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_PRINTK(x...) printk(x)
45 #define NEIGH_NOPRINTK(x...) do { ; } while(0)
46 #define NEIGH_PRINTK1 NEIGH_NOPRINTK
47 #define NEIGH_PRINTK2 NEIGH_NOPRINTK
51 #define NEIGH_PRINTK1 NEIGH_PRINTK
55 #define NEIGH_PRINTK2 NEIGH_PRINTK
58 #define PNEIGH_HASHMASK 0xF
60 static void neigh_timer_handler(unsigned long arg);
61 static void __neigh_notify(struct neighbour *n, int type, int flags);
62 static void neigh_update_notify(struct neighbour *neigh);
63 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev);
66 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
68 static struct neigh_table *neigh_tables;
70 static const struct file_operations neigh_stat_seq_fops;
74 Neighbour hash table buckets are protected with rwlock tbl->lock.
76 - All the scans/updates to hash buckets MUST be made under this lock.
77 - NOTHING clever should be made under this lock: no callbacks
78 to protocol backends, no attempts to send something to network.
79 It will result in deadlocks, if backend/driver wants to use neighbour
81 - If the entry requires some non-trivial actions, increase
82 its reference count and release table lock.
84 Neighbour entries are protected:
85 - with reference count.
86 - with rwlock neigh->lock
88 Reference count prevents destruction.
90 neigh->lock mainly serializes ll address data and its validity state.
91 However, the same lock is used to protect another entry fields:
95 Again, nothing clever shall be made under neigh->lock,
96 the most complicated procedure, which we allow is dev->hard_header.
97 It is supposed, that dev->hard_header is simplistic and does
98 not make callbacks to neighbour tables.
100 The last lock is neigh_tbl_lock. It is pure SMP lock, protecting
101 list of neighbour tables. This list is used only in process context,
104 static DEFINE_RWLOCK(neigh_tbl_lock);
106 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
112 static void neigh_cleanup_and_release(struct neighbour *neigh)
114 if (neigh->parms->neigh_cleanup)
115 neigh->parms->neigh_cleanup(neigh);
117 __neigh_notify(neigh, RTM_DELNEIGH, 0);
118 neigh_release(neigh);
122 * It is random distribution in the interval (1/2)*base...(3/2)*base.
123 * It corresponds to default IPv6 settings and is not overridable,
124 * because it is really reasonable choice.
127 unsigned long neigh_rand_reach_time(unsigned long base)
129 return base ? (net_random() % base) + (base >> 1) : 0;
131 EXPORT_SYMBOL(neigh_rand_reach_time);
134 static int neigh_forced_gc(struct neigh_table *tbl)
138 struct neigh_hash_table *nht;
140 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
142 write_lock_bh(&tbl->lock);
143 nht = rcu_dereference_protected(tbl->nht,
144 lockdep_is_held(&tbl->lock));
145 for (i = 0; i < (1 << nht->hash_shift); i++) {
147 struct neighbour __rcu **np;
149 np = &nht->hash_buckets[i];
150 while ((n = rcu_dereference_protected(*np,
151 lockdep_is_held(&tbl->lock))) != NULL) {
152 /* Neighbour record may be discarded if:
153 * - nobody refers to it.
154 * - it is not permanent
156 write_lock(&n->lock);
157 if (atomic_read(&n->refcnt) == 1 &&
158 !(n->nud_state & NUD_PERMANENT)) {
159 rcu_assign_pointer(*np,
160 rcu_dereference_protected(n->next,
161 lockdep_is_held(&tbl->lock)));
164 write_unlock(&n->lock);
165 neigh_cleanup_and_release(n);
168 write_unlock(&n->lock);
173 tbl->last_flush = jiffies;
175 write_unlock_bh(&tbl->lock);
180 static void neigh_add_timer(struct neighbour *n, unsigned long when)
183 if (unlikely(mod_timer(&n->timer, when))) {
184 printk("NEIGH: BUG, double timer add, state is %x\n",
190 static int neigh_del_timer(struct neighbour *n)
192 if ((n->nud_state & NUD_IN_TIMER) &&
193 del_timer(&n->timer)) {
200 static void pneigh_queue_purge(struct sk_buff_head *list)
204 while ((skb = skb_dequeue(list)) != NULL) {
210 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev)
213 struct neigh_hash_table *nht;
215 nht = rcu_dereference_protected(tbl->nht,
216 lockdep_is_held(&tbl->lock));
218 for (i = 0; i < (1 << nht->hash_shift); i++) {
220 struct neighbour __rcu **np = &nht->hash_buckets[i];
222 while ((n = rcu_dereference_protected(*np,
223 lockdep_is_held(&tbl->lock))) != NULL) {
224 if (dev && n->dev != dev) {
228 rcu_assign_pointer(*np,
229 rcu_dereference_protected(n->next,
230 lockdep_is_held(&tbl->lock)));
231 write_lock(&n->lock);
235 if (atomic_read(&n->refcnt) != 1) {
236 /* The most unpleasant situation.
237 We must destroy neighbour entry,
238 but someone still uses it.
240 The destroy will be delayed until
241 the last user releases us, but
242 we must kill timers etc. and move
245 skb_queue_purge(&n->arp_queue);
246 n->arp_queue_len_bytes = 0;
247 n->output = neigh_blackhole;
248 if (n->nud_state & NUD_VALID)
249 n->nud_state = NUD_NOARP;
251 n->nud_state = NUD_NONE;
252 NEIGH_PRINTK2("neigh %p is stray.\n", n);
254 write_unlock(&n->lock);
255 neigh_cleanup_and_release(n);
260 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
262 write_lock_bh(&tbl->lock);
263 neigh_flush_dev(tbl, dev);
264 write_unlock_bh(&tbl->lock);
266 EXPORT_SYMBOL(neigh_changeaddr);
268 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
270 write_lock_bh(&tbl->lock);
271 neigh_flush_dev(tbl, dev);
272 pneigh_ifdown(tbl, dev);
273 write_unlock_bh(&tbl->lock);
275 del_timer_sync(&tbl->proxy_timer);
276 pneigh_queue_purge(&tbl->proxy_queue);
279 EXPORT_SYMBOL(neigh_ifdown);
281 static struct neighbour *neigh_alloc(struct neigh_table *tbl, struct net_device *dev)
283 struct neighbour *n = NULL;
284 unsigned long now = jiffies;
287 entries = atomic_inc_return(&tbl->entries) - 1;
288 if (entries >= tbl->gc_thresh3 ||
289 (entries >= tbl->gc_thresh2 &&
290 time_after(now, tbl->last_flush + 5 * HZ))) {
291 if (!neigh_forced_gc(tbl) &&
292 entries >= tbl->gc_thresh3)
297 n = kzalloc(tbl->entry_size, GFP_ATOMIC);
299 int sz = sizeof(*n) + tbl->key_len;
301 sz = ALIGN(sz, NEIGH_PRIV_ALIGN);
302 sz += dev->neigh_priv_len;
303 n = kzalloc(sz, GFP_ATOMIC);
308 skb_queue_head_init(&n->arp_queue);
309 rwlock_init(&n->lock);
310 seqlock_init(&n->ha_lock);
311 n->updated = n->used = now;
312 n->nud_state = NUD_NONE;
313 n->output = neigh_blackhole;
314 seqlock_init(&n->hh.hh_lock);
315 n->parms = neigh_parms_clone(&tbl->parms);
316 setup_timer(&n->timer, neigh_timer_handler, (unsigned long)n);
318 NEIGH_CACHE_STAT_INC(tbl, allocs);
320 atomic_set(&n->refcnt, 1);
326 atomic_dec(&tbl->entries);
330 static void neigh_get_hash_rnd(u32 *x)
332 get_random_bytes(x, sizeof(*x));
336 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
338 size_t size = (1 << shift) * sizeof(struct neighbour *);
339 struct neigh_hash_table *ret;
340 struct neighbour __rcu **buckets;
343 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
346 if (size <= PAGE_SIZE)
347 buckets = kzalloc(size, GFP_ATOMIC);
349 buckets = (struct neighbour __rcu **)
350 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
356 ret->hash_buckets = buckets;
357 ret->hash_shift = shift;
358 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
359 neigh_get_hash_rnd(&ret->hash_rnd[i]);
363 static void neigh_hash_free_rcu(struct rcu_head *head)
365 struct neigh_hash_table *nht = container_of(head,
366 struct neigh_hash_table,
368 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
369 struct neighbour __rcu **buckets = nht->hash_buckets;
371 if (size <= PAGE_SIZE)
374 free_pages((unsigned long)buckets, get_order(size));
378 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
379 unsigned long new_shift)
381 unsigned int i, hash;
382 struct neigh_hash_table *new_nht, *old_nht;
384 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
386 old_nht = rcu_dereference_protected(tbl->nht,
387 lockdep_is_held(&tbl->lock));
388 new_nht = neigh_hash_alloc(new_shift);
392 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
393 struct neighbour *n, *next;
395 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
396 lockdep_is_held(&tbl->lock));
399 hash = tbl->hash(n->primary_key, n->dev,
402 hash >>= (32 - new_nht->hash_shift);
403 next = rcu_dereference_protected(n->next,
404 lockdep_is_held(&tbl->lock));
