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);
60 static const struct file_operations neigh_stat_seq_fops;
64 Neighbour hash table buckets are protected with rwlock tbl->lock.
66 - All the scans/updates to hash buckets MUST be made under this lock.
67 - NOTHING clever should be made under this lock: no callbacks
68 to protocol backends, no attempts to send something to network.
69 It will result in deadlocks, if backend/driver wants to use neighbour
71 - If the entry requires some non-trivial actions, increase
72 its reference count and release table lock.
74 Neighbour entries are protected:
75 - with reference count.
76 - with rwlock neigh->lock
78 Reference count prevents destruction.
80 neigh->lock mainly serializes ll address data and its validity state.
81 However, the same lock is used to protect another entry fields:
85 Again, nothing clever shall be made under neigh->lock,
86 the most complicated procedure, which we allow is dev->hard_header.
87 It is supposed, that dev->hard_header is simplistic and does
88 not make callbacks to neighbour tables.
91 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
97 static void neigh_cleanup_and_release(struct neighbour *neigh)
99 if (neigh->parms->neigh_cleanup)
100 neigh->parms->neigh_cleanup(neigh);
102 __neigh_notify(neigh, RTM_DELNEIGH, 0);
103 neigh_release(neigh);
107 * It is random distribution in the interval (1/2)*base...(3/2)*base.
108 * It corresponds to default IPv6 settings and is not overridable,
109 * because it is really reasonable choice.
112 unsigned long neigh_rand_reach_time(unsigned long base)
114 return base ? (prandom_u32() % base) + (base >> 1) : 0;
116 EXPORT_SYMBOL(neigh_rand_reach_time);
119 static int neigh_forced_gc(struct neigh_table *tbl)
123 struct neigh_hash_table *nht;
125 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
127 write_lock_bh(&tbl->lock);
128 nht = rcu_dereference_protected(tbl->nht,
129 lockdep_is_held(&tbl->lock));
130 for (i = 0; i < (1 << nht->hash_shift); i++) {
132 struct neighbour __rcu **np;
134 np = &nht->hash_buckets[i];
135 while ((n = rcu_dereference_protected(*np,
136 lockdep_is_held(&tbl->lock))) != NULL) {
137 /* Neighbour record may be discarded if:
138 * - nobody refers to it.
139 * - it is not permanent
141 write_lock(&n->lock);
142 if (atomic_read(&n->refcnt) == 1 &&
143 !(n->nud_state & NUD_PERMANENT)) {
144 rcu_assign_pointer(*np,
145 rcu_dereference_protected(n->next,
146 lockdep_is_held(&tbl->lock)));
149 write_unlock(&n->lock);
150 neigh_cleanup_and_release(n);
153 write_unlock(&n->lock);
158 tbl->last_flush = jiffies;
160 write_unlock_bh(&tbl->lock);
165 static void neigh_add_timer(struct neighbour *n, unsigned long when)
168 if (unlikely(mod_timer(&n->timer, when))) {
169 printk("NEIGH: BUG, double timer add, state is %x\n",
175 static int neigh_del_timer(struct neighbour *n)
177 if ((n->nud_state & NUD_IN_TIMER) &&
178 del_timer(&n->timer)) {
185 static void pneigh_queue_purge(struct sk_buff_head *list)
189 while ((skb = skb_dequeue(list)) != NULL) {
195 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
198 struct neigh_hash_table *nht;
200 nht = rcu_dereference_protected(tbl->nht,
201 lockdep_is_held(&tbl->lock));
203 for (i = 0; i < (1 << nht->hash_shift); i++) {
205 struct neighbour __rcu **np = &nht->hash_buckets[i];
207 while ((n = rcu_dereference_protected(*np,
208 lockdep_is_held(&tbl->lock))) != NULL) {
209 if (dev && n->dev != dev) {
213 rcu_assign_pointer(*np,
214 rcu_dereference_protected(n->next,
215 lockdep_is_held(&tbl->lock)));
216 write_lock(&n->lock);
220 if (atomic_read(&n->refcnt) != 1) {
221 /* The most unpleasant situation.
222 We must destroy neighbour entry,
223 but someone still uses it.
225 The destroy will be delayed until
226 the last user releases us, but
227 we must kill timers etc. and move
230 __skb_queue_purge(&n->arp_queue);
231 n->arp_queue_len_bytes = 0;
232 n->output = neigh_blackhole;
233 if (n->nud_state & NUD_VALID)
234 n->nud_state = NUD_NOARP;
236 n->nud_state = NUD_NONE;
237 neigh_dbg(2, "neigh %p is stray\n", n);
239 write_unlock(&n->lock);
240 neigh_cleanup_and_release(n);
245 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
247 write_lock_bh(&tbl->lock);
248 neigh_flush_dev(tbl, dev);
249 write_unlock_bh(&tbl->lock);
251 EXPORT_SYMBOL(neigh_changeaddr);
253 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
255 write_lock_bh(&tbl->lock);
256 neigh_flush_dev(tbl, dev);
257 pneigh_ifdown(tbl, dev);
258 write_unlock_bh(&tbl->lock);
260 del_timer_sync(&tbl->proxy_timer);
261 pneigh_queue_purge(&tbl->proxy_queue);
264 EXPORT_SYMBOL(neigh_ifdown);
266 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
268 struct neighbour *n = NULL;
269 unsigned long now = jiffies;
272 entries = atomic_inc_return(&tbl->entries) - 1;
273 if (entries >= tbl->gc_thresh3 ||
274 (entries >= tbl->gc_thresh2 &&
275 time_after(now, tbl->last_flush + 5 * HZ))) {
276 if (!neigh_forced_gc(tbl) &&
277 entries >= tbl->gc_thresh3)
281 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
285 __skb_queue_head_init(&n->arp_queue);
286 rwlock_init(&n->lock);
287 seqlock_init(&n->ha_lock);
288 n->updated = n->used = now;
289 n->nud_state = NUD_NONE;
290 n->output = neigh_blackhole;
291 seqlock_init(&n->hh.hh_lock);
292 n->parms = neigh_parms_clone(&tbl->parms);
293 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
295 NEIGH_CACHE_STAT_INC(tbl, allocs);
297 atomic_set(&n->refcnt, 1);
303 atomic_dec(&tbl->entries);
307 static void neigh_get_hash_rnd(u32 *x)
309 get_random_bytes(x, sizeof(*x));
313 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
315 size_t size = (1 << shift) * sizeof(struct neighbour *);
316 struct neigh_hash_table *ret;
317 struct neighbour __rcu **buckets;
320 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
323 if (size <= PAGE_SIZE)
324 buckets = kzalloc(size, GFP_ATOMIC);
326 buckets = (struct neighbour __rcu **)
327 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
333 ret->hash_buckets = buckets;
334 ret->hash_shift = shift;
335 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
336 neigh_get_hash_rnd(&ret->hash_rnd[i]);
340 static void neigh_hash_free_rcu(struct rcu_head *head)
342 struct neigh_hash_table *nht = container_of(head,
343 struct neigh_hash_table,
345 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
346 struct neighbour __rcu **buckets = nht->hash_buckets;
348 if (size <= PAGE_SIZE)
351 free_pages((unsigned long)buckets, get_order(size));
355 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
356 unsigned long new_shift)
358 unsigned int i, hash;
359 struct neigh_hash_table *new_nht, *old_nht;
361 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
363 old_nht = rcu_dereference_protected(tbl->nht,
364 lockdep_is_held(&tbl->lock));
365 new_nht = neigh_hash_alloc(new_shift);
369 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
370 struct neighbour *n, *next;
372 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
373 lockdep_is_held(&tbl->lock));
376 hash = tbl->hash(n->primary_key, n->dev,
379 hash >>= (32 - new_nht->hash_shift);
380 next = rcu_dereference_protected(n->next,
381 lockdep_is_held(&tbl->lock));
383 rcu_assign_pointer(n->next,
384 rcu_dereference_protected(
385 new_nht->hash_buckets[hash],
386 lockdep_is_held(&tbl->lock)));
387 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
391 rcu_assign_pointer(tbl->nht, new_nht);
392 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
396 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
397 struct net_device *dev)
401 NEIGH_CACHE_STAT_INC(tbl, lookups);
404 n = __neigh_lookup_noref(tbl, pkey, dev);
406 if (!atomic_inc_not_zero(&n->refcnt))
