(tbl, NULL);
1587 if (atomic_read(&tbl->entries))
1588 pr_crit("neighbour leakage\n");
1589 write_lock(&neigh_tbl_lock);
1590 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1596 write_unlock(&neigh_tbl_lock);
1598 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1599 neigh_hash_free_rcu);
1602 kfree(tbl->phash_buckets);
1603 tbl->phash_buckets = NULL;
1605 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1607 free_percpu(tbl->stats);
1612 EXPORT_SYMBOL(neigh_table_clear);
1614 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1616 struct net *net = sock_net(skb->sk);
1618 struct nlattr *dst_attr;
1619 struct neigh_table *tbl;
1620 struct net_device *dev = NULL;
1624 if (nlmsg_len(nlh) < sizeof(*ndm))
1627 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1628 if (dst_attr == NULL)
1631 ndm = nlmsg_data(nlh);
1632 if (ndm->ndm_ifindex) {
1633 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1640 read_lock(&neigh_tbl_lock);
1641 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1642 struct neighbour *neigh;
1644 if (tbl->family != ndm->ndm_family)
1646 read_unlock(&neigh_tbl_lock);
1648 if (nla_len(dst_attr) < tbl->key_len)
1651 if (ndm->ndm_flags & NTF_PROXY) {
1652 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1659 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1660 if (neigh == NULL) {
1665 err = neigh_update(neigh, NULL, NUD_FAILED,
1666 NEIGH_UPDATE_F_OVERRIDE |
1667 NEIGH_UPDATE_F_ADMIN);
1668 neigh_release(neigh);
1671 read_unlock(&neigh_tbl_lock);
1672 err = -EAFNOSUPPORT;
1678 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1680 struct net *net = sock_net(skb->sk);
1682 struct nlattr *tb[NDA_MAX+1];
1683 struct neigh_table *tbl;
1684 struct net_device *dev = NULL;
1688 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1693 if (tb[NDA_DST] == NULL)
1696 ndm = nlmsg_data(nlh);
1697 if (ndm->ndm_ifindex) {
1698 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1704 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1708 read_lock(&neigh_tbl_lock);
1709 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1710 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1711 struct neighbour *neigh;
1714 if (tbl->family != ndm->ndm_family)
1716 read_unlock(&neigh_tbl_lock);
1718 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1720 dst = nla_data(tb[NDA_DST]);
1721 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1723 if (ndm->ndm_flags & NTF_PROXY) {
1724 struct pneigh_entry *pn;
1727 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1729 pn->flags = ndm->ndm_flags;
1738 neigh = neigh_lookup(tbl, dst, dev);
1739 if (neigh == NULL) {
1740 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1745 neigh = __neigh_lookup_errno(tbl, dst, dev);
1746 if (IS_ERR(neigh)) {
1747 err = PTR_ERR(neigh);
1751 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1753 neigh_release(neigh);
1757 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1758 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1761 if (ndm->ndm_flags & NTF_USE) {
1762 neigh_event_send(neigh, NULL);
1765 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1766 neigh_release(neigh);
1770 read_unlock(&neigh_tbl_lock);
1771 err = -EAFNOSUPPORT;
1776 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1778 struct nlattr *nest;
1780 nest = nla_nest_start(skb, NDTA_PARMS);
1785 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1786 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1787 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES, parms->queue_len_bytes) ||
1788 /* approximative value for deprecated QUEUE_LEN (in packets) */
1789 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1790 parms->queue_len_bytes / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1791 nla_put_u32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen) ||
1792 nla_put_u32(skb, NDTPA_APP_PROBES, parms->app_probes) ||
1793 nla_put_u32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes) ||
1794 nla_put_u32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes) ||
1795 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1796 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1797 parms->base_reachable_time) ||
1798 nla_put_msecs(skb, NDTPA_GC_STALETIME, parms->gc_staletime) ||
1799 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1800 parms->delay_probe_time) ||
1801 nla_put_msecs(skb, NDTPA_RETRANS_TIME, parms->retrans_time) ||
1802 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay) ||
1803 nla_put_msecs(skb, NDTPA_PROXY_DELAY, parms->proxy_delay) ||
1804 nla_put_msecs(skb, NDTPA_LOCKTIME, parms->locktime))
1805 goto nla_put_failure;
1806 return nla_nest_end(skb, nest);
1809 nla_nest_cancel(skb, nest);
1813 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1814 u32 pid, u32 seq, int type, int flags)
