1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Generic address resolution entity
6 * Pedro Roque <roque@di.fc.ul.pt>
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
10 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
11 * Harald Welte Add neighbour cache statistics like rtstat
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 #include <linux/slab.h>
17 #include <linux/kmemleak.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/socket.h>
22 #include <linux/netdevice.h>
23 #include <linux/proc_fs.h>
25 #include <linux/sysctl.h>
27 #include <linux/times.h>
28 #include <net/net_namespace.h>
29 #include <net/neighbour.h>
33 #include <net/netevent.h>
34 #include <net/netlink.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/random.h>
37 #include <linux/string.h>
38 #include <linux/log2.h>
39 #include <linux/inetdevice.h>
40 #include <net/addrconf.h>
42 #include <trace/events/neigh.h>
45 #define neigh_dbg(level, fmt, ...) \
47 if (level <= NEIGH_DEBUG) \
48 pr_debug(fmt, ##__VA_ARGS__); \
51 #define PNEIGH_HASHMASK 0xF
53 static void neigh_timer_handler(struct timer_list *t);
54 static void __neigh_notify(struct neighbour *n, int type, int flags,
56 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
57 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
58 struct net_device *dev);
61 static const struct seq_operations neigh_stat_seq_ops;
65 Neighbour hash table buckets are protected with rwlock tbl->lock.
67 - All the scans/updates to hash buckets MUST be made under this lock.
68 - NOTHING clever should be made under this lock: no callbacks
69 to protocol backends, no attempts to send something to network.
70 It will result in deadlocks, if backend/driver wants to use neighbour
72 - If the entry requires some non-trivial actions, increase
73 its reference count and release table lock.
75 Neighbour entries are protected:
76 - with reference count.
77 - with rwlock neigh->lock
79 Reference count prevents destruction.
81 neigh->lock mainly serializes ll address data and its validity state.
82 However, the same lock is used to protect another entry fields:
86 Again, nothing clever shall be made under neigh->lock,
87 the most complicated procedure, which we allow is dev->hard_header.
88 It is supposed, that dev->hard_header is simplistic and does
89 not make callbacks to neighbour tables.
92 static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
98 static void neigh_cleanup_and_release(struct neighbour *neigh)
100 trace_neigh_cleanup_and_release(neigh, 0);
101 __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
102 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
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);
118 static void neigh_mark_dead(struct neighbour *n)
121 if (!list_empty(&n->gc_list)) {
122 list_del_init(&n->gc_list);
123 atomic_dec(&n->tbl->gc_entries);
127 static void neigh_update_gc_list(struct neighbour *n)
129 bool on_gc_list, exempt_from_gc;
131 write_lock_bh(&n->tbl->lock);
132 write_lock(&n->lock);
137 /* remove from the gc list if new state is permanent or if neighbor
138 * is externally learned; otherwise entry should be on the gc list
140 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
141 n->flags & NTF_EXT_LEARNED;
142 on_gc_list = !list_empty(&n->gc_list);
144 if (exempt_from_gc && on_gc_list) {
145 list_del_init(&n->gc_list);
146 atomic_dec(&n->tbl->gc_entries);
147 } else if (!exempt_from_gc && !on_gc_list) {
148 /* add entries to the tail; cleaning removes from the front */
149 list_add_tail(&n->gc_list, &n->tbl->gc_list);
150 atomic_inc(&n->tbl->gc_entries);
154 write_unlock(&n->lock);
155 write_unlock_bh(&n->tbl->lock);
158 static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
164 if (!(flags & NEIGH_UPDATE_F_ADMIN))
167 ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
168 if ((neigh->flags ^ ndm_flags) & NTF_EXT_LEARNED) {
169 if (ndm_flags & NTF_EXT_LEARNED)
170 neigh->flags |= NTF_EXT_LEARNED;
172 neigh->flags &= ~NTF_EXT_LEARNED;
180 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
181 struct neigh_table *tbl)
185 write_lock(&n->lock);
186 if (refcount_read(&n->refcnt) == 1) {
187 struct neighbour *neigh;
189 neigh = rcu_dereference_protected(n->next,
190 lockdep_is_held(&tbl->lock));
191 rcu_assign_pointer(*np, neigh);
195 write_unlock(&n->lock);
197 neigh_cleanup_and_release(n);
201 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
203 struct neigh_hash_table *nht;
204 void *pkey = ndel->primary_key;
207 struct neighbour __rcu **np;
209 nht = rcu_dereference_protected(tbl->nht,
210 lockdep_is_held(&tbl->lock));
211 hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
212 hash_val = hash_val >> (32 - nht->hash_shift);
214 np = &nht->hash_buckets[hash_val];
215 while ((n = rcu_dereference_protected(*np,
216 lockdep_is_held(&tbl->lock)))) {
218 return neigh_del(n, np, tbl);
224 static int neigh_forced_gc(struct neigh_table *tbl)
226 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
227 unsigned long tref = jiffies - 5 * HZ;
228 struct neighbour *n, *tmp;
231 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
233 write_lock_bh(&tbl->lock);
235 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
236 if (refcount_read(&n->refcnt) == 1) {
239 write_lock(&n->lock);
240 if ((n->nud_state == NUD_FAILED) ||
241 (n->nud_state == NUD_NOARP) ||
242 (tbl->is_multicast &&
243 tbl->is_multicast(n->primary_key)) ||
244 time_after(tref, n->updated))
246 write_unlock(&n->lock);
248 if (remove && neigh_remove_one(n, tbl))
250 if (shrunk >= max_clean)
255 tbl->last_flush = jiffies;
257 write_unlock_bh(&tbl->lock);
262 static void neigh_add_timer(struct neighbour *n, unsigned long when)
265 if (unlikely(mod_timer(&n->timer, when))) {
266 printk("NEIGH: BUG, double timer add, state is %x\n",
272 static int neigh_del_timer(struct neighbour *n)
274 if ((n->nud_state & NUD_IN_TIMER) &&
275 del_timer(&n->timer)) {
282 static void pneigh_queue_purge(struct sk_buff_head *list)
286 while ((skb = skb_dequeue(list)) != NULL) {
292 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
296 struct neigh_hash_table *nht;
298 nht = rcu_dereference_protected(tbl->nht,
299 lockdep_is_held(&tbl->lock));
301 for (i = 0; i < (1 << nht->hash_shift); i++) {
303 struct neighbour __rcu **np = &nht->hash_buckets[i];
305 while ((n = rcu_dereference_protected(*np,
306 lockdep_is_held(&tbl->lock))) != NULL) {
307 if (dev && n->dev != dev) {
311 if (skip_perm && n->nud_state & NUD_PERMANENT) {
315 rcu_assign_pointer(*np,
316 rcu_dereference_protected(n->next,
317 lockdep_is_held(&tbl->lock)));
318 write_lock(&n->lock);
321 if (refcount_read(&n->refcnt) != 1) {
322 /* The most unpleasant situation.
323 We must destroy neighbour entry,
324 but someone still uses it.
