Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Generic address resolution entity
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 *
9 * Fixes:
10 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
11 * Harald Welte Add neighbour cache statistics like rtstat
12 */
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
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>
24#ifdef CONFIG_SYSCTL
25#include <linux/sysctl.h>
26#endif
27#include <linux/times.h>
28#include <net/net_namespace.h>
29#include <net/neighbour.h>
30#include <net/arp.h>
31#include <net/dst.h>
32#include <net/sock.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>
41
42#include <trace/events/neigh.h>
43
44#define NEIGH_DEBUG 1
45#define neigh_dbg(level, fmt, ...) \
46do { \
47 if (level <= NEIGH_DEBUG) \
48 pr_debug(fmt, ##__VA_ARGS__); \
49} while (0)
50
51#define PNEIGH_HASHMASK 0xF
52
53static void neigh_timer_handler(struct timer_list *t);
54static void __neigh_notify(struct neighbour *n, int type, int flags,
55 u32 pid);
56static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
57static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
58 struct net_device *dev);
59
60#ifdef CONFIG_PROC_FS
61static const struct seq_operations neigh_stat_seq_ops;
62#endif
63
64/*
65 Neighbour hash table buckets are protected with rwlock tbl->lock.
66
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
71 cache.
72 - If the entry requires some non-trivial actions, increase
73 its reference count and release table lock.
74
75 Neighbour entries are protected:
76 - with reference count.
77 - with rwlock neigh->lock
78
79 Reference count prevents destruction.
80
81 neigh->lock mainly serializes ll address data and its validity state.
82 However, the same lock is used to protect another entry fields:
83 - timer
84 - resolution queue
85
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.
90 */
91
92static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
93{
94 kfree_skb(skb);
95 return -ENETDOWN;
96}
97
98static void neigh_cleanup_and_release(struct neighbour *neigh)
99{
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);
104}
105
106/*
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.
110 */
111
112unsigned long neigh_rand_reach_time(unsigned long base)
113{
114 return base ? (prandom_u32() % base) + (base >> 1) : 0;
115}
116EXPORT_SYMBOL(neigh_rand_reach_time);
117
118static void neigh_mark_dead(struct neighbour *n)
119{
120 n->dead = 1;
121 if (!list_empty(&n->gc_list)) {
122 list_del_init(&n->gc_list);
123 atomic_dec(&n->tbl->gc_entries);
124 }
125}
126
127static void neigh_update_gc_list(struct neighbour *n)
128{
129 bool on_gc_list, exempt_from_gc;
130
131 write_lock_bh(&n->tbl->lock);
132 write_lock(&n->lock);
133
134 if (n->dead)
135 goto out;
136
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
139 */
140 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
141 n->flags & NTF_EXT_LEARNED;
142 on_gc_list = !list_empty(&n->gc_list);
143
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);
151 }
152
153out:
154 write_unlock(&n->lock);
155 write_unlock_bh(&n->tbl->lock);
156}
157
158static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
159 int *notify)
160{
161 bool rc = false;
162 u8 ndm_flags;
163
164 if (!(flags & NEIGH_UPDATE_F_ADMIN))
165 return rc;
166
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;
171 else
172 neigh->flags &= ~NTF_EXT_LEARNED;
173 rc = true;
174 *notify = 1;
175 }
176
177 return rc;
178}
179
180static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
181 struct neigh_table *tbl)
182{
183 bool retval = false;
184
185 write_lock(&n->lock);
186 if (refcount_read(&n->refcnt) == 1) {
187 struct neighbour *neigh;
188
189 neigh = rcu_dereference_protected(n->next,
190 lockdep_is_held(&tbl->lock));
191 rcu_assign_pointer(*np, neigh);
192 neigh_mark_dead(n);
193 retval = true;
194 }
195 write_unlock(&n->lock);
196 if (retval)
197 neigh_cleanup_and_release(n);
198 return retval;
199}
200
201bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
202{
203 struct neigh_hash_table *nht;
204 void *pkey = ndel->primary_key;
205 u32 hash_val;
206 struct neighbour *n;
207 struct neighbour __rcu **np;
208
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);
213
214 np = &nht->hash_buckets[hash_val];
215 while ((n = rcu_dereference_protected(*np,
216 lockdep_is_held(&tbl->lock)))) {
217 if (n == ndel)
218 return neigh_del(n, np, tbl);
219 np = &n->next;
220 }
221 return false;
222}
223
224static int neigh_forced_gc(struct neigh_table *tbl)
225{
226 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
227 unsigned long tref = jiffies - 5 * HZ;
228 struct neighbour *n, *tmp;
229 int shrunk = 0;
230
231 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
232
233 write_lock_bh(&tbl->lock);
234
235 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
236 if (refcount_read(&n->refcnt) == 1) {
237 bool remove = false;
238
239 write_lock(&n->lock);
240 if ((n->nud_state == NUD_FAILED) ||
241 (n->nud_state == NUD_NOARP) ||
242 (tbl->is_multicast &&
243 tbl->is_multicast(n->primary_key)) ||
244 time_after(tref, n->updated))
245 remove = true;
246 write_unlock(&n->lock);
247
248 if (remove && neigh_remove_one(n, tbl))
249 shrunk++;
250 if (shrunk >= max_clean)
251 break;
252 }
253 }
254
255 tbl->last_flush = jiffies;
256
257 write_unlock_bh(&tbl->lock);
258
259 return shrunk;
260}
261
262static void neigh_add_timer(struct neighbour *n, unsigned long when)
263{
264 neigh_hold(n);
265 if (unlikely(mod_timer(&n->timer, when))) {
266 printk("NEIGH: BUG, double timer add, state is %x\n",
267 n->nud_state);
268 dump_stack();
269 }
270}
271
272static int neigh_del_timer(struct neighbour *n)
273{
274 if ((n->nud_state & NUD_IN_TIMER) &&
275 del_timer(&n->timer)) {
276 neigh_release(n);
277 return 1;
278 }
279 return 0;
280}
281
282static void pneigh_queue_purge(struct sk_buff_head *list)
283{
284 struct sk_buff *skb;
285
286 while ((skb = skb_dequeue(list)) != NULL) {
287 dev_put(skb->dev);
288 kfree_skb(skb);
289 }
290}
291
292static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
293 bool skip_perm)
294{
295 int i;
296 struct neigh_hash_table *nht;
297
298 nht = rcu_dereference_protected(tbl->nht,
299 lockdep_is_held(&tbl->lock));
300
301 for (i = 0; i < (1 << nht->hash_shift); i++) {
302 struct neighbour *n;
303 struct neighbour __rcu **np = &nht->hash_buckets[i];
304
305 while ((n = rcu_dereference_protected(*np,
306 lockdep_is_held(&tbl->lock))) != NULL) {
307 if (dev && n->dev != dev) {
308 np = &n->next;
309 continue;
310 }
311 if (skip_perm && n->nud_state & NUD_PERMANENT) {
312 np = &n->next;
313 continue;
314 }
315 rcu_assign_pointer(*np,
316 rcu_dereference_protected(n->next,
317 lockdep_is_held(&tbl->lock)));
318 write_lock(&n->lock);
319 neigh_del_timer(n);
320 neigh_mark_dead(n);
321 if (refcount_read(&n->refcnt) != 1) {
322 /* The most unpleasant situation.
323 We must destroy neighbour entry,
324 but someone still uses it.
325
326 The destroy will be delayed until
327 the last user releases us, but
328 we must kill timers etc. and move
329 it to safe state.
330 */
331 __skb_queue_purge(&n->arp_queue);
332 n->arp_queue_len_bytes = 0;
333 n->output = neigh_blackhole;
334 if (n->nud_state & NUD_VALID)
335 n->nud_state = NUD_NOARP;
336 else
337 n->nud_state = NUD_NONE;
338 neigh_dbg(2, "neigh %p is stray\n", n);
339 }
340 write_unlock(&n->lock);
341 neigh_cleanup_and_release(n);
342 }
343 }
344}
345
346void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
347{
348 write_lock_bh(&tbl->lock);
349 neigh_flush_dev(tbl, dev, false);
350 write_unlock_bh(&tbl->lock);
351}
352EXPORT_SYMBOL(neigh_changeaddr);
353
354static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
355 bool skip_perm)
356{
357 write_lock_bh(&tbl->lock);
358 neigh_flush_dev(tbl, dev, skip_perm);
359 pneigh_ifdown_and_unlock(tbl, dev);
360
361 del_timer_sync(&tbl->proxy_timer);
362 pneigh_queue_purge(&tbl->proxy_queue);
363 return 0;
364}
365
366int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
367{
368 __neigh_ifdown(tbl, dev, true);
369 return 0;
370}
371EXPORT_SYMBOL(neigh_carrier_down);
372
373int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
374{
375 __neigh_ifdown(tbl, dev, false);
376 return 0;
377}
378EXPORT_SYMBOL(neigh_ifdown);
379
380static struct neighbour *neigh_alloc(struct neigh_table *tbl,
381 struct net_device *dev,
382 bool exempt_from_gc)
383{
384 struct neighbour *n = NULL;
385 unsigned long now = jiffies;
386 int entries;
387
388 if (exempt_from_gc)
389 goto do_alloc;
390
391 entries = atomic_inc_return(&tbl->gc_entries) - 1;
392 if (entries >= tbl->gc_thresh3 ||
393 (entries >= tbl->gc_thresh2 &&
394 time_after(now, tbl->last_flush + 5 * HZ))) {
395 if (!neigh_forced_gc(tbl) &&
396 entries >= tbl->gc_thresh3) {
397 net_info_ratelimited("%s: neighbor table overflow!\n",
398 tbl->id);
399 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
400 goto out_entries;
401 }
402 }
403
404do_alloc:
405 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
406 if (!n)
407 goto out_entries;
408
409 __skb_queue_head_init(&n->arp_queue);
410 rwlock_init(&n->lock);
411 seqlock_init(&n->ha_lock);
412 n->updated = n->used = now;
413 n->nud_state = NUD_NONE;
414 n->output = neigh_blackhole;
415 seqlock_init(&n->hh.hh_lock);
416 n->parms = neigh_parms_clone(&tbl->parms);
417 timer_setup(&n->timer, neigh_timer_handler, 0);
418
419 NEIGH_CACHE_STAT_INC(tbl, allocs);
420 n->tbl = tbl;
421 refcount_set(&n->refcnt, 1);
422 n->dead = 1;
423 INIT_LIST_HEAD(&n->gc_list);
424
425 atomic_inc(&tbl->entries);
426out:
427 return n;
428
429out_entries:
430 if (!exempt_from_gc)
431 atomic_dec(&tbl->gc_entries);
432 goto out;
433}
434
435static void neigh_get_hash_rnd(u32 *x)
436{
437 *x = get_random_u32() | 1;
438}
439
440static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
441{
442 size_t size = (1 << shift) * sizeof(struct neighbour *);
443 struct neigh_hash_table *ret;
444 struct neighbour __rcu **buckets;
445 int i;
446
447 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
448 if (!ret)
449 return NULL;
450 if (size <= PAGE_SIZE) {
451 buckets = kzalloc(size, GFP_ATOMIC);
452 } else {
453 buckets = (struct neighbour __rcu **)
454 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
455 get_order(size));
456 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
457 }
458 if (!buckets) {
459 kfree(ret);
460 return NULL;
461 }
462 ret->hash_buckets = buckets;
463 ret->hash_shift = shift;
464 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
465 neigh_get_hash_rnd(&ret->hash_rnd[i]);
466 return ret;
467}
468
469static void neigh_hash_free_rcu(struct rcu_head *head)
470{
471 struct neigh_hash_table *nht = container_of(head,
472 struct neigh_hash_table,
473 rcu);
474 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
475 struct neighbour __rcu **buckets = nht->hash_buckets;
476
477 if (size <= PAGE_SIZE) {
478 kfree(buckets);
479 } else {
480 kmemleak_free(buckets);
481 free_pages((unsigned long)buckets, get_order(size));
482 }
483 kfree(nht);
484}
485
486static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
487 unsigned long new_shift)
488{
489 unsigned int i, hash;
490 struct neigh_hash_table *new_nht, *old_nht;
491
492 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
493
494 old_nht = rcu_dereference_protected(tbl->nht,
495 lockdep_is_held(&tbl->lock));
496 new_nht = neigh_hash_alloc(new_shift);
497 if (!new_nht)
