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_in_range(n->updated, tref, jiffies))
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)
264 /* Use safe distance from the jiffies - LONG_MAX point while timer
265 * is running in DELAY/PROBE state but still show to user space
266 * large times in the past.
268 unsigned long mint = jiffies - (LONG_MAX - 86400 * HZ);
271 if (!time_in_range(n->confirmed, mint, jiffies))
273 if (time_before(n->used, n->confirmed))
274 n->used = n->confirmed;
275 if (unlikely(mod_timer(&n->timer, when))) {
276 printk("NEIGH: BUG, double timer add, state is %x\n",
282 static int neigh_del_timer(struct neighbour *n)
284 if ((n->nud_state & NUD_IN_TIMER) &&
285 del_timer(&n->timer)) {
292 static void pneigh_queue_purge(struct sk_buff_head *list, struct net *net)
294 struct sk_buff_head tmp;
298 skb_queue_head_init(&tmp);
299 spin_lock_irqsave(&list->lock, flags);
300 skb = skb_peek(list);
301 while (skb != NULL) {
302 struct sk_buff *skb_next = skb_peek_next(skb, list);
303 if (net == NULL || net_eq(dev_net(skb->dev), net)) {
304 __skb_unlink(skb, list);
305 __skb_queue_tail(&tmp, skb);
309 spin_unlock_irqrestore(&list->lock, flags);
311 while ((skb = __skb_dequeue(&tmp))) {
317 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
321 struct neigh_hash_table *nht;
323 nht = rcu_dereference_protected(tbl->nht,
324 lockdep_is_held(&tbl->lock));
326 for (i = 0; i < (1 << nht->hash_shift); i++) {
328 struct neighbour __rcu **np = &nht->hash_buckets[i];
330 while ((n = rcu_dereference_protected(*np,
331 lockdep_is_held(&tbl->lock))) != NULL) {
332 if (dev && n->dev != dev) {
336 if (skip_perm && n->nud_state & NUD_PERMANENT) {
340 rcu_assign_pointer(*np,
341 rcu_dereference_protected(n->next,
342 lockdep_is_held(&tbl->lock)));
343 write_lock(&n->lock);
346 if (refcount_read(&n->refcnt) != 1) {
347 /* The most unpleasant situation.
348 We must destroy neighbour entry,
349 but someone still uses it.
351 The destroy will be delayed until
352 the last user releases us, but
353 we must kill timers etc. and move
356 __skb_queue_purge(&n->arp_queue);
357 n->arp_queue_len_bytes = 0;
358 n->output = neigh_blackhole;
359 if (n->nud_state & NUD_VALID)
360 n->nud_state = NUD_NOARP;
362 n->nud_state = NUD_NONE;
363 neigh_dbg(2, "neigh %p is stray\n", n);
365 write_unlock(&n->lock);
366 neigh_cleanup_and_release(n);
371 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
373 write_lock_bh(&tbl->lock);
374 neigh_flush_dev(tbl, dev, false);
375 write_unlock_bh(&tbl->lock);
377 EXPORT_SYMBOL(neigh_changeaddr);
379 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
382 write_lock_bh(&tbl->lock);
383 neigh_flush_dev(tbl, dev, skip_perm);
384 pneigh_ifdown_and_unlock(tbl, dev);
385 pneigh_queue_purge(&tbl->proxy_queue, dev ? dev_net(dev) : NULL);
386 if (skb_queue_empty_lockless(&tbl->proxy_queue))
387 del_timer_sync(&tbl->proxy_timer);
391 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
393 __neigh_ifdown(tbl, dev, true);
396 EXPORT_SYMBOL(neigh_carrier_down);
398 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
400 __neigh_ifdown(tbl, dev, false);
403 EXPORT_SYMBOL(neigh_ifdown);
405 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
406 struct net_device *dev,
407 u8 flags, bool exempt_from_gc)
409 struct neighbour *n = NULL;
410 unsigned long now = jiffies;
416 entries = atomic_inc_return(&tbl->gc_entries) - 1;
417 if (entries >= tbl->gc_thresh3 ||
418 (entries >= tbl->gc_thresh2 &&
419 time_after(now, tbl->last_flush + 5 * HZ))) {
420 if (!neigh_forced_gc(tbl) &&
421 entries >= tbl->gc_thresh3) {
422 net_info_ratelimited("%s: neighbor table overflow!\n",
424 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
430 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
434 __skb_queue_head_init(&n->arp_queue);
435 rwlock_init(&n->lock);
436 seqlock_init(&n->ha_lock);
437 n->updated = n->used = now;
438 n->nud_state = NUD_NONE;
439 n->output = neigh_blackhole;
441 seqlock_init(&n->hh.hh_lock);
442 n->parms = neigh_parms_clone(&tbl->parms);