326 The destroy will be delayed until
327 the last user releases us, but
328 we must kill timers etc. and move
331 __skb_queue_purge(&n->arp_queue);
332 n->arp_queue_len_bytes = 0;
333 n->output = neigh_blackhole;
334 if (n->nud_state & NUD_VALID)
335 n->nud_state = NUD_NOARP;
337 n->nud_state = NUD_NONE;
338 neigh_dbg(2, "neigh %p is stray\n", n);
340 write_unlock(&n->lock);
341 neigh_cleanup_and_release(n);
346 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
348 write_lock_bh(&tbl->lock);
349 neigh_flush_dev(tbl, dev, false);
350 write_unlock_bh(&tbl->lock);
352 EXPORT_SYMBOL(neigh_changeaddr);
354 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
357 write_lock_bh(&tbl->lock);
358 neigh_flush_dev(tbl, dev, skip_perm);
359 pneigh_ifdown_and_unlock(tbl, dev);
361 del_timer_sync(&tbl->proxy_timer);
362 pneigh_queue_purge(&tbl->proxy_queue);
366 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
368 __neigh_ifdown(tbl, dev, true);
371 EXPORT_SYMBOL(neigh_carrier_down);
373 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
375 __neigh_ifdown(tbl, dev, false);
378 EXPORT_SYMBOL(neigh_ifdown);
380 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
381 struct net_device *dev,
384 struct neighbour *n = NULL;
385 unsigned long now = jiffies;
391 entries = atomic_inc_return(&tbl->gc_entries) - 1;
392 if (entries >= tbl->gc_thresh3 ||
393 (entries >= tbl->gc_thresh2 &&
394 time_after(now, tbl->last_flush + 5 * HZ))) {
395 if (!neigh_forced_gc(tbl) &&
396 entries >= tbl->gc_thresh3) {
397 net_info_ratelimited("%s: neighbor table overflow!\n",
399 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
405 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
409 __skb_queue_head_init(&n->arp_queue);
410 rwlock_init(&n->lock);
411 seqlock_init(&n->ha_lock);
412 n->updated = n->used = now;
413 n->nud_state = NUD_NONE;
414 n->output = neigh_blackhole;
415 seqlock_init(&n->hh.hh_lock);
416 n->parms = neigh_parms_clone(&tbl->parms);
417 timer_setup(&n->timer, neigh_timer_handler, 0);
419 NEIGH_CACHE_STAT_INC(tbl, allocs);
421 refcount_set(&n->refcnt, 1);
423 INIT_LIST_HEAD(&n->gc_list);
425 atomic_inc(&tbl->entries);
431 atomic_dec(&tbl->gc_entries);
435 static void neigh_get_hash_rnd(u32 *x)
437 *x = get_random_u32() | 1;
440 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
442 size_t size = (1 << shift) * sizeof(struct neighbour *);
443 struct neigh_hash_table *ret;
444 struct neighbour __rcu **buckets;
447 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
450 if (size <= PAGE_SIZE) {
451 buckets = kzalloc(size, GFP_ATOMIC);
453 buckets = (struct neighbour __rcu **)
454 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
456 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
462 ret->hash_buckets = buckets;
463 ret->hash_shift = shift;
464 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
465 neigh_get_hash_rnd(&ret->hash_rnd[i]);
469 static void neigh_hash_free_rcu(struct rcu_head *head)
471 struct neigh_hash_table *nht = container_of(head,
472 struct neigh_hash_table,
474 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
475 struct neighbour __rcu **buckets = nht->hash_buckets;
477 if (size <= PAGE_SIZE) {
480 kmemleak_free(buckets);
481 free_pages((unsigned long)buckets, get_order(size));
486 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
487 unsigned long new_shift)
489 unsigned int i, hash;
490 struct neigh_hash_table *new_nht, *old_nht;
492 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
494 old_nht = rcu_dereference_protected(tbl->nht,
495 lockdep_is_held(&tbl->lock));
496 new_nht = neigh_hash_alloc(new_shift);
500 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
501 struct neighbour *n, *next;
503 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
504 lockdep_is_held(&tbl->lock));
507 hash = tbl->hash(n->primary_key, n->dev,
510 hash >>= (32 - new_nht->hash_shift);
511 next = rcu_dereference_protected(n->next,
512 lockdep_is_held(&tbl->lock));
514 rcu_assign_pointer(n->next,
515 rcu_dereference_protected(
516 new_nht->hash_buckets[hash],
517 lockdep_is_held(&tbl->lock)));
518 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
522 rcu_assign_pointer(tbl->nht, new_nht);
523 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
527 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
528 struct net_device *dev)
532 NEIGH_CACHE_STAT_INC(tbl, lookups);
535 n = __neigh_lookup_noref(tbl, pkey, dev);
537 if (!refcount_inc_not_zero(&n->refcnt))
539 NEIGH_CACHE_STAT_INC(tbl, hits);
542 rcu_read_unlock_bh();
545 EXPORT_SYMBOL(neigh_lookup);
547 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
551 unsigned int key_len = tbl->key_len;
553 struct neigh_hash_table *nht;
555 NEIGH_CACHE_STAT_INC(tbl, lookups);
558 nht = rcu_dereference_bh(tbl->nht);
559 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
561 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
563 n = rcu_dereference_bh(n->next)) {
564 if (!memcmp(n->primary_key, pkey, key_len) &&
565 net_eq(dev_net(n->dev), net)) {
566 if (!refcount_inc_not_zero(&n->refcnt))
568 NEIGH_CACHE_STAT_INC(tbl, hits);
573 rcu_read_unlock_bh();
576 EXPORT_SYMBOL(neigh_lookup_nodev);
578 static struct neighbour *___neigh_create(struct neigh_table *tbl,
580 struct net_device *dev,
581 bool exempt_from_gc, bool want_ref)
583 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
585 unsigned int key_len = tbl->key_len;
587 struct neigh_hash_table *nht;
589 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
592 rc = ERR_PTR(-ENOBUFS);
596 memcpy(n->primary_key, pkey, key_len);
600 /* Protocol specific setup. */
601 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
603 goto out_neigh_release;
606 if (dev->netdev_ops->ndo_neigh_construct) {
607 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
610 goto out_neigh_release;
614 /* Device specific setup. */
615 if (n->parms->neigh_setup &&
616 (error = n->parms->neigh_setup(n)) < 0) {
618 goto out_neigh_release;
621 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
623 write_lock_bh(&tbl->lock);
624 nht = rcu_dereference_protected(tbl->nht,
625 lockdep_is_held(&tbl->lock));
627 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
628 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
630 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
632 if (n->parms->dead) {
633 rc = ERR_PTR(-EINVAL);
637 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
638 lockdep_is_held(&tbl->lock));
640 n1 = rcu_dereference_protected(n1->next,
641 lockdep_is_held(&tbl->lock))) {
642 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
652 list_add_tail(&n->gc_list, &n->tbl->gc_list);
656 rcu_assign_pointer(n->next,
657 rcu_dereference_protected(nht->hash_buckets[hash_val],
658 lockdep_is_held(&tbl->lock)));
659 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
660 write_unlock_bh(&tbl->lock);
661 neigh_dbg(2, "neigh %p is created\n", n);
666 write_unlock_bh(&tbl->lock);
669 atomic_dec(&tbl->gc_entries);
674 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
675 struct net_device *dev, bool want_ref)
677 return ___neigh_create(tbl, pkey, dev, false, want_ref);
679 EXPORT_SYMBOL(__neigh_create);
681 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
683 u32 hash_val = *(u32 *)(pkey + key_len - 4);
684 hash_val ^= (hash_val >> 16);
685 hash_val ^= hash_val >> 8;
686 hash_val ^= hash_val >> 4;
687 hash_val &= PNEIGH_HASHMASK;
691 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
694 unsigned int key_len,
695 struct net_device *dev)
698 if (!memcmp(n->key, pkey, key_len) &&
699 net_eq(pneigh_net(n), net) &&
700 (n->dev == dev || !n->dev))
707 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
708 struct net *net, const void *pkey, struct net_device *dev)
710 unsigned int key_len = tbl->key_len;
711 u32 hash_val = pneigh_hash(pkey, key_len);
713 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
714 net, pkey, key_len, dev);
716 EXPORT_SYMBOL_GPL(__pneigh_lookup);
718 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
719 struct net *net, const void *pkey,
720 struct net_device *dev, int creat)
722 struct pneigh_entry *n;
723 unsigned int key_len = tbl->key_len;
724 u32 hash_val = pneigh_hash(pkey, key_len);
726 read_lock_bh(&tbl->lock);
727 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
728 net, pkey, key_len, dev);
729 read_unlock_bh(&tbl->lock);
736 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
741 write_pnet(&n->net, net);
742 memcpy(n->key, pkey, key_len);
746 if (tbl->pconstructor && tbl->pconstructor(n)) {
753 write_lock_bh(&tbl->lock);
754 n->next = tbl->phash_buckets[hash_val];
755 tbl->phash_buckets[hash_val] = n;
756 write_unlock_bh(&tbl->lock);
760 EXPORT_SYMBOL(pneigh_lookup);
763 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
764 struct net_device *dev)
766 struct pneigh_entry *n, **np;
767 unsigned int key_len = tbl->key_len;
768 u32 hash_val = pneigh_hash(pkey, key_len);
770 write_lock_bh(&tbl->lock);
771 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
773 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
774 net_eq(pneigh_net(n), net)) {
776 write_unlock_bh(&tbl->lock);
777 if (tbl->pdestructor)
784 write_unlock_bh(&tbl->lock);
788 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
789 struct net_device *dev)
791 struct pneigh_entry *n, **np, *freelist = NULL;
794 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
795 np = &tbl->phash_buckets[h];
796 while ((n = *np) != NULL) {
797 if (!dev || n->dev == dev) {
806 write_unlock_bh(&tbl->lock);
807 while ((n = freelist)) {
810 if (tbl->pdestructor)
818 static void neigh_parms_destroy(struct neigh_parms *parms);
820 static inline void neigh_parms_put(struct neigh_parms *parms)
822 if (refcount_dec_and_test(&parms->refcnt))
823 neigh_parms_destroy(parms);
827 * neighbour must already be out of the table;
830 void neigh_destroy(struct neighbour *neigh)
832 struct net_device *dev = neigh->dev;
834 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
837 pr_warn("Destroying alive neighbour %p\n", neigh);
842 if (neigh_del_timer(neigh))
843 pr_warn("Impossible event\n");
845 write_lock_bh(&neigh->lock);
846 __skb_queue_purge(&neigh->arp_queue);
847 write_unlock_bh(&neigh->lock);
848 neigh->arp_queue_len_bytes = 0;
850 if (dev->netdev_ops->ndo_neigh_destroy)
851 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
854 neigh_parms_put(neigh->parms);
856 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
858 atomic_dec(&neigh->tbl->entries);
859 kfree_rcu(neigh, rcu);
861 EXPORT_SYMBOL(neigh_destroy);
863 /* Neighbour state is suspicious;
866 Called with write_locked neigh.
868 static void neigh_suspect(struct neighbour *neigh)
870 neigh_dbg(2, "neigh %p is suspected\n", neigh);
872 neigh->output = neigh->ops->output;
875 /* Neighbour state is OK;
878 Called with write_locked neigh.
880 static void neigh_connect(struct neighbour *neigh)
882 neigh_dbg(2, "neigh %p is connected\n", neigh);
884 neigh->output = neigh->ops->connected_output;
887 static void neigh_periodic_work(struct work_struct *work)
889 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
891 struct neighbour __rcu **np;
893 struct neigh_hash_table *nht;
895 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
897 write_lock_bh(&tbl->lock);
898 nht = rcu_dereference_protected(tbl->nht,
899 lockdep_is_held(&tbl->lock));
902 * periodically recompute ReachableTime from random function
905 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
906 struct neigh_parms *p;
907 tbl->last_rand = jiffies;
908 list_for_each_entry(p, &tbl->parms_list, list)
910 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
913 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
916 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
917 np = &nht->hash_buckets[i];
919 while ((n = rcu_dereference_protected(*np,
920 lockdep_is_held(&tbl->lock))) != NULL) {
923 write_lock(&n->lock);
925 state = n->nud_state;
926 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
927 (n->flags & NTF_EXT_LEARNED)) {
928 write_unlock(&n->lock);
932 if (time_before(n->used, n->confirmed))
933 n->used = n->confirmed;
935 if (refcount_read(&n->refcnt) == 1 &&
936 (state == NUD_FAILED ||
937 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
940 write_unlock(&n->lock);
941 neigh_cleanup_and_release(n);
944 write_unlock(&n->lock);
950 * It's fine to release lock here, even if hash table
951 * grows while we are preempted.