1816 struct nlmsghdr *nlh;
1817 struct ndtmsg *ndtmsg;
1819 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1823 ndtmsg = nlmsg_data(nlh);
1825 read_lock_bh(&tbl->lock);
1826 ndtmsg->ndtm_family = tbl->family;
1827 ndtmsg->ndtm_pad1 = 0;
1828 ndtmsg->ndtm_pad2 = 0;
1830 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1831 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1832 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1833 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1834 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1835 goto nla_put_failure;
1837 unsigned long now = jiffies;
1838 unsigned int flush_delta = now - tbl->last_flush;
1839 unsigned int rand_delta = now - tbl->last_rand;
1840 struct neigh_hash_table *nht;
1841 struct ndt_config ndc = {
1842 .ndtc_key_len = tbl->key_len,
1843 .ndtc_entry_size = tbl->entry_size,
1844 .ndtc_entries = atomic_read(&tbl->entries),
1845 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1846 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1847 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1851 nht = rcu_dereference_bh(tbl->nht);
1852 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1853 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1854 rcu_read_unlock_bh();
1856 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1857 goto nla_put_failure;
1862 struct ndt_stats ndst;
1864 memset(&ndst, 0, sizeof(ndst));
1866 for_each_possible_cpu(cpu) {
1867 struct neigh_statistics *st;
1869 st = per_cpu_ptr(tbl->stats, cpu);
1870 ndst.ndts_allocs += st->allocs;
1871 ndst.ndts_destroys += st->destroys;
1872 ndst.ndts_hash_grows += st->hash_grows;
1873 ndst.ndts_res_failed += st->res_failed;
1874 ndst.ndts_lookups += st->lookups;
1875 ndst.ndts_hits += st->hits;
1876 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1877 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1878 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1879 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1882 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1883 goto nla_put_failure;
1886 BUG_ON(tbl->parms.dev);
1887 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1888 goto nla_put_failure;
1890 read_unlock_bh(&tbl->lock);
1891 return nlmsg_end(skb, nlh);
1894 read_unlock_bh(&tbl->lock);
1895 nlmsg_cancel(skb, nlh);
1899 static int neightbl_fill_param_info(struct sk_buff *skb,
1900 struct neigh_table *tbl,
1901 struct neigh_parms *parms,
1902 u32 pid, u32 seq, int type,
1905 struct ndtmsg *ndtmsg;
1906 struct nlmsghdr *nlh;
1908 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1912 ndtmsg = nlmsg_data(nlh);
1914 read_lock_bh(&tbl->lock);
1915 ndtmsg->ndtm_family = tbl->family;
1916 ndtmsg->ndtm_pad1 = 0;
1917 ndtmsg->ndtm_pad2 = 0;
1919 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1920 neightbl_fill_parms(skb, parms) < 0)
1923 read_unlock_bh(&tbl->lock);
1924 return nlmsg_end(skb, nlh);
1926 read_unlock_bh(&tbl->lock);
1927 nlmsg_cancel(skb, nlh);
1931 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1932 [NDTA_NAME] = { .type = NLA_STRING },
1933 [NDTA_THRESH1] = { .type = NLA_U32 },
1934 [NDTA_THRESH2] = { .type = NLA_U32 },
1935 [NDTA_THRESH3] = { .type = NLA_U32 },
1936 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1937 [NDTA_PARMS] = { .type = NLA_NESTED },
1940 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1941 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1942 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1943 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1944 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1945 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1946 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1947 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1948 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1949 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1950 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1951 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1952 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1953 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1956 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1958 struct net *net = sock_net(skb->sk);
1959 struct neigh_table *tbl;
1960 struct ndtmsg *ndtmsg;
1961 struct nlattr *tb[NDTA_MAX+1];
1964 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1965 nl_neightbl_policy);
1969 if (tb[NDTA_NAME] == NULL) {
1974 ndtmsg = nlmsg_data(nlh);
1975 read_lock(&neigh_tbl_lock);
1976 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1977 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1980 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1990 * We acquire tbl->lock to be nice to the periodic timers and
1991 * make sure they always see a consistent set of values.