408 NEIGH_CACHE_STAT_INC(tbl, hits);
411 rcu_read_unlock_bh();
414 EXPORT_SYMBOL(neigh_lookup);
416 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
420 int key_len = tbl->key_len;
422 struct neigh_hash_table *nht;
424 NEIGH_CACHE_STAT_INC(tbl, lookups);
427 nht = rcu_dereference_bh(tbl->nht);
428 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
430 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
432 n = rcu_dereference_bh(n->next)) {
433 if (!memcmp(n->primary_key, pkey, key_len) &&
434 net_eq(dev_net(n->dev), net)) {
435 if (!atomic_inc_not_zero(&n->refcnt))
437 NEIGH_CACHE_STAT_INC(tbl, hits);
442 rcu_read_unlock_bh();
445 EXPORT_SYMBOL(neigh_lookup_nodev);
447 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
448 struct net_device *dev, bool want_ref)
451 int key_len = tbl->key_len;
453 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
454 struct neigh_hash_table *nht;
457 rc = ERR_PTR(-ENOBUFS);
461 memcpy(n->primary_key, pkey, key_len);
465 /* Protocol specific setup. */
466 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
468 goto out_neigh_release;
471 if (dev->netdev_ops->ndo_neigh_construct) {
472 error = dev->netdev_ops->ndo_neigh_construct(n);
475 goto out_neigh_release;
479 /* Device specific setup. */
480 if (n->parms->neigh_setup &&
481 (error = n->parms->neigh_setup(n)) < 0) {
483 goto out_neigh_release;
486 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
488 write_lock_bh(&tbl->lock);
489 nht = rcu_dereference_protected(tbl->nht,
490 lockdep_is_held(&tbl->lock));
492 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
493 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
495 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
497 if (n->parms->dead) {
498 rc = ERR_PTR(-EINVAL);
502 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
503 lockdep_is_held(&tbl->lock));
505 n1 = rcu_dereference_protected(n1->next,
506 lockdep_is_held(&tbl->lock))) {
507 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
518 rcu_assign_pointer(n->next,
519 rcu_dereference_protected(nht->hash_buckets[hash_val],
520 lockdep_is_held(&tbl->lock)));
521 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
522 write_unlock_bh(&tbl->lock);
523 neigh_dbg(2, "neigh %p is created\n", n);
528 write_unlock_bh(&tbl->lock);
533 EXPORT_SYMBOL(__neigh_create);
535 static u32 pneigh_hash(const void *pkey, int key_len)
537 u32 hash_val = *(u32 *)(pkey + key_len - 4);
538 hash_val ^= (hash_val >> 16);
539 hash_val ^= hash_val >> 8;
540 hash_val ^= hash_val >> 4;
541 hash_val &= PNEIGH_HASHMASK;
545 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
549 struct net_device *dev)
552 if (!memcmp(n->key, pkey, key_len) &&
553 net_eq(pneigh_net(n), net) &&
554 (n->dev == dev || !n->dev))
561 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
562 struct net *net, const void *pkey, struct net_device *dev)
564 int key_len = tbl->key_len;
565 u32 hash_val = pneigh_hash(pkey, key_len);
567 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
568 net, pkey, key_len, dev);
570 EXPORT_SYMBOL_GPL(__pneigh_lookup);
572 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
573 struct net *net, const void *pkey,
574 struct net_device *dev, int creat)
576 struct pneigh_entry *n;
577 int key_len = tbl->key_len;
578 u32 hash_val = pneigh_hash(pkey, key_len);
580 read_lock_bh(&tbl->lock);
581 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
582 net, pkey, key_len, dev);
583 read_unlock_bh(&tbl->lock);
590 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
594 write_pnet(&n->net, net);
595 memcpy(n->key, pkey, key_len);
600 if (tbl->pconstructor && tbl->pconstructor(n)) {
608 write_lock_bh(&tbl->lock);
609 n->next = tbl->phash_buckets[hash_val];
610 tbl->phash_buckets[hash_val] = n;
611 write_unlock_bh(&tbl->lock);
615 EXPORT_SYMBOL(pneigh_lookup);
618 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
619 struct net_device *dev)
621 struct pneigh_entry *n, **np;
622 int key_len = tbl->key_len;
623 u32 hash_val = pneigh_hash(pkey, key_len);
625 write_lock_bh(&tbl->lock);
626 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
628 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
629 net_eq(pneigh_net(n), net)) {
631 write_unlock_bh(&tbl->lock);
632 if (tbl->pdestructor)
640 write_unlock_bh(&tbl->lock);
644 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
646 struct pneigh_entry *n, **np;
649 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
650 np = &tbl->phash_buckets[h];
651 while ((n = *np) != NULL) {
652 if (!dev || n->dev == dev) {
654 if (tbl->pdestructor)
667 static void neigh_parms_destroy(struct neigh_parms *parms);
669 static inline void neigh_parms_put(struct neigh_parms *parms)
671 if (atomic_dec_and_test(&parms->refcnt))
672 neigh_parms_destroy(parms);
676 * neighbour must already be out of the table;
679 void neigh_destroy(struct neighbour *neigh)
681 struct net_device *dev = neigh->dev;
683 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
686 pr_warn("Destroying alive neighbour %p\n", neigh);
691 if (neigh_del_timer(neigh))
692 pr_warn("Impossible event\n");
694 write_lock_bh(&neigh->lock);
695 __skb_queue_purge(&neigh->arp_queue);
696 write_unlock_bh(&neigh->lock);
697 neigh->arp_queue_len_bytes = 0;
699 if (dev->netdev_ops->ndo_neigh_destroy)
700 dev->netdev_ops->ndo_neigh_destroy(neigh);
703 neigh_parms_put(neigh->parms);
705 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
707 atomic_dec(&neigh->tbl->entries);
708 kfree_rcu(neigh, rcu);
710 EXPORT_SYMBOL(neigh_destroy);
712 /* Neighbour state is suspicious;
715 Called with write_locked neigh.
717 static void neigh_suspect(struct neighbour *neigh)
719 neigh_dbg(2, "neigh %p is suspected\n", neigh);
721 neigh->output = neigh->ops->output;
724 /* Neighbour state is OK;
727 Called with write_locked neigh.
729 static void neigh_connect(struct neighbour *neigh)
731 neigh_dbg(2, "neigh %p is connected\n", neigh);
733 neigh->output = neigh->ops->connected_output;
736 static void neigh_periodic_work(struct work_struct *work)
738 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
740 struct neighbour __rcu **np;
742 struct neigh_hash_table *nht;
744 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
746 write_lock_bh(&tbl->lock);
747 nht = rcu_dereference_protected(tbl->nht,
748 lockdep_is_held(&tbl->lock));
751 * periodically recompute ReachableTime from random function
754 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
755 struct neigh_parms *p;
756 tbl->last_rand = jiffies;
757 list_for_each_entry(p, &tbl->parms_list, list)
759 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
762 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
765 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
766 np = &nht->hash_buckets[i];
768 while ((n = rcu_dereference_protected(*np,
769 lockdep_is_held(&tbl->lock))) != NULL) {
772 write_lock(&n->lock);
774 state = n->nud_state;
775 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
776 write_unlock(&n->lock);
780 if (time_before(n->used, n->confirmed))
781 n->used = n->confirmed;
783 if (atomic_read(&n->refcnt) == 1 &&
784 (state == NUD_FAILED ||
785 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
788 write_unlock(&n->lock);
789 neigh_cleanup_and_release(n);
792 write_unlock(&n->lock);
798 * It's fine to release lock here, even if hash table
799 * grows while we are preempted.
801 write_unlock_bh(&tbl->lock);
803 write_lock_bh(&tbl->lock);
804 nht = rcu_dereference_protected(tbl->nht,
805 lockdep_is_held(&tbl->lock));
808 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
809 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
810 * BASE_REACHABLE_TIME.