953 write_unlock_bh(&tbl->lock);
955 write_lock_bh(&tbl->lock);
956 nht = rcu_dereference_protected(tbl->nht,
957 lockdep_is_held(&tbl->lock));
960 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
961 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
962 * BASE_REACHABLE_TIME.
964 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
965 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
966 write_unlock_bh(&tbl->lock);
969 static __inline__ int neigh_max_probes(struct neighbour *n)
971 struct neigh_parms *p = n->parms;
972 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
973 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
974 NEIGH_VAR(p, MCAST_PROBES));
977 static void neigh_invalidate(struct neighbour *neigh)
978 __releases(neigh->lock)
979 __acquires(neigh->lock)
983 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
984 neigh_dbg(2, "neigh %p is failed\n", neigh);
985 neigh->updated = jiffies;
987 /* It is very thin place. report_unreachable is very complicated
988 routine. Particularly, it can hit the same neighbour entry!
990 So that, we try to be accurate and avoid dead loop. --ANK
992 while (neigh->nud_state == NUD_FAILED &&
993 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
994 write_unlock(&neigh->lock);
995 neigh->ops->error_report(neigh, skb);
996 write_lock(&neigh->lock);
998 __skb_queue_purge(&neigh->arp_queue);
999 neigh->arp_queue_len_bytes = 0;
1002 static void neigh_probe(struct neighbour *neigh)
1003 __releases(neigh->lock)
1005 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1006 /* keep skb alive even if arp_queue overflows */
1008 skb = skb_clone(skb, GFP_ATOMIC);
1009 write_unlock(&neigh->lock);
1010 if (neigh->ops->solicit)
1011 neigh->ops->solicit(neigh, skb);
1012 atomic_inc(&neigh->probes);
1016 /* Called when a timer expires for a neighbour entry. */
1018 static void neigh_timer_handler(struct timer_list *t)
1020 unsigned long now, next;
1021 struct neighbour *neigh = from_timer(neigh, t, timer);
1025 write_lock(&neigh->lock);
1027 state = neigh->nud_state;
1031 if (!(state & NUD_IN_TIMER))
1034 if (state & NUD_REACHABLE) {
1035 if (time_before_eq(now,
1036 neigh->confirmed + neigh->parms->reachable_time)) {
1037 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1038 next = neigh->confirmed + neigh->parms->reachable_time;
1039 } else if (time_before_eq(now,
1041 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1042 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1043 neigh->nud_state = NUD_DELAY;
1044 neigh->updated = jiffies;
1045 neigh_suspect(neigh);
1046 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1048 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1049 neigh->nud_state = NUD_STALE;
1050 neigh->updated = jiffies;
1051 neigh_suspect(neigh);
1054 } else if (state & NUD_DELAY) {
1055 if (time_before_eq(now,
1057 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1058 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1059 neigh->nud_state = NUD_REACHABLE;
1060 neigh->updated = jiffies;
1061 neigh_connect(neigh);
1063 next = neigh->confirmed + neigh->parms->reachable_time;
1065 neigh_dbg(2, "neigh %p is probed\n", neigh);
1066 neigh->nud_state = NUD_PROBE;
1067 neigh->updated = jiffies;
1068 atomic_set(&neigh->probes, 0);
1070 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1074 /* NUD_PROBE|NUD_INCOMPLETE */
1075 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1078 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1079 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1080 neigh->nud_state = NUD_FAILED;
1082 neigh_invalidate(neigh);
1086 if (neigh->nud_state & NUD_IN_TIMER) {
1087 if (time_before(next, jiffies + HZ/100))
1088 next = jiffies + HZ/100;
1089 if (!mod_timer(&neigh->timer, next))
1092 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1096 write_unlock(&neigh->lock);
1100 neigh_update_notify(neigh, 0);
1102 trace_neigh_timer_handler(neigh, 0);
1104 neigh_release(neigh);
1107 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1110 bool immediate_probe = false;
1112 write_lock_bh(&neigh->lock);
1115 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1120 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1121 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1122 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1123 unsigned long next, now = jiffies;
1125 atomic_set(&neigh->probes,
1126 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1127 neigh_del_timer(neigh);
1128 neigh->nud_state = NUD_INCOMPLETE;
1129 neigh->updated = now;
1130 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1132 neigh_add_timer(neigh, next);
1133 immediate_probe = true;
1135 neigh->nud_state = NUD_FAILED;
1136 neigh->updated = jiffies;
1137 write_unlock_bh(&neigh->lock);
1142 } else if (neigh->nud_state & NUD_STALE) {
1143 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1144 neigh_del_timer(neigh);
1145 neigh->nud_state = NUD_DELAY;
1146 neigh->updated = jiffies;
1147 neigh_add_timer(neigh, jiffies +
1148 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1151 if (neigh->nud_state == NUD_INCOMPLETE) {
1153 while (neigh->arp_queue_len_bytes + skb->truesize >
1154 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1155 struct sk_buff *buff;
1157 buff = __skb_dequeue(&neigh->arp_queue);
1160 neigh->arp_queue_len_bytes -= buff->truesize;
1162 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1165 __skb_queue_tail(&neigh->arp_queue, skb);
1166 neigh->arp_queue_len_bytes += skb->truesize;
1171 if (immediate_probe)
1174 write_unlock(&neigh->lock);
1176 trace_neigh_event_send_done(neigh, rc);
1180 if (neigh->nud_state & NUD_STALE)
1182 write_unlock_bh(&neigh->lock);
1184 trace_neigh_event_send_dead(neigh, 1);
1187 EXPORT_SYMBOL(__neigh_event_send);
1189 static void neigh_update_hhs(struct neighbour *neigh)
1191 struct hh_cache *hh;
1192 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1195 if (neigh->dev->header_ops)
1196 update = neigh->dev->header_ops->cache_update;
1200 if (READ_ONCE(hh->hh_len)) {
1201 write_seqlock_bh(&hh->hh_lock);
1202 update(hh, neigh->dev, neigh->ha);
1203 write_sequnlock_bh(&hh->hh_lock);
1210 /* Generic update routine.
1211 -- lladdr is new lladdr or NULL, if it is not supplied.
1212 -- new is new state.
1214 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1216 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1217 lladdr instead of overriding it
1219 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1221 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1223 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1226 Caller MUST hold reference count on the entry.
1229 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1230 u8 new, u32 flags, u32 nlmsg_pid,
1231 struct netlink_ext_ack *extack)
1233 bool ext_learn_change = false;
1237 struct net_device *dev;
1238 int update_isrouter = 0;
1240 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1242 write_lock_bh(&neigh->lock);
1245 old = neigh->nud_state;
1249 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1253 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1254 (old & (NUD_NOARP | NUD_PERMANENT)))
1257 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1259 if (!(new & NUD_VALID)) {
1260 neigh_del_timer(neigh);
1261 if (old & NUD_CONNECTED)
1262 neigh_suspect(neigh);
1263 neigh->nud_state = new;
1265 notify = old & NUD_VALID;
1266 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1267 (new & NUD_FAILED)) {
1268 neigh_invalidate(neigh);
1274 /* Compare new lladdr with cached one */
1275 if (!dev->addr_len) {
1276 /* First case: device needs no address. */
1278 } else if (lladdr) {
1279 /* The second case: if something is already cached
1280 and a new address is proposed:
1282 - if they are different, check override flag
1284 if ((old & NUD_VALID) &&
1285 !memcmp(lladdr, neigh->ha, dev->addr_len))
1288 /* No address is supplied; if we know something,
1289 use it, otherwise discard the request.
1292 if (!(old & NUD_VALID)) {
1293 NL_SET_ERR_MSG(extack, "No link layer address given");
1299 /* Update confirmed timestamp for neighbour entry after we
1300 * received ARP packet even if it doesn't change IP to MAC binding.
1302 if (new & NUD_CONNECTED)
1303 neigh->confirmed = jiffies;
1305 /* If entry was valid and address is not changed,
1306 do not change entry state, if new one is STALE.
1309 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1310 if (old & NUD_VALID) {
1311 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1312 update_isrouter = 0;
1313 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1314 (old & NUD_CONNECTED)) {
1320 if (lladdr == neigh->ha && new == NUD_STALE &&
1321 !(flags & NEIGH_UPDATE_F_ADMIN))
1326 /* Update timestamp only once we know we will make a change to the
1327 * neighbour entry. Otherwise we risk to move the locktime window with
1328 * noop updates and ignore relevant ARP updates.
