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);
134 /* remove from the gc list if new state is permanent or if neighbor
135 * is externally learned; otherwise entry should be on the gc list
137 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
138 n->flags & NTF_EXT_LEARNED;
139 on_gc_list = !list_empty(&n->gc_list);
141 if (exempt_from_gc && on_gc_list) {
142 list_del_init(&n->gc_list);
143 atomic_dec(&n->tbl->gc_entries);
144 } else if (!exempt_from_gc && !on_gc_list) {
145 /* add entries to the tail; cleaning removes from the front */
146 list_add_tail(&n->gc_list, &n->tbl->gc_list);
147 atomic_inc(&n->tbl->gc_entries);
150 write_unlock(&n->lock);
151 write_unlock_bh(&n->tbl->lock);
154 static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
160 if (!(flags & NEIGH_UPDATE_F_ADMIN))
163 ndm_flags = (flags & NEIGH_UPDATE_F_EXT_LEARNED) ? NTF_EXT_LEARNED : 0;
164 if ((neigh->flags ^ ndm_flags) & NTF_EXT_LEARNED) {
165 if (ndm_flags & NTF_EXT_LEARNED)
166 neigh->flags |= NTF_EXT_LEARNED;
168 neigh->flags &= ~NTF_EXT_LEARNED;
176 static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
177 struct neigh_table *tbl)
181 write_lock(&n->lock);
182 if (refcount_read(&n->refcnt) == 1) {
183 struct neighbour *neigh;
185 neigh = rcu_dereference_protected(n->next,
186 lockdep_is_held(&tbl->lock));
187 rcu_assign_pointer(*np, neigh);
191 write_unlock(&n->lock);
193 neigh_cleanup_and_release(n);
197 bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
199 struct neigh_hash_table *nht;
200 void *pkey = ndel->primary_key;
203 struct neighbour __rcu **np;
205 nht = rcu_dereference_protected(tbl->nht,
206 lockdep_is_held(&tbl->lock));
207 hash_val = tbl->hash(pkey, ndel->dev, nht->hash_rnd);
208 hash_val = hash_val >> (32 - nht->hash_shift);
210 np = &nht->hash_buckets[hash_val];
211 while ((n = rcu_dereference_protected(*np,
212 lockdep_is_held(&tbl->lock)))) {
214 return neigh_del(n, np, tbl);
220 static int neigh_forced_gc(struct neigh_table *tbl)
222 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
223 unsigned long tref = jiffies - 5 * HZ;
224 struct neighbour *n, *tmp;
227 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
229 write_lock_bh(&tbl->lock);
231 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
232 if (refcount_read(&n->refcnt) == 1) {
235 write_lock(&n->lock);
236 if ((n->nud_state == NUD_FAILED) ||
237 (tbl->is_multicast &&
238 tbl->is_multicast(n->primary_key)) ||
239 time_after(tref, n->updated))
241 write_unlock(&n->lock);
243 if (remove && neigh_remove_one(n, tbl))
245 if (shrunk >= max_clean)
250 tbl->last_flush = jiffies;
252 write_unlock_bh(&tbl->lock);
257 static void neigh_add_timer(struct neighbour *n, unsigned long when)
260 if (unlikely(mod_timer(&n->timer, when))) {
261 printk("NEIGH: BUG, double timer add, state is %x\n",
267 static int neigh_del_timer(struct neighbour *n)
269 if ((n->nud_state & NUD_IN_TIMER) &&
270 del_timer(&n->timer)) {
277 static void pneigh_queue_purge(struct sk_buff_head *list)
281 while ((skb = skb_dequeue(list)) != NULL) {
287 static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
291 struct neigh_hash_table *nht;
293 nht = rcu_dereference_protected(tbl->nht,
294 lockdep_is_held(&tbl->lock));
296 for (i = 0; i < (1 << nht->hash_shift); i++) {
298 struct neighbour __rcu **np = &nht->hash_buckets[i];
300 while ((n = rcu_dereference_protected(*np,
301 lockdep_is_held(&tbl->lock))) != NULL) {
302 if (dev && n->dev != dev) {
306 if (skip_perm && n->nud_state & NUD_PERMANENT) {
310 rcu_assign_pointer(*np,
311 rcu_dereference_protected(n->next,
312 lockdep_is_held(&tbl->lock)));
313 write_lock(&n->lock);
316 if (refcount_read(&n->refcnt) != 1) {
317 /* The most unpleasant situation.
318 We must destroy neighbour entry,
319 but someone still uses it.
321 The destroy will be delayed until
322 the last user releases us, but
323 we must kill timers etc. and move
326 __skb_queue_purge(&n->arp_queue);
327 n->arp_queue_len_bytes = 0;
328 n->output = neigh_blackhole;
329 if (n->nud_state & NUD_VALID)
330 n->nud_state = NUD_NOARP;
332 n->nud_state = NUD_NONE;
333 neigh_dbg(2, "neigh %p is stray\n", n);
335 write_unlock(&n->lock);
336 neigh_cleanup_and_release(n);
341 void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
343 write_lock_bh(&tbl->lock);
344 neigh_flush_dev(tbl, dev, false);
345 write_unlock_bh(&tbl->lock);
347 EXPORT_SYMBOL(neigh_changeaddr);
349 static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
352 write_lock_bh(&tbl->lock);
353 neigh_flush_dev(tbl, dev, skip_perm);
354 pneigh_ifdown_and_unlock(tbl, dev);
356 del_timer_sync(&tbl->proxy_timer);
357 pneigh_queue_purge(&tbl->proxy_queue);
361 int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
363 __neigh_ifdown(tbl, dev, true);
366 EXPORT_SYMBOL(neigh_carrier_down);
368 int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
370 __neigh_ifdown(tbl, dev, false);
373 EXPORT_SYMBOL(neigh_ifdown);
375 static struct neighbour *neigh_alloc(struct neigh_table *tbl,
376 struct net_device *dev,
379 struct neighbour *n = NULL;
380 unsigned long now = jiffies;
386 entries = atomic_inc_return(&tbl->gc_entries) - 1;
387 if (entries >= tbl->gc_thresh3 ||
388 (entries >= tbl->gc_thresh2 &&
389 time_after(now, tbl->last_flush + 5 * HZ))) {
390 if (!neigh_forced_gc(tbl) &&
391 entries >= tbl->gc_thresh3) {
392 net_info_ratelimited("%s: neighbor table overflow!\n",
394 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
400 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
404 __skb_queue_head_init(&n->arp_queue);
405 rwlock_init(&n->lock);
406 seqlock_init(&n->ha_lock);
407 n->updated = n->used = now;
408 n->nud_state = NUD_NONE;
409 n->output = neigh_blackhole;
410 seqlock_init(&n->hh.hh_lock);
411 n->parms = neigh_parms_clone(&tbl->parms);
412 timer_setup(&n->timer, neigh_timer_handler, 0);
414 NEIGH_CACHE_STAT_INC(tbl, allocs);
416 refcount_set(&n->refcnt, 1);
418 INIT_LIST_HEAD(&n->gc_list);
420 atomic_inc(&tbl->entries);
426 atomic_dec(&tbl->gc_entries);
430 static void neigh_get_hash_rnd(u32 *x)
432 *x = get_random_u32() | 1;
435 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
437 size_t size = (1 << shift) * sizeof(struct neighbour *);
438 struct neigh_hash_table *ret;
439 struct neighbour __rcu **buckets;
442 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
445 if (size <= PAGE_SIZE) {
446 buckets = kzalloc(size, GFP_ATOMIC);
448 buckets = (struct neighbour __rcu **)
449 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
451 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
457 ret->hash_buckets = buckets;
458 ret->hash_shift = shift;
459 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
460 neigh_get_hash_rnd(&ret->hash_rnd[i]);
464 static void neigh_hash_free_rcu(struct rcu_head *head)
466 struct neigh_hash_table *nht = container_of(head,
467 struct neigh_hash_table,
469 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
470 struct neighbour __rcu **buckets = nht->hash_buckets;
472 if (size <= PAGE_SIZE) {
475 kmemleak_free(buckets);
476 free_pages((unsigned long)buckets, get_order(size));
481 static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
482 unsigned long new_shift)
484 unsigned int i, hash;
485 struct neigh_hash_table *new_nht, *old_nht;
487 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
489 old_nht = rcu_dereference_protected(tbl->nht,
490 lockdep_is_held(&tbl->lock));
491 new_nht = neigh_hash_alloc(new_shift);
495 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
496 struct neighbour *n, *next;
498 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
499 lockdep_is_held(&tbl->lock));
502 hash = tbl->hash(n->primary_key, n->dev,
505 hash >>= (32 - new_nht->hash_shift);
506 next = rcu_dereference_protected(n->next,
507 lockdep_is_held(&tbl->lock));
509 rcu_assign_pointer(n->next,
510 rcu_dereference_protected(
511 new_nht->hash_buckets[hash],
512 lockdep_is_held(&tbl->lock)));
513 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
517 rcu_assign_pointer(tbl->nht, new_nht);
518 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
522 struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
523 struct net_device *dev)
527 NEIGH_CACHE_STAT_INC(tbl, lookups);
530 n = __neigh_lookup_noref(tbl, pkey, dev);
532 if (!refcount_inc_not_zero(&n->refcnt))
534 NEIGH_CACHE_STAT_INC(tbl, hits);
537 rcu_read_unlock_bh();
540 EXPORT_SYMBOL(neigh_lookup);
542 struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
546 unsigned int key_len = tbl->key_len;
548 struct neigh_hash_table *nht;
550 NEIGH_CACHE_STAT_INC(tbl, lookups);
553 nht = rcu_dereference_bh(tbl->nht);
554 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