1993 write_lock_bh(&tbl->lock);
1995 if (tb[NDTA_PARMS]) {
1996 struct nlattr *tbp[NDTPA_MAX+1];
1997 struct neigh_parms *p;
2000 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2001 nl_ntbl_parm_policy);
2003 goto errout_tbl_lock;
2005 if (tbp[NDTPA_IFINDEX])
2006 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2008 p = lookup_neigh_parms(tbl, net, ifindex);
2011 goto errout_tbl_lock;
2014 for (i = 1; i <= NDTPA_MAX; i++) {
2019 case NDTPA_QUEUE_LEN:
2020 p->queue_len_bytes = nla_get_u32(tbp[i]) *
2021 SKB_TRUESIZE(ETH_FRAME_LEN);
2023 case NDTPA_QUEUE_LENBYTES:
2024 p->queue_len_bytes = nla_get_u32(tbp[i]);
2026 case NDTPA_PROXY_QLEN:
2027 p->proxy_qlen = nla_get_u32(tbp[i]);
2029 case NDTPA_APP_PROBES:
2030 p->app_probes = nla_get_u32(tbp[i]);
2032 case NDTPA_UCAST_PROBES:
2033 p->ucast_probes = nla_get_u32(tbp[i]);
2035 case NDTPA_MCAST_PROBES:
2036 p->mcast_probes = nla_get_u32(tbp[i]);
2038 case NDTPA_BASE_REACHABLE_TIME:
2039 p->base_reachable_time = nla_get_msecs(tbp[i]);
2041 case NDTPA_GC_STALETIME:
2042 p->gc_staletime = nla_get_msecs(tbp[i]);
2044 case NDTPA_DELAY_PROBE_TIME:
2045 p->delay_probe_time = nla_get_msecs(tbp[i]);
2047 case NDTPA_RETRANS_TIME:
2048 p->retrans_time = nla_get_msecs(tbp[i]);
2050 case NDTPA_ANYCAST_DELAY:
2051 p->anycast_delay = nla_get_msecs(tbp[i]);
2053 case NDTPA_PROXY_DELAY:
2054 p->proxy_delay = nla_get_msecs(tbp[i]);
2056 case NDTPA_LOCKTIME:
2057 p->locktime = nla_get_msecs(tbp[i]);
2063 if (tb[NDTA_THRESH1])
2064 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2066 if (tb[NDTA_THRESH2])
2067 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2069 if (tb[NDTA_THRESH3])
2070 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2072 if (tb[NDTA_GC_INTERVAL])
2073 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2078 write_unlock_bh(&tbl->lock);
2080 read_unlock(&neigh_tbl_lock);
2085 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2087 struct net *net = sock_net(skb->sk);
2088 int family, tidx, nidx = 0;
2089 int tbl_skip = cb->args[0];
2090 int neigh_skip = cb->args[1];
2091 struct neigh_table *tbl;
2093 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2095 read_lock(&neigh_tbl_lock);
2096 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2097 struct neigh_parms *p;
2099 if (tidx < tbl_skip || (family && tbl->family != family))
2102 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2103 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2107 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2108 if (!net_eq(neigh_parms_net(p), net))
2111 if (nidx < neigh_skip)
2114 if (neightbl_fill_param_info(skb, tbl, p,
2115 NETLINK_CB(cb->skb).portid,
2127 read_unlock(&neigh_tbl_lock);
2134 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2135 u32 pid, u32 seq, int type, unsigned int flags)
2137 unsigned long now = jiffies;
2138 struct nda_cacheinfo ci;
2139 struct nlmsghdr *nlh;
2142 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2146 ndm = nlmsg_data(nlh);
2147 ndm->ndm_family = neigh->ops->family;
2150 ndm->ndm_flags = neigh->flags;
2151 ndm->ndm_type = neigh->type;
2152 ndm->ndm_ifindex = neigh->dev->ifindex;
2154 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2155 goto nla_put_failure;
2157 read_lock_bh(&neigh->lock);
2158 ndm->ndm_state = neigh->nud_state;
2159 if (neigh->nud_state & NUD_VALID) {
2160 char haddr[MAX_ADDR_LEN];
2162 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2163 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2164 read_unlock_bh(&neigh->lock);
2165 goto nla_put_failure;