1330 if (new != old || lladdr != neigh->ha)
1331 neigh->updated = jiffies;
1334 neigh_del_timer(neigh);
1335 if (new & NUD_PROBE)
1336 atomic_set(&neigh->probes, 0);
1337 if (new & NUD_IN_TIMER)
1338 neigh_add_timer(neigh, (jiffies +
1339 ((new & NUD_REACHABLE) ?
1340 neigh->parms->reachable_time :
1342 neigh->nud_state = new;
1346 if (lladdr != neigh->ha) {
1347 write_seqlock(&neigh->ha_lock);
1348 memcpy(&neigh->ha, lladdr, dev->addr_len);
1349 write_sequnlock(&neigh->ha_lock);
1350 neigh_update_hhs(neigh);
1351 if (!(new & NUD_CONNECTED))
1352 neigh->confirmed = jiffies -
1353 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1358 if (new & NUD_CONNECTED)
1359 neigh_connect(neigh);
1361 neigh_suspect(neigh);
1362 if (!(old & NUD_VALID)) {
1363 struct sk_buff *skb;
1365 /* Again: avoid dead loop if something went wrong */
1367 while (neigh->nud_state & NUD_VALID &&
1368 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1369 struct dst_entry *dst = skb_dst(skb);
1370 struct neighbour *n2, *n1 = neigh;
1371 write_unlock_bh(&neigh->lock);
1375 /* Why not just use 'neigh' as-is? The problem is that
1376 * things such as shaper, eql, and sch_teql can end up
1377 * using alternative, different, neigh objects to output
1378 * the packet in the output path. So what we need to do
1379 * here is re-lookup the top-level neigh in the path so
1380 * we can reinject the packet there.
1383 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1384 n2 = dst_neigh_lookup_skb(dst, skb);
1388 n1->output(n1, skb);
1393 write_lock_bh(&neigh->lock);
1395 __skb_queue_purge(&neigh->arp_queue);
1396 neigh->arp_queue_len_bytes = 0;
1399 if (update_isrouter)
1400 neigh_update_is_router(neigh, flags, ¬ify);
1401 write_unlock_bh(&neigh->lock);
1403 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1404 neigh_update_gc_list(neigh);
1407 neigh_update_notify(neigh, nlmsg_pid);
1409 trace_neigh_update_done(neigh, err);
1414 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1415 u32 flags, u32 nlmsg_pid)
1417 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1419 EXPORT_SYMBOL(neigh_update);
1421 /* Update the neigh to listen temporarily for probe responses, even if it is
1422 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1424 void __neigh_set_probe_once(struct neighbour *neigh)
1428 neigh->updated = jiffies;
1429 if (!(neigh->nud_state & NUD_FAILED))
1431 neigh->nud_state = NUD_INCOMPLETE;
1432 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1433 neigh_add_timer(neigh,
1434 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1437 EXPORT_SYMBOL(__neigh_set_probe_once);
1439 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1440 u8 *lladdr, void *saddr,
1441 struct net_device *dev)
1443 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1444 lladdr || !dev->addr_len);
1446 neigh_update(neigh, lladdr, NUD_STALE,
1447 NEIGH_UPDATE_F_OVERRIDE, 0);
1450 EXPORT_SYMBOL(neigh_event_ns);
1452 /* called with read_lock_bh(&n->lock); */
1453 static void neigh_hh_init(struct neighbour *n)
1455 struct net_device *dev = n->dev;
1456 __be16 prot = n->tbl->protocol;
1457 struct hh_cache *hh = &n->hh;
1459 write_lock_bh(&n->lock);
1461 /* Only one thread can come in here and initialize the
1465 dev->header_ops->cache(n, hh, prot);
1467 write_unlock_bh(&n->lock);
1470 /* Slow and careful. */
1472 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1476 if (!neigh_event_send(neigh, skb)) {
1478 struct net_device *dev = neigh->dev;
1481 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1482 neigh_hh_init(neigh);
1485 __skb_pull(skb, skb_network_offset(skb));
1486 seq = read_seqbegin(&neigh->ha_lock);
1487 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1488 neigh->ha, NULL, skb->len);
1489 } while (read_seqretry(&neigh->ha_lock, seq));
1492 rc = dev_queue_xmit(skb);
1503 EXPORT_SYMBOL(neigh_resolve_output);
1505 /* As fast as possible without hh cache */
1507 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1509 struct net_device *dev = neigh->dev;
1514 __skb_pull(skb, skb_network_offset(skb));
1515 seq = read_seqbegin(&neigh->ha_lock);
1516 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1517 neigh->ha, NULL, skb->len);
1518 } while (read_seqretry(&neigh->ha_lock, seq));
1521 err = dev_queue_xmit(skb);
1528 EXPORT_SYMBOL(neigh_connected_output);
1530 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1532 return dev_queue_xmit(skb);
1534 EXPORT_SYMBOL(neigh_direct_output);
1536 static void neigh_proxy_process(struct timer_list *t)
1538 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1539 long sched_next = 0;
1540 unsigned long now = jiffies;
1541 struct sk_buff *skb, *n;
1543 spin_lock(&tbl->proxy_queue.lock);
1545 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1546 long tdif = NEIGH_CB(skb)->sched_next - now;
1549 struct net_device *dev = skb->dev;
1551 __skb_unlink(skb, &tbl->proxy_queue);
1552 if (tbl->proxy_redo && netif_running(dev)) {
1554 tbl->proxy_redo(skb);
1561 } else if (!sched_next || tdif < sched_next)
1564 del_timer(&tbl->proxy_timer);
1566 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1567 spin_unlock(&tbl->proxy_queue.lock);
1570 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1571 struct sk_buff *skb)
1573 unsigned long sched_next = jiffies +
1574 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1576 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1581 NEIGH_CB(skb)->sched_next = sched_next;
1582 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1584 spin_lock(&tbl->proxy_queue.lock);
1585 if (del_timer(&tbl->proxy_timer)) {
1586 if (time_before(tbl->proxy_timer.expires, sched_next))
1587 sched_next = tbl->proxy_timer.expires;
1591 __skb_queue_tail(&tbl->proxy_queue, skb);
1592 mod_timer(&tbl->proxy_timer, sched_next);
1593 spin_unlock(&tbl->proxy_queue.lock);
1595 EXPORT_SYMBOL(pneigh_enqueue);
1597 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1598 struct net *net, int ifindex)
1600 struct neigh_parms *p;
1602 list_for_each_entry(p, &tbl->parms_list, list) {
1603 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1604 (!p->dev && !ifindex && net_eq(net, &init_net)))
1611 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1612 struct neigh_table *tbl)
1614 struct neigh_parms *p;
1615 struct net *net = dev_net(dev);
1616 const struct net_device_ops *ops = dev->netdev_ops;
1618 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1621 refcount_set(&p->refcnt, 1);
1623 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1626 write_pnet(&p->net, net);
1627 p->sysctl_table = NULL;
1629 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1635 write_lock_bh(&tbl->lock);
1636 list_add(&p->list, &tbl->parms.list);
1637 write_unlock_bh(&tbl->lock);
1639 neigh_parms_data_state_cleanall(p);
1643 EXPORT_SYMBOL(neigh_parms_alloc);
1645 static void neigh_rcu_free_parms(struct rcu_head *head)
1647 struct neigh_parms *parms =
1648 container_of(head, struct neigh_parms, rcu_head);
1650 neigh_parms_put(parms);
1653 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1655 if (!parms || parms == &tbl->parms)
1657 write_lock_bh(&tbl->lock);
1658 list_del(&parms->list);
1660 write_unlock_bh(&tbl->lock);
1661 dev_put(parms->dev);
1662 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1664 EXPORT_SYMBOL(neigh_parms_release);
1666 static void neigh_parms_destroy(struct neigh_parms *parms)
1671 static struct lock_class_key neigh_table_proxy_queue_class;
1673 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1675 void neigh_table_init(int index, struct neigh_table *tbl)
1677 unsigned long now = jiffies;
1678 unsigned long phsize;
1680 INIT_LIST_HEAD(&tbl->parms_list);
1681 INIT_LIST_HEAD(&tbl->gc_list);
1682 list_add(&tbl->parms.list, &tbl->parms_list);
1683 write_pnet(&tbl->parms.net, &init_net);
1684 refcount_set(&tbl->parms.refcnt, 1);
1685 tbl->parms.reachable_time =
1686 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1688 tbl->stats = alloc_percpu(struct neigh_statistics);
1690 panic("cannot create neighbour cache statistics");
1692 #ifdef CONFIG_PROC_FS
1693 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1694 &neigh_stat_seq_ops, tbl))
1695 panic("cannot create neighbour proc dir entry");
1698 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1700 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1701 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1703 if (!tbl->nht || !tbl->phash_buckets)
1704 panic("cannot allocate neighbour cache hashes");
1706 if (!tbl->entry_size)
1707 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1708 tbl->key_len, NEIGH_PRIV_ALIGN);
1710 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1712 rwlock_init(&tbl->lock);
1713 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1714 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1715 tbl->parms.reachable_time);
1716 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1717 skb_queue_head_init_class(&tbl->proxy_queue,
1718 &neigh_table_proxy_queue_class);
1720 tbl->last_flush = now;
1721 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1723 neigh_tables[index] = tbl;
1725 EXPORT_SYMBOL(neigh_table_init);
1727 int neigh_table_clear(int index, struct neigh_table *tbl)
1729 neigh_tables[index] = NULL;
1730 /* It is not clean... Fix it to unload IPv6 module safely */
1731 cancel_delayed_work_sync(&tbl->gc_work);
1732 del_timer_sync(&tbl->proxy_timer);
1733 pneigh_queue_purge(&tbl->proxy_queue);
1734 neigh_ifdown(tbl, NULL);
1735 if (atomic_read(&tbl->entries))
1736 pr_crit("neighbour leakage\n");
1738 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1739 neigh_hash_free_rcu);
1742 kfree(tbl->phash_buckets);
1743 tbl->phash_buckets = NULL;
1745 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1747 free_percpu(tbl->stats);
1752 EXPORT_SYMBOL(neigh_table_clear);
1754 static struct neigh_table *neigh_find_table(int family)
1756 struct neigh_table *tbl = NULL;
1760 tbl = neigh_tables[NEIGH_ARP_TABLE];
1763 tbl = neigh_tables[NEIGH_ND_TABLE];
1766 tbl = neigh_tables[NEIGH_DN_TABLE];
1773 const struct nla_policy nda_policy[NDA_MAX+1] = {
1774 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1775 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1776 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1777 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1778 [NDA_PROBES] = { .type = NLA_U32 },
1779 [NDA_VLAN] = { .type = NLA_U16 },
1780 [NDA_PORT] = { .type = NLA_U16 },
1781 [NDA_VNI] = { .type = NLA_U32 },
1782 [NDA_IFINDEX] = { .type = NLA_U32 },
1783 [NDA_MASTER] = { .type = NLA_U32 },
1784 [NDA_PROTOCOL] = { .type = NLA_U8 },
1785 [NDA_NH_ID] = { .type = NLA_U32 },