443 timer_setup(&n->timer, neigh_timer_handler, 0);
445 NEIGH_CACHE_STAT_INC(tbl, allocs);
447 refcount_set(&n->refcnt, 1);
449 INIT_LIST_HEAD(&n->gc_list);
451 atomic_inc(&tbl->entries);
457 atomic_dec(&tbl->gc_entries);
461 static void neigh_get_hash_rnd(u32 *x)
463 *x = get_random_u32() | 1;
466 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
468 size_t size = (1 << shift) * sizeof(struct neighbour *);
469 struct neigh_hash_table *ret;
470 struct neighbour __rcu **buckets;
473 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
476 if (size <= PAGE_SIZE) {
477 buckets = kzalloc(size, GFP_ATOMIC);
479 buckets = (struct neighbour __rcu **)
480 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
482 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
488 ret->hash_buckets = buckets;
489 ret->hash_shift = shift;
490 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
491 neigh_get_hash_rnd(&ret->hash_rnd[i]);
495 static void neigh_hash_free_rcu(struct rcu_head *head)
497 struct neigh_hash_table *nht = container_of(head,
498 struct neigh_hash_table,
500 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
501 struct neighbour __rcu **buckets = nht->hash_buckets;
503 if (size <= PAGE_SIZE) {
506 kmemleak_free(buckets);
507 free_pages((unsigned long)buckets, get_order(size));
512 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
513 unsigned long new_shift)
515 unsigned int i, hash;
516 struct neigh_hash_table *new_nht, *old_nht;
518 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
520 old_nht = rcu_dereference_protected(tbl->nht,
521 lockdep_is_held(&tbl->lock));
522 new_nht = neigh_hash_alloc(new_shift);
526 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
527 struct neighbour *n, *next;
529 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
530 lockdep_is_held(&tbl->lock));
533 hash = tbl->hash(n->primary_key, n->dev,
536 hash >>= (32 - new_nht->hash_shift);
537 next = rcu_dereference_protected(n->next,
538 lockdep_is_held(&tbl->lock));
540 rcu_assign_pointer(n->next,
541 rcu_dereference_protected(
542 new_nht->hash_buckets[hash],
543 lockdep_is_held(&tbl->lock)));
544 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
548 rcu_assign_pointer(tbl->nht, new_nht);
549 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
553 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
554 struct net_device *dev)
558 NEIGH_CACHE_STAT_INC(tbl, lookups);
561 n = __neigh_lookup_noref(tbl, pkey, dev);
563 if (!refcount_inc_not_zero(&n->refcnt))
565 NEIGH_CACHE_STAT_INC(tbl, hits);
568 rcu_read_unlock_bh();
571 EXPORT_SYMBOL(neigh_lookup);
573 static struct neighbour *
574 ___neigh_create(struct neigh_table *tbl, const void *pkey,
575 struct net_device *dev, u8 flags,
576 bool exempt_from_gc, bool want_ref)
578 u32 hash_val, key_len = tbl->key_len;
579 struct neighbour *n1, *rc, *n;
580 struct neigh_hash_table *nht;
583 n = neigh_alloc(tbl, dev, flags, exempt_from_gc);
584 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
586 rc = ERR_PTR(-ENOBUFS);
590 memcpy(n->primary_key, pkey, key_len);
594 /* Protocol specific setup. */
595 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
597 goto out_neigh_release;
600 if (dev->netdev_ops->ndo_neigh_construct) {
601 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
604 goto out_neigh_release;
608 /* Device specific setup. */
609 if (n->parms->neigh_setup &&
610 (error = n->parms->neigh_setup(n)) < 0) {
612 goto out_neigh_release;
615 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
617 write_lock_bh(&tbl->lock);
618 nht = rcu_dereference_protected(tbl->nht,
619 lockdep_is_held(&tbl->lock));
621 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
622 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
624 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
626 if (n->parms->dead) {
627 rc = ERR_PTR(-EINVAL);
631 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
632 lockdep_is_held(&tbl->lock));
634 n1 = rcu_dereference_protected(n1->next,
635 lockdep_is_held(&tbl->lock))) {