406 rcu_assign_pointer(n->next,
407 rcu_dereference_protected(
408 new_nht->hash_buckets[hash],
409 lockdep_is_held(&tbl->lock)));
410 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
414 rcu_assign_pointer(tbl->nht, new_nht);
415 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
419 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
420 struct net_device *dev)
423 int key_len = tbl->key_len;
425 struct neigh_hash_table *nht;
427 NEIGH_CACHE_STAT_INC(tbl, lookups);
430 nht = rcu_dereference_bh(tbl->nht);
431 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
433 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
435 n = rcu_dereference_bh(n->next)) {
436 if (dev == n->dev && !memcmp(n->primary_key, pkey, key_len)) {
437 if (!atomic_inc_not_zero(&n->refcnt))
439 NEIGH_CACHE_STAT_INC(tbl, hits);
444 rcu_read_unlock_bh();
447 EXPORT_SYMBOL(neigh_lookup);
449 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
453 int key_len = tbl->key_len;
455 struct neigh_hash_table *nht;
457 NEIGH_CACHE_STAT_INC(tbl, lookups);
460 nht = rcu_dereference_bh(tbl->nht);
461 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
463 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
465 n = rcu_dereference_bh(n->next)) {
466 if (!memcmp(n->primary_key, pkey, key_len) &&
467 net_eq(dev_net(n->dev), net)) {
468 if (!atomic_inc_not_zero(&n->refcnt))
470 NEIGH_CACHE_STAT_INC(tbl, hits);
475 rcu_read_unlock_bh();
478 EXPORT_SYMBOL(neigh_lookup_nodev);
480 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
481 struct net_device *dev, bool want_ref)
484 int key_len = tbl->key_len;
486 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev);
487 struct neigh_hash_table *nht;
490 rc = ERR_PTR(-ENOBUFS);
494 memcpy(n->primary_key, pkey, key_len);
498 /* Protocol specific setup. */
499 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
501 goto out_neigh_release;
504 if (dev->netdev_ops->ndo_neigh_construct) {
505 error = dev->netdev_ops->ndo_neigh_construct(n);
508 goto out_neigh_release;
512 /* Device specific setup. */
513 if (n->parms->neigh_setup &&
514 (error = n->parms->neigh_setup(n)) < 0) {
516 goto out_neigh_release;
519 n->confirmed = jiffies - (n->parms->base_reachable_time << 1);
521 write_lock_bh(&tbl->lock);
522 nht = rcu_dereference_protected(tbl->nht,
523 lockdep_is_held(&tbl->lock));
525 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
526 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
528 hash_val = tbl->hash(pkey, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
530 if (n->parms->dead) {
531 rc = ERR_PTR(-EINVAL);
535 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
536 lockdep_is_held(&tbl->lock));
538 n1 = rcu_dereference_protected(n1->next,
539 lockdep_is_held(&tbl->lock))) {
540 if (dev == n1->dev && !memcmp(n1->primary_key, pkey, key_len)) {
551 rcu_assign_pointer(n->next,
552 rcu_dereference_protected(nht->hash_buckets[hash_val],
553 lockdep_is_held(&tbl->lock)));
554 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
555 write_unlock_bh(&tbl->lock);
556 NEIGH_PRINTK2("neigh %p is created.\n", n);
561 write_unlock_bh(&tbl->lock);
566 EXPORT_SYMBOL(__neigh_create);
568 static u32 pneigh_hash(const void *pkey, int key_len)
570 u32 hash_val = *(u32 *)(pkey + key_len - 4);
571 hash_val ^= (hash_val >> 16);
572 hash_val ^= hash_val >> 8;
573 hash_val ^= hash_val >> 4;
574 hash_val &= PNEIGH_HASHMASK;
578 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
582 struct net_device *dev)
585 if (!memcmp(n->key, pkey, key_len) &&
586 net_eq(pneigh_net(n), net) &&
587 (n->dev == dev || !n->dev))
594 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
595 struct net *net, const void *pkey, struct net_device *dev)
597 int key_len = tbl->key_len;
598 u32 hash_val = pneigh_hash(pkey, key_len);
600 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
601 net, pkey, key_len, dev);
603 EXPORT_SYMBOL_GPL(__pneigh_lookup);
605 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
606 struct net *net, const void *pkey,
607 struct net_device *dev, int creat)
609 struct pneigh_entry *n;
610 int key_len = tbl->key_len;
611 u32 hash_val = pneigh_hash(pkey, key_len);
613 read_lock_bh(&tbl->lock);
614 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
615 net, pkey, key_len, dev);
616 read_unlock_bh(&tbl->lock);
623 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
627 write_pnet(&n->net, hold_net(net));
628 memcpy(n->key, pkey, key_len);
633 if (tbl->pconstructor && tbl->pconstructor(n)) {
642 write_lock_bh(&tbl->lock);
643 n->next = tbl->phash_buckets[hash_val];
644 tbl->phash_buckets[hash_val] = n;
645 write_unlock_bh(&tbl->lock);
649 EXPORT_SYMBOL(pneigh_lookup);
652 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
653 struct net_device *dev)
655 struct pneigh_entry *n, **np;
656 int key_len = tbl->key_len;
657 u32 hash_val = pneigh_hash(pkey, key_len);
659 write_lock_bh(&tbl->lock);
660 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
662 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
663 net_eq(pneigh_net(n), net)) {
665 write_unlock_bh(&tbl->lock);
666 if (tbl->pdestructor)
670 release_net(pneigh_net(n));
675 write_unlock_bh(&tbl->lock);
679 static int pneigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
681 struct pneigh_entry *n, **np;
684 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
685 np = &tbl->phash_buckets[h];
686 while ((n = *np) != NULL) {
687 if (!dev || n->dev == dev) {
689 if (tbl->pdestructor)
693 release_net(pneigh_net(n));
703 static void neigh_parms_destroy(struct neigh_parms *parms);
705 static inline void neigh_parms_put(struct neigh_parms *parms)
707 if (atomic_dec_and_test(&parms->refcnt))
708 neigh_parms_destroy(parms);
712 * neighbour must already be out of the table;
715 void neigh_destroy(struct neighbour *neigh)
717 struct net_device *dev = neigh->dev;
719 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
722 pr_warn("Destroying alive neighbour %p\n", neigh);
727 if (neigh_del_timer(neigh))
728 pr_warn("Impossible event\n");
730 skb_queue_purge(&neigh->arp_queue);
731 neigh->arp_queue_len_bytes = 0;
733 if (dev->netdev_ops->ndo_neigh_destroy)
734 dev->netdev_ops->ndo_neigh_destroy(neigh);
737 neigh_parms_put(neigh->parms);
739 NEIGH_PRINTK2("neigh %p is destroyed.\n", neigh);
741 atomic_dec(&neigh->tbl->entries);
742 kfree_rcu(neigh, rcu);
744 EXPORT_SYMBOL(neigh_destroy);
746 /* Neighbour state is suspicious;
749 Called with write_locked neigh.
751 static void neigh_suspect(struct neighbour *neigh)
753 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
755 neigh->output = neigh->ops->output;
758 /* Neighbour state is OK;
761 Called with write_locked neigh.
763 static void neigh_connect(struct neighbour *neigh)
765 NEIGH_PRINTK2("neigh %p is connected.\n", neigh);
767 neigh->output = neigh->ops->connected_output;
770 static void neigh_periodic_work(struct work_struct *work)
772 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
774 struct neighbour __rcu **np;
776 struct neigh_hash_table *nht;
778 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
780 write_lock_bh(&tbl->lock);
781 nht = rcu_dereference_protected(tbl->nht,
782 lockdep_is_held(&tbl->lock));
785 * periodically recompute ReachableTime from random function
788 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
789 struct neigh_parms *p;
790 tbl->last_rand = jiffies;
791 for (p = &tbl->parms; p; p = p->next)
793 neigh_rand_reach_time(p->base_reachable_time);
796 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
797 np = &nht->hash_buckets[i];
799 while ((n = rcu_dereference_protected(*np,
800 lockdep_is_held(&tbl->lock))) != NULL) {
803 write_lock(&n->lock);
805 state = n->nud_state;
806 if (state & (NUD_PERMANENT | NUD_IN_TIMER)) {
807 write_unlock(&n->lock);
811 if (time_before(n->used, n->confirmed))
812 n->used = n->confirmed;
814 if (atomic_read(&n->refcnt) == 1 &&
815 (state == NUD_FAILED ||
816 time_after(jiffies, n->used + n->parms->gc_staletime))) {
819 write_unlock(&n->lock);
820 neigh_cleanup_and_release(n);
823 write_unlock(&n->lock);
829 * It's fine to release lock here, even if hash table
830 * grows while we are preempted.