1786 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1789 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1790 struct netlink_ext_ack *extack)
1792 struct net *net = sock_net(skb->sk);
1794 struct nlattr *dst_attr;
1795 struct neigh_table *tbl;
1796 struct neighbour *neigh;
1797 struct net_device *dev = NULL;
1801 if (nlmsg_len(nlh) < sizeof(*ndm))
1804 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1806 NL_SET_ERR_MSG(extack, "Network address not specified");
1810 ndm = nlmsg_data(nlh);
1811 if (ndm->ndm_ifindex) {
1812 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1819 tbl = neigh_find_table(ndm->ndm_family);
1821 return -EAFNOSUPPORT;
1823 if (nla_len(dst_attr) < (int)tbl->key_len) {
1824 NL_SET_ERR_MSG(extack, "Invalid network address");
1828 if (ndm->ndm_flags & NTF_PROXY) {
1829 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1836 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1837 if (neigh == NULL) {
1842 err = __neigh_update(neigh, NULL, NUD_FAILED,
1843 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1844 NETLINK_CB(skb).portid, extack);
1845 write_lock_bh(&tbl->lock);
1846 neigh_release(neigh);
1847 neigh_remove_one(neigh, tbl);
1848 write_unlock_bh(&tbl->lock);
1854 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1855 struct netlink_ext_ack *extack)
1857 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1858 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1859 struct net *net = sock_net(skb->sk);
1861 struct nlattr *tb[NDA_MAX+1];
1862 struct neigh_table *tbl;
1863 struct net_device *dev = NULL;
1864 struct neighbour *neigh;
1870 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1871 nda_policy, extack);
1877 NL_SET_ERR_MSG(extack, "Network address not specified");
1881 ndm = nlmsg_data(nlh);
1882 if (ndm->ndm_ifindex) {
1883 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1889 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1890 NL_SET_ERR_MSG(extack, "Invalid link address");
1895 tbl = neigh_find_table(ndm->ndm_family);
1897 return -EAFNOSUPPORT;
1899 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1900 NL_SET_ERR_MSG(extack, "Invalid network address");
1904 dst = nla_data(tb[NDA_DST]);
1905 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1907 if (tb[NDA_PROTOCOL])
1908 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1910 if (ndm->ndm_flags & NTF_PROXY) {
1911 struct pneigh_entry *pn;
1914 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1916 pn->flags = ndm->ndm_flags;
1918 pn->protocol = protocol;
1925 NL_SET_ERR_MSG(extack, "Device not specified");
1929 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1934 neigh = neigh_lookup(tbl, dst, dev);
1935 if (neigh == NULL) {
1936 bool exempt_from_gc;
1938 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1943 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1944 ndm->ndm_flags & NTF_EXT_LEARNED;
1945 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1946 if (IS_ERR(neigh)) {
1947 err = PTR_ERR(neigh);
1951 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1953 neigh_release(neigh);
1957 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1958 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1959 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1963 neigh->protocol = protocol;
1965 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1966 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1968 if (ndm->ndm_flags & NTF_ROUTER)
1969 flags |= NEIGH_UPDATE_F_ISROUTER;
1971 if (ndm->ndm_flags & NTF_USE) {
1972 neigh_event_send(neigh, NULL);
1975 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1976 NETLINK_CB(skb).portid, extack);
1978 neigh_release(neigh);
1984 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1986 struct nlattr *nest;
1988 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1993 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1994 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1995 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1996 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1997 /* approximative value for deprecated QUEUE_LEN (in packets) */
1998 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1999 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2000 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2001 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2002 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2003 NEIGH_VAR(parms, UCAST_PROBES)) ||
2004 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2005 NEIGH_VAR(parms, MCAST_PROBES)) ||
2006 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2007 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2008 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2010 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2011 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2012 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2013 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2014 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2015 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2016 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2017 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2018 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2019 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2020 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2021 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2022 nla_put_msecs(skb, NDTPA_LOCKTIME,
2023 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2024 goto nla_put_failure;
2025 return nla_nest_end(skb, nest);
2028 nla_nest_cancel(skb, nest);
2032 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2033 u32 pid, u32 seq, int type, int flags)
2035 struct nlmsghdr *nlh;
2036 struct ndtmsg *ndtmsg;
2038 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2042 ndtmsg = nlmsg_data(nlh);
2044 read_lock_bh(&tbl->lock);
2045 ndtmsg->ndtm_family = tbl->family;
2046 ndtmsg->ndtm_pad1 = 0;
2047 ndtmsg->ndtm_pad2 = 0;
2049 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2050 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2051 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2052 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2053 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2054 goto nla_put_failure;
2056 unsigned long now = jiffies;
2057 long flush_delta = now - tbl->last_flush;
2058 long rand_delta = now - tbl->last_rand;
2059 struct neigh_hash_table *nht;
2060 struct ndt_config ndc = {
2061 .ndtc_key_len = tbl->key_len,
2062 .ndtc_entry_size = tbl->entry_size,
2063 .ndtc_entries = atomic_read(&tbl->entries),
2064 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2065 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2066 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2070 nht = rcu_dereference_bh(tbl->nht);
2071 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2072 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2073 rcu_read_unlock_bh();
2075 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2076 goto nla_put_failure;
2081 struct ndt_stats ndst;
2083 memset(&ndst, 0, sizeof(ndst));
2085 for_each_possible_cpu(cpu) {
2086 struct neigh_statistics *st;
2088 st = per_cpu_ptr(tbl->stats, cpu);
2089 ndst.ndts_allocs += st->allocs;
2090 ndst.ndts_destroys += st->destroys;
2091 ndst.ndts_hash_grows += st->hash_grows;
2092 ndst.ndts_res_failed += st->res_failed;
2093 ndst.ndts_lookups += st->lookups;
2094 ndst.ndts_hits += st->hits;
2095 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2096 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2097 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2098 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2099 ndst.ndts_table_fulls += st->table_fulls;
2102 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2104 goto nla_put_failure;
2107 BUG_ON(tbl->parms.dev);
2108 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2109 goto nla_put_failure;
2111 read_unlock_bh(&tbl->lock);
2112 nlmsg_end(skb, nlh);
2116 read_unlock_bh(&tbl->lock);
2117 nlmsg_cancel(skb, nlh);
2121 static int neightbl_fill_param_info(struct sk_buff *skb,
2122 struct neigh_table *tbl,
2123 struct neigh_parms *parms,
2124 u32 pid, u32 seq, int type,
2127 struct ndtmsg *ndtmsg;
2128 struct nlmsghdr *nlh;
2130 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2134 ndtmsg = nlmsg_data(nlh);
2136 read_lock_bh(&tbl->lock);
2137 ndtmsg->ndtm_family = tbl->family;
2138 ndtmsg->ndtm_pad1 = 0;
2139 ndtmsg->ndtm_pad2 = 0;
2141 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2142 neightbl_fill_parms(skb, parms) < 0)
2145 read_unlock_bh(&tbl->lock);
2146 nlmsg_end(skb, nlh);
2149 read_unlock_bh(&tbl->lock);
2150 nlmsg_cancel(skb, nlh);
2154 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2155 [NDTA_NAME] = { .type = NLA_STRING },
2156 [NDTA_THRESH1] = { .type = NLA_U32 },
2157 [NDTA_THRESH2] = { .type = NLA_U32 },
2158 [NDTA_THRESH3] = { .type = NLA_U32 },
2159 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2160 [NDTA_PARMS] = { .type = NLA_NESTED },
2163 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2164 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2165 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2166 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2167 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2168 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2169 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2170 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2171 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2172 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2173 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2174 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2175 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2176 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2177 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2180 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2181 struct netlink_ext_ack *extack)
2183 struct net *net = sock_net(skb->sk);
2184 struct neigh_table *tbl;
2185 struct ndtmsg *ndtmsg;
2186 struct nlattr *tb[NDTA_MAX+1];
2190 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2191 nl_neightbl_policy, extack);
2195 if (tb[NDTA_NAME] == NULL) {
2200 ndtmsg = nlmsg_data(nlh);
2202 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2203 tbl = neigh_tables[tidx];
2206 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2208 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2218 * We acquire tbl->lock to be nice to the periodic timers and
2219 * make sure they always see a consistent set of values.