556 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
558 n = rcu_dereference_bh(n->next)) {
559 if (!memcmp(n->primary_key, pkey, key_len) &&
560 net_eq(dev_net(n->dev), net)) {
561 if (!refcount_inc_not_zero(&n->refcnt))
563 NEIGH_CACHE_STAT_INC(tbl, hits);
568 rcu_read_unlock_bh();
571 EXPORT_SYMBOL(neigh_lookup_nodev);
573 static struct neighbour *___neigh_create(struct neigh_table *tbl,
575 struct net_device *dev,
576 bool exempt_from_gc, bool want_ref)
578 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
580 unsigned int key_len = tbl->key_len;
582 struct neigh_hash_table *nht;
584 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
587 rc = ERR_PTR(-ENOBUFS);
591 memcpy(n->primary_key, pkey, key_len);
595 /* Protocol specific setup. */
596 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
598 goto out_neigh_release;
601 if (dev->netdev_ops->ndo_neigh_construct) {
602 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
605 goto out_neigh_release;
609 /* Device specific setup. */
610 if (n->parms->neigh_setup &&
611 (error = n->parms->neigh_setup(n)) < 0) {
613 goto out_neigh_release;
616 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
618 write_lock_bh(&tbl->lock);
619 nht = rcu_dereference_protected(tbl->nht,
620 lockdep_is_held(&tbl->lock));
622 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
623 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
625 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
627 if (n->parms->dead) {
628 rc = ERR_PTR(-EINVAL);
632 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
633 lockdep_is_held(&tbl->lock));
635 n1 = rcu_dereference_protected(n1->next,
636 lockdep_is_held(&tbl->lock))) {
637 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
647 list_add_tail(&n->gc_list, &n->tbl->gc_list);
651 rcu_assign_pointer(n->next,
652 rcu_dereference_protected(nht->hash_buckets[hash_val],
653 lockdep_is_held(&tbl->lock)));
654 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
655 write_unlock_bh(&tbl->lock);
656 neigh_dbg(2, "neigh %p is created\n", n);
661 write_unlock_bh(&tbl->lock);
664 atomic_dec(&tbl->gc_entries);
669 struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
670 struct net_device *dev, bool want_ref)
672 return ___neigh_create(tbl, pkey, dev, false, want_ref);
674 EXPORT_SYMBOL(__neigh_create);
676 static u32 pneigh_hash(const void *pkey, unsigned int key_len)
678 u32 hash_val = *(u32 *)(pkey + key_len - 4);
679 hash_val ^= (hash_val >> 16);
680 hash_val ^= hash_val >> 8;
681 hash_val ^= hash_val >> 4;
682 hash_val &= PNEIGH_HASHMASK;
686 static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
689 unsigned int key_len,
690 struct net_device *dev)
693 if (!memcmp(n->key, pkey, key_len) &&
694 net_eq(pneigh_net(n), net) &&
695 (n->dev == dev || !n->dev))
702 struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
703 struct net *net, const void *pkey, struct net_device *dev)
705 unsigned int key_len = tbl->key_len;
706 u32 hash_val = pneigh_hash(pkey, key_len);
708 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
709 net, pkey, key_len, dev);
711 EXPORT_SYMBOL_GPL(__pneigh_lookup);
713 struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
714 struct net *net, const void *pkey,
715 struct net_device *dev, int creat)
717 struct pneigh_entry *n;
718 unsigned int key_len = tbl->key_len;
719 u32 hash_val = pneigh_hash(pkey, key_len);
721 read_lock_bh(&tbl->lock);
722 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
723 net, pkey, key_len, dev);
724 read_unlock_bh(&tbl->lock);
731 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
736 write_pnet(&n->net, net);
737 memcpy(n->key, pkey, key_len);
742 if (tbl->pconstructor && tbl->pconstructor(n)) {
750 write_lock_bh(&tbl->lock);
751 n->next = tbl->phash_buckets[hash_val];
752 tbl->phash_buckets[hash_val] = n;
753 write_unlock_bh(&tbl->lock);
757 EXPORT_SYMBOL(pneigh_lookup);
760 int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
761 struct net_device *dev)
763 struct pneigh_entry *n, **np;
764 unsigned int key_len = tbl->key_len;
765 u32 hash_val = pneigh_hash(pkey, key_len);
767 write_lock_bh(&tbl->lock);
768 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
770 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
771 net_eq(pneigh_net(n), net)) {
773 write_unlock_bh(&tbl->lock);
774 if (tbl->pdestructor)
782 write_unlock_bh(&tbl->lock);
786 static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
787 struct net_device *dev)
789 struct pneigh_entry *n, **np, *freelist = NULL;
792 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
793 np = &tbl->phash_buckets[h];
794 while ((n = *np) != NULL) {
795 if (!dev || n->dev == dev) {
804 write_unlock_bh(&tbl->lock);
805 while ((n = freelist)) {
808 if (tbl->pdestructor)
817 static void neigh_parms_destroy(struct neigh_parms *parms);
819 static inline void neigh_parms_put(struct neigh_parms *parms)
821 if (refcount_dec_and_test(&parms->refcnt))
822 neigh_parms_destroy(parms);
826 * neighbour must already be out of the table;
829 void neigh_destroy(struct neighbour *neigh)
831 struct net_device *dev = neigh->dev;
833 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
836 pr_warn("Destroying alive neighbour %p\n", neigh);
841 if (neigh_del_timer(neigh))
842 pr_warn("Impossible event\n");
844 write_lock_bh(&neigh->lock);
845 __skb_queue_purge(&neigh->arp_queue);
846 write_unlock_bh(&neigh->lock);
847 neigh->arp_queue_len_bytes = 0;
849 if (dev->netdev_ops->ndo_neigh_destroy)
850 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
853 neigh_parms_put(neigh->parms);
855 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
857 atomic_dec(&neigh->tbl->entries);
858 kfree_rcu(neigh, rcu);
860 EXPORT_SYMBOL(neigh_destroy);
862 /* Neighbour state is suspicious;
865 Called with write_locked neigh.
867 static void neigh_suspect(struct neighbour *neigh)
869 neigh_dbg(2, "neigh %p is suspected\n", neigh);
871 neigh->output = neigh->ops->output;
874 /* Neighbour state is OK;
877 Called with write_locked neigh.
879 static void neigh_connect(struct neighbour *neigh)
881 neigh_dbg(2, "neigh %p is connected\n", neigh);
883 neigh->output = neigh->ops->connected_output;
886 static void neigh_periodic_work(struct work_struct *work)
888 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
890 struct neighbour __rcu **np;
892 struct neigh_hash_table *nht;
894 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
896 write_lock_bh(&tbl->lock);
897 nht = rcu_dereference_protected(tbl->nht,
898 lockdep_is_held(&tbl->lock));
901 * periodically recompute ReachableTime from random function
904 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
905 struct neigh_parms *p;
906 tbl->last_rand = jiffies;
907 list_for_each_entry(p, &tbl->parms_list, list)
909 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
912 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
915 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
916 np = &nht->hash_buckets[i];
918 while ((n = rcu_dereference_protected(*np,
919 lockdep_is_held(&tbl->lock))) != NULL) {
922 write_lock(&n->lock);
924 state = n->nud_state;
925 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
926 (n->flags & NTF_EXT_LEARNED)) {
927 write_unlock(&n->lock);
931 if (time_before(n->used, n->confirmed))
932 n->used = n->confirmed;
934 if (refcount_read(&n->refcnt) == 1 &&
935 (state == NUD_FAILED ||
936 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
939 write_unlock(&n->lock);
940 neigh_cleanup_and_release(n);
943 write_unlock(&n->lock);
949 * It's fine to release lock here, even if hash table
950 * grows while we are preempted.
952 write_unlock_bh(&tbl->lock);
954 write_lock_bh(&tbl->lock);
955 nht = rcu_dereference_protected(tbl->nht,
956 lockdep_is_held(&tbl->lock));
959 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
960 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
961 * BASE_REACHABLE_TIME.
963 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
964 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
965 write_unlock_bh(&tbl->lock);
968 static __inline__ int neigh_max_probes(struct neighbour *n)
970 struct neigh_parms *p = n->parms;
971 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
972 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
973 NEIGH_VAR(p, MCAST_PROBES));
976 static void neigh_invalidate(struct neighbour *neigh)
977 __releases(neigh->lock)
978 __acquires(neigh->lock)
982 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
983 neigh_dbg(2, "neigh %p is failed\n", neigh);
984 neigh->updated = jiffies;
986 /* It is very thin place. report_unreachable is very complicated
987 routine. Particularly, it can hit the same neighbour entry!