498 return old_nht;
499
500 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
501 struct neighbour *n, *next;
502
503 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
504 lockdep_is_held(&tbl->lock));
505 n != NULL;
506 n = next) {
507 hash = tbl->hash(n->primary_key, n->dev,
508 new_nht->hash_rnd);
509
510 hash >>= (32 - new_nht->hash_shift);
511 next = rcu_dereference_protected(n->next,
512 lockdep_is_held(&tbl->lock));
513
514 rcu_assign_pointer(n->next,
515 rcu_dereference_protected(
516 new_nht->hash_buckets[hash],
517 lockdep_is_held(&tbl->lock)));
518 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
519 }
520 }
521
522 rcu_assign_pointer(tbl->nht, new_nht);
523 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
524 return new_nht;
525}
526
527struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
528 struct net_device *dev)
529{
530 struct neighbour *n;
531
532 NEIGH_CACHE_STAT_INC(tbl, lookups);
533
534 rcu_read_lock_bh();
535 n = __neigh_lookup_noref(tbl, pkey, dev);
536 if (n) {
537 if (!refcount_inc_not_zero(&n->refcnt))
538 n = NULL;
539 NEIGH_CACHE_STAT_INC(tbl, hits);
540 }
541
542 rcu_read_unlock_bh();
543 return n;
544}
545EXPORT_SYMBOL(neigh_lookup);
546
547struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
548 const void *pkey)
549{
550 struct neighbour *n;
551 unsigned int key_len = tbl->key_len;
552 u32 hash_val;
553 struct neigh_hash_table *nht;
554
555 NEIGH_CACHE_STAT_INC(tbl, lookups);
556
557 rcu_read_lock_bh();
558 nht = rcu_dereference_bh(tbl->nht);
559 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
560
561 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
562 n != NULL;
563 n = rcu_dereference_bh(n->next)) {
564 if (!memcmp(n->primary_key, pkey, key_len) &&
565 net_eq(dev_net(n->dev), net)) {
566 if (!refcount_inc_not_zero(&n->refcnt))
567 n = NULL;
568 NEIGH_CACHE_STAT_INC(tbl, hits);
569 break;
570 }
571 }
572
573 rcu_read_unlock_bh();
574 return n;
575}
576EXPORT_SYMBOL(neigh_lookup_nodev);
577
578static struct neighbour *___neigh_create(struct neigh_table *tbl,
579 const void *pkey,
580 struct net_device *dev,
581 bool exempt_from_gc, bool want_ref)
582{
583 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
584 u32 hash_val;
585 unsigned int key_len = tbl->key_len;
586 int error;
587 struct neigh_hash_table *nht;
588
589 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
590
591 if (!n) {
592 rc = ERR_PTR(-ENOBUFS);
593 goto out;
594 }
595
596 memcpy(n->primary_key, pkey, key_len);
597 n->dev = dev;
598 dev_hold(dev);
599
600 /* Protocol specific setup. */
601 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
602 rc = ERR_PTR(error);
603 goto out_neigh_release;
604 }
605
606 if (dev->netdev_ops->ndo_neigh_construct) {
607 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
608 if (error < 0) {
609 rc = ERR_PTR(error);
610 goto out_neigh_release;
611 }
612 }
613
614 /* Device specific setup. */
615 if (n->parms->neigh_setup &&
616 (error = n->parms->neigh_setup(n)) < 0) {
617 rc = ERR_PTR(error);
618 goto out_neigh_release;
619 }
620
621 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
622
623 write_lock_bh(&tbl->lock);
624 nht = rcu_dereference_protected(tbl->nht,
625 lockdep_is_held(&tbl->lock));
626
627 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
628 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
629
630 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
631
632 if (n->parms->dead) {
633 rc = ERR_PTR(-EINVAL);
634 goto out_tbl_unlock;
635 }
636
637 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
638 lockdep_is_held(&tbl->lock));
639 n1 != NULL;
640 n1 = rcu_dereference_protected(n1->next,
641 lockdep_is_held(&tbl->lock))) {
642 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
643 if (want_ref)
644 neigh_hold(n1);
645 rc = n1;
646 goto out_tbl_unlock;
647 }
648 }
649
650 n->dead = 0;
651 if (!exempt_from_gc)
652 list_add_tail(&n->gc_list, &n->tbl->gc_list);
653
654 if (want_ref)
655 neigh_hold(n);
656 rcu_assign_pointer(n->next,
657 rcu_dereference_protected(nht->hash_buckets[hash_val],
658 lockdep_is_held(&tbl->lock)));
659 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
660 write_unlock_bh(&tbl->lock);
661 neigh_dbg(2, "neigh %p is created\n", n);
662 rc = n;
663out:
664 return rc;
665out_tbl_unlock:
666 write_unlock_bh(&tbl->lock);
667out_neigh_release:
668 if (!exempt_from_gc)
669 atomic_dec(&tbl->gc_entries);
670 neigh_release(n);
671 goto out;
672}
673
674struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
675 struct net_device *dev, bool want_ref)
676{
677 return ___neigh_create(tbl, pkey, dev, false, want_ref);
678}
679EXPORT_SYMBOL(__neigh_create);
680
681static u32 pneigh_hash(const void *pkey, unsigned int key_len)
682{
683 u32 hash_val = *(u32 *)(pkey + key_len - 4);
684 hash_val ^= (hash_val >> 16);
685 hash_val ^= hash_val >> 8;
686 hash_val ^= hash_val >> 4;
687 hash_val &= PNEIGH_HASHMASK;
688 return hash_val;
689}
690
691static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
692 struct net *net,
693 const void *pkey,
694 unsigned int key_len,
695 struct net_device *dev)
696{
697 while (n) {
698 if (!memcmp(n->key, pkey, key_len) &&
699 net_eq(pneigh_net(n), net) &&
700 (n->dev == dev || !n->dev))
701 return n;
702 n = n->next;
703 }
704 return NULL;
705}
706
707struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
708 struct net *net, const void *pkey, struct net_device *dev)
709{
710 unsigned int key_len = tbl->key_len;
711 u32 hash_val = pneigh_hash(pkey, key_len);
712
713 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
714 net, pkey, key_len, dev);
715}
716EXPORT_SYMBOL_GPL(__pneigh_lookup);
717
718struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
719 struct net *net, const void *pkey,
720 struct net_device *dev, int creat)
721{
722 struct pneigh_entry *n;
723 unsigned int key_len = tbl->key_len;
724 u32 hash_val = pneigh_hash(pkey, key_len);
725
726 read_lock_bh(&tbl->lock);
727 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
728 net, pkey, key_len, dev);
729 read_unlock_bh(&tbl->lock);
730
731 if (n || !creat)
732 goto out;
733
734 ASSERT_RTNL();
735
736 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
737 if (!n)
738 goto out;
739
740 n->protocol = 0;
741 write_pnet(&n->net, net);
742 memcpy(n->key, pkey, key_len);
743 n->dev = dev;
744 if (dev)
745 dev_hold(dev);
746
747 if (tbl->pconstructor && tbl->pconstructor(n)) {
748 if (dev)
749 dev_put(dev);
750 kfree(n);
751 n = NULL;
752 goto out;
753 }
754
755 write_lock_bh(&tbl->lock);
756 n->next = tbl->phash_buckets[hash_val];
757 tbl->phash_buckets[hash_val] = n;
758 write_unlock_bh(&tbl->lock);
759out:
760 return n;
761}
762EXPORT_SYMBOL(pneigh_lookup);
763
764
765int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
766 struct net_device *dev)
767{
768 struct pneigh_entry *n, **np;
769 unsigned int key_len = tbl->key_len;
770 u32 hash_val = pneigh_hash(pkey, key_len);
771
772 write_lock_bh(&tbl->lock);
773 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
774 np = &n->next) {
775 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
776 net_eq(pneigh_net(n), net)) {
777 *np = n->next;
778 write_unlock_bh(&tbl->lock);
779 if (tbl->pdestructor)
780 tbl->pdestructor(n);
781 if (n->dev)
782 dev_put(n->dev);
783 kfree(n);
784 return 0;
785 }
786 }
787 write_unlock_bh(&tbl->lock);
788 return -ENOENT;
789}
790
791static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
792 struct net_device *dev)
793{
794 struct pneigh_entry *n, **np, *freelist = NULL;
795 u32 h;
796
797 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
798 np = &tbl->phash_buckets[h];
799 while ((n = *np) != NULL) {
800 if (!dev || n->dev == dev) {
801 *np = n->next;
802 n->next = freelist;
803 freelist = n;
804 continue;
805 }
806 np = &n->next;
807 }
808 }
809 write_unlock_bh(&tbl->lock);
810 while ((n = freelist)) {
811 freelist = n->next;
812 n->next = NULL;
813 if (tbl->pdestructor)
814 tbl->pdestructor(n);
815 if (n->dev)
816 dev_put(n->dev);
817 kfree(n);
818 }
819 return -ENOENT;
820}
821
822static void neigh_parms_destroy(struct neigh_parms *parms);
823
824static inline void neigh_parms_put(struct neigh_parms *parms)
825{
826 if (refcount_dec_and_test(&parms->refcnt))
827 neigh_parms_destroy(parms);
828}
829
830/*
831 * neighbour must already be out of the table;
832 *
833 */
834void neigh_destroy(struct neighbour *neigh)
835{
836 struct net_device *dev = neigh->dev;
837
838 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
839
840 if (!neigh->dead) {
841 pr_warn("Destroying alive neighbour %p\n", neigh);
842 dump_stack();
843 return;
844 }
845
846 if (neigh_del_timer(neigh))
847 pr_warn("Impossible event\n");
848
849 write_lock_bh(&neigh->lock);
850 __skb_queue_purge(&neigh->arp_queue);
851 write_unlock_bh(&neigh->lock);
852 neigh->arp_queue_len_bytes = 0;
853
854 if (dev->netdev_ops->ndo_neigh_destroy)
855 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
856
857 dev_put(dev);
858 neigh_parms_put(neigh->parms);
859
860 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
861
862 atomic_dec(&neigh->tbl->entries);
863 kfree_rcu(neigh, rcu);
864}
865EXPORT_SYMBOL(neigh_destroy);
866
867/* Neighbour state is suspicious;
868 disable fast path.
869
870 Called with write_locked neigh.
871 */
872static void neigh_suspect(struct neighbour *neigh)
873{
874 neigh_dbg(2, "neigh %p is suspected\n", neigh);
875
876 neigh->output = neigh->ops->output;
877}
878
879/* Neighbour state is OK;
880 enable fast path.
881
882 Called with write_locked neigh.
883 */
884static void neigh_connect(struct neighbour *neigh)
885{
886 neigh_dbg(2, "neigh %p is connected\n", neigh);
887
888 neigh->output = neigh->ops->connected_output;
889}
890
891static void neigh_periodic_work(struct work_struct *work)
892{
893 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
894 struct neighbour *n;
895 struct neighbour __rcu **np;
896 unsigned int i;
897 struct neigh_hash_table *nht;
898
899 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
900
901 write_lock_bh(&tbl->lock);
902 nht = rcu_dereference_protected(tbl->nht,
903 lockdep_is_held(&tbl->lock));
904
905 /*
906 * periodically recompute ReachableTime from random function
907 */
908
909 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
910 struct neigh_parms *p;
911 tbl->last_rand = jiffies;
912 list_for_each_entry(p, &tbl->parms_list, list)
913 p->reachable_time =
914 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
915 }
916
917 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
918 goto out;
919
920 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
921 np = &nht->hash_buckets[i];
922
923 while ((n = rcu_dereference_protected(*np,
924 lockdep_is_held(&tbl->lock))) != NULL) {
925 unsigned int state;
926
927 write_lock(&n->lock);
928
929 state = n->nud_state;
930 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
931 (n->flags & NTF_EXT_LEARNED)) {
932 write_unlock(&n->lock);
933 goto next_elt;
934 }
935
936 if (time_before(n->used, n->confirmed))
937 n->used = n->confirmed;
938
939 if (refcount_read(&n->refcnt) == 1 &&
940 (state == NUD_FAILED ||
941 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
942 *np = n->next;
943 neigh_mark_dead(n);
944 write_unlock(&n->lock);
945 neigh_cleanup_and_release(n);
946 continue;
947 }
948 write_unlock(&n->lock);
949
950next_elt:
951 np = &n->next;
952 }
953 /*
954 * It's fine to release lock here, even if hash table
955 * grows while we are preempted.