2221 write_lock_bh(&tbl->lock);
2223 if (tb[NDTA_PARMS]) {
2224 struct nlattr *tbp[NDTPA_MAX+1];
2225 struct neigh_parms *p;
2228 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2230 nl_ntbl_parm_policy, extack);
2232 goto errout_tbl_lock;
2234 if (tbp[NDTPA_IFINDEX])
2235 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2237 p = lookup_neigh_parms(tbl, net, ifindex);
2240 goto errout_tbl_lock;
2243 for (i = 1; i <= NDTPA_MAX; i++) {
2248 case NDTPA_QUEUE_LEN:
2249 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2250 nla_get_u32(tbp[i]) *
2251 SKB_TRUESIZE(ETH_FRAME_LEN));
2253 case NDTPA_QUEUE_LENBYTES:
2254 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2255 nla_get_u32(tbp[i]));
2257 case NDTPA_PROXY_QLEN:
2258 NEIGH_VAR_SET(p, PROXY_QLEN,
2259 nla_get_u32(tbp[i]));
2261 case NDTPA_APP_PROBES:
2262 NEIGH_VAR_SET(p, APP_PROBES,
2263 nla_get_u32(tbp[i]));
2265 case NDTPA_UCAST_PROBES:
2266 NEIGH_VAR_SET(p, UCAST_PROBES,
2267 nla_get_u32(tbp[i]));
2269 case NDTPA_MCAST_PROBES:
2270 NEIGH_VAR_SET(p, MCAST_PROBES,
2271 nla_get_u32(tbp[i]));
2273 case NDTPA_MCAST_REPROBES:
2274 NEIGH_VAR_SET(p, MCAST_REPROBES,
2275 nla_get_u32(tbp[i]));
2277 case NDTPA_BASE_REACHABLE_TIME:
2278 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2279 nla_get_msecs(tbp[i]));
2280 /* update reachable_time as well, otherwise, the change will
2281 * only be effective after the next time neigh_periodic_work
2282 * decides to recompute it (can be multiple minutes)
2285 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2287 case NDTPA_GC_STALETIME:
2288 NEIGH_VAR_SET(p, GC_STALETIME,
2289 nla_get_msecs(tbp[i]));
2291 case NDTPA_DELAY_PROBE_TIME:
2292 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2293 nla_get_msecs(tbp[i]));
2294 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2296 case NDTPA_RETRANS_TIME:
2297 NEIGH_VAR_SET(p, RETRANS_TIME,
2298 nla_get_msecs(tbp[i]));
2300 case NDTPA_ANYCAST_DELAY:
2301 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2302 nla_get_msecs(tbp[i]));
2304 case NDTPA_PROXY_DELAY:
2305 NEIGH_VAR_SET(p, PROXY_DELAY,
2306 nla_get_msecs(tbp[i]));
2308 case NDTPA_LOCKTIME:
2309 NEIGH_VAR_SET(p, LOCKTIME,
2310 nla_get_msecs(tbp[i]));
2317 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2318 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2319 !net_eq(net, &init_net))
2320 goto errout_tbl_lock;
2322 if (tb[NDTA_THRESH1])
2323 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2325 if (tb[NDTA_THRESH2])
2326 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2328 if (tb[NDTA_THRESH3])
2329 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2331 if (tb[NDTA_GC_INTERVAL])
2332 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2337 write_unlock_bh(&tbl->lock);
2342 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2343 struct netlink_ext_ack *extack)
2345 struct ndtmsg *ndtm;
2347 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2348 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2352 ndtm = nlmsg_data(nlh);
2353 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2354 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2358 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2359 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2366 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2368 const struct nlmsghdr *nlh = cb->nlh;
2369 struct net *net = sock_net(skb->sk);
2370 int family, tidx, nidx = 0;
2371 int tbl_skip = cb->args[0];
2372 int neigh_skip = cb->args[1];
2373 struct neigh_table *tbl;
2375 if (cb->strict_check) {
2376 int err = neightbl_valid_dump_info(nlh, cb->extack);
2382 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2384 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2385 struct neigh_parms *p;
2387 tbl = neigh_tables[tidx];
2391 if (tidx < tbl_skip || (family && tbl->family != family))
2394 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2395 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2400 p = list_next_entry(&tbl->parms, list);
2401 list_for_each_entry_from(p, &tbl->parms_list, list) {
2402 if (!net_eq(neigh_parms_net(p), net))
2405 if (nidx < neigh_skip)
2408 if (neightbl_fill_param_info(skb, tbl, p,
2409 NETLINK_CB(cb->skb).portid,
2427 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2428 u32 pid, u32 seq, int type, unsigned int flags)
2430 unsigned long now = jiffies;
2431 struct nda_cacheinfo ci;
2432 struct nlmsghdr *nlh;
2435 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2439 ndm = nlmsg_data(nlh);
2440 ndm->ndm_family = neigh->ops->family;
2443 ndm->ndm_flags = neigh->flags;
2444 ndm->ndm_type = neigh->type;
2445 ndm->ndm_ifindex = neigh->dev->ifindex;
2447 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2448 goto nla_put_failure;
2450 read_lock_bh(&neigh->lock);
2451 ndm->ndm_state = neigh->nud_state;
2452 if (neigh->nud_state & NUD_VALID) {
2453 char haddr[MAX_ADDR_LEN];
2455 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2456 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2457 read_unlock_bh(&neigh->lock);
2458 goto nla_put_failure;
2462 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2463 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2464 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2465 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2466 read_unlock_bh(&neigh->lock);
2468 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2469 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2470 goto nla_put_failure;
2472 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2473 goto nla_put_failure;
2475 nlmsg_end(skb, nlh);
2479 nlmsg_cancel(skb, nlh);
2483 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2484 u32 pid, u32 seq, int type, unsigned int flags,
2485 struct neigh_table *tbl)
2487 struct nlmsghdr *nlh;
2490 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2494 ndm = nlmsg_data(nlh);
2495 ndm->ndm_family = tbl->family;
2498 ndm->ndm_flags = pn->flags | NTF_PROXY;
2499 ndm->ndm_type = RTN_UNICAST;
2500 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2501 ndm->ndm_state = NUD_NONE;
2503 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2504 goto nla_put_failure;
2506 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2507 goto nla_put_failure;
2509 nlmsg_end(skb, nlh);
2513 nlmsg_cancel(skb, nlh);
2517 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2519 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2520 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2523 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2525 struct net_device *master;
2530 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2532 /* 0 is already used to denote NDA_MASTER wasn't passed, therefore need another
2533 * invalid value for ifindex to denote "no master".