832 write_unlock_bh(&tbl->lock);
834 write_lock_bh(&tbl->lock);
835 nht = rcu_dereference_protected(tbl->nht,
836 lockdep_is_held(&tbl->lock));
838 /* Cycle through all hash buckets every base_reachable_time/2 ticks.
839 * ARP entry timeouts range from 1/2 base_reachable_time to 3/2
840 * base_reachable_time.
842 schedule_delayed_work(&tbl->gc_work,
843 tbl->parms.base_reachable_time >> 1);
844 write_unlock_bh(&tbl->lock);
847 static __inline__ int neigh_max_probes(struct neighbour *n)
849 struct neigh_parms *p = n->parms;
850 return (n->nud_state & NUD_PROBE) ?
852 p->ucast_probes + p->app_probes + p->mcast_probes;
855 static void neigh_invalidate(struct neighbour *neigh)
856 __releases(neigh->lock)
857 __acquires(neigh->lock)
861 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
862 NEIGH_PRINTK2("neigh %p is failed.\n", neigh);
863 neigh->updated = jiffies;
865 /* It is very thin place. report_unreachable is very complicated
866 routine. Particularly, it can hit the same neighbour entry!
868 So that, we try to be accurate and avoid dead loop. --ANK
870 while (neigh->nud_state == NUD_FAILED &&
871 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
872 write_unlock(&neigh->lock);
873 neigh->ops->error_report(neigh, skb);
874 write_lock(&neigh->lock);
876 skb_queue_purge(&neigh->arp_queue);
877 neigh->arp_queue_len_bytes = 0;
880 static void neigh_probe(struct neighbour *neigh)
881 __releases(neigh->lock)
883 struct sk_buff *skb = skb_peek(&neigh->arp_queue);
884 /* keep skb alive even if arp_queue overflows */
886 skb = skb_copy(skb, GFP_ATOMIC);
887 write_unlock(&neigh->lock);
888 neigh->ops->solicit(neigh, skb);
889 atomic_inc(&neigh->probes);
893 /* Called when a timer expires for a neighbour entry. */
895 static void neigh_timer_handler(unsigned long arg)
897 unsigned long now, next;
898 struct neighbour *neigh = (struct neighbour *)arg;
902 write_lock(&neigh->lock);
904 state = neigh->nud_state;
908 if (!(state & NUD_IN_TIMER))
911 if (state & NUD_REACHABLE) {
912 if (time_before_eq(now,
913 neigh->confirmed + neigh->parms->reachable_time)) {
914 NEIGH_PRINTK2("neigh %p is still alive.\n", neigh);
915 next = neigh->confirmed + neigh->parms->reachable_time;
916 } else if (time_before_eq(now,
917 neigh->used + neigh->parms->delay_probe_time)) {
918 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
919 neigh->nud_state = NUD_DELAY;
920 neigh->updated = jiffies;
921 neigh_suspect(neigh);
922 next = now + neigh->parms->delay_probe_time;
924 NEIGH_PRINTK2("neigh %p is suspected.\n", neigh);
925 neigh->nud_state = NUD_STALE;
926 neigh->updated = jiffies;
927 neigh_suspect(neigh);
930 } else if (state & NUD_DELAY) {
931 if (time_before_eq(now,
932 neigh->confirmed + neigh->parms->delay_probe_time)) {
933 NEIGH_PRINTK2("neigh %p is now reachable.\n", neigh);
934 neigh->nud_state = NUD_REACHABLE;
935 neigh->updated = jiffies;
936 neigh_connect(neigh);
938 next = neigh->confirmed + neigh->parms->reachable_time;
940 NEIGH_PRINTK2("neigh %p is probed.\n", neigh);
941 neigh->nud_state = NUD_PROBE;
942 neigh->updated = jiffies;
943 atomic_set(&neigh->probes, 0);
944 next = now + neigh->parms->retrans_time;
947 /* NUD_PROBE|NUD_INCOMPLETE */
948 next = now + neigh->parms->retrans_time;
951 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
952 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
953 neigh->nud_state = NUD_FAILED;
955 neigh_invalidate(neigh);
958 if (neigh->nud_state & NUD_IN_TIMER) {
959 if (time_before(next, jiffies + HZ/2))
960 next = jiffies + HZ/2;
961 if (!mod_timer(&neigh->timer, next))
964 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
968 write_unlock(&neigh->lock);
972 neigh_update_notify(neigh);
974 neigh_release(neigh);
977 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
980 bool immediate_probe = false;
982 write_lock_bh(&neigh->lock);
985 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
988 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
989 if (neigh->parms->mcast_probes + neigh->parms->app_probes) {
990 unsigned long next, now = jiffies;
992 atomic_set(&neigh->probes, neigh->parms->ucast_probes);
993 neigh->nud_state = NUD_INCOMPLETE;
994 neigh->updated = now;
995 next = now + max(neigh->parms->retrans_time, HZ/2);
996 neigh_add_timer(neigh, next);
997 immediate_probe = true;
999 neigh->nud_state = NUD_FAILED;
1000 neigh->updated = jiffies;
1001 write_unlock_bh(&neigh->lock);
1006 } else if (neigh->nud_state & NUD_STALE) {
1007 NEIGH_PRINTK2("neigh %p is delayed.\n", neigh);
1008 neigh->nud_state = NUD_DELAY;
1009 neigh->updated = jiffies;
1010 neigh_add_timer(neigh,
1011 jiffies + neigh->parms->delay_probe_time);
1014 if (neigh->nud_state == NUD_INCOMPLETE) {
1016 while (neigh->arp_queue_len_bytes + skb->truesize >
1017 neigh->parms->queue_len_bytes) {
1018 struct sk_buff *buff;
1020 buff = __skb_dequeue(&neigh->arp_queue);
1023 neigh->arp_queue_len_bytes -= buff->truesize;
1025 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1028 __skb_queue_tail(&neigh->arp_queue, skb);
1029 neigh->arp_queue_len_bytes += skb->truesize;
1034 if (immediate_probe)
1037 write_unlock(&neigh->lock);
1041 EXPORT_SYMBOL(__neigh_event_send);
1043 static void neigh_update_hhs(struct neighbour *neigh)
1045 struct hh_cache *hh;
1046 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1049 if (neigh->dev->header_ops)
1050 update = neigh->dev->header_ops->cache_update;
1055 write_seqlock_bh(&hh->hh_lock);
1056 update(hh, neigh->dev, neigh->ha);
1057 write_sequnlock_bh(&hh->hh_lock);
1064 /* Generic update routine.
1065 -- lladdr is new lladdr or NULL, if it is not supplied.
1066 -- new is new state.
1068 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1070 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1071 lladdr instead of overriding it
1073 It also allows to retain current state
1074 if lladdr is unchanged.
1075 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1077 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1079 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1082 Caller MUST hold reference count on the entry.
1085 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1091 struct net_device *dev;
1092 int update_isrouter = 0;
1094 write_lock_bh(&neigh->lock);
1097 old = neigh->nud_state;
1100 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1101 (old & (NUD_NOARP | NUD_PERMANENT)))
1104 if (!(new & NUD_VALID)) {
1105 neigh_del_timer(neigh);
1106 if (old & NUD_CONNECTED)
1107 neigh_suspect(neigh);
1108 neigh->nud_state = new;
1110 notify = old & NUD_VALID;
1111 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1112 (new & NUD_FAILED)) {
1113 neigh_invalidate(neigh);
1119 /* Compare new lladdr with cached one */
1120 if (!dev->addr_len) {
1121 /* First case: device needs no address. */
1123 } else if (lladdr) {
1124 /* The second case: if something is already cached
1125 and a new address is proposed:
1127 - if they are different, check override flag
1129 if ((old & NUD_VALID) &&
1130 !memcmp(lladdr, neigh->ha, dev->addr_len))
1133 /* No address is supplied; if we know something,
1134 use it, otherwise discard the request.
1137 if (!(old & NUD_VALID))
1142 if (new & NUD_CONNECTED)
1143 neigh->confirmed = jiffies;
1144 neigh->updated = jiffies;
1146 /* If entry was valid and address is not changed,
1147 do not change entry state, if new one is STALE.
1150 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1151 if (old & NUD_VALID) {
1152 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1153 update_isrouter = 0;
1154 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1155 (old & NUD_CONNECTED)) {
1161 if (lladdr == neigh->ha && new == NUD_STALE &&
1162 ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) ||
1163 (old & NUD_CONNECTED))
1170 neigh_del_timer(neigh);
1171 if (new & NUD_IN_TIMER)
1172 neigh_add_timer(neigh, (jiffies +
1173 ((new & NUD_REACHABLE) ?