636 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
646 list_add_tail(&n->gc_list, &n->tbl->gc_list);
650 rcu_assign_pointer(n->next,
651 rcu_dereference_protected(nht->hash_buckets[hash_val],
652 lockdep_is_held(&tbl->lock)));
653 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
654 write_unlock_bh(&tbl->lock);
655 neigh_dbg(2, "neigh %p is created\n", n);
660 write_unlock_bh(&tbl->lock);
663 atomic_dec(&tbl->gc_entries);
668 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
669 struct net_device *dev, bool want_ref)
671 return ___neigh_create(tbl, pkey, dev, 0, false, want_ref);
673 EXPORT_SYMBOL(__neigh_create);
675 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
677 u32 hash_val = *(u32 *)(pkey + key_len - 4);
678 hash_val ^= (hash_val >> 16);
679 hash_val ^= hash_val >> 8;
680 hash_val ^= hash_val >> 4;
681 hash_val &= PNEIGH_HASHMASK;
685 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
688 unsigned int key_len,
689 struct net_device *dev)
692 if (!memcmp(n->key, pkey, key_len) &&
693 net_eq(pneigh_net(n), net) &&
694 (n->dev == dev || !n->dev))
701 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
702 struct net *net, const void *pkey, struct net_device *dev)
704 unsigned int key_len = tbl->key_len;
705 u32 hash_val = pneigh_hash(pkey, key_len);
707 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
708 net, pkey, key_len, dev);
710 EXPORT_SYMBOL_GPL(__pneigh_lookup);
712 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
713 struct net *net, const void *pkey,
714 struct net_device *dev, int creat)
716 struct pneigh_entry *n;
717 unsigned int key_len = tbl->key_len;
718 u32 hash_val = pneigh_hash(pkey, key_len);
720 read_lock_bh(&tbl->lock);
721 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
722 net, pkey, key_len, dev);
723 read_unlock_bh(&tbl->lock);
730 n = kzalloc(sizeof(*n) + key_len, GFP_KERNEL);
734 write_pnet(&n->net, net);
735 memcpy(n->key, pkey, key_len);
739 if (tbl->pconstructor && tbl->pconstructor(n)) {
746 write_lock_bh(&tbl->lock);
747 n->next = tbl->phash_buckets[hash_val];
748 tbl->phash_buckets[hash_val] = n;
749 write_unlock_bh(&tbl->lock);
753 EXPORT_SYMBOL(pneigh_lookup);
756 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
757 struct net_device *dev)
759 struct pneigh_entry *n, **np;
760 unsigned int key_len = tbl->key_len;
761 u32 hash_val = pneigh_hash(pkey, key_len);
763 write_lock_bh(&tbl->lock);
764 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
766 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
767 net_eq(pneigh_net(n), net)) {
769 write_unlock_bh(&tbl->lock);
770 if (tbl->pdestructor)
777 write_unlock_bh(&tbl->lock);
781 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
782 struct net_device *dev)
784 struct pneigh_entry *n, **np, *freelist = NULL;
787 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
788 np = &tbl->phash_buckets[h];
789 while ((n = *np) != NULL) {
790 if (!dev || n->dev == dev) {
799 write_unlock_bh(&tbl->lock);
800 while ((n = freelist)) {
803 if (tbl->pdestructor)
811 static void neigh_parms_destroy(struct neigh_parms *parms);
813 static inline void neigh_parms_put(struct neigh_parms *parms)
815 if (refcount_dec_and_test(&parms->refcnt))
816 neigh_parms_destroy(parms);
820 * neighbour must already be out of the table;
823 void neigh_destroy(struct neighbour *neigh)
825 struct net_device *dev = neigh->dev;
827 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
830 pr_warn("Destroying alive neighbour %p\n", neigh);
835 if (neigh_del_timer(neigh))
836 pr_warn("Impossible event\n");
838 write_lock_bh(&neigh->lock);
839 __skb_queue_purge(&neigh->arp_queue);
840 write_unlock_bh(&neigh->lock);
841 neigh->arp_queue_len_bytes = 0;
843 if (dev->netdev_ops->ndo_neigh_destroy)
844 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
847 neigh_parms_put(neigh->parms);
849 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
851 atomic_dec(&neigh->tbl->entries);
852 kfree_rcu(neigh, rcu);
854 EXPORT_SYMBOL(neigh_destroy);
856 /* Neighbour state is suspicious;