2535 if (master_idx == -1)
2538 if (!master || master->ifindex != master_idx)
2544 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2546 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2552 struct neigh_dump_filter {
2557 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2558 struct netlink_callback *cb,
2559 struct neigh_dump_filter *filter)
2561 struct net *net = sock_net(skb->sk);
2562 struct neighbour *n;
2563 int rc, h, s_h = cb->args[1];
2564 int idx, s_idx = idx = cb->args[2];
2565 struct neigh_hash_table *nht;
2566 unsigned int flags = NLM_F_MULTI;
2568 if (filter->dev_idx || filter->master_idx)
2569 flags |= NLM_F_DUMP_FILTERED;
2572 nht = rcu_dereference_bh(tbl->nht);
2574 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2577 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2579 n = rcu_dereference_bh(n->next)) {
2580 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2582 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2583 neigh_master_filtered(n->dev, filter->master_idx))
2585 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2598 rcu_read_unlock_bh();
2604 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2605 struct netlink_callback *cb,
2606 struct neigh_dump_filter *filter)
2608 struct pneigh_entry *n;
2609 struct net *net = sock_net(skb->sk);
2610 int rc, h, s_h = cb->args[3];
2611 int idx, s_idx = idx = cb->args[4];
2612 unsigned int flags = NLM_F_MULTI;
2614 if (filter->dev_idx || filter->master_idx)
2615 flags |= NLM_F_DUMP_FILTERED;
2617 read_lock_bh(&tbl->lock);
2619 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2622 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2623 if (idx < s_idx || pneigh_net(n) != net)
2625 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2626 neigh_master_filtered(n->dev, filter->master_idx))
2628 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2630 RTM_NEWNEIGH, flags, tbl) < 0) {
2631 read_unlock_bh(&tbl->lock);
2640 read_unlock_bh(&tbl->lock);
2649 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2651 struct neigh_dump_filter *filter,
2652 struct netlink_ext_ack *extack)
2654 struct nlattr *tb[NDA_MAX + 1];
2660 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2661 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2665 ndm = nlmsg_data(nlh);
2666 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2667 ndm->ndm_state || ndm->ndm_type) {
2668 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2672 if (ndm->ndm_flags & ~NTF_PROXY) {
2673 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2677 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2678 tb, NDA_MAX, nda_policy,
2681 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2682 NDA_MAX, nda_policy, extack);
2687 for (i = 0; i <= NDA_MAX; ++i) {
2691 /* all new attributes should require strict_check */
2694 filter->dev_idx = nla_get_u32(tb[i]);
2697 filter->master_idx = nla_get_u32(tb[i]);
2701 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2710 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2712 const struct nlmsghdr *nlh = cb->nlh;
2713 struct neigh_dump_filter filter = {};
2714 struct neigh_table *tbl;
2719 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2721 /* check for full ndmsg structure presence, family member is
2722 * the same for both structures
2724 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2725 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2728 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2729 if (err < 0 && cb->strict_check)
2734 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2735 tbl = neigh_tables[t];
2739 if (t < s_t || (family && tbl->family != family))
2742 memset(&cb->args[1], 0, sizeof(cb->args) -
2743 sizeof(cb->args[0]));
2745 err = pneigh_dump_table(tbl, skb, cb, &filter);
2747 err = neigh_dump_table(tbl, skb, cb, &filter);
2756 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2757 struct neigh_table **tbl,
2758 void **dst, int *dev_idx, u8 *ndm_flags,
2759 struct netlink_ext_ack *extack)
2761 struct nlattr *tb[NDA_MAX + 1];
2765 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2766 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2770 ndm = nlmsg_data(nlh);
2771 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2773 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2777 if (ndm->ndm_flags & ~NTF_PROXY) {
2778 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2782 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2783 NDA_MAX, nda_policy, extack);
2787 *ndm_flags = ndm->ndm_flags;
2788 *dev_idx = ndm->ndm_ifindex;
2789 *tbl = neigh_find_table(ndm->ndm_family);
2791 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2792 return -EAFNOSUPPORT;
2795 for (i = 0; i <= NDA_MAX; ++i) {
2801 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2802 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2805 *dst = nla_data(tb[i]);
2808 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2816 static inline size_t neigh_nlmsg_size(void)
2818 return NLMSG_ALIGN(sizeof(struct ndmsg))
2819 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2820 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2821 + nla_total_size(sizeof(struct nda_cacheinfo))
2822 + nla_total_size(4) /* NDA_PROBES */
2823 + nla_total_size(1); /* NDA_PROTOCOL */
2826 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2829 struct sk_buff *skb;
2832 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2836 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2842 err = rtnl_unicast(skb, net, pid);
2847 static inline size_t pneigh_nlmsg_size(void)
2849 return NLMSG_ALIGN(sizeof(struct ndmsg))
2850 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2851 + nla_total_size(1); /* NDA_PROTOCOL */
2854 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2855 u32 pid, u32 seq, struct neigh_table *tbl)
2857 struct sk_buff *skb;
2860 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2864 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2870 err = rtnl_unicast(skb, net, pid);
2875 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2876 struct netlink_ext_ack *extack)
2878 struct net *net = sock_net(in_skb->sk);
2879 struct net_device *dev = NULL;
2880 struct neigh_table *tbl = NULL;
2881 struct neighbour *neigh;
2887 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2893 dev = __dev_get_by_index(net, dev_idx);
2895 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2901 NL_SET_ERR_MSG(extack, "Network address not specified");
2905 if (ndm_flags & NTF_PROXY) {
2906 struct pneigh_entry *pn;
2908 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2910 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2913 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2914 nlh->nlmsg_seq, tbl);
2918 NL_SET_ERR_MSG(extack, "No device specified");
2922 neigh = neigh_lookup(tbl, dst, dev);
2924 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2928 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2931 neigh_release(neigh);
2936 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2939 struct neigh_hash_table *nht;
2942 nht = rcu_dereference_bh(tbl->nht);
2944 read_lock(&tbl->lock); /* avoid resizes */
2945 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2946 struct neighbour *n;
2948 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2950 n = rcu_dereference_bh(n->next))
2953 read_unlock(&tbl->lock);
2954 rcu_read_unlock_bh();
2956 EXPORT_SYMBOL(neigh_for_each);
2958 /* The tbl->lock must be held as a writer and BH disabled. */
2959 void __neigh_for_each_release(struct neigh_table *tbl,
2960 int (*cb)(struct neighbour *))
2963 struct neigh_hash_table *nht;
2965 nht = rcu_dereference_protected(tbl->nht,
2966 lockdep_is_held(&tbl->lock));
2967 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2968 struct neighbour *n;
2969 struct neighbour __rcu **np;
2971 np = &nht->hash_buckets[chain];
2972 while ((n = rcu_dereference_protected(*np,
2973 lockdep_is_held(&tbl->lock))) != NULL) {
2976 write_lock(&n->lock);
2979 rcu_assign_pointer(*np,
2980 rcu_dereference_protected(n->next,
2981 lockdep_is_held(&tbl->lock)));
2985 write_unlock(&n->lock);
2987 neigh_cleanup_and_release(n);
2991 EXPORT_SYMBOL(__neigh_for_each_release);
2993 int neigh_xmit(int index, struct net_device *dev,
2994 const void *addr, struct sk_buff *skb)
2996 int err = -EAFNOSUPPORT;
2997 if (likely(index < NEIGH_NR_TABLES)) {
2998 struct neigh_table *tbl;
2999 struct neighbour *neigh;
3001 tbl = neigh_tables[index];
3005 if (index == NEIGH_ARP_TABLE) {
3006 u32 key = *((u32 *)addr);
3008 neigh = __ipv4_neigh_lookup_noref(dev, key);
3010 neigh = __neigh_lookup_noref(tbl, addr, dev);
3013 neigh = __neigh_create(tbl, addr, dev, false);
3014 err = PTR_ERR(neigh);
3015 if (IS_ERR(neigh)) {
3016 rcu_read_unlock_bh();
3019 err = neigh->output(neigh, skb);
3020 rcu_read_unlock_bh();
3022 else if (index == NEIGH_LINK_TABLE) {
3023 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3024 addr, NULL, skb->len);
3027 err = dev_queue_xmit(skb);
3035 EXPORT_SYMBOL(neigh_xmit);
3037 #ifdef CONFIG_PROC_FS
3039 static struct neighbour *neigh_get_first(struct seq_file *seq)
3041 struct neigh_seq_state *state = seq->private;
3042 struct net *net = seq_file_net(seq);
3043 struct neigh_hash_table *nht = state->nht;
3044 struct neighbour *n = NULL;
3047 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3048 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3049 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3052 if (!net_eq(dev_net(n->dev), net))
3054 if (state->neigh_sub_iter) {
3058 v = state->neigh_sub_iter(state, n, &fakep);
3062 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3064 if (n->nud_state & ~NUD_NOARP)
3067 n = rcu_dereference_bh(n->next);
3073 state->bucket = bucket;
3078 static struct neighbour *neigh_get_next(struct seq_file *seq,
3079 struct neighbour *n,
3082 struct neigh_seq_state *state = seq->private;
3083 struct net *net = seq_file_net(seq);
3084 struct neigh_hash_table *nht = state->nht;
3086 if (state->neigh_sub_iter) {
3087 void *v = state->neigh_sub_iter(state, n, pos);
3091 n = rcu_dereference_bh(n->next);
3095 if (!net_eq(dev_net(n->dev), net))
3097 if (state->neigh_sub_iter) {
3098 void *v = state->neigh_sub_iter(state, n, pos);
3103 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3106 if (n->nud_state & ~NUD_NOARP)
3109 n = rcu_dereference_bh(n->next);
3115 if (++state->bucket >= (1 << nht->hash_shift))
3118 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3126 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3128 struct neighbour *n = neigh_get_first(seq);
3133 n = neigh_get_next(seq, n, pos);
3138 return *pos ? NULL : n;
3141 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3143 struct neigh_seq_state *state = seq->private;
3144 struct net *net = seq_file_net(seq);
3145 struct neigh_table *tbl = state->tbl;
3146 struct pneigh_entry *pn = NULL;
3149 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3150 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3151 pn = tbl->phash_buckets[bucket];
3152 while (pn && !net_eq(pneigh_net(pn), net))
3157 state->bucket = bucket;
3162 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3163 struct pneigh_entry *pn,
3166 struct neigh_seq_state *state = seq->private;
3167 struct net *net = seq_file_net(seq);
3168 struct neigh_table *tbl = state->tbl;
3172 } while (pn && !net_eq(pneigh_net(pn), net));
3175 if (++state->bucket > PNEIGH_HASHMASK)
3177 pn = tbl->phash_buckets[state->bucket];
3178 while (pn && !net_eq(pneigh_net(pn), net))
3190 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3192 struct pneigh_entry *pn = pneigh_get_first(seq);
3197 pn = pneigh_get_next(seq, pn, pos);
3202 return *pos ? NULL : pn;
3205 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3207 struct neigh_seq_state *state = seq->private;
3209 loff_t idxpos = *pos;
3211 rc = neigh_get_idx(seq, &idxpos);
3212 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3213 rc = pneigh_get_idx(seq, &idxpos);
3218 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3219 __acquires(tbl->lock)