989 So that, we try to be accurate and avoid dead loop. --ANK
991 while (neigh->nud_state == NUD_FAILED &&
992 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
993 write_unlock(&neigh->lock);
994 neigh->ops->error_report(neigh, skb);
995 write_lock(&neigh->lock);
997 __skb_queue_purge(&neigh->arp_queue);
998 neigh->arp_queue_len_bytes = 0;
1001 static void neigh_probe(struct neighbour *neigh)
1002 __releases(neigh->lock)
1004 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1005 /* keep skb alive even if arp_queue overflows */
1007 skb = skb_clone(skb, GFP_ATOMIC);
1008 write_unlock(&neigh->lock);
1009 if (neigh->ops->solicit)
1010 neigh->ops->solicit(neigh, skb);
1011 atomic_inc(&neigh->probes);
1015 /* Called when a timer expires for a neighbour entry. */
1017 static void neigh_timer_handler(struct timer_list *t)
1019 unsigned long now, next;
1020 struct neighbour *neigh = from_timer(neigh, t, timer);
1024 write_lock(&neigh->lock);
1026 state = neigh->nud_state;
1030 if (!(state & NUD_IN_TIMER))
1033 if (state & NUD_REACHABLE) {
1034 if (time_before_eq(now,
1035 neigh->confirmed + neigh->parms->reachable_time)) {
1036 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1037 next = neigh->confirmed + neigh->parms->reachable_time;
1038 } else if (time_before_eq(now,
1040 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1041 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1042 neigh->nud_state = NUD_DELAY;
1043 neigh->updated = jiffies;
1044 neigh_suspect(neigh);
1045 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1047 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1048 neigh->nud_state = NUD_STALE;
1049 neigh->updated = jiffies;
1050 neigh_suspect(neigh);
1053 } else if (state & NUD_DELAY) {
1054 if (time_before_eq(now,
1056 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1057 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1058 neigh->nud_state = NUD_REACHABLE;
1059 neigh->updated = jiffies;
1060 neigh_connect(neigh);
1062 next = neigh->confirmed + neigh->parms->reachable_time;
1064 neigh_dbg(2, "neigh %p is probed\n", neigh);
1065 neigh->nud_state = NUD_PROBE;
1066 neigh->updated = jiffies;
1067 atomic_set(&neigh->probes, 0);
1069 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1073 /* NUD_PROBE|NUD_INCOMPLETE */
1074 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1077 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1078 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1079 neigh->nud_state = NUD_FAILED;
1081 neigh_invalidate(neigh);
1085 if (neigh->nud_state & NUD_IN_TIMER) {
1086 if (time_before(next, jiffies + HZ/100))
1087 next = jiffies + HZ/100;
1088 if (!mod_timer(&neigh->timer, next))
1091 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1095 write_unlock(&neigh->lock);
1099 neigh_update_notify(neigh, 0);
1101 trace_neigh_timer_handler(neigh, 0);
1103 neigh_release(neigh);
1106 int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1109 bool immediate_probe = false;
1111 write_lock_bh(&neigh->lock);
1114 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1119 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1120 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1121 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1122 unsigned long next, now = jiffies;
1124 atomic_set(&neigh->probes,
1125 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1126 neigh_del_timer(neigh);
1127 neigh->nud_state = NUD_INCOMPLETE;
1128 neigh->updated = now;
1129 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1131 neigh_add_timer(neigh, next);
1132 immediate_probe = true;
1134 neigh->nud_state = NUD_FAILED;
1135 neigh->updated = jiffies;
1136 write_unlock_bh(&neigh->lock);
1141 } else if (neigh->nud_state & NUD_STALE) {
1142 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1143 neigh_del_timer(neigh);
1144 neigh->nud_state = NUD_DELAY;
1145 neigh->updated = jiffies;
1146 neigh_add_timer(neigh, jiffies +
1147 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1150 if (neigh->nud_state == NUD_INCOMPLETE) {
1152 while (neigh->arp_queue_len_bytes + skb->truesize >
1153 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1154 struct sk_buff *buff;
1156 buff = __skb_dequeue(&neigh->arp_queue);
1159 neigh->arp_queue_len_bytes -= buff->truesize;
1161 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1164 __skb_queue_tail(&neigh->arp_queue, skb);
1165 neigh->arp_queue_len_bytes += skb->truesize;
1170 if (immediate_probe)
1173 write_unlock(&neigh->lock);
1175 trace_neigh_event_send_done(neigh, rc);
1179 if (neigh->nud_state & NUD_STALE)
1181 write_unlock_bh(&neigh->lock);
1183 trace_neigh_event_send_dead(neigh, 1);
1186 EXPORT_SYMBOL(__neigh_event_send);
1188 static void neigh_update_hhs(struct neighbour *neigh)
1190 struct hh_cache *hh;
1191 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1194 if (neigh->dev->header_ops)
1195 update = neigh->dev->header_ops->cache_update;
1199 if (READ_ONCE(hh->hh_len)) {
1200 write_seqlock_bh(&hh->hh_lock);
1201 update(hh, neigh->dev, neigh->ha);
1202 write_sequnlock_bh(&hh->hh_lock);
1209 /* Generic update routine.
1210 -- lladdr is new lladdr or NULL, if it is not supplied.
1211 -- new is new state.
1213 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1215 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1216 lladdr instead of overriding it
1218 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1220 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1222 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1225 Caller MUST hold reference count on the entry.
1228 static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1229 u8 new, u32 flags, u32 nlmsg_pid,
1230 struct netlink_ext_ack *extack)
1232 bool ext_learn_change = false;
1236 struct net_device *dev;
1237 int update_isrouter = 0;
1239 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1241 write_lock_bh(&neigh->lock);
1244 old = neigh->nud_state;
1248 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1252 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1253 (old & (NUD_NOARP | NUD_PERMANENT)))
1256 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1258 if (!(new & NUD_VALID)) {
1259 neigh_del_timer(neigh);
1260 if (old & NUD_CONNECTED)
1261 neigh_suspect(neigh);
1262 neigh->nud_state = new;
1264 notify = old & NUD_VALID;
1265 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1266 (new & NUD_FAILED)) {
1267 neigh_invalidate(neigh);
1273 /* Compare new lladdr with cached one */
1274 if (!dev->addr_len) {
1275 /* First case: device needs no address. */
1277 } else if (lladdr) {
1278 /* The second case: if something is already cached
1279 and a new address is proposed:
1281 - if they are different, check override flag
1283 if ((old & NUD_VALID) &&
1284 !memcmp(lladdr, neigh->ha, dev->addr_len))
1287 /* No address is supplied; if we know something,
1288 use it, otherwise discard the request.
1291 if (!(old & NUD_VALID)) {
1292 NL_SET_ERR_MSG(extack, "No link layer address given");
1298 /* Update confirmed timestamp for neighbour entry after we
1299 * received ARP packet even if it doesn't change IP to MAC binding.
1301 if (new & NUD_CONNECTED)
1302 neigh->confirmed = jiffies;
1304 /* If entry was valid and address is not changed,
1305 do not change entry state, if new one is STALE.
1308 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1309 if (old & NUD_VALID) {
1310 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1311 update_isrouter = 0;
1312 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1313 (old & NUD_CONNECTED)) {
1319 if (lladdr == neigh->ha && new == NUD_STALE &&
1320 !(flags & NEIGH_UPDATE_F_ADMIN))
1325 /* Update timestamp only once we know we will make a change to the
1326 * neighbour entry. Otherwise we risk to move the locktime window with
1327 * noop updates and ignore relevant ARP updates.
1329 if (new != old || lladdr != neigh->ha)
1330 neigh->updated = jiffies;
1333 neigh_del_timer(neigh);
1334 if (new & NUD_PROBE)
1335 atomic_set(&neigh->probes, 0);
1336 if (new & NUD_IN_TIMER)
1337 neigh_add_timer(neigh, (jiffies +
1338 ((new & NUD_REACHABLE) ?
1339 neigh->parms->reachable_time :
1341 neigh->nud_state = new;
1345 if (lladdr != neigh->ha) {
1346 write_seqlock(&neigh->ha_lock);
1347 memcpy(&neigh->ha, lladdr, dev->addr_len);
1348 write_sequnlock(&neigh->ha_lock);
1349 neigh_update_hhs(neigh);
1350 if (!(new & NUD_CONNECTED))
1351 neigh->confirmed = jiffies -
1352 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1357 if (new & NUD_CONNECTED)
1358 neigh_connect(neigh);
1360 neigh_suspect(neigh);
1361 if (!(old & NUD_VALID)) {
1362 struct sk_buff *skb;
1364 /* Again: avoid dead loop if something went wrong */
1366 while (neigh->nud_state & NUD_VALID &&
1367 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1368 struct dst_entry *dst = skb_dst(skb);
1369 struct neighbour *n2, *n1 = neigh;
1370 write_unlock_bh(&neigh->lock);
1374 /* Why not just use 'neigh' as-is? The problem is that
1375 * things such as shaper, eql, and sch_teql can end up
1376 * using alternative, different, neigh objects to output
1377 * the packet in the output path. So what we need to do
1378 * here is re-lookup the top-level neigh in the path so
1379 * we can reinject the packet there.