812 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
813 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
814 write_unlock_bh(&tbl->lock);
817 static __inline__ int neigh_max_probes(struct neighbour *n)
819 struct neigh_parms *p = n->parms;
820 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
821 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
822 NEIGH_VAR(p, MCAST_PROBES));
825 static void neigh_invalidate(struct neighbour *neigh)
826 __releases(neigh->lock)
827 __acquires(neigh->lock)
831 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
832 neigh_dbg(2, "neigh %p is failed\n", neigh);
833 neigh->updated = jiffies;
835 /* It is very thin place. report_unreachable is very complicated
836 routine. Particularly, it can hit the same neighbour entry!
838 So that, we try to be accurate and avoid dead loop. --ANK
840 while (neigh->nud_state == NUD_FAILED &&
841 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
842 write_unlock(&neigh->lock);
843 neigh->ops->error_report(neigh, skb);
844 write_lock(&neigh->lock);
846 __skb_queue_purge(&neigh->arp_queue);
847 neigh->arp_queue_len_bytes = 0;
850 static void neigh_probe(struct neighbour *neigh)
851 __releases(neigh->lock)
853 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
854 /* keep skb alive even if arp_queue overflows */
856 skb = skb_copy(skb, GFP_ATOMIC);
857 write_unlock(&neigh->lock);
858 neigh->ops->solicit(neigh, skb);
859 atomic_inc(&neigh->probes);
863 /* Called when a timer expires for a neighbour entry. */
865 static void neigh_timer_handler(unsigned long arg)
867 unsigned long now, next;
868 struct neighbour *neigh = (struct neighbour *)arg;
872 write_lock(&neigh->lock);
874 state = neigh->nud_state;
878 if (!(state & NUD_IN_TIMER))
881 if (state & NUD_REACHABLE) {
882 if (time_before_eq(now,
883 neigh->confirmed + neigh->parms->reachable_time)) {
884 neigh_dbg(2, "neigh %p is still alive\n", neigh);
885 next = neigh->confirmed + neigh->parms->reachable_time;
886 } else if (time_before_eq(now,
888 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
889 neigh_dbg(2, "neigh %p is delayed\n", neigh);
890 neigh->nud_state = NUD_DELAY;
891 neigh->updated = jiffies;
892 neigh_suspect(neigh);
893 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
895 neigh_dbg(2, "neigh %p is suspected\n", neigh);
896 neigh->nud_state = NUD_STALE;
897 neigh->updated = jiffies;
898 neigh_suspect(neigh);
901 } else if (state & NUD_DELAY) {
902 if (time_before_eq(now,
904 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
905 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
906 neigh->nud_state = NUD_REACHABLE;
907 neigh->updated = jiffies;
908 neigh_connect(neigh);
910 next = neigh->confirmed + neigh->parms->reachable_time;
912 neigh_dbg(2, "neigh %p is probed\n", neigh);
913 neigh->nud_state = NUD_PROBE;
914 neigh->updated = jiffies;
915 atomic_set(&neigh->probes, 0);
917 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
920 /* NUD_PROBE|NUD_INCOMPLETE */
921 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
924 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
925 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
926 neigh->nud_state = NUD_FAILED;
928 neigh_invalidate(neigh);
932 if (neigh->nud_state & NUD_IN_TIMER) {
933 if (time_before(next, jiffies + HZ/2))
934 next = jiffies + HZ/2;
935 if (!mod_timer(&neigh->timer, next))
938 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
942 write_unlock(&neigh->lock);
946 neigh_update_notify(neigh);
948 neigh_release(neigh);
951 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
954 bool immediate_probe = false;
956 write_lock_bh(&neigh->lock);
959 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
962 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
963 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
964 NEIGH_VAR(neigh->parms, APP_PROBES)) {
965 unsigned long next, now = jiffies;
967 atomic_set(&neigh->probes,
968 NEIGH_VAR(neigh->parms, UCAST_PROBES));
969 neigh->nud_state = NUD_INCOMPLETE;
970 neigh->updated = now;
971 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
973 neigh_add_timer(neigh, next);
974 immediate_probe = true;
976 neigh->nud_state = NUD_FAILED;
977 neigh->updated = jiffies;
978 write_unlock_bh(&neigh->lock);
983 } else if (neigh->nud_state & NUD_STALE) {
984 neigh_dbg(2, "neigh %p is delayed\n", neigh);
985 neigh->nud_state = NUD_DELAY;
986 neigh->updated = jiffies;
987 neigh_add_timer(neigh, jiffies +
988 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
991 if (neigh->nud_state == NUD_INCOMPLETE) {
993 while (neigh->arp_queue_len_bytes + skb->truesize >
994 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
995 struct sk_buff *buff;
997 buff = __skb_dequeue(&neigh->arp_queue);
1000 neigh->arp_queue_len_bytes -= buff->truesize;
1002 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1005 __skb_queue_tail(&neigh->arp_queue, skb);
1006 neigh->arp_queue_len_bytes += skb->truesize;
1011 if (immediate_probe)
1014 write_unlock(&neigh->lock);
1018 EXPORT_SYMBOL(__neigh_event_send);
1020 static void neigh_update_hhs(struct neighbour *neigh)
1022 struct hh_cache *hh;
1023 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1026 if (neigh->dev->header_ops)
1027 update = neigh->dev->header_ops->cache_update;
1032 write_seqlock_bh(&hh->hh_lock);
1033 update(hh, neigh->dev, neigh->ha);
1034 write_sequnlock_bh(&hh->hh_lock);
1041 /* Generic update routine.
1042 -- lladdr is new lladdr or NULL, if it is not supplied.
1043 -- new is new state.
1045 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1047 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1048 lladdr instead of overriding it
1050 It also allows to retain current state
1051 if lladdr is unchanged.
1052 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1054 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1056 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1059 Caller MUST hold reference count on the entry.
1062 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1068 struct net_device *dev;
1069 int update_isrouter = 0;
1071 write_lock_bh(&neigh->lock);
1074 old = neigh->nud_state;
1077 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1078 (old & (NUD_NOARP | NUD_PERMANENT)))
1081 if (!(new & NUD_VALID)) {
1082 neigh_del_timer(neigh);
1083 if (old & NUD_CONNECTED)
1084 neigh_suspect(neigh);
1085 neigh->nud_state = new;
1087 notify = old & NUD_VALID;
1088 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1089 (new & NUD_FAILED)) {
1090 neigh_invalidate(neigh);
1096 /* Compare new lladdr with cached one */
1097 if (!dev->addr_len) {
1098 /* First case: device needs no address. */
1100 } else if (lladdr) {
1101 /* The second case: if something is already cached
1102 and a new address is proposed:
1104 - if they are different, check override flag
1106 if ((old & NUD_VALID) &&
1107 !memcmp(lladdr, neigh->ha, dev->addr_len))
1110 /* No address is supplied; if we know something,
1111 use it, otherwise discard the request.
1114 if (!(old & NUD_VALID))
1119 if (new & NUD_CONNECTED)
1120 neigh->confirmed = jiffies;
1121 neigh->updated = jiffies;
1123 /* If entry was valid and address is not changed,
1124 do not change entry state, if new one is STALE.
1127 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1128 if (old & NUD_VALID) {
1129 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1130 update_isrouter = 0;
1131 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1132 (old & NUD_CONNECTED)) {
1138 if (lladdr == neigh->ha && new == NUD_STALE &&
1139 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1140 (old & NUD_CONNECTED))
1147 neigh_del_timer(neigh);
1148 if (new & NUD_PROBE)
1149 atomic_set(&neigh->probes, 0);
1150 if (new & NUD_IN_TIMER)
1151 neigh_add_timer(neigh, (jiffies +
1152 ((new & NUD_REACHABLE) ?
