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);
747 if (tbl->pconstructor && tbl->pconstructor(n)) {
755 write_lock_bh(&tbl->lock);
756 n->next = tbl->phash_buckets[hash_val];
757 tbl->phash_buckets[hash_val] = n;
758 write_unlock_bh(&tbl->lock);
762 EXPORT_SYMBOL(pneigh_lookup);
765 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
766 struct net_device *dev)
768 struct pneigh_entry *n, **np;
769 unsigned int key_len = tbl->key_len;
770 u32 hash_val = pneigh_hash(pkey, key_len);
772 write_lock_bh(&tbl->lock);
773 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
775 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
776 net_eq(pneigh_net(n), net)) {
778 write_unlock_bh(&tbl->lock);
779 if (tbl->pdestructor)
787 write_unlock_bh(&tbl->lock);
791 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
792 struct net_device *dev)
794 struct pneigh_entry *n, **np, *freelist = NULL;
797 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
798 np = &tbl->phash_buckets[h];
799 while ((n = *np) != NULL) {
800 if (!dev || n->dev == dev) {
809 write_unlock_bh(&tbl->lock);
810 while ((n = freelist)) {
813 if (tbl->pdestructor)
822 static void neigh_parms_destroy(struct neigh_parms *parms);
824 static inline void neigh_parms_put(struct neigh_parms *parms)
826 if (refcount_dec_and_test(&parms->refcnt))
827 neigh_parms_destroy(parms);
831 * neighbour must already be out of the table;
834 void neigh_destroy(struct neighbour *neigh)
836 struct net_device *dev = neigh->dev;
838 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
841 pr_warn("Destroying alive neighbour %p\n", neigh);
846 if (neigh_del_timer(neigh))
847 pr_warn("Impossible event\n");
849 write_lock_bh(&neigh->lock);
850 __skb_queue_purge(&neigh->arp_queue);
851 write_unlock_bh(&neigh->lock);
852 neigh->arp_queue_len_bytes = 0;
854 if (dev->netdev_ops->ndo_neigh_destroy)
855 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
858 neigh_parms_put(neigh->parms);
860 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
862 atomic_dec(&neigh->tbl->entries);
863 kfree_rcu(neigh, rcu);
865 EXPORT_SYMBOL(neigh_destroy);
867 /* Neighbour state is suspicious;
870 Called with write_locked neigh.
872 static void neigh_suspect(struct neighbour *neigh)
874 neigh_dbg(2, "neigh %p is suspected\n", neigh);
876 neigh->output = neigh->ops->output;
879 /* Neighbour state is OK;
882 Called with write_locked neigh.
884 static void neigh_connect(struct neighbour *neigh)
886 neigh_dbg(2, "neigh %p is connected\n", neigh);
888 neigh->output = neigh->ops->connected_output;
891 static void neigh_periodic_work(struct work_struct *work)
893 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
895 struct neighbour __rcu **np;
897 struct neigh_hash_table *nht;
899 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
901 write_lock_bh(&tbl->lock);
902 nht = rcu_dereference_protected(tbl->nht,
903 lockdep_is_held(&tbl->lock));
906 * periodically recompute ReachableTime from random function
909 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
910 struct neigh_parms *p;
911 tbl->last_rand = jiffies;
912 list_for_each_entry(p, &tbl->parms_list, list)
914 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
917 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
920 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
921 np = &nht->hash_buckets[i];
923 while ((n = rcu_dereference_protected(*np,
924 lockdep_is_held(&tbl->lock))) != NULL) {
927 write_lock(&n->lock);
929 state = n->nud_state;
930 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
931 (n->flags & NTF_EXT_LEARNED)) {
932 write_unlock(&n->lock);
936 if (time_before(n->used, n->confirmed))
937 n->used = n->confirmed;
939 if (refcount_read(&n->refcnt) == 1 &&
940 (state == NUD_FAILED ||
941 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
944 write_unlock(&n->lock);
945 neigh_cleanup_and_release(n);
948 write_unlock(&n->lock);
954 * It's fine to release lock here, even if hash table
955 * grows while we are preempted.
957 write_unlock_bh(&tbl->lock);
959 write_lock_bh(&tbl->lock);
960 nht = rcu_dereference_protected(tbl->nht,
961 lockdep_is_held(&tbl->lock));
964 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
965 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
966 * BASE_REACHABLE_TIME.
968 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
969 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
970 write_unlock_bh(&tbl->lock);
973 static __inline__ int neigh_max_probes(struct neighbour *n)
975 struct neigh_parms *p = n->parms;
976 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
977 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
978 NEIGH_VAR(p, MCAST_PROBES));
981 static void neigh_invalidate(struct neighbour *neigh)
982 __releases(neigh->lock)
983 __acquires(neigh->lock)
987 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
988 neigh_dbg(2, "neigh %p is failed\n", neigh);
989 neigh->updated = jiffies;
991 /* It is very thin place. report_unreachable is very complicated
992 routine. Particularly, it can hit the same neighbour entry!
994 So that, we try to be accurate and avoid dead loop. --ANK
996 while (neigh->nud_state == NUD_FAILED &&
997 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
998 write_unlock(&neigh->lock);
999 neigh->ops->error_report(neigh, skb);
1000 write_lock(&neigh->lock);
1002 __skb_queue_purge(&neigh->arp_queue);
1003 neigh->arp_queue_len_bytes = 0;
1006 static void neigh_probe(struct neighbour *neigh)
1007 __releases(neigh->lock)
1009 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1010 /* keep skb alive even if arp_queue overflows */
1012 skb = skb_clone(skb, GFP_ATOMIC);
1013 write_unlock(&neigh->lock);
1014 if (neigh->ops->solicit)
1015 neigh->ops->solicit(neigh, skb);
1016 atomic_inc(&neigh->probes);
1020 /* Called when a timer expires for a neighbour entry. */
1022 static void neigh_timer_handler(struct timer_list *t)
1024 unsigned long now, next;
1025 struct neighbour *neigh = from_timer(neigh, t, timer);
1029 write_lock(&neigh->lock);
1031 state = neigh->nud_state;
1035 if (!(state & NUD_IN_TIMER))
1038 if (state & NUD_REACHABLE) {
1039 if (time_before_eq(now,
1040 neigh->confirmed + neigh->parms->reachable_time)) {
1041 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1042 next = neigh->confirmed + neigh->parms->reachable_time;
1043 } else if (time_before_eq(now,
1045 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1046 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1047 neigh->nud_state = NUD_DELAY;
1048 neigh->updated = jiffies;
1049 neigh_suspect(neigh);
1050 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1052 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1053 neigh->nud_state = NUD_STALE;
1054 neigh->updated = jiffies;
1055 neigh_suspect(neigh);
1058 } else if (state & NUD_DELAY) {
1059 if (time_before_eq(now,
1061 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1062 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1063 neigh->nud_state = NUD_REACHABLE;
1064 neigh->updated = jiffies;
1065 neigh_connect(neigh);
1067 next = neigh->confirmed + neigh->parms->reachable_time;
1069 neigh_dbg(2, "neigh %p is probed\n", neigh);