1383 n2 = dst_neigh_lookup_skb(dst, skb);
1387 n1->output(n1, skb);
1392 write_lock_bh(&neigh->lock);
1394 __skb_queue_purge(&neigh->arp_queue);
1395 neigh->arp_queue_len_bytes = 0;
1398 if (update_isrouter)
1399 neigh_update_is_router(neigh, flags, ¬ify);
1400 write_unlock_bh(&neigh->lock);
1402 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1403 neigh_update_gc_list(neigh);
1406 neigh_update_notify(neigh, nlmsg_pid);
1408 trace_neigh_update_done(neigh, err);
1413 int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1414 u32 flags, u32 nlmsg_pid)
1416 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1418 EXPORT_SYMBOL(neigh_update);
1420 /* Update the neigh to listen temporarily for probe responses, even if it is
1421 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1423 void __neigh_set_probe_once(struct neighbour *neigh)
1427 neigh->updated = jiffies;
1428 if (!(neigh->nud_state & NUD_FAILED))
1430 neigh->nud_state = NUD_INCOMPLETE;
1431 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1432 neigh_add_timer(neigh,
1433 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1436 EXPORT_SYMBOL(__neigh_set_probe_once);
1438 struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1439 u8 *lladdr, void *saddr,
1440 struct net_device *dev)
1442 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1443 lladdr || !dev->addr_len);
1445 neigh_update(neigh, lladdr, NUD_STALE,
1446 NEIGH_UPDATE_F_OVERRIDE, 0);
1449 EXPORT_SYMBOL(neigh_event_ns);
1451 /* called with read_lock_bh(&n->lock); */
1452 static void neigh_hh_init(struct neighbour *n)
1454 struct net_device *dev = n->dev;
1455 __be16 prot = n->tbl->protocol;
1456 struct hh_cache *hh = &n->hh;
1458 write_lock_bh(&n->lock);
1460 /* Only one thread can come in here and initialize the
1464 dev->header_ops->cache(n, hh, prot);
1466 write_unlock_bh(&n->lock);
1469 /* Slow and careful. */
1471 int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1475 if (!neigh_event_send(neigh, skb)) {
1477 struct net_device *dev = neigh->dev;
1480 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1481 neigh_hh_init(neigh);
1484 __skb_pull(skb, skb_network_offset(skb));
1485 seq = read_seqbegin(&neigh->ha_lock);
1486 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1487 neigh->ha, NULL, skb->len);
1488 } while (read_seqretry(&neigh->ha_lock, seq));
1491 rc = dev_queue_xmit(skb);
1502 EXPORT_SYMBOL(neigh_resolve_output);
1504 /* As fast as possible without hh cache */
1506 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1508 struct net_device *dev = neigh->dev;
1513 __skb_pull(skb, skb_network_offset(skb));
1514 seq = read_seqbegin(&neigh->ha_lock);
1515 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1516 neigh->ha, NULL, skb->len);
1517 } while (read_seqretry(&neigh->ha_lock, seq));
1520 err = dev_queue_xmit(skb);
1527 EXPORT_SYMBOL(neigh_connected_output);
1529 int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1531 return dev_queue_xmit(skb);
1533 EXPORT_SYMBOL(neigh_direct_output);
1535 static void neigh_proxy_process(struct timer_list *t)
1537 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1538 long sched_next = 0;
1539 unsigned long now = jiffies;
1540 struct sk_buff *skb, *n;
1542 spin_lock(&tbl->proxy_queue.lock);
1544 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1545 long tdif = NEIGH_CB(skb)->sched_next - now;
1548 struct net_device *dev = skb->dev;
1550 __skb_unlink(skb, &tbl->proxy_queue);
1551 if (tbl->proxy_redo && netif_running(dev)) {
1553 tbl->proxy_redo(skb);
1560 } else if (!sched_next || tdif < sched_next)
1563 del_timer(&tbl->proxy_timer);
1565 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1566 spin_unlock(&tbl->proxy_queue.lock);
1569 void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1570 struct sk_buff *skb)
1572 unsigned long sched_next = jiffies +
1573 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1575 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1580 NEIGH_CB(skb)->sched_next = sched_next;
1581 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1583 spin_lock(&tbl->proxy_queue.lock);
1584 if (del_timer(&tbl->proxy_timer)) {
1585 if (time_before(tbl->proxy_timer.expires, sched_next))
1586 sched_next = tbl->proxy_timer.expires;
1590 __skb_queue_tail(&tbl->proxy_queue, skb);
1591 mod_timer(&tbl->proxy_timer, sched_next);
1592 spin_unlock(&tbl->proxy_queue.lock);
1594 EXPORT_SYMBOL(pneigh_enqueue);
1596 static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1597 struct net *net, int ifindex)
1599 struct neigh_parms *p;
1601 list_for_each_entry(p, &tbl->parms_list, list) {
1602 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1603 (!p->dev && !ifindex && net_eq(net, &init_net)))
1610 struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1611 struct neigh_table *tbl)
1613 struct neigh_parms *p;
1614 struct net *net = dev_net(dev);
1615 const struct net_device_ops *ops = dev->netdev_ops;
1617 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1620 refcount_set(&p->refcnt, 1);
1622 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1625 write_pnet(&p->net, net);
1626 p->sysctl_table = NULL;
1628 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1634 write_lock_bh(&tbl->lock);
1635 list_add(&p->list, &tbl->parms.list);
1636 write_unlock_bh(&tbl->lock);
1638 neigh_parms_data_state_cleanall(p);
1642 EXPORT_SYMBOL(neigh_parms_alloc);
1644 static void neigh_rcu_free_parms(struct rcu_head *head)
1646 struct neigh_parms *parms =
1647 container_of(head, struct neigh_parms, rcu_head);
1649 neigh_parms_put(parms);
1652 void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1654 if (!parms || parms == &tbl->parms)
1656 write_lock_bh(&tbl->lock);
1657 list_del(&parms->list);
1659 write_unlock_bh(&tbl->lock);
1661 dev_put(parms->dev);
1662 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1664 EXPORT_SYMBOL(neigh_parms_release);
1666 static void neigh_parms_destroy(struct neigh_parms *parms)
1671 static struct lock_class_key neigh_table_proxy_queue_class;
1673 static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1675 void neigh_table_init(int index, struct neigh_table *tbl)
1677 unsigned long now = jiffies;
1678 unsigned long phsize;
1680 INIT_LIST_HEAD(&tbl->parms_list);
1681 INIT_LIST_HEAD(&tbl->gc_list);
1682 list_add(&tbl->parms.list, &tbl->parms_list);
1683 write_pnet(&tbl->parms.net, &init_net);
1684 refcount_set(&tbl->parms.refcnt, 1);
1685 tbl->parms.reachable_time =
1686 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1688 tbl->stats = alloc_percpu(struct neigh_statistics);
1690 panic("cannot create neighbour cache statistics");
1692 #ifdef CONFIG_PROC_FS
1693 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1694 &neigh_stat_seq_ops, tbl))
1695 panic("cannot create neighbour proc dir entry");
1698 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1700 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1701 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1703 if (!tbl->nht || !tbl->phash_buckets)
1704 panic("cannot allocate neighbour cache hashes");
1706 if (!tbl->entry_size)
1707 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1708 tbl->key_len, NEIGH_PRIV_ALIGN);
1710 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1712 rwlock_init(&tbl->lock);
1713 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1714 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1715 tbl->parms.reachable_time);
1716 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1717 skb_queue_head_init_class(&tbl->proxy_queue,
1718 &neigh_table_proxy_queue_class);
1720 tbl->last_flush = now;
1721 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1723 neigh_tables[index] = tbl;
1725 EXPORT_SYMBOL(neigh_table_init);
1727 int neigh_table_clear(int index, struct neigh_table *tbl)
1729 neigh_tables[index] = NULL;
1730 /* It is not clean... Fix it to unload IPv6 module safely */
1731 cancel_delayed_work_sync(&tbl->gc_work);
1732 del_timer_sync(&tbl->proxy_timer);
1733 pneigh_queue_purge(&tbl->proxy_queue);
1734 neigh_ifdown(tbl, NULL);
1735 if (atomic_read(&tbl->entries))
1736 pr_crit("neighbour leakage\n");
1738 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1739 neigh_hash_free_rcu);
1742 kfree(tbl->phash_buckets);
1743 tbl->phash_buckets = NULL;
1745 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1747 free_percpu(tbl->stats);
1752 EXPORT_SYMBOL(neigh_table_clear);
1754 static struct neigh_table *neigh_find_table(int family)
1756 struct neigh_table *tbl = NULL;
1760 tbl = neigh_tables[NEIGH_ARP_TABLE];
1763 tbl = neigh_tables[NEIGH_ND_TABLE];
1766 tbl = neigh_tables[NEIGH_DN_TABLE];
1773 const struct nla_policy nda_policy[NDA_MAX+1] = {
1774 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1775 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1776 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1777 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1778 [NDA_PROBES] = { .type = NLA_U32 },
1779 [NDA_VLAN] = { .type = NLA_U16 },
1780 [NDA_PORT] = { .type = NLA_U16 },
1781 [NDA_VNI] = { .type = NLA_U32 },
1782 [NDA_IFINDEX] = { .type = NLA_U32 },
1783 [NDA_MASTER] = { .type = NLA_U32 },
1784 [NDA_PROTOCOL] = { .type = NLA_U8 },
1785 [NDA_NH_ID] = { .type = NLA_U32 },
1786 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1789 static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1790 struct netlink_ext_ack *extack)
1792 struct net *net = sock_net(skb->sk);
1794 struct nlattr *dst_attr;
1795 struct neigh_table *tbl;
1796 struct neighbour *neigh;
1797 struct net_device *dev = NULL;
1801 if (nlmsg_len(nlh) < sizeof(*ndm))
1804 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1806 NL_SET_ERR_MSG(extack, "Network address not specified");
1810 ndm = nlmsg_data(nlh);
1811 if (ndm->ndm_ifindex) {
1812 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1819 tbl = neigh_find_table(ndm->ndm_family);
1821 return -EAFNOSUPPORT;
1823 if (nla_len(dst_attr) < (int)tbl->key_len) {