1174 neigh->parms->reachable_time :
1176 neigh->nud_state = new;
1179 if (lladdr != neigh->ha) {
1180 write_seqlock(&neigh->ha_lock);
1181 memcpy(&neigh->ha, lladdr, dev->addr_len);
1182 write_sequnlock(&neigh->ha_lock);
1183 neigh_update_hhs(neigh);
1184 if (!(new & NUD_CONNECTED))
1185 neigh->confirmed = jiffies -
1186 (neigh->parms->base_reachable_time << 1);
1191 if (new & NUD_CONNECTED)
1192 neigh_connect(neigh);
1194 neigh_suspect(neigh);
1195 if (!(old & NUD_VALID)) {
1196 struct sk_buff *skb;
1198 /* Again: avoid dead loop if something went wrong */
1200 while (neigh->nud_state & NUD_VALID &&
1201 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1202 struct dst_entry *dst = skb_dst(skb);
1203 struct neighbour *n2, *n1 = neigh;
1204 write_unlock_bh(&neigh->lock);
1208 /* Why not just use 'neigh' as-is? The problem is that
1209 * things such as shaper, eql, and sch_teql can end up
1210 * using alternative, different, neigh objects to output
1211 * the packet in the output path. So what we need to do
1212 * here is re-lookup the top-level neigh in the path so
1213 * we can reinject the packet there.
1217 n2 = dst_neigh_lookup_skb(dst, skb);
1221 n1->output(n1, skb);
1226 write_lock_bh(&neigh->lock);
1228 skb_queue_purge(&neigh->arp_queue);
1229 neigh->arp_queue_len_bytes = 0;
1232 if (update_isrouter) {
1233 neigh->flags = (flags & NEIGH_UPDATE_F_ISROUTER) ?
1234 (neigh->flags | NTF_ROUTER) :
1235 (neigh->flags & ~NTF_ROUTER);
1237 write_unlock_bh(&neigh->lock);
1240 neigh_update_notify(neigh);
1244 EXPORT_SYMBOL(neigh_update);
1246 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1247 u8 *lladdr, void *saddr,
1248 struct net_device *dev)
1250 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1251 lladdr || !dev->addr_len);
1253 neigh_update(neigh, lladdr, NUD_STALE,
1254 NEIGH_UPDATE_F_OVERRIDE);
1257 EXPORT_SYMBOL(neigh_event_ns);
1259 /* called with read_lock_bh(&n->lock); */
1260 static void neigh_hh_init(struct neighbour *n, struct dst_entry *dst)
1262 struct net_device *dev = dst->dev;
1263 __be16 prot = dst->ops->protocol;
1264 struct hh_cache *hh = &n->hh;
1266 write_lock_bh(&n->lock);
1268 /* Only one thread can come in here and initialize the
1272 dev->header_ops->cache(n, hh, prot);
1274 write_unlock_bh(&n->lock);
1277 /* This function can be used in contexts, where only old dev_queue_xmit
1278 * worked, f.e. if you want to override normal output path (eql, shaper),
1279 * but resolution is not made yet.
1282 int neigh_compat_output(struct neighbour *neigh, struct sk_buff *skb)
1284 struct net_device *dev = skb->dev;
1286 __skb_pull(skb, skb_network_offset(skb));
1288 if (dev_hard_header(skb, dev, ntohs(skb->protocol), NULL, NULL,
1290 dev->header_ops->rebuild(skb))
1293 return dev_queue_xmit(skb);
1295 EXPORT_SYMBOL(neigh_compat_output);
1297 /* Slow and careful. */
1299 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1301 struct dst_entry *dst = skb_dst(skb);
1307 if (!neigh_event_send(neigh, skb)) {
1309 struct net_device *dev = neigh->dev;
1312 if (dev->header_ops->cache && !neigh->hh.hh_len)
1313 neigh_hh_init(neigh, dst);
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 rc = dev_queue_xmit(skb);
1330 NEIGH_PRINTK1("neigh_resolve_output: dst=%p neigh=%p\n",
1337 EXPORT_SYMBOL(neigh_resolve_output);
1339 /* As fast as possible without hh cache */
1341 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1343 struct net_device *dev = neigh->dev;
1348 __skb_pull(skb, skb_network_offset(skb));
1349 seq = read_seqbegin(&neigh->ha_lock);
1350 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1351 neigh->ha, NULL, skb->len);
1352 } while (read_seqretry(&neigh->ha_lock, seq));
1355 err = dev_queue_xmit(skb);
1362 EXPORT_SYMBOL(neigh_connected_output);
1364 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1366 return dev_queue_xmit(skb);
1368 EXPORT_SYMBOL(neigh_direct_output);
1370 static void neigh_proxy_process(unsigned long arg)
1372 struct neigh_table *tbl = (struct neigh_table *)arg;
1373 long sched_next = 0;
1374 unsigned long now = jiffies;
1375 struct sk_buff *skb, *n;
1377 spin_lock(&tbl->proxy_queue.lock);
1379 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1380 long tdif = NEIGH_CB(skb)->sched_next - now;
1383 struct net_device *dev = skb->dev;
1385 __skb_unlink(skb, &tbl->proxy_queue);
1386 if (tbl->proxy_redo && netif_running(dev)) {
1388 tbl->proxy_redo(skb);
1395 } else if (!sched_next || tdif < sched_next)
1398 del_timer(&tbl->proxy_timer);
1400 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1401 spin_unlock(&tbl->proxy_queue.lock);
1404 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1405 struct sk_buff *skb)
1407 unsigned long now = jiffies;
1408 unsigned long sched_next = now + (net_random() % p->proxy_delay);
1410 if (tbl->proxy_queue.qlen > p->proxy_qlen) {
1415 NEIGH_CB(skb)->sched_next = sched_next;
1416 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1418 spin_lock(&tbl->proxy_queue.lock);
1419 if (del_timer(&tbl->proxy_timer)) {
1420 if (time_before(tbl->proxy_timer.expires, sched_next))
1421 sched_next = tbl->proxy_timer.expires;
1425 __skb_queue_tail(&tbl->proxy_queue, skb);
1426 mod_timer(&tbl->proxy_timer, sched_next);
1427 spin_unlock(&tbl->proxy_queue.lock);
1429 EXPORT_SYMBOL(pneigh_enqueue);
1431 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1432 struct net *net, int ifindex)
1434 struct neigh_parms *p;
1436 for (p = &tbl->parms; p; p = p->next) {
1437 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1438 (!p->dev && !ifindex))
1445 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1446 struct neigh_table *tbl)
1448 struct neigh_parms *p, *ref;
1449 struct net *net = dev_net(dev);
1450 const struct net_device_ops *ops = dev->netdev_ops;
1452 ref = lookup_neigh_parms(tbl, net, 0);
1456 p = kmemdup(ref, sizeof(*p), GFP_KERNEL);
1459 atomic_set(&p->refcnt, 1);
1461 neigh_rand_reach_time(p->base_reachable_time);
1463 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1470 write_pnet(&p->net, hold_net(net));
1471 p->sysctl_table = NULL;
1472 write_lock_bh(&tbl->lock);
1473 p->next = tbl->parms.next;
1474 tbl->parms.next = p;
1475 write_unlock_bh(&tbl->lock);
1479 EXPORT_SYMBOL(neigh_parms_alloc);
1481 static void neigh_rcu_free_parms(struct rcu_head *head)
1483 struct neigh_parms *parms =
1484 container_of(head, struct neigh_parms, rcu_head);
1486 neigh_parms_put(parms);
1489 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1491 struct neigh_parms **p;
1493 if (!parms || parms == &tbl->parms)
1495 write_lock_bh(&tbl->lock);
1496 for (p = &tbl->parms.next; *p; p = &(*p)->next) {
1500 write_unlock_bh(&tbl->lock);
1502 dev_put(parms->dev);
1503 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1507 write_unlock_bh(&tbl->lock);
1508 NEIGH_PRINTK1("neigh_parms_release: not found\n");
1510 EXPORT_SYMBOL(neigh_parms_release);
1512 static void neigh_parms_destroy(struct neigh_parms *parms)
1514 release_net(neigh_parms_net(parms));
1518 static struct lock_class_key neigh_table_proxy_queue_class;
1520 static void neigh_table_init_no_netlink(struct neigh_table *tbl)
1522 unsigned long now = jiffies;
1523 unsigned long phsize;
1525 write_pnet(&tbl->parms.net, &init_net);
1526 atomic_set(&tbl->parms.refcnt, 1);
1527 tbl->parms.reachable_time =
1528 neigh_rand_reach_time(tbl->parms.base_reachable_time);
1530 tbl->stats = alloc_percpu(struct neigh_statistics);
1532 panic("cannot create neighbour cache statistics");
1534 #ifdef CONFIG_PROC_FS
1535 if (!proc_create_data(tbl->id, 0, init_net.proc_net_stat,
1536 &neigh_stat_seq_fops, tbl))
1537 panic("cannot create neighbour proc dir entry");
1540 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1542 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1543 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1545 if (!tbl->nht || !tbl->phash_buckets)
1546 panic("cannot allocate neighbour cache hashes");
1548 rwlock_init(&tbl->lock);
1549 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1550 schedule_delayed_work(&tbl->gc_work, tbl->parms.reachable_time);
1551 setup_timer(&tbl->proxy_timer, neigh_proxy_process, (unsigned long)tbl);
1552 skb_queue_head_init_class(&tbl->proxy_queue,
1553 &neigh_table_proxy_queue_class);
1555 tbl->last_flush = now;
1556 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1559 void neigh_table_init(struct neigh_table *tbl)
1561 struct neigh_table *tmp;
1563 neigh_table_init_no_netlink(tbl);
1564 write_lock(&neigh_tbl_lock);
1565 for (tmp = neigh_tables; tmp; tmp = tmp->next) {
1566 if (tmp->family == tbl->family)
1569 tbl->next = neigh_tables;
1571 write_unlock(&neigh_tbl_lock);
1573 if (unlikely(tmp)) {
1574 pr_err("Registering multiple tables for family %d\n",
1579 EXPORT_SYMBOL(neigh_table_init);
1581 int neigh_table_clear(struct neigh_table *tbl)
1583 struct neigh_table **tp;
1585 /* It is not clean... Fix it to unload IPv6 module safely */
1586 cancel_delayed_work_sync(&tbl->gc_work);
1587 del_timer_sync(&tbl->proxy_timer);
1588 pneigh_queue_purge(&tbl->proxy_queue);
1589 neigh_ifdown(tbl, NULL);
1590 if (atomic_read(&tbl->entries))
1591 pr_crit("neighbour leakage\n");
1592 write_lock(&neigh_tbl_lock);
1593 for (tp = &neigh_tables; *tp; tp = &(*tp)->next) {
1599 write_unlock(&neigh_tbl_lock);
1601 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1602 neigh_hash_free_rcu);