1070 neigh->nud_state = NUD_PROBE;
1071 neigh->updated = jiffies;
1072 atomic_set(&neigh->probes, 0);
1074 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1078 /* NUD_PROBE|NUD_INCOMPLETE */
1079 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1082 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1083 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1084 neigh->nud_state = NUD_FAILED;
1086 neigh_invalidate(neigh);
1090 if (neigh->nud_state & NUD_IN_TIMER) {
1091 if (time_before(next, jiffies + HZ/100))
1092 next = jiffies + HZ/100;
1093 if (!mod_timer(&neigh->timer, next))
1096 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1100 write_unlock(&neigh->lock);
1104 neigh_update_notify(neigh, 0);
1106 trace_neigh_timer_handler(neigh, 0);
1108 neigh_release(neigh);
1111 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1114 bool immediate_probe = false;
1116 write_lock_bh(&neigh->lock);
1119 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1124 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1125 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1126 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1127 unsigned long next, now = jiffies;
1129 atomic_set(&neigh->probes,
1130 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1131 neigh_del_timer(neigh);
1132 neigh->nud_state = NUD_INCOMPLETE;
1133 neigh->updated = now;
1134 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1136 neigh_add_timer(neigh, next);
1137 immediate_probe = true;
1139 neigh->nud_state = NUD_FAILED;
1140 neigh->updated = jiffies;
1141 write_unlock_bh(&neigh->lock);
1146 } else if (neigh->nud_state & NUD_STALE) {
1147 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1148 neigh_del_timer(neigh);
1149 neigh->nud_state = NUD_DELAY;
1150 neigh->updated = jiffies;
1151 neigh_add_timer(neigh, jiffies +
1152 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1155 if (neigh->nud_state == NUD_INCOMPLETE) {
1157 while (neigh->arp_queue_len_bytes + skb->truesize >
1158 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1159 struct sk_buff *buff;
1161 buff = __skb_dequeue(&neigh->arp_queue);
1164 neigh->arp_queue_len_bytes -= buff->truesize;
1166 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1169 __skb_queue_tail(&neigh->arp_queue, skb);
1170 neigh->arp_queue_len_bytes += skb->truesize;
1175 if (immediate_probe)
1178 write_unlock(&neigh->lock);
1180 trace_neigh_event_send_done(neigh, rc);
1184 if (neigh->nud_state & NUD_STALE)
1186 write_unlock_bh(&neigh->lock);
1188 trace_neigh_event_send_dead(neigh, 1);
1191 EXPORT_SYMBOL(__neigh_event_send);
1193 static void neigh_update_hhs(struct neighbour *neigh)
1195 struct hh_cache *hh;
1196 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1199 if (neigh->dev->header_ops)
1200 update = neigh->dev->header_ops->cache_update;
1204 if (READ_ONCE(hh->hh_len)) {
1205 write_seqlock_bh(&hh->hh_lock);
1206 update(hh, neigh->dev, neigh->ha);
1207 write_sequnlock_bh(&hh->hh_lock);
1214 /* Generic update routine.
1215 -- lladdr is new lladdr or NULL, if it is not supplied.
1216 -- new is new state.
1218 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1220 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1221 lladdr instead of overriding it
1223 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1225 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1227 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1230 Caller MUST hold reference count on the entry.
1233 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1234 u8 new, u32 flags, u32 nlmsg_pid,
1235 struct netlink_ext_ack *extack)
1237 bool ext_learn_change = false;
1241 struct net_device *dev;
1242 int update_isrouter = 0;
1244 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1246 write_lock_bh(&neigh->lock);
1249 old = neigh->nud_state;
1253 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1257 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1258 (old & (NUD_NOARP | NUD_PERMANENT)))
1261 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1263 if (!(new & NUD_VALID)) {
1264 neigh_del_timer(neigh);
1265 if (old & NUD_CONNECTED)
1266 neigh_suspect(neigh);
1267 neigh->nud_state = new;
1269 notify = old & NUD_VALID;
1270 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1271 (new & NUD_FAILED)) {
1272 neigh_invalidate(neigh);
1278 /* Compare new lladdr with cached one */
1279 if (!dev->addr_len) {
1280 /* First case: device needs no address. */
1282 } else if (lladdr) {
1283 /* The second case: if something is already cached
1284 and a new address is proposed:
1286 - if they are different, check override flag
1288 if ((old & NUD_VALID) &&
1289 !memcmp(lladdr, neigh->ha, dev->addr_len))
1292 /* No address is supplied; if we know something,
1293 use it, otherwise discard the request.
1296 if (!(old & NUD_VALID)) {
1297 NL_SET_ERR_MSG(extack, "No link layer address given");
1303 /* Update confirmed timestamp for neighbour entry after we
1304 * received ARP packet even if it doesn't change IP to MAC binding.
1306 if (new & NUD_CONNECTED)
1307 neigh->confirmed = jiffies;
1309 /* If entry was valid and address is not changed,
1310 do not change entry state, if new one is STALE.
1313 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1314 if (old & NUD_VALID) {
1315 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1316 update_isrouter = 0;
1317 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1318 (old & NUD_CONNECTED)) {
1324 if (lladdr == neigh->ha && new == NUD_STALE &&
1325 !(flags & NEIGH_UPDATE_F_ADMIN))
1330 /* Update timestamp only once we know we will make a change to the
1331 * neighbour entry. Otherwise we risk to move the locktime window with
1332 * noop updates and ignore relevant ARP updates.
1334 if (new != old || lladdr != neigh->ha)
1335 neigh->updated = jiffies;
1338 neigh_del_timer(neigh);
1339 if (new & NUD_PROBE)
1340 atomic_set(&neigh->probes, 0);
1341 if (new & NUD_IN_TIMER)
1342 neigh_add_timer(neigh, (jiffies +
1343 ((new & NUD_REACHABLE) ?
1344 neigh->parms->reachable_time :
1346 neigh->nud_state = new;
1350 if (lladdr != neigh->ha) {
1351 write_seqlock(&neigh->ha_lock);
1352 memcpy(&neigh->ha, lladdr, dev->addr_len);
1353 write_sequnlock(&neigh->ha_lock);
1354 neigh_update_hhs(neigh);
1355 if (!(new & NUD_CONNECTED))
1356 neigh->confirmed = jiffies -
1357 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1362 if (new & NUD_CONNECTED)
1363 neigh_connect(neigh);
1365 neigh_suspect(neigh);
1366 if (!(old & NUD_VALID)) {
1367 struct sk_buff *skb;
1369 /* Again: avoid dead loop if something went wrong */
1371 while (neigh->nud_state & NUD_VALID &&
1372 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1373 struct dst_entry *dst = skb_dst(skb);
1374 struct neighbour *n2, *n1 = neigh;
1375 write_unlock_bh(&neigh->lock);
1379 /* Why not just use 'neigh' as-is? The problem is that
1380 * things such as shaper, eql, and sch_teql can end up
1381 * using alternative, different, neigh objects to output
1382 * the packet in the output path. So what we need to do
1383 * here is re-lookup the top-level neigh in the path so
1384 * we can reinject the packet there.