2169 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2170 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2171 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2172 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2173 read_unlock_bh(&neigh->lock);
2175 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2176 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2177 goto nla_put_failure;
2179 return nlmsg_end(skb, nlh);
2182 nlmsg_cancel(skb, nlh);
2186 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2187 u32 pid, u32 seq, int type, unsigned int flags,
2188 struct neigh_table *tbl)
2190 struct nlmsghdr *nlh;
2193 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2197 ndm = nlmsg_data(nlh);
2198 ndm->ndm_family = tbl->family;
2201 ndm->ndm_flags = pn->flags | NTF_PROXY;
2202 ndm->ndm_type = NDA_DST;
2203 ndm->ndm_ifindex = pn->dev->ifindex;
2204 ndm->ndm_state = NUD_NONE;
2206 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2207 goto nla_put_failure;
2209 return nlmsg_end(skb, nlh);
2212 nlmsg_cancel(skb, nlh);
2216 static void neigh_update_notify(struct neighbour *neigh)
2218 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2219 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2222 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2223 struct netlink_callback *cb)
2225 struct net *net = sock_net(skb->sk);
2226 struct neighbour *n;
2227 int rc, h, s_h = cb->args[1];
2228 int idx, s_idx = idx = cb->args[2];
2229 struct neigh_hash_table *nht;
2232 nht = rcu_dereference_bh(tbl->nht);
2234 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2237 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2239 n = rcu_dereference_bh(n->next)) {
2240 if (!net_eq(dev_net(n->dev), net))
2244 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2247 NLM_F_MULTI) <= 0) {
2257 rcu_read_unlock_bh();
2263 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2264 struct netlink_callback *cb)
2266 struct pneigh_entry *n;
2267 struct net *net = sock_net(skb->sk);
2268 int rc, h, s_h = cb->args[3];
2269 int idx, s_idx = idx = cb->args[4];
2271 read_lock_bh(&tbl->lock);
2273 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2276 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2277 if (dev_net(n->dev) != net)
2281 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2284 NLM_F_MULTI, tbl) <= 0) {
2285 read_unlock_bh(&tbl->lock);
2294 read_unlock_bh(&tbl->lock);
2303 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2305 struct neigh_table *tbl;
2310 read_lock(&neigh_tbl_lock);
2311 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2313 /* check for full ndmsg structure presence, family member is
2314 * the same for both structures
2316 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2317 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2322 for (tbl = neigh_tables, t = 0; tbl;
2323 tbl = tbl->next, t++) {
2324 if (t < s_t || (family && tbl->family != family))
2327 memset(&cb->args[1], 0, sizeof(cb->args) -
2328 sizeof(cb->args[0]));
2330 err = pneigh_dump_table(tbl, skb, cb);
2332 err = neigh_dump_table(tbl, skb, cb);
2336 read_unlock(&neigh_tbl_lock);
2342 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2345 struct neigh_hash_table *nht;
2348 nht = rcu_dereference_bh(tbl->nht);
2350 read_lock(&tbl->lock); /* avoid resizes */
2351 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2352 struct neighbour *n;
2354 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2356 n = rcu_dereference_bh(n->next))
2359 read_unlock(&tbl->lock);
2360 rcu_read_unlock_bh();
2362 EXPORT_SYMBOL(neigh_for_each);
2364 /* The tbl->lock must be held as a writer and BH disabled. */
2365 void __neigh_for_each_release(struct neigh_table *tbl,