1153 neigh->parms->reachable_time :
1155 neigh->nud_state = new;
1159 if (lladdr != neigh->ha) {
1160 write_seqlock(&neigh->ha_lock);
1161 memcpy(&neigh->ha, lladdr, dev->addr_len);
1162 write_sequnlock(&neigh->ha_lock);
1163 neigh_update_hhs(neigh);
1164 if (!(new & NUD_CONNECTED))
1165 neigh->confirmed = jiffies -
1166 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1171 if (new & NUD_CONNECTED)
1172 neigh_connect(neigh);
1174 neigh_suspect(neigh);
1175 if (!(old & NUD_VALID)) {
1176 struct sk_buff *skb;
1178 /* Again: avoid dead loop if something went wrong */
1180 while (neigh->nud_state & NUD_VALID &&
1181 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1182 struct dst_entry *dst = skb_dst(skb);
1183 struct neighbour *n2, *n1 = neigh;
1184 write_unlock_bh(&neigh->lock);
1188 /* Why not just use 'neigh' as-is? The problem is that
1189 * things such as shaper, eql, and sch_teql can end up
1190 * using alternative, different, neigh objects to output
1191 * the packet in the output path. So what we need to do
1192 * here is re-lookup the top-level neigh in the path so
1193 * we can reinject the packet there.
1197 n2 = dst_neigh_lookup_skb(dst, skb);
1201 n1->output(n1, skb);
1206 write_lock_bh(&neigh->lock);
1208 __skb_queue_purge(&neigh->arp_queue);
1209 neigh->arp_queue_len_bytes = 0;
1212 if (update_isrouter) {
1213 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1214 (neigh->flags | NTF_ROUTER) :
1215 (neigh->flags & ~NTF_ROUTER);
1217 write_unlock_bh(&neigh->lock);
1220 neigh_update_notify(neigh);
1224 EXPORT_SYMBOL(neigh_update);
1226 /* Update the neigh to listen temporarily for probe responses, even if it is
1227 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1229 void __neigh_set_probe_once(struct neighbour *neigh)
1231 neigh->updated = jiffies;
1232 if (!(neigh->nud_state & NUD_FAILED))
1234 neigh->nud_state = NUD_INCOMPLETE;
1235 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1236 neigh_add_timer(neigh,
1237 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1239 EXPORT_SYMBOL(__neigh_set_probe_once);
1241 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1242 u8 *lladdr, void *saddr,
1243 struct net_device *dev)
1245 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1246 lladdr || !dev->addr_len);
1248 neigh_update(neigh, lladdr, NUD_STALE,
1249 NEIGH_UPDATE_F_OVERRIDE);
1252 EXPORT_SYMBOL(neigh_event_ns);
1254 /* called with read_lock_bh(&n->lock); */
1255 static void neigh_hh_init(struct neighbour *n)
1257 struct net_device *dev = n->dev;
1258 __be16 prot = n->tbl->protocol;
1259 struct hh_cache *hh = &n->hh;
1261 write_lock_bh(&n->lock);
1263 /* Only one thread can come in here and initialize the
1267 dev->header_ops->cache(n, hh, prot);
1269 write_unlock_bh(&n->lock);
1272 /* Slow and careful. */
1274 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1278 if (!neigh_event_send(neigh, skb)) {
1280 struct net_device *dev = neigh->dev;
1283 if (dev->header_ops->cache && !neigh->hh.hh_len)
1284 neigh_hh_init(neigh);
1287 __skb_pull(skb, skb_network_offset(skb));
1288 seq = read_seqbegin(&neigh->ha_lock);
1289 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1290 neigh->ha, NULL, skb->len);
1291 } while (read_seqretry(&neigh->ha_lock, seq));
1294 rc = dev_queue_xmit(skb);
1305 EXPORT_SYMBOL(neigh_resolve_output);
1307 /* As fast as possible without hh cache */
1309 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1311 struct net_device *dev = neigh->dev;
1316 __skb_pull(skb, skb_network_offset(skb));
1317 seq = read_seqbegin(&neigh->ha_lock);
1318 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1319 neigh->ha, NULL, skb->len);
1320 } while (read_seqretry(&neigh->ha_lock, seq));
1323 err = dev_queue_xmit(skb);
1330 EXPORT_SYMBOL(neigh_connected_output);
1332 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1334 return dev_queue_xmit(skb);
1336 EXPORT_SYMBOL(neigh_direct_output);
1338 static void neigh_proxy_process(unsigned long arg)
1340 struct neigh_table *tbl = (struct neigh_table *)arg;
1341 long sched_next = 0;
1342 unsigned long now = jiffies;
1343 struct sk_buff *skb, *n;
1345 spin_lock(&tbl->proxy_queue.lock);
1347 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1348 long tdif = NEIGH_CB(skb)->sched_next - now;
1351 struct net_device *dev = skb->dev;
1353 __skb_unlink(skb, &tbl->proxy_queue);
1354 if (tbl->proxy_redo && netif_running(dev)) {
1356 tbl->proxy_redo(skb);
1363 } else if (!sched_next || tdif < sched_next)
1366 del_timer(&tbl->proxy_timer);
1368 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1369 spin_unlock(&tbl->proxy_queue.lock);
1372 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1373 struct sk_buff *skb)
1375 unsigned long now = jiffies;
1377 unsigned long sched_next = now + (prandom_u32() %
1378 NEIGH_VAR(p, PROXY_DELAY));
1380 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1385 NEIGH_CB(skb)->sched_next = sched_next;
1386 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1388 spin_lock(&tbl->proxy_queue.lock);
1389 if (del_timer(&tbl->proxy_timer)) {
1390 if (time_before(tbl->proxy_timer.expires, sched_next))
1391 sched_next = tbl->proxy_timer.expires;
1395 __skb_queue_tail(&tbl->proxy_queue, skb);
1396 mod_timer(&tbl->proxy_timer, sched_next);
1397 spin_unlock(&tbl->proxy_queue.lock);
1399 EXPORT_SYMBOL(pneigh_enqueue);
1401 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1402 struct net *net, int ifindex)
1404 struct neigh_parms *p;
1406 list_for_each_entry(p, &tbl->parms_list, list) {
1407 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1408 (!p->dev && !ifindex && net_eq(net, &init_net)))
1415 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1416 struct neigh_table *tbl)
1418 struct neigh_parms *p;
1419 struct net *net = dev_net(dev);
1420 const struct net_device_ops *ops = dev->netdev_ops;
1422 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1425 atomic_set(&p->refcnt, 1);
1427 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1430 write_pnet(&p->net, net);
1431 p->sysctl_table = NULL;
1433 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1439 write_lock_bh(&tbl->lock);
1440 list_add(&p->list, &tbl->parms.list);
1441 write_unlock_bh(&tbl->lock);
1443 neigh_parms_data_state_cleanall(p);
1447 EXPORT_SYMBOL(neigh_parms_alloc);
1449 static void neigh_rcu_free_parms(struct rcu_head *head)
1451 struct neigh_parms *parms =
1452 container_of(head, struct neigh_parms, rcu_head);
1454 neigh_parms_put(parms);
1457 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1459 if (!parms || parms == &tbl->parms)
1461 write_lock_bh(&tbl->lock);
1462 list_del(&parms->list);
1464 write_unlock_bh(&tbl->lock);
1466 dev_put(parms->dev);
1467 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1469 EXPORT_SYMBOL(neigh_parms_release);
1471 static void neigh_parms_destroy(struct neigh_parms *parms)
1476 static struct lock_class_key neigh_table_proxy_queue_class;
1478 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1480 void neigh_table_init(int index, struct neigh_table *tbl)
1482 unsigned long now = jiffies;
1483 unsigned long phsize;
1485 INIT_LIST_HEAD(&tbl->parms_list);
1486 list_add(&tbl->parms.list, &tbl->parms_list);
1487 write_pnet(&tbl->parms.net, &init_net);
1488 atomic_set(&tbl->parms.refcnt, 1);
1489 tbl->parms.reachable_time =
1490 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1492 tbl->stats = alloc_percpu(struct neigh_statistics);
1494 panic("cannot create neighbour cache statistics");
1496 #ifdef CONFIG_PROC_FS
1497 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1498 &neigh_stat_seq_fops, tbl))
1499 panic("cannot create neighbour proc dir entry");
1502 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1504 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1505 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1507 if (!tbl->nht || !tbl->phash_buckets)
1508 panic("cannot allocate neighbour cache hashes");
1510 if (!tbl->entry_size)
1511 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1512 tbl->key_len, NEIGH_PRIV_ALIGN);
1514 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1516 rwlock_init(&tbl->lock);
1517 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1518 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1519 tbl->parms.reachable_time);
1520 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1521 skb_queue_head_init_class(&tbl->proxy_queue,
1522 &neigh_table_proxy_queue_class);
1524 tbl->last_flush = now;
1525 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1527 neigh_tables[index] = tbl;
1529 EXPORT_SYMBOL(neigh_table_init);
1531 int neigh_table_clear(int index, struct neigh_table *tbl)
1533 neigh_tables[index] = NULL;
1534 /* It is not clean... Fix it to unload IPv6 module safely */
1535 cancel_delayed_work_sync(&tbl->gc_work);
1536 del_timer_sync(&tbl->proxy_timer);
1537 pneigh_queue_purge(&tbl->proxy_queue);
1538 neigh_ifdown(tbl, NULL);
1539 if (atomic_read(&tbl->entries))
1540 pr_crit("neighbour leakage\n");
1542 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1543 neigh_hash_free_rcu);
1546 kfree(tbl->phash_buckets);
1547 tbl->phash_buckets = NULL;
1549 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1551 free_percpu(tbl->stats);
1556 EXPORT_SYMBOL(neigh_table_clear);
1558 static struct neigh_table *neigh_find_table(int family)