859 Called with write_locked neigh.
861 static void neigh_suspect(struct neighbour *neigh)
863 neigh_dbg(2, "neigh %p is suspected\n", neigh);
865 neigh->output = neigh->ops->output;
868 /* Neighbour state is OK;
871 Called with write_locked neigh.
873 static void neigh_connect(struct neighbour *neigh)
875 neigh_dbg(2, "neigh %p is connected\n", neigh);
877 neigh->output = neigh->ops->connected_output;
880 static void neigh_periodic_work(struct work_struct *work)
882 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
884 struct neighbour __rcu **np;
886 struct neigh_hash_table *nht;
888 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
890 write_lock_bh(&tbl->lock);
891 nht = rcu_dereference_protected(tbl->nht,
892 lockdep_is_held(&tbl->lock));
895 * periodically recompute ReachableTime from random function
898 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
899 struct neigh_parms *p;
900 tbl->last_rand = jiffies;
901 list_for_each_entry(p, &tbl->parms_list, list)
903 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
906 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
909 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
910 np = &nht->hash_buckets[i];
912 while ((n = rcu_dereference_protected(*np,
913 lockdep_is_held(&tbl->lock))) != NULL) {
916 write_lock(&n->lock);
918 state = n->nud_state;
919 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
920 (n->flags & NTF_EXT_LEARNED)) {
921 write_unlock(&n->lock);
925 if (time_before(n->used, n->confirmed) &&
926 time_is_before_eq_jiffies(n->confirmed))
927 n->used = n->confirmed;
929 if (refcount_read(&n->refcnt) == 1 &&
930 (state == NUD_FAILED ||
931 !time_in_range_open(jiffies, n->used,
932 n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
933 rcu_assign_pointer(*np,
934 rcu_dereference_protected(n->next,
935 lockdep_is_held(&tbl->lock)));
937 write_unlock(&n->lock);
938 neigh_cleanup_and_release(n);
941 write_unlock(&n->lock);
947 * It's fine to release lock here, even if hash table
948 * grows while we are preempted.
950 write_unlock_bh(&tbl->lock);
952 write_lock_bh(&tbl->lock);
953 nht = rcu_dereference_protected(tbl->nht,
954 lockdep_is_held(&tbl->lock));
957 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
958 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
959 * BASE_REACHABLE_TIME.
961 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
962 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
963 write_unlock_bh(&tbl->lock);
966 static __inline__ int neigh_max_probes(struct neighbour *n)
968 struct neigh_parms *p = n->parms;
969 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
970 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
971 NEIGH_VAR(p, MCAST_PROBES));
974 static void neigh_invalidate(struct neighbour *neigh)
975 __releases(neigh->lock)
976 __acquires(neigh->lock)
980 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
981 neigh_dbg(2, "neigh %p is failed\n", neigh);
982 neigh->updated = jiffies;
984 /* It is very thin place. report_unreachable is very complicated
985 routine. Particularly, it can hit the same neighbour entry!
987 So that, we try to be accurate and avoid dead loop. --ANK
989 while (neigh->nud_state == NUD_FAILED &&
990 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
991 write_unlock(&neigh->lock);
992 neigh->ops->error_report(neigh, skb);
993 write_lock(&neigh->lock);
995 __skb_queue_purge(&neigh->arp_queue);
996 neigh->arp_queue_len_bytes = 0;
999 static void neigh_probe(struct neighbour *neigh)
1000 __releases(neigh->lock)
1002 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1003 /* keep skb alive even if arp_queue overflows */
1005 skb = skb_clone(skb, GFP_ATOMIC);
1006 write_unlock(&neigh->lock);
1007 if (neigh->ops->solicit)
1008 neigh->ops->solicit(neigh, skb);
1009 atomic_inc(&neigh->probes);
1013 /* Called when a timer expires for a neighbour entry. */
1015 static void neigh_timer_handler(struct timer_list *t)
1017 unsigned long now, next;
1018 struct neighbour *neigh = from_timer(neigh, t, timer);
1022 write_lock(&neigh->lock);
1024 state = neigh->nud_state;
1028 if (!(state & NUD_IN_TIMER))
1031 if (state & NUD_REACHABLE) {
1032 if (time_before_eq(now,
1033 neigh->confirmed + neigh->parms->reachable_time)) {
1034 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1035 next = neigh->confirmed + neigh->parms->reachable_time;