2366 int (*cb)(struct neighbour *))
2369 struct neigh_hash_table *nht;
2371 nht = rcu_dereference_protected(tbl->nht,
2372 lockdep_is_held(&tbl->lock));
2373 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2374 struct neighbour *n;
2375 struct neighbour __rcu **np;
2377 np = &nht->hash_buckets[chain];
2378 while ((n = rcu_dereference_protected(*np,
2379 lockdep_is_held(&tbl->lock))) != NULL) {
2382 write_lock(&n->lock);
2385 rcu_assign_pointer(*np,
2386 rcu_dereference_protected(n->next,
2387 lockdep_is_held(&tbl->lock)));
2391 write_unlock(&n->lock);
2393 neigh_cleanup_and_release(n);
2397 EXPORT_SYMBOL(__neigh_for_each_release);
2399 #ifdef CONFIG_PROC_FS
2401 static struct neighbour *neigh_get_first(struct seq_file *seq)
2403 struct neigh_seq_state *state = seq->private;
2404 struct net *net = seq_file_net(seq);
2405 struct neigh_hash_table *nht = state->nht;
2406 struct neighbour *n = NULL;
2407 int bucket = state->bucket;
2409 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2410 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2411 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2414 if (!net_eq(dev_net(n->dev), net))
2416 if (state->neigh_sub_iter) {
2420 v = state->neigh_sub_iter(state, n, &fakep);
2424 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2426 if (n->nud_state & ~NUD_NOARP)
2429 n = rcu_dereference_bh(n->next);
2435 state->bucket = bucket;
2440 static struct neighbour *neigh_get_next(struct seq_file *seq,
2441 struct neighbour *n,
2444 struct neigh_seq_state *state = seq->private;
2445 struct net *net = seq_file_net(seq);
2446 struct neigh_hash_table *nht = state->nht;
2448 if (state->neigh_sub_iter) {
2449 void *v = state->neigh_sub_iter(state, n, pos);
2453 n = rcu_dereference_bh(n->next);
2457 if (!net_eq(dev_net(n->dev), net))
2459 if (state->neigh_sub_iter) {
2460 void *v = state->neigh_sub_iter(state, n, pos);
2465 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2468 if (n->nud_state & ~NUD_NOARP)
2471 n = rcu_dereference_bh(n->next);
2477 if (++state->bucket >= (1 << nht->hash_shift))
2480 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2488 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2490 struct neighbour *n = neigh_get_first(seq);
2495 n = neigh_get_next(seq, n, pos);
2500 return *pos ? NULL : n;
2503 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2505 struct neigh_seq_state *state = seq->private;
2506 struct net *net = seq_file_net(seq);
2507 struct neigh_table *tbl = state->tbl;
2508 struct pneigh_entry *pn = NULL;
2509 int bucket = state->bucket;
2511 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2512 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2513 pn = tbl->phash_buckets[bucket];
2514 while (pn && !net_eq(pneigh_net(pn), net))
2519 state->bucket = bucket;
2524 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2525 struct pneigh_entry *pn,
2528 struct neigh_seq_state *state = seq->private;
2529 struct net *net = seq_file_net(seq);
2530 struct neigh_table *tbl = state->tbl;
2534 } while (pn && !net_eq(pneigh_net(pn), net));
2537 if (++state->bucket > PNEIGH_HASHMASK)
2539 pn = tbl->phash_buckets[state->bucket];
2540 while (pn && !net_eq(pneigh_net(pn), net))
2552 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2554 struct pneigh_entry *pn = pneigh_get_first(seq);
2559 pn = pneigh_get_next(seq, pn, pos);
2564 return *pos ? NULL : pn;
2567 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2569 struct neigh_seq_state *state = seq->private;
2571 loff_t idxpos = *pos;
2573 rc = neigh_get_idx(seq, &idxpos);