1824 NL_SET_ERR_MSG(extack, "Invalid network address");
1828 if (ndm->ndm_flags & NTF_PROXY) {
1829 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1836 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1837 if (neigh == NULL) {
1842 err = __neigh_update(neigh, NULL, NUD_FAILED,
1843 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1844 NETLINK_CB(skb).portid, extack);
1845 write_lock_bh(&tbl->lock);
1846 neigh_release(neigh);
1847 neigh_remove_one(neigh, tbl);
1848 write_unlock_bh(&tbl->lock);
1854 static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1855 struct netlink_ext_ack *extack)
1857 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1858 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1859 struct net *net = sock_net(skb->sk);
1861 struct nlattr *tb[NDA_MAX+1];
1862 struct neigh_table *tbl;
1863 struct net_device *dev = NULL;
1864 struct neighbour *neigh;
1870 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1871 nda_policy, extack);
1877 NL_SET_ERR_MSG(extack, "Network address not specified");
1881 ndm = nlmsg_data(nlh);
1882 if (ndm->ndm_ifindex) {
1883 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1889 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1890 NL_SET_ERR_MSG(extack, "Invalid link address");
1895 tbl = neigh_find_table(ndm->ndm_family);
1897 return -EAFNOSUPPORT;
1899 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1900 NL_SET_ERR_MSG(extack, "Invalid network address");
1904 dst = nla_data(tb[NDA_DST]);
1905 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1907 if (tb[NDA_PROTOCOL])
1908 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1910 if (ndm->ndm_flags & NTF_PROXY) {
1911 struct pneigh_entry *pn;
1914 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1916 pn->flags = ndm->ndm_flags;
1918 pn->protocol = protocol;
1925 NL_SET_ERR_MSG(extack, "Device not specified");
1929 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1934 neigh = neigh_lookup(tbl, dst, dev);
1935 if (neigh == NULL) {
1936 bool exempt_from_gc;
1938 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1943 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1944 ndm->ndm_flags & NTF_EXT_LEARNED;
1945 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1946 if (IS_ERR(neigh)) {
1947 err = PTR_ERR(neigh);
1951 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1953 neigh_release(neigh);
1957 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1958 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1959 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1963 neigh->protocol = protocol;
1965 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1966 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1968 if (ndm->ndm_flags & NTF_ROUTER)
1969 flags |= NEIGH_UPDATE_F_ISROUTER;
1971 if (ndm->ndm_flags & NTF_USE) {
1972 neigh_event_send(neigh, NULL);
1975 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1976 NETLINK_CB(skb).portid, extack);
1978 neigh_release(neigh);
1984 static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1986 struct nlattr *nest;
1988 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1993 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1994 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1995 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1996 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1997 /* approximative value for deprecated QUEUE_LEN (in packets) */
1998 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1999 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2000 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2001 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2002 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2003 NEIGH_VAR(parms, UCAST_PROBES)) ||
2004 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2005 NEIGH_VAR(parms, MCAST_PROBES)) ||
2006 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2007 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2008 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2010 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2011 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2012 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2013 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2014 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2015 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2016 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2017 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2018 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2019 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2020 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2021 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2022 nla_put_msecs(skb, NDTPA_LOCKTIME,
2023 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2024 goto nla_put_failure;
2025 return nla_nest_end(skb, nest);
2028 nla_nest_cancel(skb, nest);
2032 static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2033 u32 pid, u32 seq, int type, int flags)
2035 struct nlmsghdr *nlh;
2036 struct ndtmsg *ndtmsg;
2038 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2042 ndtmsg = nlmsg_data(nlh);
2044 read_lock_bh(&tbl->lock);
2045 ndtmsg->ndtm_family = tbl->family;
2046 ndtmsg->ndtm_pad1 = 0;
2047 ndtmsg->ndtm_pad2 = 0;
2049 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2050 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2051 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2052 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2053 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2054 goto nla_put_failure;
2056 unsigned long now = jiffies;
2057 long flush_delta = now - tbl->last_flush;
2058 long rand_delta = now - tbl->last_rand;
2059 struct neigh_hash_table *nht;
2060 struct ndt_config ndc = {
2061 .ndtc_key_len = tbl->key_len,
2062 .ndtc_entry_size = tbl->entry_size,
2063 .ndtc_entries = atomic_read(&tbl->entries),
2064 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2065 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2066 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2070 nht = rcu_dereference_bh(tbl->nht);
2071 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2072 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2073 rcu_read_unlock_bh();
2075 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2076 goto nla_put_failure;
2081 struct ndt_stats ndst;
2083 memset(&ndst, 0, sizeof(ndst));
2085 for_each_possible_cpu(cpu) {
2086 struct neigh_statistics *st;
2088 st = per_cpu_ptr(tbl->stats, cpu);
2089 ndst.ndts_allocs += st->allocs;
2090 ndst.ndts_destroys += st->destroys;
2091 ndst.ndts_hash_grows += st->hash_grows;
2092 ndst.ndts_res_failed += st->res_failed;
2093 ndst.ndts_lookups += st->lookups;
2094 ndst.ndts_hits += st->hits;
2095 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2096 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2097 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2098 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2099 ndst.ndts_table_fulls += st->table_fulls;
2102 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2104 goto nla_put_failure;
2107 BUG_ON(tbl->parms.dev);
2108 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2109 goto nla_put_failure;
2111 read_unlock_bh(&tbl->lock);
2112 nlmsg_end(skb, nlh);
2116 read_unlock_bh(&tbl->lock);
2117 nlmsg_cancel(skb, nlh);
2121 static int neightbl_fill_param_info(struct sk_buff *skb,
2122 struct neigh_table *tbl,
2123 struct neigh_parms *parms,
2124 u32 pid, u32 seq, int type,
2127 struct ndtmsg *ndtmsg;
2128 struct nlmsghdr *nlh;
2130 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2134 ndtmsg = nlmsg_data(nlh);
2136 read_lock_bh(&tbl->lock);
2137 ndtmsg->ndtm_family = tbl->family;
2138 ndtmsg->ndtm_pad1 = 0;
2139 ndtmsg->ndtm_pad2 = 0;
2141 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2142 neightbl_fill_parms(skb, parms) < 0)
2145 read_unlock_bh(&tbl->lock);
2146 nlmsg_end(skb, nlh);
2149 read_unlock_bh(&tbl->lock);
2150 nlmsg_cancel(skb, nlh);
2154 static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2155 [NDTA_NAME] = { .type = NLA_STRING },
2156 [NDTA_THRESH1] = { .type = NLA_U32 },
2157 [NDTA_THRESH2] = { .type = NLA_U32 },
2158 [NDTA_THRESH3] = { .type = NLA_U32 },
2159 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2160 [NDTA_PARMS] = { .type = NLA_NESTED },
2163 static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2164 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2165 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2166 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2167 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2168 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2169 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2170 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2171 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2172 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2173 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2174 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2175 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2176 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2177 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2180 static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2181 struct netlink_ext_ack *extack)
2183 struct net *net = sock_net(skb->sk);
2184 struct neigh_table *tbl;
2185 struct ndtmsg *ndtmsg;
2186 struct nlattr *tb[NDTA_MAX+1];
2190 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2191 nl_neightbl_policy, extack);
2195 if (tb[NDTA_NAME] == NULL) {
2200 ndtmsg = nlmsg_data(nlh);
2202 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2203 tbl = neigh_tables[tidx];
2206 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2208 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2218 * We acquire tbl->lock to be nice to the periodic timers and
2219 * make sure they always see a consistent set of values.