3222 struct neigh_seq_state *state = seq->private;
3226 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3229 state->nht = rcu_dereference_bh(tbl->nht);
3230 read_lock(&tbl->lock);
3232 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3234 EXPORT_SYMBOL(neigh_seq_start);
3236 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3238 struct neigh_seq_state *state;
3241 if (v == SEQ_START_TOKEN) {
3242 rc = neigh_get_first(seq);
3246 state = seq->private;
3247 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3248 rc = neigh_get_next(seq, v, NULL);
3251 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3252 rc = pneigh_get_first(seq);
3254 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3255 rc = pneigh_get_next(seq, v, NULL);
3261 EXPORT_SYMBOL(neigh_seq_next);
3263 void neigh_seq_stop(struct seq_file *seq, void *v)
3264 __releases(tbl->lock)
3267 struct neigh_seq_state *state = seq->private;
3268 struct neigh_table *tbl = state->tbl;
3270 read_unlock(&tbl->lock);
3271 rcu_read_unlock_bh();
3273 EXPORT_SYMBOL(neigh_seq_stop);
3275 /* statistics via seq_file */
3277 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3279 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3283 return SEQ_START_TOKEN;
3285 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3286 if (!cpu_possible(cpu))
3289 return per_cpu_ptr(tbl->stats, cpu);
3294 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3296 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3299 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3300 if (!cpu_possible(cpu))
3303 return per_cpu_ptr(tbl->stats, cpu);
3309 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3314 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3316 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3317 struct neigh_statistics *st = v;
3319 if (v == SEQ_START_TOKEN) {
3320 seq_puts(seq, "entries allocs destroys hash_grows lookups hits res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
3324 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3325 "%08lx %08lx %08lx "
3326 "%08lx %08lx %08lx\n",
3327 atomic_read(&tbl->entries),
3338 st->rcv_probes_mcast,
3339 st->rcv_probes_ucast,
3341 st->periodic_gc_runs,
3350 static const struct seq_operations neigh_stat_seq_ops = {
3351 .start = neigh_stat_seq_start,
3352 .next = neigh_stat_seq_next,
3353 .stop = neigh_stat_seq_stop,
3354 .show = neigh_stat_seq_show,
3356 #endif /* CONFIG_PROC_FS */
3358 static void __neigh_notify(struct neighbour *n, int type, int flags,
3361 struct net *net = dev_net(n->dev);
3362 struct sk_buff *skb;
3365 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3369 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3371 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3372 WARN_ON(err == -EMSGSIZE);
3376 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3380 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3383 void neigh_app_ns(struct neighbour *n)
3385 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3387 EXPORT_SYMBOL(neigh_app_ns);
3389 #ifdef CONFIG_SYSCTL
3390 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3392 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3393 void *buffer, size_t *lenp, loff_t *ppos)
3396 struct ctl_table tmp = *ctl;
3398 tmp.extra1 = SYSCTL_ZERO;
3399 tmp.extra2 = &unres_qlen_max;
3402 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3403 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3406 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3410 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3415 return __in_dev_arp_parms_get_rcu(dev);
3417 return __in6_dev_nd_parms_get_rcu(dev);
3422 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3425 struct net_device *dev;
3426 int family = neigh_parms_family(p);
3429 for_each_netdev_rcu(net, dev) {
3430 struct neigh_parms *dst_p =
3431 neigh_get_dev_parms_rcu(dev, family);
3433 if (dst_p && !test_bit(index, dst_p->data_state))
3434 dst_p->data[index] = p->data[index];
3439 static void neigh_proc_update(struct ctl_table *ctl, int write)
3441 struct net_device *dev = ctl->extra1;
3442 struct neigh_parms *p = ctl->extra2;
3443 struct net *net = neigh_parms_net(p);
3444 int index = (int *) ctl->data - p->data;
3449 set_bit(index, p->data_state);
3450 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3451 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3452 if (!dev) /* NULL dev means this is default value */
3453 neigh_copy_dflt_parms(net, p, index);
3456 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3457 void *buffer, size_t *lenp,
3460 struct ctl_table tmp = *ctl;
3463 tmp.extra1 = SYSCTL_ZERO;
3464 tmp.extra2 = SYSCTL_INT_MAX;
3466 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3467 neigh_proc_update(ctl, write);
3471 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3472 size_t *lenp, loff_t *ppos)
3474 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3476 neigh_proc_update(ctl, write);
3479 EXPORT_SYMBOL(neigh_proc_dointvec);
3481 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3482 size_t *lenp, loff_t *ppos)
3484 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3486 neigh_proc_update(ctl, write);
3489 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3491 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3492 void *buffer, size_t *lenp,
3495 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3497 neigh_proc_update(ctl, write);
3501 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3502 void *buffer, size_t *lenp, loff_t *ppos)
3504 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3506 neigh_proc_update(ctl, write);
3509 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3511 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3512 void *buffer, size_t *lenp,
3515 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3517 neigh_proc_update(ctl, write);
3521 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3522 void *buffer, size_t *lenp,
3525 struct neigh_parms *p = ctl->extra2;
3528 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3529 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3530 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3531 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3535 if (write && ret == 0) {
3536 /* update reachable_time as well, otherwise, the change will
3537 * only be effective after the next time neigh_periodic_work
3538 * decides to recompute it
3541 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3546 #define NEIGH_PARMS_DATA_OFFSET(index) \
3547 (&((struct neigh_parms *) 0)->data[index])
3549 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3550 [NEIGH_VAR_ ## attr] = { \
3552 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3553 .maxlen = sizeof(int), \
3555 .proc_handler = proc, \
3558 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3559 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3561 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3562 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3564 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3565 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3567 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3568 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3570 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3571 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3573 static struct neigh_sysctl_table {
3574 struct ctl_table_header *sysctl_header;
3575 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3576 } neigh_sysctl_template __read_mostly = {
3578 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3579 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3580 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3581 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3582 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3583 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3584 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3585 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3586 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3587 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3588 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3589 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3590 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3591 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3592 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3593 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3594 [NEIGH_VAR_GC_INTERVAL] = {
3595 .procname = "gc_interval",
3596 .maxlen = sizeof(int),
3598 .proc_handler = proc_dointvec_jiffies,
3600 [NEIGH_VAR_GC_THRESH1] = {
3601 .procname = "gc_thresh1",
3602 .maxlen = sizeof(int),
3604 .extra1 = SYSCTL_ZERO,
3605 .extra2 = SYSCTL_INT_MAX,
3606 .proc_handler = proc_dointvec_minmax,
3608 [NEIGH_VAR_GC_THRESH2] = {
3609 .procname = "gc_thresh2",
3610 .maxlen = sizeof(int),
3612 .extra1 = SYSCTL_ZERO,
3613 .extra2 = SYSCTL_INT_MAX,
3614 .proc_handler = proc_dointvec_minmax,
3616 [NEIGH_VAR_GC_THRESH3] = {
3617 .procname = "gc_thresh3",
3618 .maxlen = sizeof(int),
3620 .extra1 = SYSCTL_ZERO,
3621 .extra2 = SYSCTL_INT_MAX,
3622 .proc_handler = proc_dointvec_minmax,
3628 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3629 proc_handler *handler)
3632 struct neigh_sysctl_table *t;
3633 const char *dev_name_source;
3634 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3637 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3641 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3642 t->neigh_vars[i].data += (long) p;
3643 t->neigh_vars[i].extra1 = dev;
3644 t->neigh_vars[i].extra2 = p;
3648 dev_name_source = dev->name;
3649 /* Terminate the table early */
3650 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3651 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3653 struct neigh_table *tbl = p->tbl;
3654 dev_name_source = "default";
3655 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3656 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3657 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3658 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3663 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3665 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3666 /* RetransTime (in milliseconds)*/
3667 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3668 /* ReachableTime (in milliseconds) */
3669 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3671 /* Those handlers will update p->reachable_time after
3672 * base_reachable_time(_ms) is set to ensure the new timer starts being
3673 * applied after the next neighbour update instead of waiting for
3674 * neigh_periodic_work to update its value (can be multiple minutes)
3675 * So any handler that replaces them should do this as well
3678 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3679 neigh_proc_base_reachable_time;
3680 /* ReachableTime (in milliseconds) */
3681 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3682 neigh_proc_base_reachable_time;
3685 /* Don't export sysctls to unprivileged users */
3686 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3687 t->neigh_vars[0].procname = NULL;
3689 switch (neigh_parms_family(p)) {
3700 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3701 p_name, dev_name_source);
3703 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3704 if (!t->sysctl_header)
3707 p->sysctl_table = t;
3715 EXPORT_SYMBOL(neigh_sysctl_register);
3717 void neigh_sysctl_unregister(struct neigh_parms *p)
3719 if (p->sysctl_table) {
3720 struct neigh_sysctl_table *t = p->sysctl_table;
3721 p->sysctl_table = NULL;
3722 unregister_net_sysctl_table(t->sysctl_header);
3726 EXPORT_SYMBOL(neigh_sysctl_unregister);
3728 #endif /* CONFIG_SYSCTL */
3730 static int __init neigh_init(void)
3732 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3733 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3734 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3736 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3738 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3743 subsys_initcall(neigh_init);