1387 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1388 n2 = dst_neigh_lookup_skb(dst, skb);
1392 n1->output(n1, skb);
1397 write_lock_bh(&neigh->lock);
1399 __skb_queue_purge(&neigh->arp_queue);
1400 neigh->arp_queue_len_bytes = 0;
1403 if (update_isrouter)
1404 neigh_update_is_router(neigh, flags, ¬ify);
1405 write_unlock_bh(&neigh->lock);
1407 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1408 neigh_update_gc_list(neigh);
1411 neigh_update_notify(neigh, nlmsg_pid);
1413 trace_neigh_update_done(neigh, err);
1418 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1419 u32 flags, u32 nlmsg_pid)
1421 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1423 EXPORT_SYMBOL(neigh_update);
1425 /* Update the neigh to listen temporarily for probe responses, even if it is
1426 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1428 void __neigh_set_probe_once(struct neighbour *neigh)
1432 neigh->updated = jiffies;
1433 if (!(neigh->nud_state & NUD_FAILED))
1435 neigh->nud_state = NUD_INCOMPLETE;
1436 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1437 neigh_add_timer(neigh,
1438 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1441 EXPORT_SYMBOL(__neigh_set_probe_once);
1443 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1444 u8 *lladdr, void *saddr,
1445 struct net_device *dev)
1447 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1448 lladdr || !dev->addr_len);
1450 neigh_update(neigh, lladdr, NUD_STALE,
1451 NEIGH_UPDATE_F_OVERRIDE, 0);
1454 EXPORT_SYMBOL(neigh_event_ns);
1456 /* called with read_lock_bh(&n->lock); */
1457 static void neigh_hh_init(struct neighbour *n)
1459 struct net_device *dev = n->dev;
1460 __be16 prot = n->tbl->protocol;
1461 struct hh_cache *hh = &n->hh;
1463 write_lock_bh(&n->lock);
1465 /* Only one thread can come in here and initialize the
1469 dev->header_ops->cache(n, hh, prot);
1471 write_unlock_bh(&n->lock);
1474 /* Slow and careful. */
1476 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1480 if (!neigh_event_send(neigh, skb)) {
1482 struct net_device *dev = neigh->dev;
1485 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1486 neigh_hh_init(neigh);
1489 __skb_pull(skb, skb_network_offset(skb));
1490 seq = read_seqbegin(&neigh->ha_lock);
1491 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1492 neigh->ha, NULL, skb->len);
1493 } while (read_seqretry(&neigh->ha_lock, seq));
1496 rc = dev_queue_xmit(skb);
1507 EXPORT_SYMBOL(neigh_resolve_output);
1509 /* As fast as possible without hh cache */
1511 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1513 struct net_device *dev = neigh->dev;
1518 __skb_pull(skb, skb_network_offset(skb));
1519 seq = read_seqbegin(&neigh->ha_lock);
1520 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1521 neigh->ha, NULL, skb->len);
1522 } while (read_seqretry(&neigh->ha_lock, seq));
1525 err = dev_queue_xmit(skb);
1532 EXPORT_SYMBOL(neigh_connected_output);
1534 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1536 return dev_queue_xmit(skb);
1538 EXPORT_SYMBOL(neigh_direct_output);
1540 static void neigh_proxy_process(struct timer_list *t)
1542 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1543 long sched_next = 0;
1544 unsigned long now = jiffies;
1545 struct sk_buff *skb, *n;
1547 spin_lock(&tbl->proxy_queue.lock);
1549 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1550 long tdif = NEIGH_CB(skb)->sched_next - now;
1553 struct net_device *dev = skb->dev;
1555 __skb_unlink(skb, &tbl->proxy_queue);
1556 if (tbl->proxy_redo && netif_running(dev)) {
1558 tbl->proxy_redo(skb);
1565 } else if (!sched_next || tdif < sched_next)
1568 del_timer(&tbl->proxy_timer);
1570 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1571 spin_unlock(&tbl->proxy_queue.lock);
1574 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1575 struct sk_buff *skb)
1577 unsigned long sched_next = jiffies +
1578 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1580 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1585 NEIGH_CB(skb)->sched_next = sched_next;
1586 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1588 spin_lock(&tbl->proxy_queue.lock);
1589 if (del_timer(&tbl->proxy_timer)) {
1590 if (time_before(tbl->proxy_timer.expires, sched_next))
1591 sched_next = tbl->proxy_timer.expires;
1595 __skb_queue_tail(&tbl->proxy_queue, skb);
1596 mod_timer(&tbl->proxy_timer, sched_next);
1597 spin_unlock(&tbl->proxy_queue.lock);
1599 EXPORT_SYMBOL(pneigh_enqueue);
1601 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1602 struct net *net, int ifindex)
1604 struct neigh_parms *p;
1606 list_for_each_entry(p, &tbl->parms_list, list) {
1607 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1608 (!p->dev && !ifindex && net_eq(net, &init_net)))
1615 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1616 struct neigh_table *tbl)
1618 struct neigh_parms *p;
1619 struct net *net = dev_net(dev);
1620 const struct net_device_ops *ops = dev->netdev_ops;
1622 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1625 refcount_set(&p->refcnt, 1);
1627 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1630 write_pnet(&p->net, net);
1631 p->sysctl_table = NULL;
1633 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1639 write_lock_bh(&tbl->lock);
1640 list_add(&p->list, &tbl->parms.list);
1641 write_unlock_bh(&tbl->lock);
1643 neigh_parms_data_state_cleanall(p);
1647 EXPORT_SYMBOL(neigh_parms_alloc);
1649 static void neigh_rcu_free_parms(struct rcu_head *head)
1651 struct neigh_parms *parms =
1652 container_of(head, struct neigh_parms, rcu_head);
1654 neigh_parms_put(parms);
1657 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1659 if (!parms || parms == &tbl->parms)
1661 write_lock_bh(&tbl->lock);
1662 list_del(&parms->list);
1664 write_unlock_bh(&tbl->lock);
1666 dev_put(parms->dev);
1667 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1669 EXPORT_SYMBOL(neigh_parms_release);
1671 static void neigh_parms_destroy(struct neigh_parms *parms)
1676 static struct lock_class_key neigh_table_proxy_queue_class;
1678 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1680 void neigh_table_init(int index, struct neigh_table *tbl)
1682 unsigned long now = jiffies;
1683 unsigned long phsize;
1685 INIT_LIST_HEAD(&tbl->parms_list);
1686 INIT_LIST_HEAD(&tbl->gc_list);
1687 list_add(&tbl->parms.list, &tbl->parms_list);
1688 write_pnet(&tbl->parms.net, &init_net);
1689 refcount_set(&tbl->parms.refcnt, 1);
1690 tbl->parms.reachable_time =
1691 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1693 tbl->stats = alloc_percpu(struct neigh_statistics);
1695 panic("cannot create neighbour cache statistics");
1697 #ifdef CONFIG_PROC_FS
1698 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1699 &neigh_stat_seq_ops, tbl))
1700 panic("cannot create neighbour proc dir entry");
1703 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1705 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1706 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1708 if (!tbl->nht || !tbl->phash_buckets)
1709 panic("cannot allocate neighbour cache hashes");
1711 if (!tbl->entry_size)
1712 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1713 tbl->key_len, NEIGH_PRIV_ALIGN);
1715 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1717 rwlock_init(&tbl->lock);
1718 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1719 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1720 tbl->parms.reachable_time);
1721 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1722 skb_queue_head_init_class(&tbl->proxy_queue,
1723 &neigh_table_proxy_queue_class);
1725 tbl->last_flush = now;
1726 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1728 neigh_tables[index] = tbl;
1730 EXPORT_SYMBOL(neigh_table_init);
1732 int neigh_table_clear(int index, struct neigh_table *tbl)
1734 neigh_tables[index] = NULL;
1735 /* It is not clean... Fix it to unload IPv6 module safely */
1736 cancel_delayed_work_sync(&tbl->gc_work);
1737 del_timer_sync(&tbl->proxy_timer);
1738 pneigh_queue_purge(&tbl->proxy_queue);
1739 neigh_ifdown(tbl, NULL);
1740 if (atomic_read(&tbl->entries))
1741 pr_crit("neighbour leakage\n");
1743 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1744 neigh_hash_free_rcu);
1747 kfree(tbl->phash_buckets);
1748 tbl->phash_buckets = NULL;
1750 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1752 free_percpu(tbl->stats);
1757 EXPORT_SYMBOL(neigh_table_clear);
1759 static struct neigh_table *neigh_find_table(int family)