1036 } else if (time_before_eq(now,
1038 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1039 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1040 neigh->nud_state = NUD_DELAY;
1041 neigh->updated = jiffies;
1042 neigh_suspect(neigh);
1043 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1045 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1046 neigh->nud_state = NUD_STALE;
1047 neigh->updated = jiffies;
1048 neigh_suspect(neigh);
1051 } else if (state & NUD_DELAY) {
1052 if (time_before_eq(now,
1054 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1055 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1056 neigh->nud_state = NUD_REACHABLE;
1057 neigh->updated = jiffies;
1058 neigh_connect(neigh);
1060 next = neigh->confirmed + neigh->parms->reachable_time;
1062 neigh_dbg(2, "neigh %p is probed\n", neigh);
1063 neigh->nud_state = NUD_PROBE;
1064 neigh->updated = jiffies;
1065 atomic_set(&neigh->probes, 0);
1067 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1071 /* NUD_PROBE|NUD_INCOMPLETE */
1072 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1075 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1076 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1077 neigh->nud_state = NUD_FAILED;
1079 neigh_invalidate(neigh);
1083 if (neigh->nud_state & NUD_IN_TIMER) {
1084 if (time_before(next, jiffies + HZ/100))
1085 next = jiffies + HZ/100;
1086 if (!mod_timer(&neigh->timer, next))
1089 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1093 write_unlock(&neigh->lock);
1097 neigh_update_notify(neigh, 0);
1099 trace_neigh_timer_handler(neigh, 0);
1101 neigh_release(neigh);
1104 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1107 bool immediate_probe = false;
1109 write_lock_bh(&neigh->lock);
1112 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1117 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1118 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1119 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1120 unsigned long next, now = jiffies;
1122 atomic_set(&neigh->probes,
1123 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1124 neigh_del_timer(neigh);
1125 neigh->nud_state = NUD_INCOMPLETE;
1126 neigh->updated = now;
1127 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1129 neigh_add_timer(neigh, next);
1130 immediate_probe = true;
1132 neigh->nud_state = NUD_FAILED;
1133 neigh->updated = jiffies;
1134 write_unlock_bh(&neigh->lock);
1139 } else if (neigh->nud_state & NUD_STALE) {
1140 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1141 neigh_del_timer(neigh);
1142 neigh->nud_state = NUD_DELAY;
1143 neigh->updated = jiffies;
1144 neigh_add_timer(neigh, jiffies +
1145 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1148 if (neigh->nud_state == NUD_INCOMPLETE) {
1150 while (neigh->arp_queue_len_bytes + skb->truesize >
1151 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1152 struct sk_buff *buff;
1154 buff = __skb_dequeue(&neigh->arp_queue);
1157 neigh->arp_queue_len_bytes -= buff->truesize;
1159 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1162 __skb_queue_tail(&neigh->arp_queue, skb);
1163 neigh->arp_queue_len_bytes += skb->truesize;
1168 if (immediate_probe)
1171 write_unlock(&neigh->lock);
1173 trace_neigh_event_send_done(neigh, rc);
1177 if (neigh->nud_state & NUD_STALE)
1179 write_unlock_bh(&neigh->lock);
1181 trace_neigh_event_send_dead(neigh, 1);
1184 EXPORT_SYMBOL(__neigh_event_send);
1186 static void neigh_update_hhs(struct neighbour *neigh)
1188 struct hh_cache *hh;
1189 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1192 if (neigh->dev->header_ops)
1193 update = neigh->dev->header_ops->cache_update;
1197 if (READ_ONCE(hh->hh_len)) {
1198 write_seqlock_bh(&hh->hh_lock);
1199 update(hh, neigh->dev, neigh->ha);
1200 write_sequnlock_bh(&hh->hh_lock);
1207 /* Generic update routine.
1208 -- lladdr is new lladdr or NULL, if it is not supplied.
1209 -- new is new state.
1211 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1213 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1214 lladdr instead of overriding it
1216 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1217 NEIGH_UPDATE_F_USE means that the entry is user triggered.
1218 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1220 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1223 Caller MUST hold reference count on the entry.