956 */
957 write_unlock_bh(&tbl->lock);
958 cond_resched();
959 write_lock_bh(&tbl->lock);
960 nht = rcu_dereference_protected(tbl->nht,
961 lockdep_is_held(&tbl->lock));
962 }
963out:
964 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
965 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
966 * BASE_REACHABLE_TIME.
967 */
968 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
969 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
970 write_unlock_bh(&tbl->lock);
971}
972
973static __inline__ int neigh_max_probes(struct neighbour *n)
974{
975 struct neigh_parms *p = n->parms;
976 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
977 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
978 NEIGH_VAR(p, MCAST_PROBES));
979}
980
981static void neigh_invalidate(struct neighbour *neigh)
982 __releases(neigh->lock)
983 __acquires(neigh->lock)
984{
985 struct sk_buff *skb;
986
987 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
988 neigh_dbg(2, "neigh %p is failed\n", neigh);
989 neigh->updated = jiffies;
990
991 /* It is very thin place. report_unreachable is very complicated
992 routine. Particularly, it can hit the same neighbour entry!
993
994 So that, we try to be accurate and avoid dead loop. --ANK
995 */
996 while (neigh->nud_state == NUD_FAILED &&
997 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
998 write_unlock(&neigh->lock);
999 neigh->ops->error_report(neigh, skb);
1000 write_lock(&neigh->lock);
1001 }
1002 __skb_queue_purge(&neigh->arp_queue);
1003 neigh->arp_queue_len_bytes = 0;
1004}
1005
1006static void neigh_probe(struct neighbour *neigh)
1007 __releases(neigh->lock)
1008{
1009 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1010 /* keep skb alive even if arp_queue overflows */
1011 if (skb)
1012 skb = skb_clone(skb, GFP_ATOMIC);
1013 write_unlock(&neigh->lock);
1014 if (neigh->ops->solicit)
1015 neigh->ops->solicit(neigh, skb);
1016 atomic_inc(&neigh->probes);
1017 consume_skb(skb);
1018}
1019
1020/* Called when a timer expires for a neighbour entry. */
1021
1022static void neigh_timer_handler(struct timer_list *t)
1023{
1024 unsigned long now, next;
1025 struct neighbour *neigh = from_timer(neigh, t, timer);
1026 unsigned int state;
1027 int notify = 0;
1028
1029 write_lock(&neigh->lock);
1030
1031 state = neigh->nud_state;
1032 now = jiffies;
1033 next = now + HZ;
1034
1035 if (!(state & NUD_IN_TIMER))
1036 goto out;
1037
1038 if (state & NUD_REACHABLE) {
1039 if (time_before_eq(now,
1040 neigh->confirmed + neigh->parms->reachable_time)) {
1041 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1042 next = neigh->confirmed + neigh->parms->reachable_time;
1043 } else if (time_before_eq(now,
1044 neigh->used +
1045 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1046 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1047 neigh->nud_state = NUD_DELAY;
1048 neigh->updated = jiffies;
1049 neigh_suspect(neigh);
1050 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1051 } else {
1052 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1053 neigh->nud_state = NUD_STALE;
1054 neigh->updated = jiffies;
1055 neigh_suspect(neigh);
1056 notify = 1;
1057 }
1058 } else if (state & NUD_DELAY) {
1059 if (time_before_eq(now,
1060 neigh->confirmed +
1061 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1062 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1063 neigh->nud_state = NUD_REACHABLE;
1064 neigh->updated = jiffies;
1065 neigh_connect(neigh);
1066 notify = 1;
1067 next = neigh->confirmed + neigh->parms->reachable_time;
1068 } else {
1069 neigh_dbg(2, "neigh %p is probed\n", neigh);
1070 neigh->nud_state = NUD_PROBE;
1071 neigh->updated = jiffies;
1072 atomic_set(&neigh->probes, 0);
1073 notify = 1;
1074 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1075 HZ/100);
1076 }
1077 } else {
1078 /* NUD_PROBE|NUD_INCOMPLETE */
1079 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME), HZ/100);
1080 }
1081
1082 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1083 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1084 neigh->nud_state = NUD_FAILED;
1085 notify = 1;
1086 neigh_invalidate(neigh);
1087 goto out;
1088 }
1089
1090 if (neigh->nud_state & NUD_IN_TIMER) {
1091 if (time_before(next, jiffies + HZ/100))
1092 next = jiffies + HZ/100;
1093 if (!mod_timer(&neigh->timer, next))
1094 neigh_hold(neigh);
1095 }
1096 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1097 neigh_probe(neigh);
1098 } else {
1099out:
1100 write_unlock(&neigh->lock);
1101 }
1102
1103 if (notify)
1104 neigh_update_notify(neigh, 0);
1105
1106 trace_neigh_timer_handler(neigh, 0);
1107
1108 neigh_release(neigh);
1109}
1110
1111int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1112{
1113 int rc;
1114 bool immediate_probe = false;
1115
1116 write_lock_bh(&neigh->lock);
1117
1118 rc = 0;
1119 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1120 goto out_unlock_bh;
1121 if (neigh->dead)
1122 goto out_dead;
1123
1124 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1125 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1126 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1127 unsigned long next, now = jiffies;
1128
1129 atomic_set(&neigh->probes,
1130 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1131 neigh_del_timer(neigh);
1132 neigh->nud_state = NUD_INCOMPLETE;
1133 neigh->updated = now;
1134 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1135 HZ/100);
1136 neigh_add_timer(neigh, next);
1137 immediate_probe = true;
1138 } else {
1139 neigh->nud_state = NUD_FAILED;
1140 neigh->updated = jiffies;
1141 write_unlock_bh(&neigh->lock);
1142
1143 kfree_skb(skb);
1144 return 1;
1145 }
1146 } else if (neigh->nud_state & NUD_STALE) {
1147 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1148 neigh_del_timer(neigh);
1149 neigh->nud_state = NUD_DELAY;
1150 neigh->updated = jiffies;
1151 neigh_add_timer(neigh, jiffies +
1152 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1153 }
1154
1155 if (neigh->nud_state == NUD_INCOMPLETE) {
1156 if (skb) {
1157 while (neigh->arp_queue_len_bytes + skb->truesize >
1158 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1159 struct sk_buff *buff;
1160
1161 buff = __skb_dequeue(&neigh->arp_queue);
1162 if (!buff)
1163 break;
1164 neigh->arp_queue_len_bytes -= buff->truesize;
1165 kfree_skb(buff);
1166 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1167 }
1168 skb_dst_force(skb);
1169 __skb_queue_tail(&neigh->arp_queue, skb);
1170 neigh->arp_queue_len_bytes += skb->truesize;
1171 }
1172 rc = 1;
1173 }
1174out_unlock_bh:
1175 if (immediate_probe)
1176 neigh_probe(neigh);
1177 else
1178 write_unlock(&neigh->lock);
1179 local_bh_enable();
1180 trace_neigh_event_send_done(neigh, rc);
1181 return rc;
1182
1183out_dead:
1184 if (neigh->nud_state & NUD_STALE)
1185 goto out_unlock_bh;
1186 write_unlock_bh(&neigh->lock);
1187 kfree_skb(skb);
1188 trace_neigh_event_send_dead(neigh, 1);
1189 return 1;
1190}
1191EXPORT_SYMBOL(__neigh_event_send);
1192
1193static void neigh_update_hhs(struct neighbour *neigh)
1194{
1195 struct hh_cache *hh;
1196 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1197 = NULL;
1198
1199 if (neigh->dev->header_ops)
1200 update = neigh->dev->header_ops->cache_update;
1201
1202 if (update) {
1203 hh = &neigh->hh;
1204 if (READ_ONCE(hh->hh_len)) {
1205 write_seqlock_bh(&hh->hh_lock);
1206 update(hh, neigh->dev, neigh->ha);
1207 write_sequnlock_bh(&hh->hh_lock);
1208 }
1209 }
1210}
1211
1212
1213
1214/* Generic update routine.
1215 -- lladdr is new lladdr or NULL, if it is not supplied.
1216 -- new is new state.
1217 -- flags
1218 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1219 if it is different.
1220 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1221 lladdr instead of overriding it
1222 if it is different.
1223 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1224
1225 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1226 NTF_ROUTER flag.
1227 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1228 a router.
1229
1230 Caller MUST hold reference count on the entry.
1231 */
1232
1233static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1234 u8 new, u32 flags, u32 nlmsg_pid,
1235 struct netlink_ext_ack *extack)
1236{
1237 bool ext_learn_change = false;
1238 u8 old;
1239 int err;
1240 int notify = 0;
1241 struct net_device *dev;
1242 int update_isrouter = 0;
1243
1244 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1245
1246 write_lock_bh(&neigh->lock);
1247
1248 dev = neigh->dev;
1249 old = neigh->nud_state;
1250 err = -EPERM;
1251
1252 if (neigh->dead) {
1253 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1254 new = old;
1255 goto out;
1256 }
1257 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1258 (old & (NUD_NOARP | NUD_PERMANENT)))
1259 goto out;
1260
1261 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1262
1263 if (!(new & NUD_VALID)) {
1264 neigh_del_timer(neigh);
1265 if (old & NUD_CONNECTED)
1266 neigh_suspect(neigh);
1267 neigh->nud_state = new;
1268 err = 0;
1269 notify = old & NUD_VALID;
1270 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1271 (new & NUD_FAILED)) {
1272 neigh_invalidate(neigh);
1273 notify = 1;
1274 }
1275 goto out;
1276 }
1277
1278 /* Compare new lladdr with cached one */
1279 if (!dev->addr_len) {
1280 /* First case: device needs no address. */
1281 lladdr = neigh->ha;
1282 } else if (lladdr) {
1283 /* The second case: if something is already cached
1284 and a new address is proposed:
1285 - compare new & old
1286 - if they are different, check override flag
1287 */
1288 if ((old & NUD_VALID) &&
1289 !memcmp(lladdr, neigh->ha, dev->addr_len))
1290 lladdr = neigh->ha;
1291 } else {
1292 /* No address is supplied; if we know something,
1293 use it, otherwise discard the request.
1294 */
1295 err = -EINVAL;
1296 if (!(old & NUD_VALID)) {
1297 NL_SET_ERR_MSG(extack, "No link layer address given");
1298 goto out;
1299 }
1300 lladdr = neigh->ha;
1301 }
1302
1303 /* Update confirmed timestamp for neighbour entry after we
1304 * received ARP packet even if it doesn't change IP to MAC binding.
1305 */
1306 if (new & NUD_CONNECTED)
1307 neigh->confirmed = jiffies;
1308
1309 /* If entry was valid and address is not changed,
1310 do not change entry state, if new one is STALE.
1311 */
1312 err = 0;
1313 update_isrouter = flags & NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1314 if (old & NUD_VALID) {
1315 if (lladdr != neigh->ha && !(flags & NEIGH_UPDATE_F_OVERRIDE)) {
1316 update_isrouter = 0;
1317 if ((flags & NEIGH_UPDATE_F_WEAK_OVERRIDE) &&
1318 (old & NUD_CONNECTED)) {
1319 lladdr = neigh->ha;
1320 new = NUD_STALE;
1321 } else
1322 goto out;
1323 } else {
1324 if (lladdr == neigh->ha && new == NUD_STALE &&
1325 !(flags & NEIGH_UPDATE_F_ADMIN))
1326 new = old;
1327 }
1328 }
1329
1330 /* Update timestamp only once we know we will make a change to the
1331 * neighbour entry. Otherwise we risk to move the locktime window with
1332 * noop updates and ignore relevant ARP updates.