2221 write_lock_bh(&tbl->lock);
2223 if (tb[NDTA_PARMS]) {
2224 struct nlattr *tbp[NDTPA_MAX+1];
2225 struct neigh_parms *p;
2228 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2230 nl_ntbl_parm_policy, extack);
2232 goto errout_tbl_lock;
2234 if (tbp[NDTPA_IFINDEX])
2235 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2237 p = lookup_neigh_parms(tbl, net, ifindex);
2240 goto errout_tbl_lock;
2243 for (i = 1; i <= NDTPA_MAX; i++) {
2248 case NDTPA_QUEUE_LEN:
2249 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2250 nla_get_u32(tbp[i]) *
2251 SKB_TRUESIZE(ETH_FRAME_LEN));
2253 case NDTPA_QUEUE_LENBYTES:
2254 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2255 nla_get_u32(tbp[i]));
2257 case NDTPA_PROXY_QLEN:
2258 NEIGH_VAR_SET(p, PROXY_QLEN,
2259 nla_get_u32(tbp[i]));
2261 case NDTPA_APP_PROBES:
2262 NEIGH_VAR_SET(p, APP_PROBES,
2263 nla_get_u32(tbp[i]));
2265 case NDTPA_UCAST_PROBES:
2266 NEIGH_VAR_SET(p, UCAST_PROBES,
2267 nla_get_u32(tbp[i]));
2269 case NDTPA_MCAST_PROBES:
2270 NEIGH_VAR_SET(p, MCAST_PROBES,
2271 nla_get_u32(tbp[i]));
2273 case NDTPA_MCAST_REPROBES:
2274 NEIGH_VAR_SET(p, MCAST_REPROBES,
2275 nla_get_u32(tbp[i]));
2277 case NDTPA_BASE_REACHABLE_TIME:
2278 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2279 nla_get_msecs(tbp[i]));
2280 /* update reachable_time as well, otherwise, the change will
2281 * only be effective after the next time neigh_periodic_work
2282 * decides to recompute it (can be multiple minutes)
2285 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2287 case NDTPA_GC_STALETIME:
2288 NEIGH_VAR_SET(p, GC_STALETIME,
2289 nla_get_msecs(tbp[i]));
2291 case NDTPA_DELAY_PROBE_TIME:
2292 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2293 nla_get_msecs(tbp[i]));
2294 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2296 case NDTPA_RETRANS_TIME:
2297 NEIGH_VAR_SET(p, RETRANS_TIME,
2298 nla_get_msecs(tbp[i]));
2300 case NDTPA_ANYCAST_DELAY:
2301 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2302 nla_get_msecs(tbp[i]));
2304 case NDTPA_PROXY_DELAY:
2305 NEIGH_VAR_SET(p, PROXY_DELAY,
2306 nla_get_msecs(tbp[i]));
2308 case NDTPA_LOCKTIME:
2309 NEIGH_VAR_SET(p, LOCKTIME,
2310 nla_get_msecs(tbp[i]));
2317 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2318 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2319 !net_eq(net, &init_net))
2320 goto errout_tbl_lock;
2322 if (tb[NDTA_THRESH1])
2323 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2325 if (tb[NDTA_THRESH2])
2326 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2328 if (tb[NDTA_THRESH3])
2329 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2331 if (tb[NDTA_GC_INTERVAL])
2332 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2337 write_unlock_bh(&tbl->lock);
2342 static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2343 struct netlink_ext_ack *extack)
2345 struct ndtmsg *ndtm;
2347 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2348 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2352 ndtm = nlmsg_data(nlh);
2353 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2354 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2358 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2359 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2366 static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2368 const struct nlmsghdr *nlh = cb->nlh;
2369 struct net *net = sock_net(skb->sk);
2370 int family, tidx, nidx = 0;
2371 int tbl_skip = cb->args[0];
2372 int neigh_skip = cb->args[1];
2373 struct neigh_table *tbl;
2375 if (cb->strict_check) {
2376 int err = neightbl_valid_dump_info(nlh, cb->extack);
2382 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2384 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2385 struct neigh_parms *p;
2387 tbl = neigh_tables[tidx];
2391 if (tidx < tbl_skip || (family && tbl->family != family))
2394 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2395 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2400 p = list_next_entry(&tbl->parms, list);
2401 list_for_each_entry_from(p, &tbl->parms_list, list) {
2402 if (!net_eq(neigh_parms_net(p), net))
2405 if (nidx < neigh_skip)
2408 if (neightbl_fill_param_info(skb, tbl, p,
2409 NETLINK_CB(cb->skb).portid,
2427 static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2428 u32 pid, u32 seq, int type, unsigned int flags)
2430 unsigned long now = jiffies;
2431 struct nda_cacheinfo ci;
2432 struct nlmsghdr *nlh;
2435 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2439 ndm = nlmsg_data(nlh);
2440 ndm->ndm_family = neigh->ops->family;
2443 ndm->ndm_flags = neigh->flags;
2444 ndm->ndm_type = neigh->type;
2445 ndm->ndm_ifindex = neigh->dev->ifindex;
2447 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2448 goto nla_put_failure;
2450 read_lock_bh(&neigh->lock);
2451 ndm->ndm_state = neigh->nud_state;
2452 if (neigh->nud_state & NUD_VALID) {
2453 char haddr[MAX_ADDR_LEN];
2455 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2456 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2457 read_unlock_bh(&neigh->lock);
2458 goto nla_put_failure;
2462 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2463 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2464 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2465 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2466 read_unlock_bh(&neigh->lock);
2468 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2469 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2470 goto nla_put_failure;
2472 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2473 goto nla_put_failure;
2475 nlmsg_end(skb, nlh);
2479 nlmsg_cancel(skb, nlh);
2483 static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2484 u32 pid, u32 seq, int type, unsigned int flags,
2485 struct neigh_table *tbl)
2487 struct nlmsghdr *nlh;
2490 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2494 ndm = nlmsg_data(nlh);
2495 ndm->ndm_family = tbl->family;
2498 ndm->ndm_flags = pn->flags | NTF_PROXY;
2499 ndm->ndm_type = RTN_UNICAST;
2500 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2501 ndm->ndm_state = NUD_NONE;
2503 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2504 goto nla_put_failure;
2506 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2507 goto nla_put_failure;
2509 nlmsg_end(skb, nlh);
2513 nlmsg_cancel(skb, nlh);
2517 static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2519 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2520 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2523 static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2525 struct net_device *master;
2530 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2531 if (!master || master->ifindex != master_idx)
2537 static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2539 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2545 struct neigh_dump_filter {
2550 static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2551 struct netlink_callback *cb,
2552 struct neigh_dump_filter *filter)
2554 struct net *net = sock_net(skb->sk);
2555 struct neighbour *n;
2556 int rc, h, s_h = cb->args[1];
2557 int idx, s_idx = idx = cb->args[2];
2558 struct neigh_hash_table *nht;
2559 unsigned int flags = NLM_F_MULTI;
2561 if (filter->dev_idx || filter->master_idx)
2562 flags |= NLM_F_DUMP_FILTERED;
2565 nht = rcu_dereference_bh(tbl->nht);
2567 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2570 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2572 n = rcu_dereference_bh(n->next)) {
2573 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2575 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2576 neigh_master_filtered(n->dev, filter->master_idx))
2578 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2591 rcu_read_unlock_bh();
2597 static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2598 struct netlink_callback *cb,
2599 struct neigh_dump_filter *filter)
2601 struct pneigh_entry *n;
2602 struct net *net = sock_net(skb->sk);
2603 int rc, h, s_h = cb->args[3];
2604 int idx, s_idx = idx = cb->args[4];
2605 unsigned int flags = NLM_F_MULTI;
2607 if (filter->dev_idx || filter->master_idx)
2608 flags |= NLM_F_DUMP_FILTERED;
2610 read_lock_bh(&tbl->lock);
2612 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2615 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2616 if (idx < s_idx || pneigh_net(n) != net)
2618 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2619 neigh_master_filtered(n->dev, filter->master_idx))
2621 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2623 RTM_NEWNEIGH, flags, tbl) < 0) {
2624 read_unlock_bh(&tbl->lock);
2633 read_unlock_bh(&tbl->lock);
2642 static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2644 struct neigh_dump_filter *filter,
2645 struct netlink_ext_ack *extack)
2647 struct nlattr *tb[NDA_MAX + 1];
2653 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2654 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2658 ndm = nlmsg_data(nlh);
2659 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2660 ndm->ndm_state || ndm->ndm_type) {
2661 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2665 if (ndm->ndm_flags & ~NTF_PROXY) {
2666 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2670 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2671 tb, NDA_MAX, nda_policy,
2674 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2675 NDA_MAX, nda_policy, extack);
2680 for (i = 0; i <= NDA_MAX; ++i) {
2684 /* all new attributes should require strict_check */
2687 filter->dev_idx = nla_get_u32(tb[i]);
2690 filter->master_idx = nla_get_u32(tb[i]);
2694 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2703 static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2705 const struct nlmsghdr *nlh = cb->nlh;
2706 struct neigh_dump_filter filter = {};
2707 struct neigh_table *tbl;
2712 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2714 /* check for full ndmsg structure presence, family member is
2715 * the same for both structures
2717 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2718 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2721 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2722 if (err < 0 && cb->strict_check)
2727 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2728 tbl = neigh_tables[t];
2732 if (t < s_t || (family && tbl->family != family))