1226 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1227 u8 new, u32 flags, u32 nlmsg_pid,
1228 struct netlink_ext_ack *extack)
1230 bool ext_learn_change = false;
1234 struct net_device *dev;
1235 int update_isrouter = 0;
1237 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1239 write_lock_bh(&neigh->lock);
1242 old = neigh->nud_state;
1246 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1250 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1251 (old & (NUD_NOARP | NUD_PERMANENT)))
1254 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1255 if (flags & NEIGH_UPDATE_F_USE) {
1256 new = old & ~NUD_PERMANENT;
1257 neigh->nud_state = new;
1262 if (!(new & NUD_VALID)) {
1263 neigh_del_timer(neigh);
1264 if (old & NUD_CONNECTED)
1265 neigh_suspect(neigh);
1266 neigh->nud_state = new;
1268 notify = old & NUD_VALID;
1269 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1270 (new & NUD_FAILED)) {
1271 neigh_invalidate(neigh);
1277 /* Compare new lladdr with cached one */
1278 if (!dev->addr_len) {
1279 /* First case: device needs no address. */
1281 } else if (lladdr) {
1282 /* The second case: if something is already cached
1283 and a new address is proposed:
1285 - if they are different, check override flag
1287 if ((old & NUD_VALID) &&
1288 !memcmp(lladdr, neigh->ha, dev->addr_len))
1291 /* No address is supplied; if we know something,
1292 use it, otherwise discard the request.
1295 if (!(old & NUD_VALID)) {
1296 NL_SET_ERR_MSG(extack, "No link layer address given");
1302 /* Update confirmed timestamp for neighbour entry after we
1303 * received ARP packet even if it doesn't change IP to MAC binding.
1305 if (new & NUD_CONNECTED)
1306 neigh->confirmed = jiffies;
1308 /* If entry was valid and address is not changed,
1309 do not change entry state, if new one is STALE.
1312 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1313 if (old & NUD_VALID) {
1314 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1315 update_isrouter = 0;
1316 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1317 (old & NUD_CONNECTED)) {
1323 if (lladdr == neigh->ha && new == NUD_STALE &&
1324 !(flags & NEIGH_UPDATE_F_ADMIN))
1329 /* Update timestamp only once we know we will make a change to the
1330 * neighbour entry. Otherwise we risk to move the locktime window with
1331 * noop updates and ignore relevant ARP updates.
1333 if (new != old || lladdr != neigh->ha)
1334 neigh->updated = jiffies;
1337 neigh_del_timer(neigh);
1338 if (new & NUD_PROBE)
1339 atomic_set(&neigh->probes, 0);
1340 if (new & NUD_IN_TIMER)
1341 neigh_add_timer(neigh, (jiffies +
1342 ((new & NUD_REACHABLE) ?
1343 neigh->parms->reachable_time :
1345 neigh->nud_state = new;
1349 if (lladdr != neigh->ha) {
1350 write_seqlock(&neigh->ha_lock);
1351 memcpy(&neigh->ha, lladdr, dev->addr_len);
1352 write_sequnlock(&neigh->ha_lock);
1353 neigh_update_hhs(neigh);
1354 if (!(new & NUD_CONNECTED))
1355 neigh->confirmed = jiffies -
1356 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1361 if (new & NUD_CONNECTED)
1362 neigh_connect(neigh);
1364 neigh_suspect(neigh);
1365 if (!(old & NUD_VALID)) {
1366 struct sk_buff *skb;
1368 /* Again: avoid dead loop if something went wrong */
1370 while (neigh->nud_state & NUD_VALID &&
1371 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1372 struct dst_entry *dst = skb_dst(skb);
1373 struct neighbour *n2, *n1 = neigh;
1374 write_unlock_bh(&neigh->lock);
1378 /* Why not just use 'neigh' as-is? The problem is that
1379 * things such as shaper, eql, and sch_teql can end up
1380 * using alternative, different, neigh objects to output
1381 * the packet in the output path. So what we need to do
1382 * here is re-lookup the top-level neigh in the path so
1383 * we can reinject the packet there.