1333 */
1334 if (new != old || lladdr != neigh->ha)
1335 neigh->updated = jiffies;
1336
1337 if (new != old) {
1338 neigh_del_timer(neigh);
1339 if (new & NUD_PROBE)
1340 atomic_set(&neigh->probes, 0);
1341 if (new & NUD_IN_TIMER)
1342 neigh_add_timer(neigh, (jiffies +
1343 ((new & NUD_REACHABLE) ?
1344 neigh->parms->reachable_time :
1345 0)));
1346 neigh->nud_state = new;
1347 notify = 1;
1348 }
1349
1350 if (lladdr != neigh->ha) {
1351 write_seqlock(&neigh->ha_lock);
1352 memcpy(&neigh->ha, lladdr, dev->addr_len);
1353 write_sequnlock(&neigh->ha_lock);
1354 neigh_update_hhs(neigh);
1355 if (!(new & NUD_CONNECTED))
1356 neigh->confirmed = jiffies -
1357 (NEIGH_VAR(neigh->parms, BASE_REACHABLE_TIME) << 1);
1358 notify = 1;
1359 }
1360 if (new == old)
1361 goto out;
1362 if (new & NUD_CONNECTED)
1363 neigh_connect(neigh);
1364 else
1365 neigh_suspect(neigh);
1366 if (!(old & NUD_VALID)) {
1367 struct sk_buff *skb;
1368
1369 /* Again: avoid dead loop if something went wrong */
1370
1371 while (neigh->nud_state & NUD_VALID &&
1372 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
1373 struct dst_entry *dst = skb_dst(skb);
1374 struct neighbour *n2, *n1 = neigh;
1375 write_unlock_bh(&neigh->lock);
1376
1377 rcu_read_lock();
1378
1379 /* Why not just use 'neigh' as-is? The problem is that
1380 * things such as shaper, eql, and sch_teql can end up
1381 * using alternative, different, neigh objects to output
1382 * the packet in the output path. So what we need to do
1383 * here is re-lookup the top-level neigh in the path so
1384 * we can reinject the packet there.
1385 */
1386 n2 = NULL;
1387 if (dst && dst->obsolete != DST_OBSOLETE_DEAD) {
1388 n2 = dst_neigh_lookup_skb(dst, skb);
1389 if (n2)
1390 n1 = n2;
1391 }
1392 n1->output(n1, skb);
1393 if (n2)
1394 neigh_release(n2);
1395 rcu_read_unlock();
1396
1397 write_lock_bh(&neigh->lock);
1398 }
1399 __skb_queue_purge(&neigh->arp_queue);
1400 neigh->arp_queue_len_bytes = 0;
1401 }
1402out:
1403 if (update_isrouter)
1404 neigh_update_is_router(neigh, flags, ¬ify);
1405 write_unlock_bh(&neigh->lock);
1406
1407 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1408 neigh_update_gc_list(neigh);
1409
1410 if (notify)
1411 neigh_update_notify(neigh, nlmsg_pid);
1412
1413 trace_neigh_update_done(neigh, err);
1414
1415 return err;
1416}
1417
1418int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1419 u32 flags, u32 nlmsg_pid)
1420{
1421 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1422}
1423EXPORT_SYMBOL(neigh_update);
1424
1425/* Update the neigh to listen temporarily for probe responses, even if it is
1426 * in a NUD_FAILED state. The caller has to hold neigh->lock for writing.
1427 */
1428void __neigh_set_probe_once(struct neighbour *neigh)
1429{
1430 if (neigh->dead)
1431 return;
1432 neigh->updated = jiffies;
1433 if (!(neigh->nud_state & NUD_FAILED))
1434 return;
1435 neigh->nud_state = NUD_INCOMPLETE;
1436 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1437 neigh_add_timer(neigh,
1438 jiffies + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1439 HZ/100));
1440}
1441EXPORT_SYMBOL(__neigh_set_probe_once);
1442
1443struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1444 u8 *lladdr, void *saddr,
1445 struct net_device *dev)
1446{
1447 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1448 lladdr || !dev->addr_len);
1449 if (neigh)
1450 neigh_update(neigh, lladdr, NUD_STALE,
1451 NEIGH_UPDATE_F_OVERRIDE, 0);
1452 return neigh;
1453}
1454EXPORT_SYMBOL(neigh_event_ns);
1455
1456/* called with read_lock_bh(&n->lock); */
1457static void neigh_hh_init(struct neighbour *n)
1458{
1459 struct net_device *dev = n->dev;
1460 __be16 prot = n->tbl->protocol;
1461 struct hh_cache *hh = &n->hh;
1462
1463 write_lock_bh(&n->lock);
1464
1465 /* Only one thread can come in here and initialize the
1466 * hh_cache entry.
1467 */
1468 if (!hh->hh_len)
1469 dev->header_ops->cache(n, hh, prot);
1470
1471 write_unlock_bh(&n->lock);
1472}
1473
1474/* Slow and careful. */
1475
1476int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1477{
1478 int rc = 0;
1479
1480 if (!neigh_event_send(neigh, skb)) {
1481 int err;
1482 struct net_device *dev = neigh->dev;
1483 unsigned int seq;
1484
1485 if (dev->header_ops->cache && !READ_ONCE(neigh->hh.hh_len))
1486 neigh_hh_init(neigh);
1487
1488 do {
1489 __skb_pull(skb, skb_network_offset(skb));
1490 seq = read_seqbegin(&neigh->ha_lock);
1491 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1492 neigh->ha, NULL, skb->len);
1493 } while (read_seqretry(&neigh->ha_lock, seq));
1494
1495 if (err >= 0)
1496 rc = dev_queue_xmit(skb);
1497 else
1498 goto out_kfree_skb;
1499 }
1500out:
1501 return rc;
1502out_kfree_skb:
1503 rc = -EINVAL;
1504 kfree_skb(skb);
1505 goto out;
1506}
1507EXPORT_SYMBOL(neigh_resolve_output);
1508
1509/* As fast as possible without hh cache */
1510
1511int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1512{
1513 struct net_device *dev = neigh->dev;
1514 unsigned int seq;
1515 int err;
1516
1517 do {
1518 __skb_pull(skb, skb_network_offset(skb));
1519 seq = read_seqbegin(&neigh->ha_lock);
1520 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1521 neigh->ha, NULL, skb->len);
1522 } while (read_seqretry(&neigh->ha_lock, seq));
1523
1524 if (err >= 0)
1525 err = dev_queue_xmit(skb);
1526 else {
1527 err = -EINVAL;
1528 kfree_skb(skb);
1529 }
1530 return err;
1531}
1532EXPORT_SYMBOL(neigh_connected_output);
1533
1534int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1535{
1536 return dev_queue_xmit(skb);
1537}
1538EXPORT_SYMBOL(neigh_direct_output);
1539
1540static void neigh_proxy_process(struct timer_list *t)
1541{
1542 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1543 long sched_next = 0;
1544 unsigned long now = jiffies;
1545 struct sk_buff *skb, *n;
1546
1547 spin_lock(&tbl->proxy_queue.lock);
1548
1549 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1550 long tdif = NEIGH_CB(skb)->sched_next - now;
1551
1552 if (tdif <= 0) {
1553 struct net_device *dev = skb->dev;
1554
1555 __skb_unlink(skb, &tbl->proxy_queue);
1556 if (tbl->proxy_redo && netif_running(dev)) {
1557 rcu_read_lock();
1558 tbl->proxy_redo(skb);
1559 rcu_read_unlock();
1560 } else {
1561 kfree_skb(skb);
1562 }
1563
1564 dev_put(dev);
1565 } else if (!sched_next || tdif < sched_next)
1566 sched_next = tdif;
1567 }
1568 del_timer(&tbl->proxy_timer);
1569 if (sched_next)
1570 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1571 spin_unlock(&tbl->proxy_queue.lock);
1572}
1573
1574void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1575 struct sk_buff *skb)
1576{
1577 unsigned long sched_next = jiffies +
1578 prandom_u32_max(NEIGH_VAR(p, PROXY_DELAY));
1579
1580 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1581 kfree_skb(skb);
1582 return;
1583 }
1584
1585 NEIGH_CB(skb)->sched_next = sched_next;
1586 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1587
1588 spin_lock(&tbl->proxy_queue.lock);
1589 if (del_timer(&tbl->proxy_timer)) {
1590 if (time_before(tbl->proxy_timer.expires, sched_next))
1591 sched_next = tbl->proxy_timer.expires;
1592 }
1593 skb_dst_drop(skb);
1594 dev_hold(skb->dev);
1595 __skb_queue_tail(&tbl->proxy_queue, skb);
1596 mod_timer(&tbl->proxy_timer, sched_next);
1597 spin_unlock(&tbl->proxy_queue.lock);
1598}
1599EXPORT_SYMBOL(pneigh_enqueue);
1600
1601static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1602 struct net *net, int ifindex)
1603{
1604 struct neigh_parms *p;
1605
1606 list_for_each_entry(p, &tbl->parms_list, list) {
1607 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1608 (!p->dev && !ifindex && net_eq(net, &init_net)))
1609 return p;
1610 }
1611
1612 return NULL;
1613}
1614
1615struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1616 struct neigh_table *tbl)
1617{
1618 struct neigh_parms *p;
1619 struct net *net = dev_net(dev);
1620 const struct net_device_ops *ops = dev->netdev_ops;
1621
1622 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1623 if (p) {
1624 p->tbl = tbl;
1625 refcount_set(&p->refcnt, 1);
1626 p->reachable_time =
1627 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1628 dev_hold(dev);
1629 p->dev = dev;
1630 write_pnet(&p->net, net);
1631 p->sysctl_table = NULL;
1632
1633 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1634 dev_put(dev);
1635 kfree(p);
1636 return NULL;
1637 }
1638
1639 write_lock_bh(&tbl->lock);
1640 list_add(&p->list, &tbl->parms.list);
1641 write_unlock_bh(&tbl->lock);
1642
1643 neigh_parms_data_state_cleanall(p);
1644 }
1645 return p;
1646}
1647EXPORT_SYMBOL(neigh_parms_alloc);
1648
1649static void neigh_rcu_free_parms(struct rcu_head *head)
1650{
1651 struct neigh_parms *parms =
1652 container_of(head, struct neigh_parms, rcu_head);
1653
1654 neigh_parms_put(parms);
1655}
1656
1657void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1658{
1659 if (!parms || parms == &tbl->parms)
1660 return;
1661 write_lock_bh(&tbl->lock);
1662 list_del(&parms->list);
1663 parms->dead = 1;
1664 write_unlock_bh(&tbl->lock);
1665 if (parms->dev)
1666 dev_put(parms->dev);
1667 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1668}
1669EXPORT_SYMBOL(neigh_parms_release);
1670
1671static void neigh_parms_destroy(struct neigh_parms *parms)
1672{
1673 kfree(parms);
1674}
1675
1676static struct lock_class_key neigh_table_proxy_queue_class;
1677
1678static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1679
1680void neigh_table_init(int index, struct neigh_table *tbl)
1681{
1682 unsigned long now = jiffies;
1683 unsigned long phsize;
1684
1685 INIT_LIST_HEAD(&tbl->parms_list);
1686 INIT_LIST_HEAD(&tbl->gc_list);
1687 list_add(&tbl->parms.list, &tbl->parms_list);
1688 write_pnet(&tbl->parms.net, &init_net);
1689 refcount_set(&tbl->parms.refcnt, 1);
1690 tbl->parms.reachable_time =
1691 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1692
1693 tbl->stats = alloc_percpu(struct neigh_statistics);
1694 if (!tbl->stats)
1695 panic("cannot create neighbour cache statistics");
1696
1697#ifdef CONFIG_PROC_FS
1698 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1699 &neigh_stat_seq_ops, tbl))
1700 panic("cannot create neighbour proc dir entry");
1701#endif
1702
1703 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1704
1705 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1706 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1707
1708 if (!tbl->nht || !tbl->phash_buckets)
1709 panic("cannot allocate neighbour cache hashes");
1710
1711 if (!tbl->entry_size)
1712 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1713 tbl->key_len, NEIGH_PRIV_ALIGN);
1714 else
1715 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1716
1717 rwlock_init(&tbl->lock);
1718 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1719 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1720 tbl->parms.reachable_time);
1721 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1722 skb_queue_head_init_class(&tbl->proxy_queue,
1723 &neigh_table_proxy_queue_class);
1724
1725 tbl->last_flush = now;
1726 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1727
1728 neigh_tables[index] = tbl;
1729}
1730EXPORT_SYMBOL(neigh_table_init);
1731
1732int neigh_table_clear(int index, struct neigh_table *tbl)
1733{
1734 neigh_tables[index] = NULL;
1735 /* It is not clean... Fix it to unload IPv6 module safely */
1736 cancel_delayed_work_sync(&tbl->gc_work);
1737 del_timer_sync(&tbl->proxy_timer);
1738 pneigh_queue_purge(&tbl->proxy_queue);
1739 neigh_ifdown(tbl, NULL);
1740 if (atomic_read(&tbl->entries))
1741 pr_crit("neighbour leakage\n");
1742
1743 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1744 neigh_hash_free_rcu);
1745 tbl->nht = NULL;
1746
1747 kfree(tbl->phash_buckets);
1748 tbl->phash_buckets = NULL;
1749
1750 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1751
1752 free_percpu(tbl->stats);
1753 tbl->stats = NULL;
1754
1755 return 0;
1756}
1757EXPORT_SYMBOL(neigh_table_clear);
1758
1759static struct neigh_table *neigh_find_table(int family)
1760{
1761 struct neigh_table *tbl = NULL;
1762
1763 switch (family) {
1764 case AF_INET:
1765 tbl = neigh_tables[NEIGH_ARP_TABLE];
1766 break;
1767 case AF_INET6:
1768 tbl = neigh_tables[NEIGH_ND_TABLE];
1769 break;
1770 case AF_DECnet:
1771 tbl = neigh_tables[NEIGH_DN_TABLE];
1772 break;
1773 }
1774
1775 return tbl;
1776}
1777
1778const struct nla_policy nda_policy[NDA_MAX+1] = {
1779 [NDA_UNSPEC] = { .strict_start_type = NDA_NH_ID },
1780 [NDA_DST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1781 [NDA_LLADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1782 [NDA_CACHEINFO] = { .len = sizeof(struct nda_cacheinfo) },
1783 [NDA_PROBES] = { .type = NLA_U32 },
1784 [NDA_VLAN] = { .type = NLA_U16 },
1785 [NDA_PORT] = { .type = NLA_U16 },
1786 [NDA_VNI] = { .type = NLA_U32 },
1787 [NDA_IFINDEX] = { .type = NLA_U32 },
1788 [NDA_MASTER] = { .type = NLA_U32 },
1789 [NDA_PROTOCOL] = { .type = NLA_U8 },
1790 [NDA_NH_ID] = { .type = NLA_U32 },