1761 struct neigh_table *tbl = NULL;
1765 tbl = neigh_tables[NEIGH_ARP_TABLE];
1768 tbl = neigh_tables[NEIGH_ND_TABLE];
1771 tbl = neigh_tables[NEIGH_DN_TABLE];
1778 const struct nla_policy nda_policy[NDA_MAX+1] = {
1779 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1780 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1781 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1782 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1783 [NDA_PROBES] = { .type = NLA_U32 },
1784 [NDA_VLAN] = { .type = NLA_U16 },
1785 [NDA_PORT] = { .type = NLA_U16 },
1786 [NDA_VNI] = { .type = NLA_U32 },
1787 [NDA_IFINDEX] = { .type = NLA_U32 },
1788 [NDA_MASTER] = { .type = NLA_U32 },
1789 [NDA_PROTOCOL] = { .type = NLA_U8 },
1790 [NDA_NH_ID] = { .type = NLA_U32 },
1791 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1794 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1795 struct netlink_ext_ack *extack)
1797 struct net *net = sock_net(skb->sk);
1799 struct nlattr *dst_attr;
1800 struct neigh_table *tbl;
1801 struct neighbour *neigh;
1802 struct net_device *dev = NULL;
1806 if (nlmsg_len(nlh) < sizeof(*ndm))
1809 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1811 NL_SET_ERR_MSG(extack, "Network address not specified");
1815 ndm = nlmsg_data(nlh);
1816 if (ndm->ndm_ifindex) {
1817 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1824 tbl = neigh_find_table(ndm->ndm_family);
1826 return -EAFNOSUPPORT;
1828 if (nla_len(dst_attr) < (int)tbl->key_len) {
1829 NL_SET_ERR_MSG(extack, "Invalid network address");
1833 if (ndm->ndm_flags & NTF_PROXY) {
1834 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1841 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1842 if (neigh == NULL) {
1847 err = __neigh_update(neigh, NULL, NUD_FAILED,
1848 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1849 NETLINK_CB(skb).portid, extack);
1850 write_lock_bh(&tbl->lock);
1851 neigh_release(neigh);
1852 neigh_remove_one(neigh, tbl);
1853 write_unlock_bh(&tbl->lock);
1859 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1860 struct netlink_ext_ack *extack)
1862 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1863 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1864 struct net *net = sock_net(skb->sk);
1866 struct nlattr *tb[NDA_MAX+1];
1867 struct neigh_table *tbl;
1868 struct net_device *dev = NULL;
1869 struct neighbour *neigh;
1875 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1876 nda_policy, extack);
1882 NL_SET_ERR_MSG(extack, "Network address not specified");
1886 ndm = nlmsg_data(nlh);
1887 if (ndm->ndm_ifindex) {
1888 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1894 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1895 NL_SET_ERR_MSG(extack, "Invalid link address");
1900 tbl = neigh_find_table(ndm->ndm_family);
1902 return -EAFNOSUPPORT;
1904 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1905 NL_SET_ERR_MSG(extack, "Invalid network address");
1909 dst = nla_data(tb[NDA_DST]);
1910 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1912 if (tb[NDA_PROTOCOL])
1913 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1915 if (ndm->ndm_flags & NTF_PROXY) {
1916 struct pneigh_entry *pn;
1919 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1921 pn->flags = ndm->ndm_flags;
1923 pn->protocol = protocol;
1930 NL_SET_ERR_MSG(extack, "Device not specified");
1934 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1939 neigh = neigh_lookup(tbl, dst, dev);
1940 if (neigh == NULL) {
1941 bool exempt_from_gc;
1943 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1948 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1949 ndm->ndm_flags & NTF_EXT_LEARNED;
1950 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1951 if (IS_ERR(neigh)) {
1952 err = PTR_ERR(neigh);
1956 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1958 neigh_release(neigh);
1962 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1963 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1964 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1968 neigh->protocol = protocol;
1970 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1971 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1973 if (ndm->ndm_flags & NTF_ROUTER)
1974 flags |= NEIGH_UPDATE_F_ISROUTER;
1976 if (ndm->ndm_flags & NTF_USE) {
1977 neigh_event_send(neigh, NULL);
1980 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1981 NETLINK_CB(skb).portid, extack);
1983 neigh_release(neigh);
1989 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1991 struct nlattr *nest;
1993 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1998 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1999 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
2000 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
2001 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
2002 /* approximative value for deprecated QUEUE_LEN (in packets) */
2003 nla_put_u32(skb, NDTPA_QUEUE_LEN,
2004 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2005 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2006 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2007 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2008 NEIGH_VAR(parms, UCAST_PROBES)) ||
2009 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2010 NEIGH_VAR(parms, MCAST_PROBES)) ||
2011 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2012 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2013 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2015 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2016 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2017 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2018 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2019 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2020 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2021 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2022 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2023 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2024 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2025 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2026 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2027 nla_put_msecs(skb, NDTPA_LOCKTIME,
2028 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2029 goto nla_put_failure;
2030 return nla_nest_end(skb, nest);
2033 nla_nest_cancel(skb, nest);
2037 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2038 u32 pid, u32 seq, int type, int flags)
2040 struct nlmsghdr *nlh;
2041 struct ndtmsg *ndtmsg;
2043 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2047 ndtmsg = nlmsg_data(nlh);
2049 read_lock_bh(&tbl->lock);
2050 ndtmsg->ndtm_family = tbl->family;
2051 ndtmsg->ndtm_pad1 = 0;
2052 ndtmsg->ndtm_pad2 = 0;
2054 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2055 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2056 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2057 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2058 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2059 goto nla_put_failure;
2061 unsigned long now = jiffies;
2062 long flush_delta = now - tbl->last_flush;
2063 long rand_delta = now - tbl->last_rand;
2064 struct neigh_hash_table *nht;
2065 struct ndt_config ndc = {
2066 .ndtc_key_len = tbl->key_len,
2067 .ndtc_entry_size = tbl->entry_size,
2068 .ndtc_entries = atomic_read(&tbl->entries),
2069 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2070 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2071 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2075 nht = rcu_dereference_bh(tbl->nht);
2076 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2077 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2078 rcu_read_unlock_bh();
2080 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2081 goto nla_put_failure;
2086 struct ndt_stats ndst;
2088 memset(&ndst, 0, sizeof(ndst));
2090 for_each_possible_cpu(cpu) {
2091 struct neigh_statistics *st;
2093 st = per_cpu_ptr(tbl->stats, cpu);
2094 ndst.ndts_allocs += st->allocs;
2095 ndst.ndts_destroys += st->destroys;
2096 ndst.ndts_hash_grows += st->hash_grows;
2097 ndst.ndts_res_failed += st->res_failed;
2098 ndst.ndts_lookups += st->lookups;
2099 ndst.ndts_hits += st->hits;
2100 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2101 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2102 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2103 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2104 ndst.ndts_table_fulls += st->table_fulls;