1386 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1387 n2 = dst_neigh_lookup_skb(dst, skb);
1391 n1->output(n1, skb);
1396 write_lock_bh(&neigh->lock);
1398 __skb_queue_purge(&neigh->arp_queue);
1399 neigh->arp_queue_len_bytes = 0;
1402 if (update_isrouter)
1403 neigh_update_is_router(neigh, flags, ¬ify);
1404 write_unlock_bh(&neigh->lock);
1406 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1407 neigh_update_gc_list(neigh);
1410 neigh_update_notify(neigh, nlmsg_pid);
1412 trace_neigh_update_done(neigh, err);
1417 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1418 u32 flags, u32 nlmsg_pid)
1420 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1422 EXPORT_SYMBOL(neigh_update);
1424 /* Update the neigh to listen temporarily for probe responses, even if it is
1425 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1427 void __neigh_set_probe_once(struct neighbour *neigh)
1431 neigh->updated = jiffies;
1432 if (!(neigh->nud_state & NUD_FAILED))
1434 neigh->nud_state = NUD_INCOMPLETE;
1435 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1436 neigh_add_timer(neigh,
1437 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1440 EXPORT_SYMBOL(__neigh_set_probe_once);
1442 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1443 u8 *lladdr, void *saddr,
1444 struct net_device *dev)
1446 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1447 lladdr || !dev->addr_len);
1449 neigh_update(neigh, lladdr, NUD_STALE,
1450 NEIGH_UPDATE_F_OVERRIDE, 0);
1453 EXPORT_SYMBOL(neigh_event_ns);
1455 /* called with read_lock_bh(&n->lock); */
1456 static void neigh_hh_init(struct neighbour *n)
1458 struct net_device *dev = n->dev;
1459 __be16 prot = n->tbl->protocol;
1460 struct hh_cache *hh = &n->hh;
1462 write_lock_bh(&n->lock);
1464 /* Only one thread can come in here and initialize the
1468 dev->header_ops->cache(n, hh, prot);
1470 write_unlock_bh(&n->lock);
1473 /* Slow and careful. */
1475 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1479 if (!neigh_event_send(neigh, skb)) {
1481 struct net_device *dev = neigh->dev;
1484 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1485 neigh_hh_init(neigh);
1488 __skb_pull(skb, skb_network_offset(skb));
1489 seq = read_seqbegin(&neigh->ha_lock);
1490 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1491 neigh->ha, NULL, skb->len);
1492 } while (read_seqretry(&neigh->ha_lock, seq));
1495 rc = dev_queue_xmit(skb);
1506 EXPORT_SYMBOL(neigh_resolve_output);
1508 /* As fast as possible without hh cache */
1510 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1512 struct net_device *dev = neigh->dev;
1517 __skb_pull(skb, skb_network_offset(skb));
1518 seq = read_seqbegin(&neigh->ha_lock);
1519 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1520 neigh->ha, NULL, skb->len);
1521 } while (read_seqretry(&neigh->ha_lock, seq));
1524 err = dev_queue_xmit(skb);
1531 EXPORT_SYMBOL(neigh_connected_output);
1533 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1535 return dev_queue_xmit(skb);
1537 EXPORT_SYMBOL(neigh_direct_output);
1539 static void neigh_proxy_process(struct timer_list *t)
1541 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1542 long sched_next = 0;
1543 unsigned long now = jiffies;
1544 struct sk_buff *skb, *n;
1546 spin_lock(&tbl->proxy_queue.lock);
1548 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1549 long tdif = NEIGH_CB(skb)->sched_next - now;
1552 struct net_device *dev = skb->dev;
1554 __skb_unlink(skb, &tbl->proxy_queue);
1555 if (tbl->proxy_redo && netif_running(dev)) {
1557 tbl->proxy_redo(skb);
1564 } else if (!sched_next || tdif < sched_next)
1567 del_timer(&tbl->proxy_timer);
1569 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1570 spin_unlock(&tbl->proxy_queue.lock);
1573 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1574 struct sk_buff *skb)
1576 unsigned long sched_next = jiffies +
1577 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1579 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1584 NEIGH_CB(skb)->sched_next = sched_next;
1585 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1587 spin_lock(&tbl->proxy_queue.lock);
1588 if (del_timer(&tbl->proxy_timer)) {
1589 if (time_before(tbl->proxy_timer.expires, sched_next))
1590 sched_next = tbl->proxy_timer.expires;
1594 __skb_queue_tail(&tbl->proxy_queue, skb);
1595 mod_timer(&tbl->proxy_timer, sched_next);
1596 spin_unlock(&tbl->proxy_queue.lock);
1598 EXPORT_SYMBOL(pneigh_enqueue);
1600 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1601 struct net *net, int ifindex)
1603 struct neigh_parms *p;
1605 list_for_each_entry(p, &tbl->parms_list, list) {
1606 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1607 (!p->dev && !ifindex && net_eq(net, &init_net)))