1791 [NDA_FDB_EXT_ATTRS] = { .type = NLA_NESTED },
1792};
1793
1794static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1795 struct netlink_ext_ack *extack)
1796{
1797 struct net *net = sock_net(skb->sk);
1798 struct ndmsg *ndm;
1799 struct nlattr *dst_attr;
1800 struct neigh_table *tbl;
1801 struct neighbour *neigh;
1802 struct net_device *dev = NULL;
1803 int err = -EINVAL;
1804
1805 ASSERT_RTNL();
1806 if (nlmsg_len(nlh) < sizeof(*ndm))
1807 goto out;
1808
1809 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1810 if (!dst_attr) {
1811 NL_SET_ERR_MSG(extack, "Network address not specified");
1812 goto out;
1813 }
1814
1815 ndm = nlmsg_data(nlh);
1816 if (ndm->ndm_ifindex) {
1817 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1818 if (dev == NULL) {
1819 err = -ENODEV;
1820 goto out;
1821 }
1822 }
1823
1824 tbl = neigh_find_table(ndm->ndm_family);
1825 if (tbl == NULL)
1826 return -EAFNOSUPPORT;
1827
1828 if (nla_len(dst_attr) < (int)tbl->key_len) {
1829 NL_SET_ERR_MSG(extack, "Invalid network address");
1830 goto out;
1831 }
1832
1833 if (ndm->ndm_flags & NTF_PROXY) {
1834 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1835 goto out;
1836 }
1837
1838 if (dev == NULL)
1839 goto out;
1840
1841 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1842 if (neigh == NULL) {
1843 err = -ENOENT;
1844 goto out;
1845 }
1846
1847 err = __neigh_update(neigh, NULL, NUD_FAILED,
1848 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1849 NETLINK_CB(skb).portid, extack);
1850 write_lock_bh(&tbl->lock);
1851 neigh_release(neigh);
1852 neigh_remove_one(neigh, tbl);
1853 write_unlock_bh(&tbl->lock);
1854
1855out:
1856 return err;
1857}
1858
1859static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1860 struct netlink_ext_ack *extack)
1861{
1862 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1863 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1864 struct net *net = sock_net(skb->sk);
1865 struct ndmsg *ndm;
1866 struct nlattr *tb[NDA_MAX+1];
1867 struct neigh_table *tbl;
1868 struct net_device *dev = NULL;
1869 struct neighbour *neigh;
1870 void *dst, *lladdr;
1871 u8 protocol = 0;
1872 int err;
1873
1874 ASSERT_RTNL();
1875 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1876 nda_policy, extack);
1877 if (err < 0)
1878 goto out;
1879
1880 err = -EINVAL;
1881 if (!tb[NDA_DST]) {
1882 NL_SET_ERR_MSG(extack, "Network address not specified");
1883 goto out;
1884 }
1885
1886 ndm = nlmsg_data(nlh);
1887 if (ndm->ndm_ifindex) {
1888 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1889 if (dev == NULL) {
1890 err = -ENODEV;
1891 goto out;
1892 }
1893
1894 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1895 NL_SET_ERR_MSG(extack, "Invalid link address");
1896 goto out;
1897 }
1898 }
1899
1900 tbl = neigh_find_table(ndm->ndm_family);
1901 if (tbl == NULL)
1902 return -EAFNOSUPPORT;
1903
1904 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1905 NL_SET_ERR_MSG(extack, "Invalid network address");
1906 goto out;
1907 }
1908
1909 dst = nla_data(tb[NDA_DST]);
1910 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1911
1912 if (tb[NDA_PROTOCOL])
1913 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1914
1915 if (ndm->ndm_flags & NTF_PROXY) {
1916 struct pneigh_entry *pn;
1917
1918 err = -ENOBUFS;
1919 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1920 if (pn) {
1921 pn->flags = ndm->ndm_flags;
1922 if (protocol)
1923 pn->protocol = protocol;
1924 err = 0;
1925 }
1926 goto out;
1927 }
1928
1929 if (!dev) {
1930 NL_SET_ERR_MSG(extack, "Device not specified");
1931 goto out;
1932 }
1933
1934 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1935 err = -EINVAL;
1936 goto out;
1937 }
1938
1939 neigh = neigh_lookup(tbl, dst, dev);
1940 if (neigh == NULL) {
1941 bool exempt_from_gc;
1942
1943 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1944 err = -ENOENT;
1945 goto out;
1946 }
1947
1948 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1949 ndm->ndm_flags & NTF_EXT_LEARNED;
1950 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1951 if (IS_ERR(neigh)) {
1952 err = PTR_ERR(neigh);
1953 goto out;
1954 }
1955 } else {
1956 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1957 err = -EEXIST;
1958 neigh_release(neigh);
1959 goto out;
1960 }
1961
1962 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1963 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1964 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1965 }
1966
1967 if (protocol)
1968 neigh->protocol = protocol;
1969
1970 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1971 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1972
1973 if (ndm->ndm_flags & NTF_ROUTER)
1974 flags |= NEIGH_UPDATE_F_ISROUTER;
1975
1976 if (ndm->ndm_flags & NTF_USE) {
1977 neigh_event_send(neigh, NULL);
1978 err = 0;
1979 } else
1980 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1981 NETLINK_CB(skb).portid, extack);
1982
1983 neigh_release(neigh);
1984
1985out:
1986 return err;
1987}
1988
1989static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1990{
1991 struct nlattr *nest;
1992
1993 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1994 if (nest == NULL)
1995 return -ENOBUFS;
1996
1997 if ((parms->dev &&
1998 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1999 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
2000 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
2001 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
2002 /* approximative value for deprecated QUEUE_LEN (in packets) */
2003 nla_put_u32(skb, NDTPA_QUEUE_LEN,
2004 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
2005 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
2006 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2007 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2008 NEIGH_VAR(parms, UCAST_PROBES)) ||
2009 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2010 NEIGH_VAR(parms, MCAST_PROBES)) ||
2011 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2012 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2013 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2014 NDTPA_PAD) ||
2015 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2016 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2017 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2018 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2019 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2020 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2021 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2022 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2023 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2024 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2025 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2026 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2027 nla_put_msecs(skb, NDTPA_LOCKTIME,
2028 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2029 goto nla_put_failure;
2030 return nla_nest_end(skb, nest);
2031
2032nla_put_failure:
2033 nla_nest_cancel(skb, nest);
2034 return -EMSGSIZE;
2035}
2036
2037static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2038 u32 pid, u32 seq, int type, int flags)
2039{
2040 struct nlmsghdr *nlh;
2041 struct ndtmsg *ndtmsg;
2042
2043 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2044 if (nlh == NULL)
2045 return -EMSGSIZE;
2046
2047 ndtmsg = nlmsg_data(nlh);
2048
2049 read_lock_bh(&tbl->lock);
2050 ndtmsg->ndtm_family = tbl->family;
2051 ndtmsg->ndtm_pad1 = 0;
2052 ndtmsg->ndtm_pad2 = 0;
2053
2054 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2055 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2056 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2057 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2058 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2059 goto nla_put_failure;
2060 {
2061 unsigned long now = jiffies;
2062 long flush_delta = now - tbl->last_flush;
2063 long rand_delta = now - tbl->last_rand;
2064 struct neigh_hash_table *nht;
2065 struct ndt_config ndc = {
2066 .ndtc_key_len = tbl->key_len,
2067 .ndtc_entry_size = tbl->entry_size,
2068 .ndtc_entries = atomic_read(&tbl->entries),
2069 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2070 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2071 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2072 };
2073
2074 rcu_read_lock_bh();
2075 nht = rcu_dereference_bh(tbl->nht);
2076 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2077 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2078 rcu_read_unlock_bh();
2079
2080 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2081 goto nla_put_failure;
2082 }
2083
2084 {
2085 int cpu;
2086 struct ndt_stats ndst;
2087
2088 memset(&ndst, 0, sizeof(ndst));
2089
2090 for_each_possible_cpu(cpu) {
2091 struct neigh_statistics *st;
2092
2093 st = per_cpu_ptr(tbl->stats, cpu);
2094 ndst.ndts_allocs += st->allocs;
2095 ndst.ndts_destroys += st->destroys;
2096 ndst.ndts_hash_grows += st->hash_grows;
2097 ndst.ndts_res_failed += st->res_failed;
2098 ndst.ndts_lookups += st->lookups;
2099 ndst.ndts_hits += st->hits;
2100 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2101 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2102 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2103 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2104 ndst.ndts_table_fulls += st->table_fulls;
2105 }
2106
2107 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2108 NDTA_PAD))
2109 goto nla_put_failure;
2110 }
2111
2112 BUG_ON(tbl->parms.dev);
2113 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2114 goto nla_put_failure;
2115
2116 read_unlock_bh(&tbl->lock);
2117 nlmsg_end(skb, nlh);
2118 return 0;
2119
2120nla_put_failure:
2121 read_unlock_bh(&tbl->lock);
2122 nlmsg_cancel(skb, nlh);
2123 return -EMSGSIZE;
2124}
2125
2126static int neightbl_fill_param_info(struct sk_buff *skb,
2127 struct neigh_table *tbl,
2128 struct neigh_parms *parms,
2129 u32 pid, u32 seq, int type,
2130 unsigned int flags)
2131{
2132 struct ndtmsg *ndtmsg;
2133 struct nlmsghdr *nlh;
2134
2135 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2136 if (nlh == NULL)
2137 return -EMSGSIZE;
2138
2139 ndtmsg = nlmsg_data(nlh);
2140
2141 read_lock_bh(&tbl->lock);
2142 ndtmsg->ndtm_family = tbl->family;
2143 ndtmsg->ndtm_pad1 = 0;
2144 ndtmsg->ndtm_pad2 = 0;
2145
2146 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2147 neightbl_fill_parms(skb, parms) < 0)
2148 goto errout;
2149
2150 read_unlock_bh(&tbl->lock);
2151 nlmsg_end(skb, nlh);
2152 return 0;
2153errout:
2154 read_unlock_bh(&tbl->lock);
2155 nlmsg_cancel(skb, nlh);
2156 return -EMSGSIZE;
2157}
2158
2159static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2160 [NDTA_NAME] = { .type = NLA_STRING },
2161 [NDTA_THRESH1] = { .type = NLA_U32 },
2162 [NDTA_THRESH2] = { .type = NLA_U32 },
2163 [NDTA_THRESH3] = { .type = NLA_U32 },
2164 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2165 [NDTA_PARMS] = { .type = NLA_NESTED },
2166};
2167
2168static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2169 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2170 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2171 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2172 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2173 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2174 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2175 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2176 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2177 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2178 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2179 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2180 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2181 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2182 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2183};
2184
2185static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2186 struct netlink_ext_ack *extack)
2187{
2188 struct net *net = sock_net(skb->sk);
2189 struct neigh_table *tbl;
2190 struct ndtmsg *ndtmsg;
2191 struct nlattr *tb[NDTA_MAX+1];
2192 bool found = false;
2193 int err, tidx;
2194
2195 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2196 nl_neightbl_policy, extack);
2197 if (err < 0)
2198 goto errout;
2199
2200 if (tb[NDTA_NAME] == NULL) {
2201 err = -EINVAL;
2202 goto errout;
2203 }
2204
2205 ndtmsg = nlmsg_data(nlh);
2206
2207 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2208 tbl = neigh_tables[tidx];
2209 if (!tbl)
2210 continue;
2211 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2212 continue;
2213 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2214 found = true;
2215 break;
2216 }
2217 }
2218
2219 if (!found)
2220 return -ENOENT;
2221
2222 /*
2223 * We acquire tbl->lock to be nice to the periodic timers and
2224 * make sure they always see a consistent set of values.