2735 memset(&cb->args[1], 0, sizeof(cb->args) -
2736 sizeof(cb->args[0]));
2738 err = pneigh_dump_table(tbl, skb, cb, &filter);
2740 err = neigh_dump_table(tbl, skb, cb, &filter);
2749 static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2750 struct neigh_table **tbl,
2751 void **dst, int *dev_idx, u8 *ndm_flags,
2752 struct netlink_ext_ack *extack)
2754 struct nlattr *tb[NDA_MAX + 1];
2758 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2759 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2763 ndm = nlmsg_data(nlh);
2764 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2766 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2770 if (ndm->ndm_flags & ~NTF_PROXY) {
2771 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2775 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2776 NDA_MAX, nda_policy, extack);
2780 *ndm_flags = ndm->ndm_flags;
2781 *dev_idx = ndm->ndm_ifindex;
2782 *tbl = neigh_find_table(ndm->ndm_family);
2784 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2785 return -EAFNOSUPPORT;
2788 for (i = 0; i <= NDA_MAX; ++i) {
2794 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2795 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2798 *dst = nla_data(tb[i]);
2801 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2809 static inline size_t neigh_nlmsg_size(void)
2811 return NLMSG_ALIGN(sizeof(struct ndmsg))
2812 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2813 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2814 + nla_total_size(sizeof(struct nda_cacheinfo))
2815 + nla_total_size(4) /* NDA_PROBES */
2816 + nla_total_size(1); /* NDA_PROTOCOL */
2819 static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2822 struct sk_buff *skb;
2825 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2829 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2835 err = rtnl_unicast(skb, net, pid);
2840 static inline size_t pneigh_nlmsg_size(void)
2842 return NLMSG_ALIGN(sizeof(struct ndmsg))
2843 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2844 + nla_total_size(1); /* NDA_PROTOCOL */
2847 static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2848 u32 pid, u32 seq, struct neigh_table *tbl)
2850 struct sk_buff *skb;
2853 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2857 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2863 err = rtnl_unicast(skb, net, pid);
2868 static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2869 struct netlink_ext_ack *extack)
2871 struct net *net = sock_net(in_skb->sk);
2872 struct net_device *dev = NULL;
2873 struct neigh_table *tbl = NULL;
2874 struct neighbour *neigh;
2880 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2886 dev = __dev_get_by_index(net, dev_idx);
2888 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2894 NL_SET_ERR_MSG(extack, "Network address not specified");
2898 if (ndm_flags & NTF_PROXY) {
2899 struct pneigh_entry *pn;
2901 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2903 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2906 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2907 nlh->nlmsg_seq, tbl);
2911 NL_SET_ERR_MSG(extack, "No device specified");
2915 neigh = neigh_lookup(tbl, dst, dev);
2917 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2921 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2924 neigh_release(neigh);
2929 void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2932 struct neigh_hash_table *nht;
2935 nht = rcu_dereference_bh(tbl->nht);
2937 read_lock(&tbl->lock); /* avoid resizes */
2938 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2939 struct neighbour *n;
2941 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2943 n = rcu_dereference_bh(n->next))
2946 read_unlock(&tbl->lock);
2947 rcu_read_unlock_bh();
2949 EXPORT_SYMBOL(neigh_for_each);
2951 /* The tbl->lock must be held as a writer and BH disabled. */
2952 void __neigh_for_each_release(struct neigh_table *tbl,
2953 int (*cb)(struct neighbour *))
2956 struct neigh_hash_table *nht;
2958 nht = rcu_dereference_protected(tbl->nht,
2959 lockdep_is_held(&tbl->lock));
2960 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2961 struct neighbour *n;
2962 struct neighbour __rcu **np;
2964 np = &nht->hash_buckets[chain];
2965 while ((n = rcu_dereference_protected(*np,
2966 lockdep_is_held(&tbl->lock))) != NULL) {
2969 write_lock(&n->lock);
2972 rcu_assign_pointer(*np,
2973 rcu_dereference_protected(n->next,
2974 lockdep_is_held(&tbl->lock)));
2978 write_unlock(&n->lock);
2980 neigh_cleanup_and_release(n);
2984 EXPORT_SYMBOL(__neigh_for_each_release);
2986 int neigh_xmit(int index, struct net_device *dev,
2987 const void *addr, struct sk_buff *skb)
2989 int err = -EAFNOSUPPORT;
2990 if (likely(index < NEIGH_NR_TABLES)) {
2991 struct neigh_table *tbl;
2992 struct neighbour *neigh;
2994 tbl = neigh_tables[index];
2998 if (index == NEIGH_ARP_TABLE) {
2999 u32 key = *((u32 *)addr);
3001 neigh = __ipv4_neigh_lookup_noref(dev, key);
3003 neigh = __neigh_lookup_noref(tbl, addr, dev);
3006 neigh = __neigh_create(tbl, addr, dev, false);
3007 err = PTR_ERR(neigh);
3008 if (IS_ERR(neigh)) {
3009 rcu_read_unlock_bh();
3012 err = neigh->output(neigh, skb);
3013 rcu_read_unlock_bh();
3015 else if (index == NEIGH_LINK_TABLE) {
3016 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3017 addr, NULL, skb->len);
3020 err = dev_queue_xmit(skb);
3028 EXPORT_SYMBOL(neigh_xmit);
3030 #ifdef CONFIG_PROC_FS
3032 static struct neighbour *neigh_get_first(struct seq_file *seq)
3034 struct neigh_seq_state *state = seq->private;
3035 struct net *net = seq_file_net(seq);
3036 struct neigh_hash_table *nht = state->nht;
3037 struct neighbour *n = NULL;
3040 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3041 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3042 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3045 if (!net_eq(dev_net(n->dev), net))
3047 if (state->neigh_sub_iter) {
3051 v = state->neigh_sub_iter(state, n, &fakep);
3055 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3057 if (n->nud_state & ~NUD_NOARP)
3060 n = rcu_dereference_bh(n->next);
3066 state->bucket = bucket;
3071 static struct neighbour *neigh_get_next(struct seq_file *seq,
3072 struct neighbour *n,
3075 struct neigh_seq_state *state = seq->private;
3076 struct net *net = seq_file_net(seq);
3077 struct neigh_hash_table *nht = state->nht;
3079 if (state->neigh_sub_iter) {
3080 void *v = state->neigh_sub_iter(state, n, pos);
3084 n = rcu_dereference_bh(n->next);
3088 if (!net_eq(dev_net(n->dev), net))
3090 if (state->neigh_sub_iter) {
3091 void *v = state->neigh_sub_iter(state, n, pos);
3096 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3099 if (n->nud_state & ~NUD_NOARP)
3102 n = rcu_dereference_bh(n->next);
3108 if (++state->bucket >= (1 << nht->hash_shift))
3111 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3119 static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3121 struct neighbour *n = neigh_get_first(seq);
3126 n = neigh_get_next(seq, n, pos);
3131 return *pos ? NULL : n;
3134 static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3136 struct neigh_seq_state *state = seq->private;
3137 struct net *net = seq_file_net(seq);
3138 struct neigh_table *tbl = state->tbl;
3139 struct pneigh_entry *pn = NULL;
3140 int bucket = state->bucket;
3142 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3143 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3144 pn = tbl->phash_buckets[bucket];
3145 while (pn && !net_eq(pneigh_net(pn), net))
3150 state->bucket = bucket;
3155 static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3156 struct pneigh_entry *pn,
3159 struct neigh_seq_state *state = seq->private;
3160 struct net *net = seq_file_net(seq);
3161 struct neigh_table *tbl = state->tbl;
3165 } while (pn && !net_eq(pneigh_net(pn), net));
3168 if (++state->bucket > PNEIGH_HASHMASK)
3170 pn = tbl->phash_buckets[state->bucket];
3171 while (pn && !net_eq(pneigh_net(pn), net))
3183 static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3185 struct pneigh_entry *pn = pneigh_get_first(seq);
3190 pn = pneigh_get_next(seq, pn, pos);
3195 return *pos ? NULL : pn;
3198 static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3200 struct neigh_seq_state *state = seq->private;
3202 loff_t idxpos = *pos;
3204 rc = neigh_get_idx(seq, &idxpos);
3205 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3206 rc = pneigh_get_idx(seq, &idxpos);
3211 void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3212 __acquires(tbl->lock)
3215 struct neigh_seq_state *state = seq->private;
3219 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3222 state->nht = rcu_dereference_bh(tbl->nht);
3223 read_lock(&tbl->lock);
3225 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3227 EXPORT_SYMBOL(neigh_seq_start);
3229 void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3231 struct neigh_seq_state *state;
3234 if (v == SEQ_START_TOKEN) {
3235 rc = neigh_get_first(seq);
3239 state = seq->private;
3240 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3241 rc = neigh_get_next(seq, v, NULL);
3244 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3245 rc = pneigh_get_first(seq);
3247 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3248 rc = pneigh_get_next(seq, v, NULL);
3254 EXPORT_SYMBOL(neigh_seq_next);
3256 void neigh_seq_stop(struct seq_file *seq, void *v)
3257 __releases(tbl->lock)
3260 struct neigh_seq_state *state = seq->private;
3261 struct neigh_table *tbl = state->tbl;
3263 read_unlock(&tbl->lock);
3264 rcu_read_unlock_bh();
3266 EXPORT_SYMBOL(neigh_seq_stop);
3268 /* statistics via seq_file */
3270 static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3272 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3276 return SEQ_START_TOKEN;
3278 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3279 if (!cpu_possible(cpu))
3282 return per_cpu_ptr(tbl->stats, cpu);
3287 static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3289 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3292 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3293 if (!cpu_possible(cpu))
3296 return per_cpu_ptr(tbl->stats, cpu);
3302 static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3307 static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3309 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3310 struct neigh_statistics *st = v;