1605 kfree(tbl->phash_buckets);
1606 tbl->phash_buckets = NULL;
1608 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1610 free_percpu(tbl->stats);
1615 EXPORT_SYMBOL(neigh_table_clear);
1617 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1619 struct net *net = sock_net(skb->sk);
1621 struct nlattr *dst_attr;
1622 struct neigh_table *tbl;
1623 struct net_device *dev = NULL;
1627 if (nlmsg_len(nlh) < sizeof(*ndm))
1630 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1631 if (dst_attr == NULL)
1634 ndm = nlmsg_data(nlh);
1635 if (ndm->ndm_ifindex) {
1636 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1643 read_lock(&neigh_tbl_lock);
1644 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1645 struct neighbour *neigh;
1647 if (tbl->family != ndm->ndm_family)
1649 read_unlock(&neigh_tbl_lock);
1651 if (nla_len(dst_attr) < tbl->key_len)
1654 if (ndm->ndm_flags & NTF_PROXY) {
1655 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1662 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1663 if (neigh == NULL) {
1668 err = neigh_update(neigh, NULL, NUD_FAILED,
1669 NEIGH_UPDATE_F_OVERRIDE |
1670 NEIGH_UPDATE_F_ADMIN);
1671 neigh_release(neigh);
1674 read_unlock(&neigh_tbl_lock);
1675 err = -EAFNOSUPPORT;
1681 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1683 struct net *net = sock_net(skb->sk);
1685 struct nlattr *tb[NDA_MAX+1];
1686 struct neigh_table *tbl;
1687 struct net_device *dev = NULL;
1691 err = nlmsg_parse(nlh, sizeof(*ndm), tb, NDA_MAX, NULL);
1696 if (tb[NDA_DST] == NULL)
1699 ndm = nlmsg_data(nlh);
1700 if (ndm->ndm_ifindex) {
1701 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1707 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len)
1711 read_lock(&neigh_tbl_lock);
1712 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1713 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE;
1714 struct neighbour *neigh;
1717 if (tbl->family != ndm->ndm_family)
1719 read_unlock(&neigh_tbl_lock);
1721 if (nla_len(tb[NDA_DST]) < tbl->key_len)
1723 dst = nla_data(tb[NDA_DST]);
1724 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1726 if (ndm->ndm_flags & NTF_PROXY) {
1727 struct pneigh_entry *pn;
1730 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1732 pn->flags = ndm->ndm_flags;
1741 neigh = neigh_lookup(tbl, dst, dev);
1742 if (neigh == NULL) {
1743 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1748 neigh = __neigh_lookup_errno(tbl, dst, dev);
1749 if (IS_ERR(neigh)) {
1750 err = PTR_ERR(neigh);
1754 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1756 neigh_release(neigh);
1760 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1761 flags &= ~NEIGH_UPDATE_F_OVERRIDE;
1764 if (ndm->ndm_flags & NTF_USE) {
1765 neigh_event_send(neigh, NULL);
1768 err = neigh_update(neigh, lladdr, ndm->ndm_state, flags);
1769 neigh_release(neigh);
1773 read_unlock(&neigh_tbl_lock);
1774 err = -EAFNOSUPPORT;
1779 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1781 struct nlattr *nest;
1783 nest = nla_nest_start(skb, NDTA_PARMS);
1788 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1789 nla_put_u32(skb, NDTPA_REFCNT, atomic_read(&parms->refcnt)) ||
1790 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES, parms->queue_len_bytes) ||
1791 /* approximative value for deprecated QUEUE_LEN (in packets) */
1792 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1793 parms->queue_len_bytes / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1794 nla_put_u32(skb, NDTPA_PROXY_QLEN, parms->proxy_qlen) ||
1795 nla_put_u32(skb, NDTPA_APP_PROBES, parms->app_probes) ||
1796 nla_put_u32(skb, NDTPA_UCAST_PROBES, parms->ucast_probes) ||
1797 nla_put_u32(skb, NDTPA_MCAST_PROBES, parms->mcast_probes) ||
1798 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time) ||
1799 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
1800 parms->base_reachable_time) ||
1801 nla_put_msecs(skb, NDTPA_GC_STALETIME, parms->gc_staletime) ||
1802 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
1803 parms->delay_probe_time) ||
1804 nla_put_msecs(skb, NDTPA_RETRANS_TIME, parms->retrans_time) ||
1805 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY, parms->anycast_delay) ||
1806 nla_put_msecs(skb, NDTPA_PROXY_DELAY, parms->proxy_delay) ||
1807 nla_put_msecs(skb, NDTPA_LOCKTIME, parms->locktime))
1808 goto nla_put_failure;
1809 return nla_nest_end(skb, nest);
1812 nla_nest_cancel(skb, nest);
1816 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
1817 u32 pid, u32 seq, int type, int flags)
1819 struct nlmsghdr *nlh;
1820 struct ndtmsg *ndtmsg;
1822 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1826 ndtmsg = nlmsg_data(nlh);
1828 read_lock_bh(&tbl->lock);
1829 ndtmsg->ndtm_family = tbl->family;
1830 ndtmsg->ndtm_pad1 = 0;
1831 ndtmsg->ndtm_pad2 = 0;
1833 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
1834 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval) ||
1835 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
1836 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
1837 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
1838 goto nla_put_failure;
1840 unsigned long now = jiffies;
1841 unsigned int flush_delta = now - tbl->last_flush;
1842 unsigned int rand_delta = now - tbl->last_rand;
1843 struct neigh_hash_table *nht;
1844 struct ndt_config ndc = {
1845 .ndtc_key_len = tbl->key_len,
1846 .ndtc_entry_size = tbl->entry_size,
1847 .ndtc_entries = atomic_read(&tbl->entries),
1848 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
1849 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
1850 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
1854 nht = rcu_dereference_bh(tbl->nht);
1855 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
1856 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
1857 rcu_read_unlock_bh();
1859 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
1860 goto nla_put_failure;
1865 struct ndt_stats ndst;
1867 memset(&ndst, 0, sizeof(ndst));
1869 for_each_possible_cpu(cpu) {
1870 struct neigh_statistics *st;
1872 st = per_cpu_ptr(tbl->stats, cpu);
1873 ndst.ndts_allocs += st->allocs;
1874 ndst.ndts_destroys += st->destroys;
1875 ndst.ndts_hash_grows += st->hash_grows;
1876 ndst.ndts_res_failed += st->res_failed;
1877 ndst.ndts_lookups += st->lookups;
1878 ndst.ndts_hits += st->hits;
1879 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
1880 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
1881 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
1882 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
1885 if (nla_put(skb, NDTA_STATS, sizeof(ndst), &ndst))
1886 goto nla_put_failure;
1889 BUG_ON(tbl->parms.dev);
1890 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
1891 goto nla_put_failure;
1893 read_unlock_bh(&tbl->lock);
1894 return nlmsg_end(skb, nlh);
1897 read_unlock_bh(&tbl->lock);
1898 nlmsg_cancel(skb, nlh);
1902 static int neightbl_fill_param_info(struct sk_buff *skb,
1903 struct neigh_table *tbl,
1904 struct neigh_parms *parms,
1905 u32 pid, u32 seq, int type,
1908 struct ndtmsg *ndtmsg;
1909 struct nlmsghdr *nlh;
1911 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
1915 ndtmsg = nlmsg_data(nlh);
1917 read_lock_bh(&tbl->lock);
1918 ndtmsg->ndtm_family = tbl->family;
1919 ndtmsg->ndtm_pad1 = 0;
1920 ndtmsg->ndtm_pad2 = 0;
1922 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
1923 neightbl_fill_parms(skb, parms) < 0)
1926 read_unlock_bh(&tbl->lock);
1927 return nlmsg_end(skb, nlh);
1929 read_unlock_bh(&tbl->lock);
1930 nlmsg_cancel(skb, nlh);
1934 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
1935 [NDTA_NAME] = { .type = NLA_STRING },
1936 [NDTA_THRESH1] = { .type = NLA_U32 },
1937 [NDTA_THRESH2] = { .type = NLA_U32 },
1938 [NDTA_THRESH3] = { .type = NLA_U32 },
1939 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
1940 [NDTA_PARMS] = { .type = NLA_NESTED },
1943 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
1944 [NDTPA_IFINDEX] = { .type = NLA_U32 },
1945 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
1946 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
1947 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
1948 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
1949 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
1950 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
1951 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
1952 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
1953 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
1954 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
1955 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
1956 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
1959 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
1961 struct net *net = sock_net(skb->sk);
1962 struct neigh_table *tbl;
1963 struct ndtmsg *ndtmsg;
1964 struct nlattr *tb[NDTA_MAX+1];
1967 err = nlmsg_parse(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
1968 nl_neightbl_policy);
1972 if (tb[NDTA_NAME] == NULL) {
1977 ndtmsg = nlmsg_data(nlh);
1978 read_lock(&neigh_tbl_lock);
1979 for (tbl = neigh_tables; tbl; tbl = tbl->next) {
1980 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
1983 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0)
1993 * We acquire tbl->lock to be nice to the periodic timers and
1994 * make sure they always see a consistent set of values.