1560 struct neigh_table *tbl = NULL;
1564 tbl = neigh_tables[NEIGH_ARP_TABLE];
1567 tbl = neigh_tables[NEIGH_ND_TABLE];
1570 tbl = neigh_tables[NEIGH_DN_TABLE];
1577 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh)
1579 struct net *net = sock_net(skb->sk);
1581 struct nlattr *dst_attr;
1582 struct neigh_table *tbl;
1583 struct neighbour *neigh;
1584 struct net_device *dev = NULL;
1588 if (nlmsg_len(nlh) < sizeof(*ndm))
1591 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1592 if (dst_attr == NULL)
1595 ndm = nlmsg_data(nlh);
1596 if (ndm->ndm_ifindex) {
1597 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1604 tbl = neigh_find_table(ndm->ndm_family);
1606 return -EAFNOSUPPORT;
1608 if (nla_len(dst_attr) < tbl->key_len)
1611 if (ndm->ndm_flags & NTF_PROXY) {
1612 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1619 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1620 if (neigh == NULL) {
1625 err = neigh_update(neigh, NULL, NUD_FAILED,
1626 NEIGH_UPDATE_F_OVERRIDE |
1627 NEIGH_UPDATE_F_ADMIN);
1628 neigh_release(neigh);
1634 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh)
1636 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1637 struct net *net = sock_net(skb->sk);
1639 struct nlattr *tb[NDA_MAX+1];
1640 struct neigh_table *tbl;
1641 struct net_device *dev = NULL;
1642 struct neighbour *neigh;
1647 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1652 if (tb[NDA_DST] == NULL)
1655 ndm = nlmsg_data(nlh);
1656 if (ndm->ndm_ifindex) {
1657 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1663 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1667 tbl = neigh_find_table(ndm->ndm_family);
1669 return -EAFNOSUPPORT;
1671 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1673 dst = nla_data(tb[NDA_DST]);
1674 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1676 if (ndm->ndm_flags & NTF_PROXY) {
1677 struct pneigh_entry *pn;
1680 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1682 pn->flags = ndm->ndm_flags;
1691 neigh = neigh_lookup(tbl, dst, dev);
1692 if (neigh == NULL) {
1693 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1698 neigh = __neigh_lookup_errno(tbl, dst, dev);
1699 if (IS_ERR(neigh)) {
1700 err = PTR_ERR(neigh);
1704 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1706 neigh_release(neigh);
1710 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1711 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1714 if (ndm->ndm_flags & NTF_USE) {
1715 neigh_event_send(neigh, NULL);
1718 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1719 neigh_release(neigh);
1725 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1727 struct nlattr *nest;
1729 nest = nla_nest_start(skb, NDTA_PARMS);
1734 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1735 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1736 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1737 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1738 /* approximative value for deprecated QUEUE_LEN (in packets) */
1739 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1740 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1741 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1742 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
1743 nla_put_u32(skb, NDTPA_UCAST_PROBES,
1744 NEIGH_VAR(parms, UCAST_PROBES)) ||
1745 nla_put_u32(skb, NDTPA_MCAST_PROBES,
1746 NEIGH_VAR(parms, MCAST_PROBES)) ||
1747 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
1748 NEIGH_VAR(parms, MCAST_REPROBES)) ||
1749 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1750 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1751 NEIGH_VAR(parms, BASE_REACHABLE_TIME)) ||
1752 nla_put_msecs(skb, NDTPA_GC_STALETIME,
1753 NEIGH_VAR(parms, GC_STALETIME)) ||
1754 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1755 NEIGH_VAR(parms, DELAY_PROBE_TIME)) ||
1756 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
1757 NEIGH_VAR(parms, RETRANS_TIME)) ||
1758 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
1759 NEIGH_VAR(parms, ANYCAST_DELAY)) ||
1760 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
1761 NEIGH_VAR(parms, PROXY_DELAY)) ||
1762 nla_put_msecs(skb, NDTPA_LOCKTIME,
1763 NEIGH_VAR(parms, LOCKTIME)))
1764 goto nla_put_failure;
1765 return nla_nest_end(skb, nest);
1768 nla_nest_cancel(skb, nest);
1772 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1773 u32 pid, u32 seq, int type, int flags)
1775 struct nlmsghdr *nlh;
1776 struct ndtmsg *ndtmsg;
1778 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1782 ndtmsg = nlmsg_data(nlh);
1784 read_lock_bh(&tbl->lock);
1785 ndtmsg->ndtm_family = tbl->family;
1786 ndtmsg->ndtm_pad1 = 0;
1787 ndtmsg->ndtm_pad2 = 0;
1789 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1790 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1791 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1792 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1793 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1794 goto nla_put_failure;
1796 unsigned long now = jiffies;
1797 unsigned int flush_delta = now - tbl->last_flush;
1798 unsigned int rand_delta = now - tbl->last_rand;
1799 struct neigh_hash_table *nht;
1800 struct ndt_config ndc = {
1801 .ndtc_key_len = tbl->key_len,
1802 .ndtc_entry_size = tbl->entry_size,
1803 .ndtc_entries = atomic_read(&tbl->entries),
1804 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1805 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1806 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1810 nht = rcu_dereference_bh(tbl->nht);
1811 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1812 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1813 rcu_read_unlock_bh();
1815 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1816 goto nla_put_failure;
1821 struct ndt_stats ndst;
1823 memset(&ndst, 0, sizeof(ndst));
1825 for_each_possible_cpu(cpu) {
1826 struct neigh_statistics *st;
1828 st = per_cpu_ptr(tbl->stats, cpu);
1829 ndst.ndts_allocs += st->allocs;
1830 ndst.ndts_destroys += st->destroys;
1831 ndst.ndts_hash_grows += st->hash_grows;
1832 ndst.ndts_res_failed += st->res_failed;
1833 ndst.ndts_lookups += st->lookups;
1834 ndst.ndts_hits += st->hits;
1835 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1836 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1837 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1838 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1841 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1842 goto nla_put_failure;
1845 BUG_ON(tbl->parms.dev);
1846 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1847 goto nla_put_failure;
1849 read_unlock_bh(&tbl->lock);
1850 nlmsg_end(skb, nlh);
1854 read_unlock_bh(&tbl->lock);
1855 nlmsg_cancel(skb, nlh);
1859 static int neightbl_fill_param_info(struct sk_buff *skb,
1860 struct neigh_table *tbl,
1861 struct neigh_parms *parms,
1862 u32 pid, u32 seq, int type,
1865 struct ndtmsg *ndtmsg;
1866 struct nlmsghdr *nlh;
1868 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1872 ndtmsg = nlmsg_data(nlh);
1874 read_lock_bh(&tbl->lock);
1875 ndtmsg->ndtm_family = tbl->family;
1876 ndtmsg->ndtm_pad1 = 0;
1877 ndtmsg->ndtm_pad2 = 0;
1879 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1880 neightbl_fill_parms(skb, parms) < 0)
1883 read_unlock_bh(&tbl->lock);
1884 nlmsg_end(skb, nlh);
1887 read_unlock_bh(&tbl->lock);
1888 nlmsg_cancel(skb, nlh);
1892 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1893 [NDTA_NAME] = { .type = NLA_STRING },
1894 [NDTA_THRESH1] = { .type = NLA_U32 },
1895 [NDTA_THRESH2] = { .type = NLA_U32 },
1896 [NDTA_THRESH3] = { .type = NLA_U32 },
1897 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1898 [NDTA_PARMS] = { .type = NLA_NESTED },
1901 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1902 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1903 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1904 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1905 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1906 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1907 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1908 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
1909 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1910 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1911 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1912 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1913 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1914 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1915 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1918 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh)
1920 struct net *net = sock_net(skb->sk);
1921 struct neigh_table *tbl;
1922 struct ndtmsg *ndtmsg;
1923 struct nlattr *tb[NDTA_MAX+1];
1927 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1928 nl_neightbl_policy);
1932 if (tb[NDTA_NAME] == NULL) {
1937 ndtmsg = nlmsg_data(nlh);
1939 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
1940 tbl = neigh_tables[tidx];
1943 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1945 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
1955 * We acquire tbl->lock to be nice to the periodic timers and
1956 * make sure they always see a consistent set of values.