2574 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2575 rc = pneigh_get_idx(seq, &idxpos);
2580 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2583 struct neigh_seq_state *state = seq->private;
2587 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2590 state->nht = rcu_dereference_bh(tbl->nht);
2592 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2594 EXPORT_SYMBOL(neigh_seq_start);
2596 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2598 struct neigh_seq_state *state;
2601 if (v == SEQ_START_TOKEN) {
2602 rc = neigh_get_first(seq);
2606 state = seq->private;
2607 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2608 rc = neigh_get_next(seq, v, NULL);
2611 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2612 rc = pneigh_get_first(seq);
2614 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2615 rc = pneigh_get_next(seq, v, NULL);
2621 EXPORT_SYMBOL(neigh_seq_next);
2623 void neigh_seq_stop(struct seq_file *seq, void *v)
2626 rcu_read_unlock_bh();
2628 EXPORT_SYMBOL(neigh_seq_stop);
2630 /* statistics via seq_file */
2632 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2634 struct neigh_table *tbl = seq->private;
2638 return SEQ_START_TOKEN;
2640 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2641 if (!cpu_possible(cpu))
2644 return per_cpu_ptr(tbl->stats, cpu);
2649 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2651 struct neigh_table *tbl = seq->private;
2654 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2655 if (!cpu_possible(cpu))
2658 return per_cpu_ptr(tbl->stats, cpu);
2663 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2668 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2670 struct neigh_table *tbl = seq->private;
2671 struct neigh_statistics *st = v;
2673 if (v == SEQ_START_TOKEN) {
2674 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");
2678 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2679 "%08lx %08lx %08lx %08lx %08lx\n",
2680 atomic_read(&tbl->entries),
2691 st->rcv_probes_mcast,
2692 st->rcv_probes_ucast,
2694 st->periodic_gc_runs,
2702 static const struct seq_operations neigh_stat_seq_ops = {
2703 .start = neigh_stat_seq_start,
2704 .next = neigh_stat_seq_next,
2705 .stop = neigh_stat_seq_stop,
2706 .show = neigh_stat_seq_show,
2709 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2711 int ret = seq_open(file, &neigh_stat_seq_ops);
2714 struct seq_file *sf = file->private_data;
2715 sf->private = PDE_DATA(inode);
2720 static const struct file_operations neigh_stat_seq_fops = {
2721 .owner = THIS_MODULE,
2722 .open = neigh_stat_seq_open,
2724 .llseek = seq_lseek,
2725 .release = seq_release,
2728 #endif /* CONFIG_PROC_FS */
2730 static inline size_t neigh_nlmsg_size(void)
2732 return NLMSG_ALIGN(sizeof(struct ndmsg))
2733 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2734 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2735 + nla_total_size(sizeof(struct nda_cacheinfo))
2736 + nla_total_size(4); /* NDA_PROBES */
2739 static void __neigh_notify(struct neighbour *n, int type, int flags)
2741 struct net *net = dev_net(n->dev);
2742 struct sk_buff *skb;
2745 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2749 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2751 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2752 WARN_ON(err == -EMSGSIZE);
2756 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2760 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2764 void neigh_app_ns(struct neighbour *n)
2766 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2768 EXPORT_SYMBOL(neigh_app_ns);
2769 #endif /* CONFIG_ARPD */
2771 #ifdef CONFIG_SYSCTL