2225 */
2226 write_lock_bh(&tbl->lock);
2227
2228 if (tb[NDTA_PARMS]) {
2229 struct nlattr *tbp[NDTPA_MAX+1];
2230 struct neigh_parms *p;
2231 int i, ifindex = 0;
2232
2233 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2234 tb[NDTA_PARMS],
2235 nl_ntbl_parm_policy, extack);
2236 if (err < 0)
2237 goto errout_tbl_lock;
2238
2239 if (tbp[NDTPA_IFINDEX])
2240 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2241
2242 p = lookup_neigh_parms(tbl, net, ifindex);
2243 if (p == NULL) {
2244 err = -ENOENT;
2245 goto errout_tbl_lock;
2246 }
2247
2248 for (i = 1; i <= NDTPA_MAX; i++) {
2249 if (tbp[i] == NULL)
2250 continue;
2251
2252 switch (i) {
2253 case NDTPA_QUEUE_LEN:
2254 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2255 nla_get_u32(tbp[i]) *
2256 SKB_TRUESIZE(ETH_FRAME_LEN));
2257 break;
2258 case NDTPA_QUEUE_LENBYTES:
2259 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2260 nla_get_u32(tbp[i]));
2261 break;
2262 case NDTPA_PROXY_QLEN:
2263 NEIGH_VAR_SET(p, PROXY_QLEN,
2264 nla_get_u32(tbp[i]));
2265 break;
2266 case NDTPA_APP_PROBES:
2267 NEIGH_VAR_SET(p, APP_PROBES,
2268 nla_get_u32(tbp[i]));
2269 break;
2270 case NDTPA_UCAST_PROBES:
2271 NEIGH_VAR_SET(p, UCAST_PROBES,
2272 nla_get_u32(tbp[i]));
2273 break;
2274 case NDTPA_MCAST_PROBES:
2275 NEIGH_VAR_SET(p, MCAST_PROBES,
2276 nla_get_u32(tbp[i]));
2277 break;
2278 case NDTPA_MCAST_REPROBES:
2279 NEIGH_VAR_SET(p, MCAST_REPROBES,
2280 nla_get_u32(tbp[i]));
2281 break;
2282 case NDTPA_BASE_REACHABLE_TIME:
2283 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2284 nla_get_msecs(tbp[i]));
2285 /* update reachable_time as well, otherwise, the change will
2286 * only be effective after the next time neigh_periodic_work
2287 * decides to recompute it (can be multiple minutes)
2288 */
2289 p->reachable_time =
2290 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2291 break;
2292 case NDTPA_GC_STALETIME:
2293 NEIGH_VAR_SET(p, GC_STALETIME,
2294 nla_get_msecs(tbp[i]));
2295 break;
2296 case NDTPA_DELAY_PROBE_TIME:
2297 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2298 nla_get_msecs(tbp[i]));
2299 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2300 break;
2301 case NDTPA_RETRANS_TIME:
2302 NEIGH_VAR_SET(p, RETRANS_TIME,
2303 nla_get_msecs(tbp[i]));
2304 break;
2305 case NDTPA_ANYCAST_DELAY:
2306 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2307 nla_get_msecs(tbp[i]));
2308 break;
2309 case NDTPA_PROXY_DELAY:
2310 NEIGH_VAR_SET(p, PROXY_DELAY,
2311 nla_get_msecs(tbp[i]));
2312 break;
2313 case NDTPA_LOCKTIME:
2314 NEIGH_VAR_SET(p, LOCKTIME,
2315 nla_get_msecs(tbp[i]));
2316 break;
2317 }
2318 }
2319 }
2320
2321 err = -ENOENT;
2322 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2323 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2324 !net_eq(net, &init_net))
2325 goto errout_tbl_lock;
2326
2327 if (tb[NDTA_THRESH1])
2328 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2329
2330 if (tb[NDTA_THRESH2])
2331 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2332
2333 if (tb[NDTA_THRESH3])
2334 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2335
2336 if (tb[NDTA_GC_INTERVAL])
2337 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2338
2339 err = 0;
2340
2341errout_tbl_lock:
2342 write_unlock_bh(&tbl->lock);
2343errout:
2344 return err;
2345}
2346
2347static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2348 struct netlink_ext_ack *extack)
2349{
2350 struct ndtmsg *ndtm;
2351
2352 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2353 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2354 return -EINVAL;
2355 }
2356
2357 ndtm = nlmsg_data(nlh);
2358 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2359 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2360 return -EINVAL;
2361 }
2362
2363 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2364 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2365 return -EINVAL;
2366 }
2367
2368 return 0;
2369}
2370
2371static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2372{
2373 const struct nlmsghdr *nlh = cb->nlh;
2374 struct net *net = sock_net(skb->sk);
2375 int family, tidx, nidx = 0;
2376 int tbl_skip = cb->args[0];
2377 int neigh_skip = cb->args[1];
2378 struct neigh_table *tbl;
2379
2380 if (cb->strict_check) {
2381 int err = neightbl_valid_dump_info(nlh, cb->extack);
2382
2383 if (err < 0)
2384 return err;
2385 }
2386
2387 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2388
2389 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2390 struct neigh_parms *p;
2391
2392 tbl = neigh_tables[tidx];
2393 if (!tbl)
2394 continue;
2395
2396 if (tidx < tbl_skip || (family && tbl->family != family))
2397 continue;
2398
2399 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2400 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2401 NLM_F_MULTI) < 0)
2402 break;
2403
2404 nidx = 0;
2405 p = list_next_entry(&tbl->parms, list);
2406 list_for_each_entry_from(p, &tbl->parms_list, list) {
2407 if (!net_eq(neigh_parms_net(p), net))
2408 continue;
2409
2410 if (nidx < neigh_skip)
2411 goto next;
2412
2413 if (neightbl_fill_param_info(skb, tbl, p,
2414 NETLINK_CB(cb->skb).portid,
2415 nlh->nlmsg_seq,
2416 RTM_NEWNEIGHTBL,
2417 NLM_F_MULTI) < 0)
2418 goto out;
2419 next:
2420 nidx++;
2421 }
2422
2423 neigh_skip = 0;
2424 }
2425out:
2426 cb->args[0] = tidx;
2427 cb->args[1] = nidx;
2428
2429 return skb->len;
2430}
2431
2432static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2433 u32 pid, u32 seq, int type, unsigned int flags)
2434{
2435 unsigned long now = jiffies;
2436 struct nda_cacheinfo ci;
2437 struct nlmsghdr *nlh;
2438 struct ndmsg *ndm;
2439
2440 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2441 if (nlh == NULL)
2442 return -EMSGSIZE;
2443
2444 ndm = nlmsg_data(nlh);
2445 ndm->ndm_family = neigh->ops->family;
2446 ndm->ndm_pad1 = 0;
2447 ndm->ndm_pad2 = 0;
2448 ndm->ndm_flags = neigh->flags;
2449 ndm->ndm_type = neigh->type;
2450 ndm->ndm_ifindex = neigh->dev->ifindex;
2451
2452 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2453 goto nla_put_failure;
2454
2455 read_lock_bh(&neigh->lock);
2456 ndm->ndm_state = neigh->nud_state;
2457 if (neigh->nud_state & NUD_VALID) {
2458 char haddr[MAX_ADDR_LEN];
2459
2460 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2461 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2462 read_unlock_bh(&neigh->lock);
2463 goto nla_put_failure;
2464 }
2465 }
2466
2467 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2468 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2469 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2470 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2471 read_unlock_bh(&neigh->lock);
2472
2473 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2474 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2475 goto nla_put_failure;
2476
2477 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2478 goto nla_put_failure;
2479
2480 nlmsg_end(skb, nlh);
2481 return 0;
2482
2483nla_put_failure:
2484 nlmsg_cancel(skb, nlh);
2485 return -EMSGSIZE;
2486}
2487
2488static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2489 u32 pid, u32 seq, int type, unsigned int flags,
2490 struct neigh_table *tbl)
2491{
2492 struct nlmsghdr *nlh;
2493 struct ndmsg *ndm;
2494
2495 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2496 if (nlh == NULL)
2497 return -EMSGSIZE;
2498
2499 ndm = nlmsg_data(nlh);
2500 ndm->ndm_family = tbl->family;
2501 ndm->ndm_pad1 = 0;
2502 ndm->ndm_pad2 = 0;
2503 ndm->ndm_flags = pn->flags | NTF_PROXY;
2504 ndm->ndm_type = RTN_UNICAST;
2505 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2506 ndm->ndm_state = NUD_NONE;
2507
2508 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2509 goto nla_put_failure;
2510
2511 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2512 goto nla_put_failure;
2513
2514 nlmsg_end(skb, nlh);
2515 return 0;
2516
2517nla_put_failure:
2518 nlmsg_cancel(skb, nlh);
2519 return -EMSGSIZE;
2520}
2521
2522static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2523{
2524 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2525 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2526}
2527
2528static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2529{
2530 struct net_device *master;
2531
2532 if (!master_idx)
2533 return false;
2534
2535 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2536 if (!master || master->ifindex != master_idx)
2537 return true;
2538
2539 return false;
2540}
2541
2542static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2543{
2544 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2545 return true;
2546
2547 return false;
2548}
2549
2550struct neigh_dump_filter {
2551 int master_idx;
2552 int dev_idx;
2553};
2554
2555static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2556 struct netlink_callback *cb,
2557 struct neigh_dump_filter *filter)
2558{
2559 struct net *net = sock_net(skb->sk);
2560 struct neighbour *n;
2561 int rc, h, s_h = cb->args[1];
2562 int idx, s_idx = idx = cb->args[2];
2563 struct neigh_hash_table *nht;
2564 unsigned int flags = NLM_F_MULTI;
2565
2566 if (filter->dev_idx || filter->master_idx)
2567 flags |= NLM_F_DUMP_FILTERED;
2568
2569 rcu_read_lock_bh();
2570 nht = rcu_dereference_bh(tbl->nht);
2571
2572 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2573 if (h > s_h)
2574 s_idx = 0;
2575 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2576 n != NULL;
2577 n = rcu_dereference_bh(n->next)) {
2578 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2579 goto next;
2580 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2581 neigh_master_filtered(n->dev, filter->master_idx))
2582 goto next;
2583 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2584 cb->nlh->nlmsg_seq,
2585 RTM_NEWNEIGH,
2586 flags) < 0) {
2587 rc = -1;
2588 goto out;
2589 }
2590next:
2591 idx++;
2592 }
2593 }
2594 rc = skb->len;
2595out:
2596 rcu_read_unlock_bh();
2597 cb->args[1] = h;
2598 cb->args[2] = idx;
2599 return rc;
2600}
2601
2602static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2603 struct netlink_callback *cb,
2604 struct neigh_dump_filter *filter)
2605{
2606 struct pneigh_entry *n;
2607 struct net *net = sock_net(skb->sk);
2608 int rc, h, s_h = cb->args[3];
2609 int idx, s_idx = idx = cb->args[4];
2610 unsigned int flags = NLM_F_MULTI;
2611
2612 if (filter->dev_idx || filter->master_idx)
2613 flags |= NLM_F_DUMP_FILTERED;
2614
2615 read_lock_bh(&tbl->lock);
2616
2617 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2618 if (h > s_h)
2619 s_idx = 0;
2620 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2621 if (idx < s_idx || pneigh_net(n) != net)
2622 goto next;