1958 write_lock_bh(&tbl->lock);
1960 if (tb[NDTA_PARMS]) {
1961 struct nlattr *tbp[NDTPA_MAX+1];
1962 struct neigh_parms *p;
1965 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
1966 nl_ntbl_parm_policy);
1968 goto errout_tbl_lock;
1970 if (tbp[NDTPA_IFINDEX])
1971 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
1973 p = lookup_neigh_parms(tbl, net, ifindex);
1976 goto errout_tbl_lock;
1979 for (i = 1; i <= NDTPA_MAX; i++) {
1984 case NDTPA_QUEUE_LEN:
1985 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
1986 nla_get_u32(tbp[i]) *
1987 SKB_TRUESIZE(ETH_FRAME_LEN));
1989 case NDTPA_QUEUE_LENBYTES:
1990 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
1991 nla_get_u32(tbp[i]));
1993 case NDTPA_PROXY_QLEN:
1994 NEIGH_VAR_SET(p, PROXY_QLEN,
1995 nla_get_u32(tbp[i]));
1997 case NDTPA_APP_PROBES:
1998 NEIGH_VAR_SET(p, APP_PROBES,
1999 nla_get_u32(tbp[i]));
2001 case NDTPA_UCAST_PROBES:
2002 NEIGH_VAR_SET(p, UCAST_PROBES,
2003 nla_get_u32(tbp[i]));
2005 case NDTPA_MCAST_PROBES:
2006 NEIGH_VAR_SET(p, MCAST_PROBES,
2007 nla_get_u32(tbp[i]));
2009 case NDTPA_MCAST_REPROBES:
2010 NEIGH_VAR_SET(p, MCAST_REPROBES,
2011 nla_get_u32(tbp[i]));
2013 case NDTPA_BASE_REACHABLE_TIME:
2014 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2015 nla_get_msecs(tbp[i]));
2016 /* update reachable_time as well, otherwise, the change will
2017 * only be effective after the next time neigh_periodic_work
2018 * decides to recompute it (can be multiple minutes)
2021 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2023 case NDTPA_GC_STALETIME:
2024 NEIGH_VAR_SET(p, GC_STALETIME,
2025 nla_get_msecs(tbp[i]));
2027 case NDTPA_DELAY_PROBE_TIME:
2028 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2029 nla_get_msecs(tbp[i]));
2031 case NDTPA_RETRANS_TIME:
2032 NEIGH_VAR_SET(p, RETRANS_TIME,
2033 nla_get_msecs(tbp[i]));
2035 case NDTPA_ANYCAST_DELAY:
2036 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2037 nla_get_msecs(tbp[i]));
2039 case NDTPA_PROXY_DELAY:
2040 NEIGH_VAR_SET(p, PROXY_DELAY,
2041 nla_get_msecs(tbp[i]));
2043 case NDTPA_LOCKTIME:
2044 NEIGH_VAR_SET(p, LOCKTIME,
2045 nla_get_msecs(tbp[i]));
2052 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2053 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2054 !net_eq(net, &init_net))
2055 goto errout_tbl_lock;
2057 if (tb[NDTA_THRESH1])
2058 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2060 if (tb[NDTA_THRESH2])
2061 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2063 if (tb[NDTA_THRESH3])
2064 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2066 if (tb[NDTA_GC_INTERVAL])
2067 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2072 write_unlock_bh(&tbl->lock);
2077 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2079 struct net *net = sock_net(skb->sk);
2080 int family, tidx, nidx = 0;
2081 int tbl_skip = cb->args[0];
2082 int neigh_skip = cb->args[1];
2083 struct neigh_table *tbl;
2085 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2087 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2088 struct neigh_parms *p;
2090 tbl = neigh_tables[tidx];
2094 if (tidx < tbl_skip || (family && tbl->family != family))
2097 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2098 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2103 p = list_next_entry(&tbl->parms, list);
2104 list_for_each_entry_from(p, &tbl->parms_list, list) {
2105 if (!net_eq(neigh_parms_net(p), net))
2108 if (nidx < neigh_skip)
2111 if (neightbl_fill_param_info(skb, tbl, p,
2112 NETLINK_CB(cb->skb).portid,
2130 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2131 u32 pid, u32 seq, int type, unsigned int flags)
2133 unsigned long now = jiffies;
2134 struct nda_cacheinfo ci;
2135 struct nlmsghdr *nlh;
2138 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2142 ndm = nlmsg_data(nlh);
2143 ndm->ndm_family = neigh->ops->family;
2146 ndm->ndm_flags = neigh->flags;
2147 ndm->ndm_type = neigh->type;
2148 ndm->ndm_ifindex = neigh->dev->ifindex;
2150 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2151 goto nla_put_failure;
2153 read_lock_bh(&neigh->lock);
2154 ndm->ndm_state = neigh->nud_state;
2155 if (neigh->nud_state & NUD_VALID) {
2156 char haddr[MAX_ADDR_LEN];
2158 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2159 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2160 read_unlock_bh(&neigh->lock);
2161 goto nla_put_failure;
2165 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2166 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2167 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2168 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2169 read_unlock_bh(&neigh->lock);
2171 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2172 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2173 goto nla_put_failure;
2175 nlmsg_end(skb, nlh);
2179 nlmsg_cancel(skb, nlh);
2183 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2184 u32 pid, u32 seq, int type, unsigned int flags,
2185 struct neigh_table *tbl)
2187 struct nlmsghdr *nlh;
2190 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2194 ndm = nlmsg_data(nlh);
2195 ndm->ndm_family = tbl->family;
2198 ndm->ndm_flags = pn->flags | NTF_PROXY;
2199 ndm->ndm_type = RTN_UNICAST;
2200 ndm->ndm_ifindex = pn->dev->ifindex;
2201 ndm->ndm_state = NUD_NONE;
2203 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2204 goto nla_put_failure;
2206 nlmsg_end(skb, nlh);
2210 nlmsg_cancel(skb, nlh);
2214 static void neigh_update_notify(struct neighbour *neigh)
2216 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2217 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2220 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2221 struct netlink_callback *cb)
2223 struct net *net = sock_net(skb->sk);
2224 struct neighbour *n;
2225 int rc, h, s_h = cb->args[1];
2226 int idx, s_idx = idx = cb->args[2];
2227 struct neigh_hash_table *nht;
2230 nht = rcu_dereference_bh(tbl->nht);
2232 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2235 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2237 n = rcu_dereference_bh(n->next)) {
2238 if (!net_eq(dev_net(n->dev), net))
2242 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2255 rcu_read_unlock_bh();
2261 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2262 struct netlink_callback *cb)
2264 struct pneigh_entry *n;
2265 struct net *net = sock_net(skb->sk);
2266 int rc, h, s_h = cb->args[3];
2267 int idx, s_idx = idx = cb->args[4];
2269 read_lock_bh(&tbl->lock);
2271 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2274 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2275 if (dev_net(n->dev) != net)
2279 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2282 NLM_F_MULTI, tbl) < 0) {
2283 read_unlock_bh(&tbl->lock);
2292 read_unlock_bh(&tbl->lock);
2301 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2303 struct neigh_table *tbl;
2308 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2310 /* check for full ndmsg structure presence, family member is
2311 * the same for both structures
2313 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2314 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2319 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2320 tbl = neigh_tables[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);
2341 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2344 struct neigh_hash_table *nht;
2347 nht = rcu_dereference_bh(tbl->nht);
2349 read_lock(&tbl->lock); /* avoid resizes */
2350 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2351 struct neighbour *n;
2353 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2355 n = rcu_dereference_bh(n->next))
2358 read_unlock(&tbl->lock);
2359 rcu_read_unlock_bh();
2361 EXPORT_SYMBOL(neigh_for_each);
2363 /* The tbl->lock must be held as a writer and BH disabled. */
2364 void __neigh_for_each_release(struct neigh_table *tbl,
2365 int (*cb)(struct neighbour *))
2368 struct neigh_hash_table *nht;
2370 nht = rcu_dereference_protected(tbl->nht,
2371 lockdep_is_held(&tbl->lock));
2372 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2373 struct neighbour *n;
2374 struct neighbour __rcu **np;
2376 np = &nht->hash_buckets[chain];
2377 while ((n = rcu_dereference_protected(*np,