2107 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2109 goto nla_put_failure;
2112 BUG_ON(tbl->parms.dev);
2113 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2114 goto nla_put_failure;
2116 read_unlock_bh(&tbl->lock);
2117 nlmsg_end(skb, nlh);
2121 read_unlock_bh(&tbl->lock);
2122 nlmsg_cancel(skb, nlh);
2126 static int neightbl_fill_param_info(struct sk_buff *skb,
2127 struct neigh_table *tbl,
2128 struct neigh_parms *parms,
2129 u32 pid, u32 seq, int type,
2132 struct ndtmsg *ndtmsg;
2133 struct nlmsghdr *nlh;
2135 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2139 ndtmsg = nlmsg_data(nlh);
2141 read_lock_bh(&tbl->lock);
2142 ndtmsg->ndtm_family = tbl->family;
2143 ndtmsg->ndtm_pad1 = 0;
2144 ndtmsg->ndtm_pad2 = 0;
2146 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2147 neightbl_fill_parms(skb, parms) < 0)
2150 read_unlock_bh(&tbl->lock);
2151 nlmsg_end(skb, nlh);
2154 read_unlock_bh(&tbl->lock);
2155 nlmsg_cancel(skb, nlh);
2159 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2160 [NDTA_NAME] = { .type = NLA_STRING },
2161 [NDTA_THRESH1] = { .type = NLA_U32 },
2162 [NDTA_THRESH2] = { .type = NLA_U32 },
2163 [NDTA_THRESH3] = { .type = NLA_U32 },
2164 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2165 [NDTA_PARMS] = { .type = NLA_NESTED },
2168 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2169 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2170 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2171 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2172 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2173 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2174 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2175 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2176 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2177 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2178 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2179 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2180 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2181 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2182 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2185 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2186 struct netlink_ext_ack *extack)
2188 struct net *net = sock_net(skb->sk);
2189 struct neigh_table *tbl;
2190 struct ndtmsg *ndtmsg;
2191 struct nlattr *tb[NDTA_MAX+1];
2195 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2196 nl_neightbl_policy, extack);
2200 if (tb[NDTA_NAME] == NULL) {
2205 ndtmsg = nlmsg_data(nlh);
2207 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2208 tbl = neigh_tables[tidx];
2211 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2213 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2223 * We acquire tbl->lock to be nice to the periodic timers and
2224 * make sure they always see a consistent set of values.
2226 write_lock_bh(&tbl->lock);
2228 if (tb[NDTA_PARMS]) {
2229 struct nlattr *tbp[NDTPA_MAX+1];
2230 struct neigh_parms *p;
2233 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2235 nl_ntbl_parm_policy, extack);
2237 goto errout_tbl_lock;
2239 if (tbp[NDTPA_IFINDEX])
2240 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2242 p = lookup_neigh_parms(tbl, net, ifindex);
2245 goto errout_tbl_lock;
2248 for (i = 1; i <= NDTPA_MAX; i++) {
2253 case NDTPA_QUEUE_LEN:
2254 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2255 nla_get_u32(tbp[i]) *
2256 SKB_TRUESIZE(ETH_FRAME_LEN));
2258 case NDTPA_QUEUE_LENBYTES:
2259 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2260 nla_get_u32(tbp[i]));
2262 case NDTPA_PROXY_QLEN:
2263 NEIGH_VAR_SET(p, PROXY_QLEN,
2264 nla_get_u32(tbp[i]));
2266 case NDTPA_APP_PROBES:
2267 NEIGH_VAR_SET(p, APP_PROBES,
2268 nla_get_u32(tbp[i]));
2270 case NDTPA_UCAST_PROBES:
2271 NEIGH_VAR_SET(p, UCAST_PROBES,
2272 nla_get_u32(tbp[i]));
2274 case NDTPA_MCAST_PROBES:
2275 NEIGH_VAR_SET(p, MCAST_PROBES,
2276 nla_get_u32(tbp[i]));
2278 case NDTPA_MCAST_REPROBES:
2279 NEIGH_VAR_SET(p, MCAST_REPROBES,
2280 nla_get_u32(tbp[i]));
2282 case NDTPA_BASE_REACHABLE_TIME:
2283 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2284 nla_get_msecs(tbp[i]));
2285 /* update reachable_time as well, otherwise, the change will
2286 * only be effective after the next time neigh_periodic_work
2287 * decides to recompute it (can be multiple minutes)
2290 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2292 case NDTPA_GC_STALETIME:
2293 NEIGH_VAR_SET(p, GC_STALETIME,
2294 nla_get_msecs(tbp[i]));
2296 case NDTPA_DELAY_PROBE_TIME:
2297 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2298 nla_get_msecs(tbp[i]));
2299 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2301 case NDTPA_RETRANS_TIME:
2302 NEIGH_VAR_SET(p, RETRANS_TIME,
2303 nla_get_msecs(tbp[i]));
2305 case NDTPA_ANYCAST_DELAY:
2306 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2307 nla_get_msecs(tbp[i]));
2309 case NDTPA_PROXY_DELAY:
2310 NEIGH_VAR_SET(p, PROXY_DELAY,
2311 nla_get_msecs(tbp[i]));
2313 case NDTPA_LOCKTIME:
2314 NEIGH_VAR_SET(p, LOCKTIME,
2315 nla_get_msecs(tbp[i]));
2322 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2323 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2324 !net_eq(net, &init_net))
2325 goto errout_tbl_lock;
2327 if (tb[NDTA_THRESH1])
2328 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2330 if (tb[NDTA_THRESH2])
2331 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2333 if (tb[NDTA_THRESH3])
2334 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2336 if (tb[NDTA_GC_INTERVAL])
2337 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2342 write_unlock_bh(&tbl->lock);
2347 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2348 struct netlink_ext_ack *extack)
2350 struct ndtmsg *ndtm;
2352 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2353 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2357 ndtm = nlmsg_data(nlh);
2358 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2359 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2363 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2364 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2371 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2373 const struct nlmsghdr *nlh = cb->nlh;
2374 struct net *net = sock_net(skb->sk);
2375 int family, tidx, nidx = 0;
2376 int tbl_skip = cb->args[0];
2377 int neigh_skip = cb->args[1];
2378 struct neigh_table *tbl;
2380 if (cb->strict_check) {
2381 int err = neightbl_valid_dump_info(nlh, cb->extack);
2387 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2389 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2390 struct neigh_parms *p;
2392 tbl = neigh_tables[tidx];
2396 if (tidx < tbl_skip || (family && tbl->family != family))
2399 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2400 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2405 p = list_next_entry(&tbl->parms, list);
2406 list_for_each_entry_from(p, &tbl->parms_list, list) {
2407 if (!net_eq(neigh_parms_net(p), net))
2410 if (nidx < neigh_skip)
2413 if (neightbl_fill_param_info(skb, tbl, p,
2414 NETLINK_CB(cb->skb).portid,
2432 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2433 u32 pid, u32 seq, int type, unsigned int flags)
2435 unsigned long now = jiffies;
2436 struct nda_cacheinfo ci;
2437 struct nlmsghdr *nlh;
2440 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2444 ndm = nlmsg_data(nlh);
2445 ndm->ndm_family = neigh->ops->family;
2448 ndm->ndm_flags = neigh->flags;
2449 ndm->ndm_type = neigh->type;
2450 ndm->ndm_ifindex = neigh->dev->ifindex;
2452 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2453 goto nla_put_failure;
2455 read_lock_bh(&neigh->lock);
2456 ndm->ndm_state = neigh->nud_state;
2457 if (neigh->nud_state & NUD_VALID) {
2458 char haddr[MAX_ADDR_LEN];
2460 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2461 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2462 read_unlock_bh(&neigh->lock);
2463 goto nla_put_failure;
2467 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2468 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2469 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2470 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2471 read_unlock_bh(&neigh->lock);
2473 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2474 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2475 goto nla_put_failure;