2623 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2624 neigh_master_filtered(n->dev, filter->master_idx))
2625 goto next;
2626 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2627 cb->nlh->nlmsg_seq,
2628 RTM_NEWNEIGH, flags, tbl) < 0) {
2629 read_unlock_bh(&tbl->lock);
2630 rc = -1;
2631 goto out;
2632 }
2633 next:
2634 idx++;
2635 }
2636 }
2637
2638 read_unlock_bh(&tbl->lock);
2639 rc = skb->len;
2640out:
2641 cb->args[3] = h;
2642 cb->args[4] = idx;
2643 return rc;
2644
2645}
2646
2647static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2648 bool strict_check,
2649 struct neigh_dump_filter *filter,
2650 struct netlink_ext_ack *extack)
2651{
2652 struct nlattr *tb[NDA_MAX + 1];
2653 int err, i;
2654
2655 if (strict_check) {
2656 struct ndmsg *ndm;
2657
2658 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2659 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2660 return -EINVAL;
2661 }
2662
2663 ndm = nlmsg_data(nlh);
2664 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2665 ndm->ndm_state || ndm->ndm_type) {
2666 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2667 return -EINVAL;
2668 }
2669
2670 if (ndm->ndm_flags & ~NTF_PROXY) {
2671 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2672 return -EINVAL;
2673 }
2674
2675 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2676 tb, NDA_MAX, nda_policy,
2677 extack);
2678 } else {
2679 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2680 NDA_MAX, nda_policy, extack);
2681 }
2682 if (err < 0)
2683 return err;
2684
2685 for (i = 0; i <= NDA_MAX; ++i) {
2686 if (!tb[i])
2687 continue;
2688
2689 /* all new attributes should require strict_check */
2690 switch (i) {
2691 case NDA_IFINDEX:
2692 filter->dev_idx = nla_get_u32(tb[i]);
2693 break;
2694 case NDA_MASTER:
2695 filter->master_idx = nla_get_u32(tb[i]);
2696 break;
2697 default:
2698 if (strict_check) {
2699 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2700 return -EINVAL;
2701 }
2702 }
2703 }
2704
2705 return 0;
2706}
2707
2708static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2709{
2710 const struct nlmsghdr *nlh = cb->nlh;
2711 struct neigh_dump_filter filter = {};
2712 struct neigh_table *tbl;
2713 int t, family, s_t;
2714 int proxy = 0;
2715 int err;
2716
2717 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2718
2719 /* check for full ndmsg structure presence, family member is
2720 * the same for both structures
2721 */
2722 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2723 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2724 proxy = 1;
2725
2726 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2727 if (err < 0 && cb->strict_check)
2728 return err;
2729
2730 s_t = cb->args[0];
2731
2732 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2733 tbl = neigh_tables[t];
2734
2735 if (!tbl)
2736 continue;
2737 if (t < s_t || (family && tbl->family != family))
2738 continue;
2739 if (t > s_t)
2740 memset(&cb->args[1], 0, sizeof(cb->args) -
2741 sizeof(cb->args[0]));
2742 if (proxy)
2743 err = pneigh_dump_table(tbl, skb, cb, &filter);
2744 else
2745 err = neigh_dump_table(tbl, skb, cb, &filter);
2746 if (err < 0)
2747 break;
2748 }
2749
2750 cb->args[0] = t;
2751 return skb->len;
2752}
2753
2754static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2755 struct neigh_table **tbl,
2756 void **dst, int *dev_idx, u8 *ndm_flags,
2757 struct netlink_ext_ack *extack)
2758{
2759 struct nlattr *tb[NDA_MAX + 1];
2760 struct ndmsg *ndm;
2761 int err, i;
2762
2763 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2764 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2765 return -EINVAL;
2766 }
2767
2768 ndm = nlmsg_data(nlh);
2769 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2770 ndm->ndm_type) {
2771 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2772 return -EINVAL;
2773 }
2774
2775 if (ndm->ndm_flags & ~NTF_PROXY) {
2776 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2777 return -EINVAL;
2778 }
2779
2780 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2781 NDA_MAX, nda_policy, extack);
2782 if (err < 0)
2783 return err;
2784
2785 *ndm_flags = ndm->ndm_flags;
2786 *dev_idx = ndm->ndm_ifindex;
2787 *tbl = neigh_find_table(ndm->ndm_family);
2788 if (*tbl == NULL) {
2789 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2790 return -EAFNOSUPPORT;
2791 }
2792
2793 for (i = 0; i <= NDA_MAX; ++i) {
2794 if (!tb[i])
2795 continue;
2796
2797 switch (i) {
2798 case NDA_DST:
2799 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2800 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2801 return -EINVAL;
2802 }
2803 *dst = nla_data(tb[i]);
2804 break;
2805 default:
2806 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2807 return -EINVAL;
2808 }
2809 }
2810
2811 return 0;
2812}
2813
2814static inline size_t neigh_nlmsg_size(void)
2815{
2816 return NLMSG_ALIGN(sizeof(struct ndmsg))
2817 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2818 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2819 + nla_total_size(sizeof(struct nda_cacheinfo))
2820 + nla_total_size(4) /* NDA_PROBES */
2821 + nla_total_size(1); /* NDA_PROTOCOL */
2822}
2823
2824static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2825 u32 pid, u32 seq)
2826{
2827 struct sk_buff *skb;
2828 int err = 0;
2829
2830 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2831 if (!skb)
2832 return -ENOBUFS;
2833
2834 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2835 if (err) {
2836 kfree_skb(skb);
2837 goto errout;
2838 }
2839
2840 err = rtnl_unicast(skb, net, pid);
2841errout:
2842 return err;
2843}
2844
2845static inline size_t pneigh_nlmsg_size(void)
2846{
2847 return NLMSG_ALIGN(sizeof(struct ndmsg))
2848 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2849 + nla_total_size(1); /* NDA_PROTOCOL */
2850}
2851
2852static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2853 u32 pid, u32 seq, struct neigh_table *tbl)
2854{
2855 struct sk_buff *skb;
2856 int err = 0;
2857
2858 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2859 if (!skb)
2860 return -ENOBUFS;
2861
2862 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2863 if (err) {
2864 kfree_skb(skb);
2865 goto errout;
2866 }
2867
2868 err = rtnl_unicast(skb, net, pid);
2869errout:
2870 return err;
2871}
2872
2873static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2874 struct netlink_ext_ack *extack)
2875{
2876 struct net *net = sock_net(in_skb->sk);
2877 struct net_device *dev = NULL;
2878 struct neigh_table *tbl = NULL;
2879 struct neighbour *neigh;
2880 void *dst = NULL;
2881 u8 ndm_flags = 0;
2882 int dev_idx = 0;
2883 int err;
2884
2885 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2886 extack);
2887 if (err < 0)
2888 return err;
2889
2890 if (dev_idx) {
2891 dev = __dev_get_by_index(net, dev_idx);
2892 if (!dev) {
2893 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2894 return -ENODEV;
2895 }
2896 }
2897
2898 if (!dst) {
2899 NL_SET_ERR_MSG(extack, "Network address not specified");
2900 return -EINVAL;
2901 }
2902
2903 if (ndm_flags & NTF_PROXY) {
2904 struct pneigh_entry *pn;
2905
2906 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2907 if (!pn) {
2908 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2909 return -ENOENT;
2910 }
2911 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2912 nlh->nlmsg_seq, tbl);
2913 }
2914
2915 if (!dev) {
2916 NL_SET_ERR_MSG(extack, "No device specified");
2917 return -EINVAL;
2918 }
2919
2920 neigh = neigh_lookup(tbl, dst, dev);
2921 if (!neigh) {
2922 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2923 return -ENOENT;
2924 }
2925
2926 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2927 nlh->nlmsg_seq);
2928
2929 neigh_release(neigh);
2930
2931 return err;
2932}
2933
2934void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2935{
2936 int chain;
2937 struct neigh_hash_table *nht;
2938
2939 rcu_read_lock_bh();
2940 nht = rcu_dereference_bh(tbl->nht);
2941
2942 read_lock(&tbl->lock); /* avoid resizes */
2943 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2944 struct neighbour *n;
2945
2946 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2947 n != NULL;
2948 n = rcu_dereference_bh(n->next))
2949 cb(n, cookie);
2950 }
2951 read_unlock(&tbl->lock);
2952 rcu_read_unlock_bh();
2953}
2954EXPORT_SYMBOL(neigh_for_each);
2955
2956/* The tbl->lock must be held as a writer and BH disabled. */
2957void __neigh_for_each_release(struct neigh_table *tbl,
2958 int (*cb)(struct neighbour *))
2959{
2960 int chain;
2961 struct neigh_hash_table *nht;
2962
2963 nht = rcu_dereference_protected(tbl->nht,
2964 lockdep_is_held(&tbl->lock));
2965 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2966 struct neighbour *n;
2967 struct neighbour __rcu **np;
2968
2969 np = &nht->hash_buckets[chain];
2970 while ((n = rcu_dereference_protected(*np,
2971 lockdep_is_held(&tbl->lock))) != NULL) {
2972 int release;
2973
2974 write_lock(&n->lock);
2975 release = cb(n);
2976 if (release) {
2977 rcu_assign_pointer(*np,
2978 rcu_dereference_protected(n->next,
2979 lockdep_is_held(&tbl->lock)));
2980 neigh_mark_dead(n);
2981 } else
2982 np = &n->next;
2983 write_unlock(&n->lock);
2984 if (release)
2985 neigh_cleanup_and_release(n);
2986 }
2987 }
2988}
2989EXPORT_SYMBOL(__neigh_for_each_release);
2990
2991int neigh_xmit(int index, struct net_device *dev,
2992 const void *addr, struct sk_buff *skb)
2993{
2994 int err = -EAFNOSUPPORT;
2995 if (likely(index < NEIGH_NR_TABLES)) {
2996 struct neigh_table *tbl;
2997 struct neighbour *neigh;
2998
2999 tbl = neigh_tables[index];
3000 if (!tbl)
3001 goto out;
3002 rcu_read_lock_bh();
3003 if (index == NEIGH_ARP_TABLE) {
3004 u32 key = *((u32 *)addr);
3005
3006 neigh = __ipv4_neigh_lookup_noref(dev, key);
3007 } else {
3008 neigh = __neigh_lookup_noref(tbl, addr, dev);
3009 }
3010 if (!neigh)
3011 neigh = __neigh_create(tbl, addr, dev, false);
3012 err = PTR_ERR(neigh);
3013 if (IS_ERR(neigh)) {
3014 rcu_read_unlock_bh();
3015 goto out_kfree_skb;
3016 }
3017 err = neigh->output(neigh, skb);
3018 rcu_read_unlock_bh();
3019 }
3020 else if (index == NEIGH_LINK_TABLE) {
3021 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3022 addr, NULL, skb->len);
3023 if (err < 0)
3024 goto out_kfree_skb;
3025 err = dev_queue_xmit(skb);
3026 }
3027out:
3028 return err;
3029out_kfree_skb:
3030 kfree_skb(skb);
3031 goto out;
3032}
3033EXPORT_SYMBOL(neigh_xmit);
3034
3035#ifdef CONFIG_PROC_FS
3036
3037static struct neighbour *neigh_get_first(struct seq_file *seq)
3038{
3039 struct neigh_seq_state *state = seq->private;
3040 struct net *net = seq_file_net(seq);
3041 struct neigh_hash_table *nht = state->nht;
3042 struct neighbour *n = NULL;
3043 int bucket;
3044