2773 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2775 static int proc_unres_qlen(ctl_table *ctl, int write, void __user *buffer,
2776 size_t *lenp, loff_t *ppos)
2779 ctl_table tmp = *ctl;
2782 tmp.extra2 = &unres_qlen_max;
2785 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2786 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2789 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2794 NEIGH_VAR_MCAST_PROBE,
2795 NEIGH_VAR_UCAST_PROBE,
2796 NEIGH_VAR_APP_PROBE,
2797 NEIGH_VAR_RETRANS_TIME,
2798 NEIGH_VAR_BASE_REACHABLE_TIME,
2799 NEIGH_VAR_DELAY_PROBE_TIME,
2800 NEIGH_VAR_GC_STALETIME,
2801 NEIGH_VAR_QUEUE_LEN,
2802 NEIGH_VAR_QUEUE_LEN_BYTES,
2803 NEIGH_VAR_PROXY_QLEN,
2804 NEIGH_VAR_ANYCAST_DELAY,
2805 NEIGH_VAR_PROXY_DELAY,
2807 NEIGH_VAR_RETRANS_TIME_MS,
2808 NEIGH_VAR_BASE_REACHABLE_TIME_MS,
2809 NEIGH_VAR_GC_INTERVAL,
2810 NEIGH_VAR_GC_THRESH1,
2811 NEIGH_VAR_GC_THRESH2,
2812 NEIGH_VAR_GC_THRESH3,
2816 static struct neigh_sysctl_table {
2817 struct ctl_table_header *sysctl_header;
2818 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2819 } neigh_sysctl_template __read_mostly = {
2821 [NEIGH_VAR_MCAST_PROBE] = {
2822 .procname = "mcast_solicit",
2823 .maxlen = sizeof(int),
2825 .proc_handler = proc_dointvec,
2827 [NEIGH_VAR_UCAST_PROBE] = {
2828 .procname = "ucast_solicit",
2829 .maxlen = sizeof(int),
2831 .proc_handler = proc_dointvec,
2833 [NEIGH_VAR_APP_PROBE] = {
2834 .procname = "app_solicit",
2835 .maxlen = sizeof(int),
2837 .proc_handler = proc_dointvec,
2839 [NEIGH_VAR_RETRANS_TIME] = {
2840 .procname = "retrans_time",
2841 .maxlen = sizeof(int),
2843 .proc_handler = proc_dointvec_userhz_jiffies,
2845 [NEIGH_VAR_BASE_REACHABLE_TIME] = {
2846 .procname = "base_reachable_time",
2847 .maxlen = sizeof(int),
2849 .proc_handler = proc_dointvec_jiffies,
2851 [NEIGH_VAR_DELAY_PROBE_TIME] = {
2852 .procname = "delay_first_probe_time",
2853 .maxlen = sizeof(int),
2855 .proc_handler = proc_dointvec_jiffies,
2857 [NEIGH_VAR_GC_STALETIME] = {
2858 .procname = "gc_stale_time",
2859 .maxlen = sizeof(int),
2861 .proc_handler = proc_dointvec_jiffies,
2863 [NEIGH_VAR_QUEUE_LEN] = {
2864 .procname = "unres_qlen",
2865 .maxlen = sizeof(int),
2867 .proc_handler = proc_unres_qlen,
2869 [NEIGH_VAR_QUEUE_LEN_BYTES] = {
2870 .procname = "unres_qlen_bytes",
2871 .maxlen = sizeof(int),
2874 .proc_handler = proc_dointvec_minmax,
2876 [NEIGH_VAR_PROXY_QLEN] = {
2877 .procname = "proxy_qlen",
2878 .maxlen = sizeof(int),
2880 .proc_handler = proc_dointvec,
2882 [NEIGH_VAR_ANYCAST_DELAY] = {
2883 .procname = "anycast_delay",
2884 .maxlen = sizeof(int),
2886 .proc_handler = proc_dointvec_userhz_jiffies,
2888 [NEIGH_VAR_PROXY_DELAY] = {
2889 .procname = "proxy_delay",
2890 .maxlen = sizeof(int),
2892 .proc_handler = proc_dointvec_userhz_jiffies,
2894 [NEIGH_VAR_LOCKTIME] = {
2895 .procname = "locktime",
2896 .maxlen = sizeof(int),
2898 .proc_handler = proc_dointvec_userhz_jiffies,
2900 [NEIGH_VAR_RETRANS_TIME_MS] = {
2901 .procname = "retrans_time_ms",
2902 .maxlen = sizeof(int),
2904 .proc_handler = proc_dointvec_ms_jiffies,
2906 [NEIGH_VAR_BASE_REACHABLE_TIME_MS] = {
2907 .procname = "base_reachable_time_ms",
2908 .maxlen = sizeof(int),
2910 .proc_handler = proc_dointvec_ms_jiffies,
2912 [NEIGH_VAR_GC_INTERVAL] = {
2913 .procname = "gc_interval",
2914 .maxlen = sizeof(int),
2916 .proc_handler = proc_dointvec_jiffies,
2918 [NEIGH_VAR_GC_THRESH1] = {
2919 .procname = "gc_thresh1",
2920 .maxlen = sizeof(int),
2922 .proc_handler = proc_dointvec,
2924 [NEIGH_VAR_GC_THRESH2] = {
2925 .procname = "gc_thresh2",
2926 .maxlen = sizeof(int),
2928 .proc_handler = proc_dointvec,
2930 [NEIGH_VAR_GC_THRESH3] = {