2477 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2478 goto nla_put_failure;
2480 nlmsg_end(skb, nlh);
2484 nlmsg_cancel(skb, nlh);
2488 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2489 u32 pid, u32 seq, int type, unsigned int flags,
2490 struct neigh_table *tbl)
2492 struct nlmsghdr *nlh;
2495 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2499 ndm = nlmsg_data(nlh);
2500 ndm->ndm_family = tbl->family;
2503 ndm->ndm_flags = pn->flags | NTF_PROXY;
2504 ndm->ndm_type = RTN_UNICAST;
2505 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2506 ndm->ndm_state = NUD_NONE;
2508 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2509 goto nla_put_failure;
2511 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2512 goto nla_put_failure;
2514 nlmsg_end(skb, nlh);
2518 nlmsg_cancel(skb, nlh);
2522 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2524 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2525 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2528 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2530 struct net_device *master;
2535 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2536 if (!master || master->ifindex != master_idx)
2542 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2544 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2550 struct neigh_dump_filter {
2555 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2556 struct netlink_callback *cb,
2557 struct neigh_dump_filter *filter)
2559 struct net *net = sock_net(skb->sk);
2560 struct neighbour *n;
2561 int rc, h, s_h = cb->args[1];
2562 int idx, s_idx = idx = cb->args[2];
2563 struct neigh_hash_table *nht;
2564 unsigned int flags = NLM_F_MULTI;
2566 if (filter->dev_idx || filter->master_idx)
2567 flags |= NLM_F_DUMP_FILTERED;
2570 nht = rcu_dereference_bh(tbl->nht);
2572 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2575 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2577 n = rcu_dereference_bh(n->next)) {
2578 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2580 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2581 neigh_master_filtered(n->dev, filter->master_idx))
2583 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2596 rcu_read_unlock_bh();
2602 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2603 struct netlink_callback *cb,
2604 struct neigh_dump_filter *filter)
2606 struct pneigh_entry *n;
2607 struct net *net = sock_net(skb->sk);
2608 int rc, h, s_h = cb->args[3];
2609 int idx, s_idx = idx = cb->args[4];
2610 unsigned int flags = NLM_F_MULTI;
2612 if (filter->dev_idx || filter->master_idx)
2613 flags |= NLM_F_DUMP_FILTERED;
2615 read_lock_bh(&tbl->lock);
2617 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2620 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2621 if (idx < s_idx || pneigh_net(n) != net)
2623 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2624 neigh_master_filtered(n->dev, filter->master_idx))
2626 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2628 RTM_NEWNEIGH, flags, tbl) < 0) {
2629 read_unlock_bh(&tbl->lock);
2638 read_unlock_bh(&tbl->lock);
2647 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2649 struct neigh_dump_filter *filter,
2650 struct netlink_ext_ack *extack)
2652 struct nlattr *tb[NDA_MAX + 1];
2658 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2659 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2663 ndm = nlmsg_data(nlh);
2664 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2665 ndm->ndm_state || ndm->ndm_type) {
2666 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2670 if (ndm->ndm_flags & ~NTF_PROXY) {
2671 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2675 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2676 tb, NDA_MAX, nda_policy,
2679 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2680 NDA_MAX, nda_policy, extack);
2685 for (i = 0; i <= NDA_MAX; ++i) {
2689 /* all new attributes should require strict_check */
2692 filter->dev_idx = nla_get_u32(tb[i]);
2695 filter->master_idx = nla_get_u32(tb[i]);
2699 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2708 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2710 const struct nlmsghdr *nlh = cb->nlh;
2711 struct neigh_dump_filter filter = {};
2712 struct neigh_table *tbl;
2717 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2719 /* check for full ndmsg structure presence, family member is
2720 * the same for both structures
2722 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2723 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2726 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2727 if (err < 0 && cb->strict_check)
2732 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2733 tbl = neigh_tables[t];
2737 if (t < s_t || (family && tbl->family != family))
2740 memset(&cb->args[1], 0, sizeof(cb->args) -
2741 sizeof(cb->args[0]));
2743 err = pneigh_dump_table(tbl, skb, cb, &filter);
2745 err = neigh_dump_table(tbl, skb, cb, &filter);
2754 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2755 struct neigh_table **tbl,
2756 void **dst, int *dev_idx, u8 *ndm_flags,
2757 struct netlink_ext_ack *extack)
2759 struct nlattr *tb[NDA_MAX + 1];
2763 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2764 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2768 ndm = nlmsg_data(nlh);
2769 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2771 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2775 if (ndm->ndm_flags & ~NTF_PROXY) {
2776 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2780 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2781 NDA_MAX, nda_policy, extack);
2785 *ndm_flags = ndm->ndm_flags;
2786 *dev_idx = ndm->ndm_ifindex;
2787 *tbl = neigh_find_table(ndm->ndm_family);
2789 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2790 return -EAFNOSUPPORT;
2793 for (i = 0; i <= NDA_MAX; ++i) {
2799 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2800 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2803 *dst = nla_data(tb[i]);
2806 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2814 static inline size_t neigh_nlmsg_size(void)
2816 return NLMSG_ALIGN(sizeof(struct ndmsg))
2817 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2818 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2819 + nla_total_size(sizeof(struct nda_cacheinfo))
2820 + nla_total_size(4) /* NDA_PROBES */
2821 + nla_total_size(1); /* NDA_PROTOCOL */
2824 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2827 struct sk_buff *skb;
2830 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2834 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2840 err = rtnl_unicast(skb, net, pid);
2845 static inline size_t pneigh_nlmsg_size(void)
2847 return NLMSG_ALIGN(sizeof(struct ndmsg))
2848 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2849 + nla_total_size(1); /* NDA_PROTOCOL */
2852 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2853 u32 pid, u32 seq, struct neigh_table *tbl)
2855 struct sk_buff *skb;
2858 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2862 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2868 err = rtnl_unicast(skb, net, pid);
2873 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2874 struct netlink_ext_ack *extack)
2876 struct net *net = sock_net(in_skb->sk);
2877 struct net_device *dev = NULL;
2878 struct neigh_table *tbl = NULL;
2879 struct neighbour *neigh;
2885 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2891 dev = __dev_get_by_index(net, dev_idx);
2893 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2899 NL_SET_ERR_MSG(extack, "Network address not specified");
2903 if (ndm_flags & NTF_PROXY) {
2904 struct pneigh_entry *pn;
2906 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2908 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2911 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2912 nlh->nlmsg_seq, tbl);
2916 NL_SET_ERR_MSG(extack, "No device specified");
2920 neigh = neigh_lookup(tbl, dst, dev);
2922 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2926 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2929 neigh_release(neigh);
2934 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2937 struct neigh_hash_table *nht;
2940 nht = rcu_dereference_bh(tbl->nht);
2942 read_lock(&tbl->lock); /* avoid resizes */
2943 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2944 struct neighbour *n;