1996 write_lock_bh(&tbl->lock);
1998 if (tb[NDTA_PARMS]) {
1999 struct nlattr *tbp[NDTPA_MAX+1];
2000 struct neigh_parms *p;
2003 err = nla_parse_nested(tbp, NDTPA_MAX, tb[NDTA_PARMS],
2004 nl_ntbl_parm_policy);
2006 goto errout_tbl_lock;
2008 if (tbp[NDTPA_IFINDEX])
2009 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2011 p = lookup_neigh_parms(tbl, net, ifindex);
2014 goto errout_tbl_lock;
2017 for (i = 1; i <= NDTPA_MAX; i++) {
2022 case NDTPA_QUEUE_LEN:
2023 p->queue_len_bytes = nla_get_u32(tbp[i]) *
2024 SKB_TRUESIZE(ETH_FRAME_LEN);
2026 case NDTPA_QUEUE_LENBYTES:
2027 p->queue_len_bytes = nla_get_u32(tbp[i]);
2029 case NDTPA_PROXY_QLEN:
2030 p->proxy_qlen = nla_get_u32(tbp[i]);
2032 case NDTPA_APP_PROBES:
2033 p->app_probes = nla_get_u32(tbp[i]);
2035 case NDTPA_UCAST_PROBES:
2036 p->ucast_probes = nla_get_u32(tbp[i]);
2038 case NDTPA_MCAST_PROBES:
2039 p->mcast_probes = nla_get_u32(tbp[i]);
2041 case NDTPA_BASE_REACHABLE_TIME:
2042 p->base_reachable_time = nla_get_msecs(tbp[i]);
2044 case NDTPA_GC_STALETIME:
2045 p->gc_staletime = nla_get_msecs(tbp[i]);
2047 case NDTPA_DELAY_PROBE_TIME:
2048 p->delay_probe_time = nla_get_msecs(tbp[i]);
2050 case NDTPA_RETRANS_TIME:
2051 p->retrans_time = nla_get_msecs(tbp[i]);
2053 case NDTPA_ANYCAST_DELAY:
2054 p->anycast_delay = nla_get_msecs(tbp[i]);
2056 case NDTPA_PROXY_DELAY:
2057 p->proxy_delay = nla_get_msecs(tbp[i]);
2059 case NDTPA_LOCKTIME:
2060 p->locktime = nla_get_msecs(tbp[i]);
2066 if (tb[NDTA_THRESH1])
2067 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2069 if (tb[NDTA_THRESH2])
2070 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2072 if (tb[NDTA_THRESH3])
2073 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2075 if (tb[NDTA_GC_INTERVAL])
2076 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2081 write_unlock_bh(&tbl->lock);
2083 read_unlock(&neigh_tbl_lock);
2088 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2090 struct net *net = sock_net(skb->sk);
2091 int family, tidx, nidx = 0;
2092 int tbl_skip = cb->args[0];
2093 int neigh_skip = cb->args[1];
2094 struct neigh_table *tbl;
2096 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2098 read_lock(&neigh_tbl_lock);
2099 for (tbl = neigh_tables, tidx = 0; tbl; tbl = tbl->next, tidx++) {
2100 struct neigh_parms *p;
2102 if (tidx < tbl_skip || (family && tbl->family != family))
2105 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2106 cb->nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2110 for (nidx = 0, p = tbl->parms.next; p; p = p->next) {
2111 if (!net_eq(neigh_parms_net(p), net))
2114 if (nidx < neigh_skip)
2117 if (neightbl_fill_param_info(skb, tbl, p,
2118 NETLINK_CB(cb->skb).portid,
2130 read_unlock(&neigh_tbl_lock);
2137 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2138 u32 pid, u32 seq, int type, unsigned int flags)
2140 unsigned long now = jiffies;
2141 struct nda_cacheinfo ci;
2142 struct nlmsghdr *nlh;
2145 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2149 ndm = nlmsg_data(nlh);
2150 ndm->ndm_family = neigh->ops->family;
2153 ndm->ndm_flags = neigh->flags;
2154 ndm->ndm_type = neigh->type;
2155 ndm->ndm_ifindex = neigh->dev->ifindex;
2157 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2158 goto nla_put_failure;
2160 read_lock_bh(&neigh->lock);
2161 ndm->ndm_state = neigh->nud_state;
2162 if (neigh->nud_state & NUD_VALID) {
2163 char haddr[MAX_ADDR_LEN];
2165 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2166 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2167 read_unlock_bh(&neigh->lock);
2168 goto nla_put_failure;
2172 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2173 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2174 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2175 ci.ndm_refcnt = atomic_read(&neigh->refcnt) - 1;
2176 read_unlock_bh(&neigh->lock);
2178 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2179 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2180 goto nla_put_failure;
2182 return nlmsg_end(skb, nlh);
2185 nlmsg_cancel(skb, nlh);
2189 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2190 u32 pid, u32 seq, int type, unsigned int flags,
2191 struct neigh_table *tbl)
2193 struct nlmsghdr *nlh;
2196 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2200 ndm = nlmsg_data(nlh);
2201 ndm->ndm_family = tbl->family;
2204 ndm->ndm_flags = pn->flags | NTF_PROXY;
2205 ndm->ndm_type = NDA_DST;
2206 ndm->ndm_ifindex = pn->dev->ifindex;
2207 ndm->ndm_state = NUD_NONE;
2209 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2210 goto nla_put_failure;
2212 return nlmsg_end(skb, nlh);
2215 nlmsg_cancel(skb, nlh);
2219 static void neigh_update_notify(struct neighbour *neigh)
2221 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2222 __neigh_notify(neigh, RTM_NEWNEIGH, 0);
2225 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2226 struct netlink_callback *cb)
2228 struct net *net = sock_net(skb->sk);
2229 struct neighbour *n;
2230 int rc, h, s_h = cb->args[1];
2231 int idx, s_idx = idx = cb->args[2];
2232 struct neigh_hash_table *nht;
2235 nht = rcu_dereference_bh(tbl->nht);
2237 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2240 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2242 n = rcu_dereference_bh(n->next)) {
2243 if (!net_eq(dev_net(n->dev), net))
2247 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2250 NLM_F_MULTI) <= 0) {
2260 rcu_read_unlock_bh();
2266 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2267 struct netlink_callback *cb)
2269 struct pneigh_entry *n;
2270 struct net *net = sock_net(skb->sk);
2271 int rc, h, s_h = cb->args[3];
2272 int idx, s_idx = idx = cb->args[4];
2274 read_lock_bh(&tbl->lock);
2276 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2279 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2280 if (dev_net(n->dev) != net)
2284 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2287 NLM_F_MULTI, tbl) <= 0) {
2288 read_unlock_bh(&tbl->lock);
2297 read_unlock_bh(&tbl->lock);
2306 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2308 struct neigh_table *tbl;
2313 read_lock(&neigh_tbl_lock);
2314 family = ((struct rtgenmsg *) nlmsg_data(cb->nlh))->rtgen_family;
2316 /* check for full ndmsg structure presence, family member is
2317 * the same for both structures
2319 if (nlmsg_len(cb->nlh) >= sizeof(struct ndmsg) &&
2320 ((struct ndmsg *) nlmsg_data(cb->nlh))->ndm_flags == NTF_PROXY)
2325 for (tbl = neigh_tables, t = 0; tbl;
2326 tbl = tbl->next, t++) {
2327 if (t < s_t || (family && tbl->family != family))
2330 memset(&cb->args[1], 0, sizeof(cb->args) -
2331 sizeof(cb->args[0]));
2333 err = pneigh_dump_table(tbl, skb, cb);
2335 err = neigh_dump_table(tbl, skb, cb);
2339 read_unlock(&neigh_tbl_lock);
2345 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2348 struct neigh_hash_table *nht;
2351 nht = rcu_dereference_bh(tbl->nht);
2353 read_lock(&tbl->lock); /* avoid resizes */
2354 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2355 struct neighbour *n;
2357 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2359 n = rcu_dereference_bh(n->next))
2362 read_unlock(&tbl->lock);
2363 rcu_read_unlock_bh();
2365 EXPORT_SYMBOL(neigh_for_each);
2367 /* The tbl->lock must be held as a writer and BH disabled. */
2368 void __neigh_for_each_release(struct neigh_table *tbl,