2378 lockdep_is_held(&tbl->lock))) != NULL) {
2381 write_lock(&n->lock);
2384 rcu_assign_pointer(*np,
2385 rcu_dereference_protected(n->next,
2386 lockdep_is_held(&tbl->lock)));
2390 write_unlock(&n->lock);
2392 neigh_cleanup_and_release(n);
2396 EXPORT_SYMBOL(__neigh_for_each_release);
2398 int neigh_xmit(int index, struct net_device *dev,
2399 const void *addr, struct sk_buff *skb)
2401 int err = -EAFNOSUPPORT;
2402 if (likely(index < NEIGH_NR_TABLES)) {
2403 struct neigh_table *tbl;
2404 struct neighbour *neigh;
2406 tbl = neigh_tables[index];
2409 neigh = __neigh_lookup_noref(tbl, addr, dev);
2411 neigh = __neigh_create(tbl, addr, dev, false);
2412 err = PTR_ERR(neigh);
2415 err = neigh->output(neigh, skb);
2417 else if (index == NEIGH_LINK_TABLE) {
2418 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
2419 addr, NULL, skb->len);
2422 err = dev_queue_xmit(skb);
2430 EXPORT_SYMBOL(neigh_xmit);
2432 #ifdef CONFIG_PROC_FS
2434 static struct neighbour *neigh_get_first(struct seq_file *seq)
2436 struct neigh_seq_state *state = seq->private;
2437 struct net *net = seq_file_net(seq);
2438 struct neigh_hash_table *nht = state->nht;
2439 struct neighbour *n = NULL;
2440 int bucket = state->bucket;
2442 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2443 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2444 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2447 if (!net_eq(dev_net(n->dev), net))
2449 if (state->neigh_sub_iter) {
2453 v = state->neigh_sub_iter(state, n, &fakep);
2457 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2459 if (n->nud_state & ~NUD_NOARP)
2462 n = rcu_dereference_bh(n->next);
2468 state->bucket = bucket;
2473 static struct neighbour *neigh_get_next(struct seq_file *seq,
2474 struct neighbour *n,
2477 struct neigh_seq_state *state = seq->private;
2478 struct net *net = seq_file_net(seq);
2479 struct neigh_hash_table *nht = state->nht;
2481 if (state->neigh_sub_iter) {
2482 void *v = state->neigh_sub_iter(state, n, pos);
2486 n = rcu_dereference_bh(n->next);
2490 if (!net_eq(dev_net(n->dev), net))
2492 if (state->neigh_sub_iter) {
2493 void *v = state->neigh_sub_iter(state, n, pos);
2498 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2501 if (n->nud_state & ~NUD_NOARP)
2504 n = rcu_dereference_bh(n->next);
2510 if (++state->bucket >= (1 << nht->hash_shift))
2513 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2521 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2523 struct neighbour *n = neigh_get_first(seq);
2528 n = neigh_get_next(seq, n, pos);
2533 return *pos ? NULL : n;
2536 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2538 struct neigh_seq_state *state = seq->private;
2539 struct net *net = seq_file_net(seq);
2540 struct neigh_table *tbl = state->tbl;
2541 struct pneigh_entry *pn = NULL;
2542 int bucket = state->bucket;
2544 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2545 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2546 pn = tbl->phash_buckets[bucket];
2547 while (pn && !net_eq(pneigh_net(pn), net))
2552 state->bucket = bucket;
2557 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2558 struct pneigh_entry *pn,
2561 struct neigh_seq_state *state = seq->private;
2562 struct net *net = seq_file_net(seq);
2563 struct neigh_table *tbl = state->tbl;
2567 } while (pn && !net_eq(pneigh_net(pn), net));
2570 if (++state->bucket > PNEIGH_HASHMASK)
2572 pn = tbl->phash_buckets[state->bucket];
2573 while (pn && !net_eq(pneigh_net(pn), net))
2585 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2587 struct pneigh_entry *pn = pneigh_get_first(seq);
2592 pn = pneigh_get_next(seq, pn, pos);
2597 return *pos ? NULL : pn;
2600 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2602 struct neigh_seq_state *state = seq->private;
2604 loff_t idxpos = *pos;
2606 rc = neigh_get_idx(seq, &idxpos);
2607 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2608 rc = pneigh_get_idx(seq, &idxpos);
2613 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2616 struct neigh_seq_state *state = seq->private;
2620 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2623 state->nht = rcu_dereference_bh(tbl->nht);
2625 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2627 EXPORT_SYMBOL(neigh_seq_start);
2629 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2631 struct neigh_seq_state *state;
2634 if (v == SEQ_START_TOKEN) {
2635 rc = neigh_get_first(seq);
2639 state = seq->private;
2640 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2641 rc = neigh_get_next(seq, v, NULL);
2644 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2645 rc = pneigh_get_first(seq);
2647 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2648 rc = pneigh_get_next(seq, v, NULL);
2654 EXPORT_SYMBOL(neigh_seq_next);
2656 void neigh_seq_stop(struct seq_file *seq, void *v)
2659 rcu_read_unlock_bh();
2661 EXPORT_SYMBOL(neigh_seq_stop);
2663 /* statistics via seq_file */
2665 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2667 struct neigh_table *tbl = seq->private;
2671 return SEQ_START_TOKEN;
2673 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2674 if (!cpu_possible(cpu))
2677 return per_cpu_ptr(tbl->stats, cpu);
2682 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2684 struct neigh_table *tbl = seq->private;
2687 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2688 if (!cpu_possible(cpu))
2691 return per_cpu_ptr(tbl->stats, cpu);
2696 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2701 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2703 struct neigh_table *tbl = seq->private;
2704 struct neigh_statistics *st = v;
2706 if (v == SEQ_START_TOKEN) {
2707 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");
2711 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2712 "%08lx %08lx %08lx %08lx %08lx\n",
2713 atomic_read(&tbl->entries),
2724 st->rcv_probes_mcast,
2725 st->rcv_probes_ucast,
2727 st->periodic_gc_runs,
2735 static const struct seq_operations neigh_stat_seq_ops = {
2736 .start = neigh_stat_seq_start,
2737 .next = neigh_stat_seq_next,
2738 .stop = neigh_stat_seq_stop,
2739 .show = neigh_stat_seq_show,
2742 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2744 int ret = seq_open(file, &neigh_stat_seq_ops);
2747 struct seq_file *sf = file->private_data;
2748 sf->private = PDE_DATA(inode);
2753 static const struct file_operations neigh_stat_seq_fops = {
2754 .owner = THIS_MODULE,
2755 .open = neigh_stat_seq_open,
2757 .llseek = seq_lseek,
2758 .release = seq_release,
2761 #endif /* CONFIG_PROC_FS */
2763 static inline size_t neigh_nlmsg_size(void)
2765 return NLMSG_ALIGN(sizeof(struct ndmsg))
2766 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2767 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2768 + nla_total_size(sizeof(struct nda_cacheinfo))
2769 + nla_total_size(4); /* NDA_PROBES */
2772 static void __neigh_notify(struct neighbour *n, int type, int flags)
2774 struct net *net = dev_net(n->dev);
2775 struct sk_buff *skb;
2778 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2782 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2784 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2785 WARN_ON(err == -EMSGSIZE);
2789 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2793 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2796 void neigh_app_ns(struct neighbour *n)
2798 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2800 EXPORT_SYMBOL(neigh_app_ns);
2802 #ifdef CONFIG_SYSCTL
2804 static int int_max = INT_MAX;
2805 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
2807 static int proc_unres_qlen(struct ctl_table *ctl, int write,
2808 void __user *buffer, size_t *lenp, loff_t *ppos)
2811 struct ctl_table tmp = *ctl;
2814 tmp.extra2 = &unres_qlen_max;
2817 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2818 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2821 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2825 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
2830 return __in_dev_arp_parms_get_rcu(dev);
2832 return __in6_dev_nd_parms_get_rcu(dev);
2837 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
2840 struct net_device *dev;
2841 int family = neigh_parms_family(p);
2844 for_each_netdev_rcu(net, dev) {
2845 struct neigh_parms *dst_p =