2946 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2948 n = rcu_dereference_bh(n->next))
2951 read_unlock(&tbl->lock);
2952 rcu_read_unlock_bh();
2954 EXPORT_SYMBOL(neigh_for_each);
2956 /* The tbl->lock must be held as a writer and BH disabled. */
2957 void __neigh_for_each_release(struct neigh_table *tbl,
2958 int (*cb)(struct neighbour *))
2961 struct neigh_hash_table *nht;
2963 nht = rcu_dereference_protected(tbl->nht,
2964 lockdep_is_held(&tbl->lock));
2965 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2966 struct neighbour *n;
2967 struct neighbour __rcu **np;
2969 np = &nht->hash_buckets[chain];
2970 while ((n = rcu_dereference_protected(*np,
2971 lockdep_is_held(&tbl->lock))) != NULL) {
2974 write_lock(&n->lock);
2977 rcu_assign_pointer(*np,
2978 rcu_dereference_protected(n->next,
2979 lockdep_is_held(&tbl->lock)));
2983 write_unlock(&n->lock);
2985 neigh_cleanup_and_release(n);
2989 EXPORT_SYMBOL(__neigh_for_each_release);
2991 int neigh_xmit(int index, struct net_device *dev,
2992 const void *addr, struct sk_buff *skb)
2994 int err = -EAFNOSUPPORT;
2995 if (likely(index < NEIGH_NR_TABLES)) {
2996 struct neigh_table *tbl;
2997 struct neighbour *neigh;
2999 tbl = neigh_tables[index];
3003 if (index == NEIGH_ARP_TABLE) {
3004 u32 key = *((u32 *)addr);
3006 neigh = __ipv4_neigh_lookup_noref(dev, key);
3008 neigh = __neigh_lookup_noref(tbl, addr, dev);
3011 neigh = __neigh_create(tbl, addr, dev, false);
3012 err = PTR_ERR(neigh);
3013 if (IS_ERR(neigh)) {
3014 rcu_read_unlock_bh();
3017 err = neigh->output(neigh, skb);
3018 rcu_read_unlock_bh();
3020 else if (index == NEIGH_LINK_TABLE) {
3021 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3022 addr, NULL, skb->len);
3025 err = dev_queue_xmit(skb);
3033 EXPORT_SYMBOL(neigh_xmit);
3035 #ifdef CONFIG_PROC_FS
3037 static struct neighbour *neigh_get_first(struct seq_file *seq)
3039 struct neigh_seq_state *state = seq->private;
3040 struct net *net = seq_file_net(seq);
3041 struct neigh_hash_table *nht = state->nht;
3042 struct neighbour *n = NULL;
3045 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3046 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3047 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3050 if (!net_eq(dev_net(n->dev), net))
3052 if (state->neigh_sub_iter) {
3056 v = state->neigh_sub_iter(state, n, &fakep);
3060 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3062 if (n->nud_state & ~NUD_NOARP)
3065 n = rcu_dereference_bh(n->next);
3071 state->bucket = bucket;
3076 static struct neighbour *neigh_get_next(struct seq_file *seq,
3077 struct neighbour *n,
3080 struct neigh_seq_state *state = seq->private;
3081 struct net *net = seq_file_net(seq);
3082 struct neigh_hash_table *nht = state->nht;
3084 if (state->neigh_sub_iter) {
3085 void *v = state->neigh_sub_iter(state, n, pos);
3089 n = rcu_dereference_bh(n->next);
3093 if (!net_eq(dev_net(n->dev), net))
3095 if (state->neigh_sub_iter) {
3096 void *v = state->neigh_sub_iter(state, n, pos);
3101 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3104 if (n->nud_state & ~NUD_NOARP)
3107 n = rcu_dereference_bh(n->next);
3113 if (++state->bucket >= (1 << nht->hash_shift))
3116 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3124 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3126 struct neighbour *n = neigh_get_first(seq);
3131 n = neigh_get_next(seq, n, pos);
3136 return *pos ? NULL : n;
3139 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3141 struct neigh_seq_state *state = seq->private;
3142 struct net *net = seq_file_net(seq);
3143 struct neigh_table *tbl = state->tbl;
3144 struct pneigh_entry *pn = NULL;
3145 int bucket = state->bucket;
3147 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3148 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3149 pn = tbl->phash_buckets[bucket];
3150 while (pn && !net_eq(pneigh_net(pn), net))
3155 state->bucket = bucket;
3160 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3161 struct pneigh_entry *pn,
3164 struct neigh_seq_state *state = seq->private;
3165 struct net *net = seq_file_net(seq);
3166 struct neigh_table *tbl = state->tbl;
3170 } while (pn && !net_eq(pneigh_net(pn), net));
3173 if (++state->bucket > PNEIGH_HASHMASK)
3175 pn = tbl->phash_buckets[state->bucket];
3176 while (pn && !net_eq(pneigh_net(pn), net))
3188 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3190 struct pneigh_entry *pn = pneigh_get_first(seq);
3195 pn = pneigh_get_next(seq, pn, pos);
3200 return *pos ? NULL : pn;
3203 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3205 struct neigh_seq_state *state = seq->private;
3207 loff_t idxpos = *pos;
3209 rc = neigh_get_idx(seq, &idxpos);
3210 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3211 rc = pneigh_get_idx(seq, &idxpos);
3216 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3217 __acquires(tbl->lock)
3220 struct neigh_seq_state *state = seq->private;
3224 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3227 state->nht = rcu_dereference_bh(tbl->nht);
3228 read_lock(&tbl->lock);
3230 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3232 EXPORT_SYMBOL(neigh_seq_start);
3234 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3236 struct neigh_seq_state *state;
3239 if (v == SEQ_START_TOKEN) {
3240 rc = neigh_get_first(seq);
3244 state = seq->private;
3245 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3246 rc = neigh_get_next(seq, v, NULL);
3249 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3250 rc = pneigh_get_first(seq);
3252 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3253 rc = pneigh_get_next(seq, v, NULL);
3259 EXPORT_SYMBOL(neigh_seq_next);
3261 void neigh_seq_stop(struct seq_file *seq, void *v)
3262 __releases(tbl->lock)
3265 struct neigh_seq_state *state = seq->private;
3266 struct neigh_table *tbl = state->tbl;
3268 read_unlock(&tbl->lock);
3269 rcu_read_unlock_bh();
3271 EXPORT_SYMBOL(neigh_seq_stop);
3273 /* statistics via seq_file */
3275 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3277 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3281 return SEQ_START_TOKEN;
3283 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3284 if (!cpu_possible(cpu))
3287 return per_cpu_ptr(tbl->stats, cpu);
3292 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3294 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3297 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3298 if (!cpu_possible(cpu))
3301 return per_cpu_ptr(tbl->stats, cpu);
3307 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3312 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3314 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3315 struct neigh_statistics *st = v;
3317 if (v == SEQ_START_TOKEN) {
3318 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 table_fulls\n");
3322 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3323 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
3324 atomic_read(&tbl->entries),
3335 st->rcv_probes_mcast,
3336 st->rcv_probes_ucast,
3338 st->periodic_gc_runs,
3347 static const struct seq_operations neigh_stat_seq_ops = {
3348 .start = neigh_stat_seq_start,
3349 .next = neigh_stat_seq_next,
3350 .stop = neigh_stat_seq_stop,
3351 .show = neigh_stat_seq_show,
3353 #endif /* CONFIG_PROC_FS */
3355 static void __neigh_notify(struct neighbour *n, int type, int flags,
3358 struct net *net = dev_net(n->dev);
3359 struct sk_buff *skb;
3362 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3366 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3368 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3369 WARN_ON(err == -EMSGSIZE);
3373 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3377 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3380 void neigh_app_ns(struct neighbour *n)
3382 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3384 EXPORT_SYMBOL(neigh_app_ns);
3386 #ifdef CONFIG_SYSCTL
3387 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3389 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3390 void *buffer, size_t *lenp, loff_t *ppos)
3393 struct ctl_table tmp = *ctl;
3395 tmp.extra1 = SYSCTL_ZERO;
3396 tmp.extra2 = &unres_qlen_max;
3399 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3400 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3403 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3407 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3412 return __in_dev_arp_parms_get_rcu(dev);
3414 return __in6_dev_nd_parms_get_rcu(dev);