2931 .procname = "gc_thresh3",
2932 .maxlen = sizeof(int),
2934 .proc_handler = proc_dointvec,
2940 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2941 char *p_name, proc_handler *handler)
2943 struct neigh_sysctl_table *t;
2944 const char *dev_name_source = NULL;
2945 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
2947 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2951 t->neigh_vars[NEIGH_VAR_MCAST_PROBE].data = &p->mcast_probes;
2952 t->neigh_vars[NEIGH_VAR_UCAST_PROBE].data = &p->ucast_probes;
2953 t->neigh_vars[NEIGH_VAR_APP_PROBE].data = &p->app_probes;
2954 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].data = &p->retrans_time;
2955 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].data = &p->base_reachable_time;
2956 t->neigh_vars[NEIGH_VAR_DELAY_PROBE_TIME].data = &p->delay_probe_time;
2957 t->neigh_vars[NEIGH_VAR_GC_STALETIME].data = &p->gc_staletime;
2958 t->neigh_vars[NEIGH_VAR_QUEUE_LEN].data = &p->queue_len_bytes;
2959 t->neigh_vars[NEIGH_VAR_QUEUE_LEN_BYTES].data = &p->queue_len_bytes;
2960 t->neigh_vars[NEIGH_VAR_PROXY_QLEN].data = &p->proxy_qlen;
2961 t->neigh_vars[NEIGH_VAR_ANYCAST_DELAY].data = &p->anycast_delay;
2962 t->neigh_vars[NEIGH_VAR_PROXY_DELAY].data = &p->proxy_delay;
2963 t->neigh_vars[NEIGH_VAR_LOCKTIME].data = &p->locktime;
2964 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].data = &p->retrans_time;
2965 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].data = &p->base_reachable_time;
2968 dev_name_source = dev->name;
2969 /* Terminate the table early */
2970 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
2971 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
2973 dev_name_source = "default";
2974 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1);
2975 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1;
2976 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2;
2977 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3;
2983 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
2984 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].extra1 = dev;
2986 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
2987 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].extra1 = dev;
2988 /* RetransTime (in milliseconds)*/
2989 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
2990 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].extra1 = dev;
2991 /* ReachableTime (in milliseconds) */
2992 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
2993 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].extra1 = dev;
2996 /* Don't export sysctls to unprivileged users */
2997 if (neigh_parms_net(p)->user_ns != &init_user_ns)
2998 t->neigh_vars[0].procname = NULL;
3000 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3001 p_name, dev_name_source);
3003 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3004 if (!t->sysctl_header)
3007 p->sysctl_table = t;
3015 EXPORT_SYMBOL(neigh_sysctl_register);
3017 void neigh_sysctl_unregister(struct neigh_parms *p)
3019 if (p->sysctl_table) {
3020 struct neigh_sysctl_table *t = p->sysctl_table;
3021 p->sysctl_table = NULL;
3022 unregister_net_sysctl_table(t->sysctl_header);
3026 EXPORT_SYMBOL(neigh_sysctl_unregister);
3028 #endif /* CONFIG_SYSCTL */
3030 static int __init neigh_init(void)
3032 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3033 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3034 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3036 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3038 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3043 subsys_initcall(neigh_init);