3312 if (v == SEQ_START_TOKEN) {
3313 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");
3317 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3318 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
3319 atomic_read(&tbl->entries),
3330 st->rcv_probes_mcast,
3331 st->rcv_probes_ucast,
3333 st->periodic_gc_runs,
3342 static const struct seq_operations neigh_stat_seq_ops = {
3343 .start = neigh_stat_seq_start,
3344 .next = neigh_stat_seq_next,
3345 .stop = neigh_stat_seq_stop,
3346 .show = neigh_stat_seq_show,
3348 #endif /* CONFIG_PROC_FS */
3350 static void __neigh_notify(struct neighbour *n, int type, int flags,
3353 struct net *net = dev_net(n->dev);
3354 struct sk_buff *skb;
3357 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3361 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3363 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3364 WARN_ON(err == -EMSGSIZE);
3368 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3372 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3375 void neigh_app_ns(struct neighbour *n)
3377 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3379 EXPORT_SYMBOL(neigh_app_ns);
3381 #ifdef CONFIG_SYSCTL
3382 static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3384 static int proc_unres_qlen(struct ctl_table *ctl, int write,
3385 void *buffer, size_t *lenp, loff_t *ppos)
3388 struct ctl_table tmp = *ctl;
3390 tmp.extra1 = SYSCTL_ZERO;
3391 tmp.extra2 = &unres_qlen_max;
3394 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3395 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3398 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3402 static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3407 return __in_dev_arp_parms_get_rcu(dev);
3409 return __in6_dev_nd_parms_get_rcu(dev);
3414 static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3417 struct net_device *dev;
3418 int family = neigh_parms_family(p);
3421 for_each_netdev_rcu(net, dev) {
3422 struct neigh_parms *dst_p =
3423 neigh_get_dev_parms_rcu(dev, family);
3425 if (dst_p && !test_bit(index, dst_p->data_state))
3426 dst_p->data[index] = p->data[index];
3431 static void neigh_proc_update(struct ctl_table *ctl, int write)
3433 struct net_device *dev = ctl->extra1;
3434 struct neigh_parms *p = ctl->extra2;
3435 struct net *net = neigh_parms_net(p);
3436 int index = (int *) ctl->data - p->data;
3441 set_bit(index, p->data_state);
3442 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3443 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3444 if (!dev) /* NULL dev means this is default value */
3445 neigh_copy_dflt_parms(net, p, index);
3448 static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3449 void *buffer, size_t *lenp,
3452 struct ctl_table tmp = *ctl;
3455 tmp.extra1 = SYSCTL_ZERO;
3456 tmp.extra2 = SYSCTL_INT_MAX;
3458 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3459 neigh_proc_update(ctl, write);
3463 int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3464 size_t *lenp, loff_t *ppos)
3466 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3468 neigh_proc_update(ctl, write);
3471 EXPORT_SYMBOL(neigh_proc_dointvec);
3473 int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3474 size_t *lenp, loff_t *ppos)
3476 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3478 neigh_proc_update(ctl, write);
3481 EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3483 static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3484 void *buffer, size_t *lenp,
3487 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3489 neigh_proc_update(ctl, write);
3493 int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3494 void *buffer, size_t *lenp, loff_t *ppos)
3496 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3498 neigh_proc_update(ctl, write);
3501 EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3503 static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3504 void *buffer, size_t *lenp,
3507 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3509 neigh_proc_update(ctl, write);
3513 static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3514 void *buffer, size_t *lenp,
3517 struct neigh_parms *p = ctl->extra2;
3520 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3521 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3522 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3523 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3527 if (write && ret == 0) {
3528 /* update reachable_time as well, otherwise, the change will
3529 * only be effective after the next time neigh_periodic_work
3530 * decides to recompute it
3533 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3538 #define NEIGH_PARMS_DATA_OFFSET(index) \
3539 (&((struct neigh_parms *) 0)->data[index])
3541 #define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3542 [NEIGH_VAR_ ## attr] = { \
3544 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3545 .maxlen = sizeof(int), \
3547 .proc_handler = proc, \
3550 #define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3551 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3553 #define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3554 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3556 #define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3557 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3559 #define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3560 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3562 #define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3563 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3565 static struct neigh_sysctl_table {
3566 struct ctl_table_header *sysctl_header;
3567 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3568 } neigh_sysctl_template __read_mostly = {
3570 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3571 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3572 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3573 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3574 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3575 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3576 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3577 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3578 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3579 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3580 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3581 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3582 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3583 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3584 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3585 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3586 [NEIGH_VAR_GC_INTERVAL] = {
3587 .procname = "gc_interval",
3588 .maxlen = sizeof(int),
3590 .proc_handler = proc_dointvec_jiffies,
3592 [NEIGH_VAR_GC_THRESH1] = {
3593 .procname = "gc_thresh1",
3594 .maxlen = sizeof(int),
3596 .extra1 = SYSCTL_ZERO,
3597 .extra2 = SYSCTL_INT_MAX,
3598 .proc_handler = proc_dointvec_minmax,
3600 [NEIGH_VAR_GC_THRESH2] = {
3601 .procname = "gc_thresh2",
3602 .maxlen = sizeof(int),
3604 .extra1 = SYSCTL_ZERO,
3605 .extra2 = SYSCTL_INT_MAX,
3606 .proc_handler = proc_dointvec_minmax,
3608 [NEIGH_VAR_GC_THRESH3] = {
3609 .procname = "gc_thresh3",
3610 .maxlen = sizeof(int),
3612 .extra1 = SYSCTL_ZERO,
3613 .extra2 = SYSCTL_INT_MAX,
3614 .proc_handler = proc_dointvec_minmax,
3620 int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3621 proc_handler *handler)
3624 struct neigh_sysctl_table *t;
3625 const char *dev_name_source;
3626 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3629 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3633 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3634 t->neigh_vars[i].data += (long) p;
3635 t->neigh_vars[i].extra1 = dev;
3636 t->neigh_vars[i].extra2 = p;
3640 dev_name_source = dev->name;
3641 /* Terminate the table early */
3642 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3643 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3645 struct neigh_table *tbl = p->tbl;
3646 dev_name_source = "default";
3647 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3648 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3649 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3650 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3655 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3657 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3658 /* RetransTime (in milliseconds)*/
3659 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3660 /* ReachableTime (in milliseconds) */
3661 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3663 /* Those handlers will update p->reachable_time after
3664 * base_reachable_time(_ms) is set to ensure the new timer starts being
3665 * applied after the next neighbour update instead of waiting for
3666 * neigh_periodic_work to update its value (can be multiple minutes)
3667 * So any handler that replaces them should do this as well
3670 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3671 neigh_proc_base_reachable_time;
3672 /* ReachableTime (in milliseconds) */
3673 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3674 neigh_proc_base_reachable_time;
3677 /* Don't export sysctls to unprivileged users */
3678 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3679 t->neigh_vars[0].procname = NULL;
3681 switch (neigh_parms_family(p)) {
3692 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3693 p_name, dev_name_source);
3695 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3696 if (!t->sysctl_header)
3699 p->sysctl_table = t;
3707 EXPORT_SYMBOL(neigh_sysctl_register);
3709 void neigh_sysctl_unregister(struct neigh_parms *p)
3711 if (p->sysctl_table) {
3712 struct neigh_sysctl_table *t = p->sysctl_table;
3713 p->sysctl_table = NULL;
3714 unregister_net_sysctl_table(t->sysctl_header);
3718 EXPORT_SYMBOL(neigh_sysctl_unregister);
3720 #endif /* CONFIG_SYSCTL */
3722 static int __init neigh_init(void)
3724 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3725 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3726 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3728 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3730 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3735 subsys_initcall(neigh_init);