1614 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1615 struct neigh_table *tbl)
1617 struct neigh_parms *p;
1618 struct net *net = dev_net(dev);
1619 const struct net_device_ops *ops = dev->netdev_ops;
1621 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1624 refcount_set(&p->refcnt, 1);
1626 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1629 write_pnet(&p->net, net);
1630 p->sysctl_table = NULL;
1632 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1638 write_lock_bh(&tbl->lock);
1639 list_add(&p->list, &tbl->parms.list);
1640 write_unlock_bh(&tbl->lock);
1642 neigh_parms_data_state_cleanall(p);
1646 EXPORT_SYMBOL(neigh_parms_alloc);
1648 static void neigh_rcu_free_parms(struct rcu_head *head)
1650 struct neigh_parms *parms =
1651 container_of(head, struct neigh_parms, rcu_head);
1653 neigh_parms_put(parms);
1656 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1658 if (!parms || parms == &tbl->parms)
1660 write_lock_bh(&tbl->lock);
1661 list_del(&parms->list);
1663 write_unlock_bh(&tbl->lock);
1664 dev_put(parms->dev);
1665 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1667 EXPORT_SYMBOL(neigh_parms_release);
1669 static void neigh_parms_destroy(struct neigh_parms *parms)
1674 static struct lock_class_key neigh_table_proxy_queue_class;
1676 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1678 void neigh_table_init(int index, struct neigh_table *tbl)
1680 unsigned long now = jiffies;
1681 unsigned long phsize;
1683 INIT_LIST_HEAD(&tbl->parms_list);
1684 INIT_LIST_HEAD(&tbl->gc_list);
1685 list_add(&tbl->parms.list, &tbl->parms_list);
1686 write_pnet(&tbl->parms.net, &init_net);
1687 refcount_set(&tbl->parms.refcnt, 1);
1688 tbl->parms.reachable_time =
1689 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1691 tbl->stats = alloc_percpu(struct neigh_statistics);
1693 panic("cannot create neighbour cache statistics");
1695 #ifdef CONFIG_PROC_FS
1696 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1697 &neigh_stat_seq_ops, tbl))
1698 panic("cannot create neighbour proc dir entry");
1701 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1703 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1704 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1706 if (!tbl->nht || !tbl->phash_buckets)
1707 panic("cannot allocate neighbour cache hashes");
1709 if (!tbl->entry_size)
1710 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1711 tbl->key_len, NEIGH_PRIV_ALIGN);
1713 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1715 rwlock_init(&tbl->lock);
1716 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1717 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1718 tbl->parms.reachable_time);
1719 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1720 skb_queue_head_init_class(&tbl->proxy_queue,
1721 &neigh_table_proxy_queue_class);
1723 tbl->last_flush = now;
1724 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1726 neigh_tables[index] = tbl;
1728 EXPORT_SYMBOL(neigh_table_init);
1730 int neigh_table_clear(int index, struct neigh_table *tbl)
1732 neigh_tables[index] = NULL;
1733 /* It is not clean... Fix it to unload IPv6 module safely */
1734 cancel_delayed_work_sync(&tbl->gc_work);
1735 del_timer_sync(&tbl->proxy_timer);
1736 pneigh_queue_purge(&tbl->proxy_queue, NULL);
1737 neigh_ifdown(tbl, NULL);
1738 if (atomic_read(&tbl->entries))
1739 pr_crit("neighbour leakage\n");
1741 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1742 neigh_hash_free_rcu);
1745 kfree(tbl->phash_buckets);
1746 tbl->phash_buckets = NULL;
1748 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1750 free_percpu(tbl->stats);
1755 EXPORT_SYMBOL(neigh_table_clear);
1757 static struct neigh_table *neigh_find_table(int family)
1759 struct neigh_table *tbl = NULL;
1763 tbl = neigh_tables[NEIGH_ARP_TABLE];
1766 tbl = neigh_tables[NEIGH_ND_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,
1946 ndm->ndm_flags & NTF_EXT_LEARNED,
1947 exempt_from_gc, true);
1948 if (IS_ERR(neigh)) {
1949 err = PTR_ERR(neigh);
1953 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1955 neigh_release(neigh);
1959 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1960 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1961 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1965 neigh->protocol = protocol;
1966 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1967 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1968 if (ndm->ndm_flags & NTF_ROUTER)
1969 flags |= NEIGH_UPDATE_F_ISROUTER;
1970 if (ndm->ndm_flags & NTF_USE)
1971 flags |= NEIGH_UPDATE_F_USE;
1973 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1974 NETLINK_CB(skb).portid, extack);
1975 if (!err && ndm->ndm_flags & NTF_USE) {
1976 neigh_event_send(neigh, NULL);
1979 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);