2369 int (*cb)(struct neighbour *))
2372 struct neigh_hash_table *nht;
2374 nht = rcu_dereference_protected(tbl->nht,
2375 lockdep_is_held(&tbl->lock));
2376 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2377 struct neighbour *n;
2378 struct neighbour __rcu **np;
2380 np = &nht->hash_buckets[chain];
2381 while ((n = rcu_dereference_protected(*np,
2382 lockdep_is_held(&tbl->lock))) != NULL) {
2385 write_lock(&n->lock);
2388 rcu_assign_pointer(*np,
2389 rcu_dereference_protected(n->next,
2390 lockdep_is_held(&tbl->lock)));
2394 write_unlock(&n->lock);
2396 neigh_cleanup_and_release(n);
2400 EXPORT_SYMBOL(__neigh_for_each_release);
2402 #ifdef CONFIG_PROC_FS
2404 static struct neighbour *neigh_get_first(struct seq_file *seq)
2406 struct neigh_seq_state *state = seq->private;
2407 struct net *net = seq_file_net(seq);
2408 struct neigh_hash_table *nht = state->nht;
2409 struct neighbour *n = NULL;
2410 int bucket = state->bucket;
2412 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
2413 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
2414 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
2417 if (!net_eq(dev_net(n->dev), net))
2419 if (state->neigh_sub_iter) {
2423 v = state->neigh_sub_iter(state, n, &fakep);
2427 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2429 if (n->nud_state & ~NUD_NOARP)
2432 n = rcu_dereference_bh(n->next);
2438 state->bucket = bucket;
2443 static struct neighbour *neigh_get_next(struct seq_file *seq,
2444 struct neighbour *n,
2447 struct neigh_seq_state *state = seq->private;
2448 struct net *net = seq_file_net(seq);
2449 struct neigh_hash_table *nht = state->nht;
2451 if (state->neigh_sub_iter) {
2452 void *v = state->neigh_sub_iter(state, n, pos);
2456 n = rcu_dereference_bh(n->next);
2460 if (!net_eq(dev_net(n->dev), net))
2462 if (state->neigh_sub_iter) {
2463 void *v = state->neigh_sub_iter(state, n, pos);
2468 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
2471 if (n->nud_state & ~NUD_NOARP)
2474 n = rcu_dereference_bh(n->next);
2480 if (++state->bucket >= (1 << nht->hash_shift))
2483 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
2491 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
2493 struct neighbour *n = neigh_get_first(seq);
2498 n = neigh_get_next(seq, n, pos);
2503 return *pos ? NULL : n;
2506 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
2508 struct neigh_seq_state *state = seq->private;
2509 struct net *net = seq_file_net(seq);
2510 struct neigh_table *tbl = state->tbl;
2511 struct pneigh_entry *pn = NULL;
2512 int bucket = state->bucket;
2514 state->flags |= NEIGH_SEQ_IS_PNEIGH;
2515 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
2516 pn = tbl->phash_buckets[bucket];
2517 while (pn && !net_eq(pneigh_net(pn), net))
2522 state->bucket = bucket;
2527 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
2528 struct pneigh_entry *pn,
2531 struct neigh_seq_state *state = seq->private;
2532 struct net *net = seq_file_net(seq);
2533 struct neigh_table *tbl = state->tbl;
2537 } while (pn && !net_eq(pneigh_net(pn), net));
2540 if (++state->bucket > PNEIGH_HASHMASK)
2542 pn = tbl->phash_buckets[state->bucket];
2543 while (pn && !net_eq(pneigh_net(pn), net))
2555 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
2557 struct pneigh_entry *pn = pneigh_get_first(seq);
2562 pn = pneigh_get_next(seq, pn, pos);
2567 return *pos ? NULL : pn;
2570 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
2572 struct neigh_seq_state *state = seq->private;
2574 loff_t idxpos = *pos;
2576 rc = neigh_get_idx(seq, &idxpos);
2577 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2578 rc = pneigh_get_idx(seq, &idxpos);
2583 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
2586 struct neigh_seq_state *state = seq->private;
2590 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
2593 state->nht = rcu_dereference_bh(tbl->nht);
2595 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
2597 EXPORT_SYMBOL(neigh_seq_start);
2599 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2601 struct neigh_seq_state *state;
2604 if (v == SEQ_START_TOKEN) {
2605 rc = neigh_get_first(seq);
2609 state = seq->private;
2610 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
2611 rc = neigh_get_next(seq, v, NULL);
2614 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
2615 rc = pneigh_get_first(seq);
2617 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
2618 rc = pneigh_get_next(seq, v, NULL);
2624 EXPORT_SYMBOL(neigh_seq_next);
2626 void neigh_seq_stop(struct seq_file *seq, void *v)
2629 rcu_read_unlock_bh();
2631 EXPORT_SYMBOL(neigh_seq_stop);
2633 /* statistics via seq_file */
2635 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
2637 struct neigh_table *tbl = seq->private;
2641 return SEQ_START_TOKEN;
2643 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
2644 if (!cpu_possible(cpu))
2647 return per_cpu_ptr(tbl->stats, cpu);
2652 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2654 struct neigh_table *tbl = seq->private;
2657 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
2658 if (!cpu_possible(cpu))
2661 return per_cpu_ptr(tbl->stats, cpu);
2666 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
2671 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
2673 struct neigh_table *tbl = seq->private;
2674 struct neigh_statistics *st = v;
2676 if (v == SEQ_START_TOKEN) {
2677 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");
2681 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
2682 "%08lx %08lx %08lx %08lx %08lx\n",
2683 atomic_read(&tbl->entries),
2694 st->rcv_probes_mcast,
2695 st->rcv_probes_ucast,
2697 st->periodic_gc_runs,
2705 static const struct seq_operations neigh_stat_seq_ops = {
2706 .start = neigh_stat_seq_start,
2707 .next = neigh_stat_seq_next,
2708 .stop = neigh_stat_seq_stop,
2709 .show = neigh_stat_seq_show,
2712 static int neigh_stat_seq_open(struct inode *inode, struct file *file)
2714 int ret = seq_open(file, &neigh_stat_seq_ops);
2717 struct seq_file *sf = file->private_data;
2718 sf->private = PDE(inode)->data;
2723 static const struct file_operations neigh_stat_seq_fops = {
2724 .owner = THIS_MODULE,
2725 .open = neigh_stat_seq_open,
2727 .llseek = seq_lseek,
2728 .release = seq_release,
2731 #endif /* CONFIG_PROC_FS */
2733 static inline size_t neigh_nlmsg_size(void)
2735 return NLMSG_ALIGN(sizeof(struct ndmsg))
2736 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2737 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2738 + nla_total_size(sizeof(struct nda_cacheinfo))
2739 + nla_total_size(4); /* NDA_PROBES */
2742 static void __neigh_notify(struct neighbour *n, int type, int flags)
2744 struct net *net = dev_net(n->dev);
2745 struct sk_buff *skb;
2748 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
2752 err = neigh_fill_info(skb, n, 0, 0, type, flags);
2754 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
2755 WARN_ON(err == -EMSGSIZE);
2759 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
2763 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
2767 void neigh_app_ns(struct neighbour *n)
2769 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST);
2771 EXPORT_SYMBOL(neigh_app_ns);
2772 #endif /* CONFIG_ARPD */
2774 #ifdef CONFIG_SYSCTL
2776 static int proc_unres_qlen(ctl_table *ctl, int write, void __user *buffer,