3419 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3422 struct net_device *dev;
3423 int family = neigh_parms_family(p);
3426 for_each_netdev_rcu(net, dev) {
3427 struct neigh_parms *dst_p =
3428 neigh_get_dev_parms_rcu(dev, family);
3430 if (dst_p && !test_bit(index, dst_p->data_state))
3431 dst_p->data[index] = p->data[index];
3436 static void neigh_proc_update(struct ctl_table *ctl, int write)
3438 struct net_device *dev = ctl->extra1;
3439 struct neigh_parms *p = ctl->extra2;
3440 struct net *net = neigh_parms_net(p);
3441 int index = (int *) ctl->data - p->data;
3446 set_bit(index, p->data_state);
3447 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3448 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3449 if (!dev) /* NULL dev means this is default value */
3450 neigh_copy_dflt_parms(net, p, index);
3453 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3454 void *buffer, size_t *lenp,
3457 struct ctl_table tmp = *ctl;
3460 tmp.extra1 = SYSCTL_ZERO;
3461 tmp.extra2 = SYSCTL_INT_MAX;
3463 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3464 neigh_proc_update(ctl, write);
3468 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3469 size_t *lenp, loff_t *ppos)
3471 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3473 neigh_proc_update(ctl, write);
3476 EXPORT_SYMBOL(neigh_proc_dointvec);
3478 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3479 size_t *lenp, loff_t *ppos)
3481 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3483 neigh_proc_update(ctl, write);
3486 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3488 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3489 void *buffer, size_t *lenp,
3492 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3494 neigh_proc_update(ctl, write);
3498 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3499 void *buffer, size_t *lenp, loff_t *ppos)
3501 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3503 neigh_proc_update(ctl, write);
3506 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3508 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3509 void *buffer, size_t *lenp,
3512 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3514 neigh_proc_update(ctl, write);
3518 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3519 void *buffer, size_t *lenp,
3522 struct neigh_parms *p = ctl->extra2;
3525 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3526 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3527 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3528 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3532 if (write && ret == 0) {
3533 /* update reachable_time as well, otherwise, the change will
3534 * only be effective after the next time neigh_periodic_work
3535 * decides to recompute it
3538 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3543 #define NEIGH_PARMS_DATA_OFFSET(index) \
3544 (&((struct neigh_parms *) 0)->data[index])
3546 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3547 [NEIGH_VAR_ ## attr] = { \
3549 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3550 .maxlen = sizeof(int), \
3552 .proc_handler = proc, \
3555 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3556 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3558 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3559 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3561 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3562 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3564 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3565 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3567 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3568 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3570 static struct neigh_sysctl_table {
3571 struct ctl_table_header *sysctl_header;
3572 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3573 } neigh_sysctl_template __read_mostly = {
3575 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3576 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3577 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3578 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3579 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3580 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3581 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3582 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3583 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3584 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3585 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3586 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3587 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3588 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3589 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3590 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3591 [NEIGH_VAR_GC_INTERVAL] = {
3592 .procname = "gc_interval",
3593 .maxlen = sizeof(int),
3595 .proc_handler = proc_dointvec_jiffies,
3597 [NEIGH_VAR_GC_THRESH1] = {
3598 .procname = "gc_thresh1",
3599 .maxlen = sizeof(int),
3601 .extra1 = SYSCTL_ZERO,
3602 .extra2 = SYSCTL_INT_MAX,
3603 .proc_handler = proc_dointvec_minmax,
3605 [NEIGH_VAR_GC_THRESH2] = {
3606 .procname = "gc_thresh2",
3607 .maxlen = sizeof(int),
3609 .extra1 = SYSCTL_ZERO,
3610 .extra2 = SYSCTL_INT_MAX,
3611 .proc_handler = proc_dointvec_minmax,
3613 [NEIGH_VAR_GC_THRESH3] = {
3614 .procname = "gc_thresh3",
3615 .maxlen = sizeof(int),
3617 .extra1 = SYSCTL_ZERO,
3618 .extra2 = SYSCTL_INT_MAX,
3619 .proc_handler = proc_dointvec_minmax,
3625 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3626 proc_handler *handler)
3629 struct neigh_sysctl_table *t;
3630 const char *dev_name_source;
3631 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3634 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3638 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3639 t->neigh_vars[i].data += (long) p;
3640 t->neigh_vars[i].extra1 = dev;
3641 t->neigh_vars[i].extra2 = p;
3645 dev_name_source = dev->name;
3646 /* Terminate the table early */
3647 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3648 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3650 struct neigh_table *tbl = p->tbl;
3651 dev_name_source = "default";
3652 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3653 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3654 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3655 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3660 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3662 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3663 /* RetransTime (in milliseconds)*/
3664 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3665 /* ReachableTime (in milliseconds) */
3666 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3668 /* Those handlers will update p->reachable_time after
3669 * base_reachable_time(_ms) is set to ensure the new timer starts being
3670 * applied after the next neighbour update instead of waiting for
3671 * neigh_periodic_work to update its value (can be multiple minutes)
3672 * So any handler that replaces them should do this as well
3675 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3676 neigh_proc_base_reachable_time;
3677 /* ReachableTime (in milliseconds) */
3678 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3679 neigh_proc_base_reachable_time;
3682 /* Don't export sysctls to unprivileged users */
3683 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3684 t->neigh_vars[0].procname = NULL;
3686 switch (neigh_parms_family(p)) {
3697 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3698 p_name, dev_name_source);
3700 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3701 if (!t->sysctl_header)
3704 p->sysctl_table = t;
3712 EXPORT_SYMBOL(neigh_sysctl_register);
3714 void neigh_sysctl_unregister(struct neigh_parms *p)
3716 if (p->sysctl_table) {
3717 struct neigh_sysctl_table *t = p->sysctl_table;
3718 p->sysctl_table = NULL;
3719 unregister_net_sysctl_table(t->sysctl_header);
3723 EXPORT_SYMBOL(neigh_sysctl_unregister);
3725 #endif /* CONFIG_SYSCTL */
3727 static int __init neigh_init(void)
3729 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3730 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3731 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3733 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3735 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3740 subsys_initcall(neigh_init);