2846 neigh_get_dev_parms_rcu(dev, family);
2848 if (dst_p && !test_bit(index, dst_p->data_state))
2849 dst_p->data[index] = p->data[index];
2854 static void neigh_proc_update(struct ctl_table *ctl, int write)
2856 struct net_device *dev = ctl->extra1;
2857 struct neigh_parms *p = ctl->extra2;
2858 struct net *net = neigh_parms_net(p);
2859 int index = (int *) ctl->data - p->data;
2864 set_bit(index, p->data_state);
2865 if (!dev) /* NULL dev means this is default value */
2866 neigh_copy_dflt_parms(net, p, index);
2869 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
2870 void __user *buffer,
2871 size_t *lenp, loff_t *ppos)
2873 struct ctl_table tmp = *ctl;
2877 tmp.extra2 = &int_max;
2879 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2880 neigh_proc_update(ctl, write);
2884 int neigh_proc_dointvec(struct ctl_table *ctl, int write,
2885 void __user *buffer, size_t *lenp, loff_t *ppos)
2887 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
2889 neigh_proc_update(ctl, write);
2892 EXPORT_SYMBOL(neigh_proc_dointvec);
2894 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
2895 void __user *buffer,
2896 size_t *lenp, loff_t *ppos)
2898 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2900 neigh_proc_update(ctl, write);
2903 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
2905 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
2906 void __user *buffer,
2907 size_t *lenp, loff_t *ppos)
2909 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
2911 neigh_proc_update(ctl, write);
2915 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
2916 void __user *buffer,
2917 size_t *lenp, loff_t *ppos)
2919 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2921 neigh_proc_update(ctl, write);
2924 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
2926 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
2927 void __user *buffer,
2928 size_t *lenp, loff_t *ppos)
2930 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
2932 neigh_proc_update(ctl, write);
2936 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
2937 void __user *buffer,
2938 size_t *lenp, loff_t *ppos)
2940 struct neigh_parms *p = ctl->extra2;
2943 if (strcmp(ctl->procname, "base_reachable_time") == 0)
2944 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
2945 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
2946 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
2950 if (write && ret == 0) {
2951 /* update reachable_time as well, otherwise, the change will
2952 * only be effective after the next time neigh_periodic_work
2953 * decides to recompute it
2956 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2961 #define NEIGH_PARMS_DATA_OFFSET(index) \
2962 (&((struct neigh_parms *) 0)->data[index])
2964 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
2965 [NEIGH_VAR_ ## attr] = { \
2967 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
2968 .maxlen = sizeof(int), \
2970 .proc_handler = proc, \
2973 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
2974 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
2976 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
2977 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
2979 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
2980 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
2982 #define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
2983 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2985 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
2986 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
2988 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
2989 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
2991 static struct neigh_sysctl_table {
2992 struct ctl_table_header *sysctl_header;
2993 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2994 } neigh_sysctl_template __read_mostly = {
2996 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
2997 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
2998 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
2999 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3000 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3001 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3002 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3003 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3004 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3005 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3006 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3007 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3008 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3009 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3010 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3011 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3012 [NEIGH_VAR_GC_INTERVAL] = {
3013 .procname = "gc_interval",
3014 .maxlen = sizeof(int),
3016 .proc_handler = proc_dointvec_jiffies,
3018 [NEIGH_VAR_GC_THRESH1] = {
3019 .procname = "gc_thresh1",
3020 .maxlen = sizeof(int),
3024 .proc_handler = proc_dointvec_minmax,
3026 [NEIGH_VAR_GC_THRESH2] = {
3027 .procname = "gc_thresh2",
3028 .maxlen = sizeof(int),
3032 .proc_handler = proc_dointvec_minmax,
3034 [NEIGH_VAR_GC_THRESH3] = {
3035 .procname = "gc_thresh3",
3036 .maxlen = sizeof(int),
3040 .proc_handler = proc_dointvec_minmax,
3046 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3047 proc_handler *handler)
3050 struct neigh_sysctl_table *t;
3051 const char *dev_name_source;
3052 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3055 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3059 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3060 t->neigh_vars[i].data += (long) p;
3061 t->neigh_vars[i].extra1 = dev;
3062 t->neigh_vars[i].extra2 = p;
3066 dev_name_source = dev->name;
3067 /* Terminate the table early */
3068 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3069 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3071 struct neigh_table *tbl = p->tbl;
3072 dev_name_source = "default";
3073 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3074 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3075 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3076 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3081 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3083 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3084 /* RetransTime (in milliseconds)*/
3085 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3086 /* ReachableTime (in milliseconds) */
3087 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3089 /* Those handlers will update p->reachable_time after
3090 * base_reachable_time(_ms) is set to ensure the new timer starts being
3091 * applied after the next neighbour update instead of waiting for
3092 * neigh_periodic_work to update its value (can be multiple minutes)
3093 * So any handler that replaces them should do this as well
3096 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3097 neigh_proc_base_reachable_time;
3098 /* ReachableTime (in milliseconds) */
3099 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3100 neigh_proc_base_reachable_time;
3103 /* Don't export sysctls to unprivileged users */
3104 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3105 t->neigh_vars[0].procname = NULL;
3107 switch (neigh_parms_family(p)) {
3118 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3119 p_name, dev_name_source);
3121 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3122 if (!t->sysctl_header)
3125 p->sysctl_table = t;
3133 EXPORT_SYMBOL(neigh_sysctl_register);
3135 void neigh_sysctl_unregister(struct neigh_parms *p)
3137 if (p->sysctl_table) {
3138 struct neigh_sysctl_table *t = p->sysctl_table;
3139 p->sysctl_table = NULL;
3140 unregister_net_sysctl_table(t->sysctl_header);
3144 EXPORT_SYMBOL(neigh_sysctl_unregister);
3146 #endif /* CONFIG_SYSCTL */
3148 static int __init neigh_init(void)
3150 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3151 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3152 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3154 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3156 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3161 subsys_initcall(neigh_init);