2777 size_t *lenp, loff_t *ppos)
2780 ctl_table tmp = *ctl;
2783 tmp.extra2 = &unres_qlen_max;
2786 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
2787 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
2790 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
2795 NEIGH_VAR_MCAST_PROBE,
2796 NEIGH_VAR_UCAST_PROBE,
2797 NEIGH_VAR_APP_PROBE,
2798 NEIGH_VAR_RETRANS_TIME,
2799 NEIGH_VAR_BASE_REACHABLE_TIME,
2800 NEIGH_VAR_DELAY_PROBE_TIME,
2801 NEIGH_VAR_GC_STALETIME,
2802 NEIGH_VAR_QUEUE_LEN,
2803 NEIGH_VAR_QUEUE_LEN_BYTES,
2804 NEIGH_VAR_PROXY_QLEN,
2805 NEIGH_VAR_ANYCAST_DELAY,
2806 NEIGH_VAR_PROXY_DELAY,
2808 NEIGH_VAR_RETRANS_TIME_MS,
2809 NEIGH_VAR_BASE_REACHABLE_TIME_MS,
2810 NEIGH_VAR_GC_INTERVAL,
2811 NEIGH_VAR_GC_THRESH1,
2812 NEIGH_VAR_GC_THRESH2,
2813 NEIGH_VAR_GC_THRESH3,
2817 static struct neigh_sysctl_table {
2818 struct ctl_table_header *sysctl_header;
2819 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
2820 } neigh_sysctl_template __read_mostly = {
2822 [NEIGH_VAR_MCAST_PROBE] = {
2823 .procname = "mcast_solicit",
2824 .maxlen = sizeof(int),
2826 .proc_handler = proc_dointvec,
2828 [NEIGH_VAR_UCAST_PROBE] = {
2829 .procname = "ucast_solicit",
2830 .maxlen = sizeof(int),
2832 .proc_handler = proc_dointvec,
2834 [NEIGH_VAR_APP_PROBE] = {
2835 .procname = "app_solicit",
2836 .maxlen = sizeof(int),
2838 .proc_handler = proc_dointvec,
2840 [NEIGH_VAR_RETRANS_TIME] = {
2841 .procname = "retrans_time",
2842 .maxlen = sizeof(int),
2844 .proc_handler = proc_dointvec_userhz_jiffies,
2846 [NEIGH_VAR_BASE_REACHABLE_TIME] = {
2847 .procname = "base_reachable_time",
2848 .maxlen = sizeof(int),
2850 .proc_handler = proc_dointvec_jiffies,
2852 [NEIGH_VAR_DELAY_PROBE_TIME] = {
2853 .procname = "delay_first_probe_time",
2854 .maxlen = sizeof(int),
2856 .proc_handler = proc_dointvec_jiffies,
2858 [NEIGH_VAR_GC_STALETIME] = {
2859 .procname = "gc_stale_time",
2860 .maxlen = sizeof(int),
2862 .proc_handler = proc_dointvec_jiffies,
2864 [NEIGH_VAR_QUEUE_LEN] = {
2865 .procname = "unres_qlen",
2866 .maxlen = sizeof(int),
2868 .proc_handler = proc_unres_qlen,
2870 [NEIGH_VAR_QUEUE_LEN_BYTES] = {
2871 .procname = "unres_qlen_bytes",
2872 .maxlen = sizeof(int),
2875 .proc_handler = proc_dointvec_minmax,
2877 [NEIGH_VAR_PROXY_QLEN] = {
2878 .procname = "proxy_qlen",
2879 .maxlen = sizeof(int),
2881 .proc_handler = proc_dointvec,
2883 [NEIGH_VAR_ANYCAST_DELAY] = {
2884 .procname = "anycast_delay",
2885 .maxlen = sizeof(int),
2887 .proc_handler = proc_dointvec_userhz_jiffies,
2889 [NEIGH_VAR_PROXY_DELAY] = {
2890 .procname = "proxy_delay",
2891 .maxlen = sizeof(int),
2893 .proc_handler = proc_dointvec_userhz_jiffies,
2895 [NEIGH_VAR_LOCKTIME] = {
2896 .procname = "locktime",
2897 .maxlen = sizeof(int),
2899 .proc_handler = proc_dointvec_userhz_jiffies,
2901 [NEIGH_VAR_RETRANS_TIME_MS] = {
2902 .procname = "retrans_time_ms",
2903 .maxlen = sizeof(int),
2905 .proc_handler = proc_dointvec_ms_jiffies,
2907 [NEIGH_VAR_BASE_REACHABLE_TIME_MS] = {
2908 .procname = "base_reachable_time_ms",
2909 .maxlen = sizeof(int),
2911 .proc_handler = proc_dointvec_ms_jiffies,
2913 [NEIGH_VAR_GC_INTERVAL] = {
2914 .procname = "gc_interval",
2915 .maxlen = sizeof(int),
2917 .proc_handler = proc_dointvec_jiffies,
2919 [NEIGH_VAR_GC_THRESH1] = {
2920 .procname = "gc_thresh1",
2921 .maxlen = sizeof(int),
2923 .proc_handler = proc_dointvec,
2925 [NEIGH_VAR_GC_THRESH2] = {
2926 .procname = "gc_thresh2",
2927 .maxlen = sizeof(int),
2929 .proc_handler = proc_dointvec,
2931 [NEIGH_VAR_GC_THRESH3] = {
2932 .procname = "gc_thresh3",
2933 .maxlen = sizeof(int),
2935 .proc_handler = proc_dointvec,
2941 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
2942 char *p_name, proc_handler *handler)
2944 struct neigh_sysctl_table *t;
2945 const char *dev_name_source = NULL;
2946 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
2948 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
2952 t->neigh_vars[NEIGH_VAR_MCAST_PROBE].data = &p->mcast_probes;
2953 t->neigh_vars[NEIGH_VAR_UCAST_PROBE].data = &p->ucast_probes;
2954 t->neigh_vars[NEIGH_VAR_APP_PROBE].data = &p->app_probes;
2955 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].data = &p->retrans_time;
2956 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].data = &p->base_reachable_time;
2957 t->neigh_vars[NEIGH_VAR_DELAY_PROBE_TIME].data = &p->delay_probe_time;
2958 t->neigh_vars[NEIGH_VAR_GC_STALETIME].data = &p->gc_staletime;
2959 t->neigh_vars[NEIGH_VAR_QUEUE_LEN].data = &p->queue_len_bytes;
2960 t->neigh_vars[NEIGH_VAR_QUEUE_LEN_BYTES].data = &p->queue_len_bytes;
2961 t->neigh_vars[NEIGH_VAR_PROXY_QLEN].data = &p->proxy_qlen;
2962 t->neigh_vars[NEIGH_VAR_ANYCAST_DELAY].data = &p->anycast_delay;
2963 t->neigh_vars[NEIGH_VAR_PROXY_DELAY].data = &p->proxy_delay;
2964 t->neigh_vars[NEIGH_VAR_LOCKTIME].data = &p->locktime;
2965 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].data = &p->retrans_time;
2966 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].data = &p->base_reachable_time;
2969 dev_name_source = dev->name;
2970 /* Terminate the table early */
2971 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
2972 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
2974 dev_name_source = "default";
2975 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = (int *)(p + 1);
2976 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = (int *)(p + 1) + 1;
2977 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = (int *)(p + 1) + 2;
2978 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = (int *)(p + 1) + 3;
2984 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
2985 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].extra1 = dev;
2987 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
2988 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].extra1 = dev;
2989 /* RetransTime (in milliseconds)*/
2990 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
2991 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].extra1 = dev;
2992 /* ReachableTime (in milliseconds) */
2993 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
2994 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].extra1 = dev;
2997 /* Don't export sysctls to unprivileged users */
2998 if (neigh_parms_net(p)->user_ns != &init_user_ns)
2999 t->neigh_vars[0].procname = NULL;
3001 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3002 p_name, dev_name_source);
3004 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3005 if (!t->sysctl_header)
3008 p->sysctl_table = t;
3016 EXPORT_SYMBOL(neigh_sysctl_register);
3018 void neigh_sysctl_unregister(struct neigh_parms *p)
3020 if (p->sysctl_table) {
3021 struct neigh_sysctl_table *t = p->sysctl_table;
3022 p->sysctl_table = NULL;
3023 unregister_net_sysctl_table(t->sysctl_header);
3027 EXPORT_SYMBOL(neigh_sysctl_unregister);
3029 #endif /* CONFIG_SYSCTL */
3031 static int __init neigh_init(void)
3033 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, NULL);
3034 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, NULL);
3035 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, NULL, neigh_dump_info, NULL);
3037 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3039 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, NULL);
3044 subsys_initcall(neigh_init);