3045 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3046 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3047 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3048
3049 while (n) {
3050 if (!net_eq(dev_net(n->dev), net))
3051 goto next;
3052 if (state->neigh_sub_iter) {
3053 loff_t fakep = 0;
3054 void *v;
3055
3056 v = state->neigh_sub_iter(state, n, &fakep);
3057 if (!v)
3058 goto next;
3059 }
3060 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3061 break;
3062 if (n->nud_state & ~NUD_NOARP)
3063 break;
3064next:
3065 n = rcu_dereference_bh(n->next);
3066 }
3067
3068 if (n)
3069 break;
3070 }
3071 state->bucket = bucket;
3072
3073 return n;
3074}
3075
3076static struct neighbour *neigh_get_next(struct seq_file *seq,
3077 struct neighbour *n,
3078 loff_t *pos)
3079{
3080 struct neigh_seq_state *state = seq->private;
3081 struct net *net = seq_file_net(seq);
3082 struct neigh_hash_table *nht = state->nht;
3083
3084 if (state->neigh_sub_iter) {
3085 void *v = state->neigh_sub_iter(state, n, pos);
3086 if (v)
3087 return n;
3088 }
3089 n = rcu_dereference_bh(n->next);
3090
3091 while (1) {
3092 while (n) {
3093 if (!net_eq(dev_net(n->dev), net))
3094 goto next;
3095 if (state->neigh_sub_iter) {
3096 void *v = state->neigh_sub_iter(state, n, pos);
3097 if (v)
3098 return n;
3099 goto next;
3100 }
3101 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3102 break;
3103
3104 if (n->nud_state & ~NUD_NOARP)
3105 break;
3106next:
3107 n = rcu_dereference_bh(n->next);
3108 }
3109
3110 if (n)
3111 break;
3112
3113 if (++state->bucket >= (1 << nht->hash_shift))
3114 break;
3115
3116 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3117 }
3118
3119 if (n && pos)
3120 --(*pos);
3121 return n;
3122}
3123
3124static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3125{
3126 struct neighbour *n = neigh_get_first(seq);
3127
3128 if (n) {
3129 --(*pos);
3130 while (*pos) {
3131 n = neigh_get_next(seq, n, pos);
3132 if (!n)
3133 break;
3134 }
3135 }
3136 return *pos ? NULL : n;
3137}
3138
3139static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3140{
3141 struct neigh_seq_state *state = seq->private;
3142 struct net *net = seq_file_net(seq);
3143 struct neigh_table *tbl = state->tbl;
3144 struct pneigh_entry *pn = NULL;
3145 int bucket;
3146
3147 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3148 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3149 pn = tbl->phash_buckets[bucket];
3150 while (pn && !net_eq(pneigh_net(pn), net))
3151 pn = pn->next;
3152 if (pn)
3153 break;
3154 }
3155 state->bucket = bucket;
3156
3157 return pn;
3158}
3159
3160static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3161 struct pneigh_entry *pn,
3162 loff_t *pos)
3163{
3164 struct neigh_seq_state *state = seq->private;
3165 struct net *net = seq_file_net(seq);
3166 struct neigh_table *tbl = state->tbl;
3167
3168 do {
3169 pn = pn->next;
3170 } while (pn && !net_eq(pneigh_net(pn), net));
3171
3172 while (!pn) {
3173 if (++state->bucket > PNEIGH_HASHMASK)
3174 break;
3175 pn = tbl->phash_buckets[state->bucket];
3176 while (pn && !net_eq(pneigh_net(pn), net))
3177 pn = pn->next;
3178 if (pn)
3179 break;
3180 }
3181
3182 if (pn && pos)
3183 --(*pos);
3184
3185 return pn;
3186}
3187
3188static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3189{
3190 struct pneigh_entry *pn = pneigh_get_first(seq);
3191
3192 if (pn) {
3193 --(*pos);
3194 while (*pos) {
3195 pn = pneigh_get_next(seq, pn, pos);
3196 if (!pn)
3197 break;
3198 }
3199 }
3200 return *pos ? NULL : pn;
3201}
3202
3203static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3204{
3205 struct neigh_seq_state *state = seq->private;
3206 void *rc;
3207 loff_t idxpos = *pos;
3208
3209 rc = neigh_get_idx(seq, &idxpos);
3210 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3211 rc = pneigh_get_idx(seq, &idxpos);
3212
3213 return rc;
3214}
3215
3216void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3217 __acquires(tbl->lock)
3218 __acquires(rcu_bh)
3219{
3220 struct neigh_seq_state *state = seq->private;
3221
3222 state->tbl = tbl;
3223 state->bucket = 0;
3224 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3225
3226 rcu_read_lock_bh();
3227 state->nht = rcu_dereference_bh(tbl->nht);
3228 read_lock(&tbl->lock);
3229
3230 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3231}
3232EXPORT_SYMBOL(neigh_seq_start);
3233
3234void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3235{
3236 struct neigh_seq_state *state;
3237 void *rc;
3238
3239 if (v == SEQ_START_TOKEN) {
3240 rc = neigh_get_first(seq);
3241 goto out;
3242 }
3243
3244 state = seq->private;
3245 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3246 rc = neigh_get_next(seq, v, NULL);
3247 if (rc)
3248 goto out;
3249 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3250 rc = pneigh_get_first(seq);
3251 } else {
3252 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3253 rc = pneigh_get_next(seq, v, NULL);
3254 }
3255out:
3256 ++(*pos);
3257 return rc;
3258}
3259EXPORT_SYMBOL(neigh_seq_next);
3260
3261void neigh_seq_stop(struct seq_file *seq, void *v)
3262 __releases(tbl->lock)
3263 __releases(rcu_bh)
3264{
3265 struct neigh_seq_state *state = seq->private;
3266 struct neigh_table *tbl = state->tbl;
3267
3268 read_unlock(&tbl->lock);
3269 rcu_read_unlock_bh();
3270}
3271EXPORT_SYMBOL(neigh_seq_stop);
3272
3273/* statistics via seq_file */
3274
3275static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3276{
3277 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3278 int cpu;
3279
3280 if (*pos == 0)
3281 return SEQ_START_TOKEN;
3282
3283 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3284 if (!cpu_possible(cpu))
3285 continue;
3286 *pos = cpu+1;
3287 return per_cpu_ptr(tbl->stats, cpu);
3288 }
3289 return NULL;
3290}
3291
3292static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3293{
3294 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3295 int cpu;
3296
3297 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3298 if (!cpu_possible(cpu))
3299 continue;
3300 *pos = cpu+1;
3301 return per_cpu_ptr(tbl->stats, cpu);
3302 }
3303 (*pos)++;
3304 return NULL;
3305}
3306
3307static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3308{
3309
3310}
3311
3312static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3313{
3314 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3315 struct neigh_statistics *st = v;
3316
3317 if (v == SEQ_START_TOKEN) {
3318 seq_printf(seq, "entries allocs destroys hash_grows lookups hits res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards table_fulls\n");
3319 return 0;
3320 }
3321
3322 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3323 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
3324 atomic_read(&tbl->entries),
3325
3326 st->allocs,
3327 st->destroys,
3328 st->hash_grows,
3329
3330 st->lookups,
3331 st->hits,
3332
3333 st->res_failed,
3334
3335 st->rcv_probes_mcast,
3336 st->rcv_probes_ucast,
3337
3338 st->periodic_gc_runs,
3339 st->forced_gc_runs,
3340 st->unres_discards,
3341 st->table_fulls
3342 );
3343
3344 return 0;
3345}
3346
3347static const struct seq_operations neigh_stat_seq_ops = {
3348 .start = neigh_stat_seq_start,
3349 .next = neigh_stat_seq_next,
3350 .stop = neigh_stat_seq_stop,
3351 .show = neigh_stat_seq_show,
3352};
3353#endif /* CONFIG_PROC_FS */
3354
3355static void __neigh_notify(struct neighbour *n, int type, int flags,
3356 u32 pid)
3357{
3358 struct net *net = dev_net(n->dev);
3359 struct sk_buff *skb;
3360 int err = -ENOBUFS;
3361
3362 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3363 if (skb == NULL)
3364 goto errout;
3365
3366 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3367 if (err < 0) {
3368 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3369 WARN_ON(err == -EMSGSIZE);
3370 kfree_skb(skb);
3371 goto errout;
3372 }
3373 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3374 return;
3375errout:
3376 if (err < 0)
3377 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3378}
3379
3380void neigh_app_ns(struct neighbour *n)
3381{
3382 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3383}
3384EXPORT_SYMBOL(neigh_app_ns);
3385
3386#ifdef CONFIG_SYSCTL
3387static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3388
3389static int proc_unres_qlen(struct ctl_table *ctl, int write,
3390 void *buffer, size_t *lenp, loff_t *ppos)
3391{
3392 int size, ret;
3393 struct ctl_table tmp = *ctl;
3394
3395 tmp.extra1 = SYSCTL_ZERO;
3396 tmp.extra2 = &unres_qlen_max;
3397 tmp.data = &size;
3398
3399 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3400 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3401
3402 if (write && !ret)
3403 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3404 return ret;
3405}
3406
3407static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3408 int family)
3409{
3410 switch (family) {
3411 case AF_INET:
3412 return __in_dev_arp_parms_get_rcu(dev);
3413 case AF_INET6:
3414 return __in6_dev_nd_parms_get_rcu(dev);
3415 }
3416 return NULL;
3417}
3418
3419static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3420 int index)
3421{
3422 struct net_device *dev;
3423 int family = neigh_parms_family(p);
3424
3425 rcu_read_lock();
3426 for_each_netdev_rcu(net, dev) {
3427 struct neigh_parms *dst_p =
3428 neigh_get_dev_parms_rcu(dev, family);
3429
3430 if (dst_p && !test_bit(index, dst_p->data_state))
3431 dst_p->data[index] = p->data[index];
3432 }
3433 rcu_read_unlock();
3434}
3435
3436static void neigh_proc_update(struct ctl_table *ctl, int write)
3437{
3438 struct net_device *dev = ctl->extra1;
3439 struct neigh_parms *p = ctl->extra2;
3440 struct net *net = neigh_parms_net(p);
3441 int index = (int *) ctl->data - p->data;
3442
3443 if (!write)
3444 return;
3445
3446 set_bit(index, p->data_state);
3447 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3448 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3449 if (!dev) /* NULL dev means this is default value */
3450 neigh_copy_dflt_parms(net, p, index);
3451}
3452
3453static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3454 void *buffer, size_t *lenp,
3455 loff_t *ppos)
3456{
3457 struct ctl_table tmp = *ctl;
3458 int ret;
3459
3460 tmp.extra1 = SYSCTL_ZERO;
3461 tmp.extra2 = SYSCTL_INT_MAX;
3462
3463 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3464 neigh_proc_update(ctl, write);
3465 return ret;
3466}
3467
3468int neigh_proc_dointvec(struct ctl_table *ctl, int write, void *buffer,
3469 size_t *lenp, loff_t *ppos)
3470{
3471 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3472
3473 neigh_proc_update(ctl, write);
3474 return ret;
3475}
3476EXPORT_SYMBOL(neigh_proc_dointvec);
3477
3478int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write, void *buffer,
3479 size_t *lenp, loff_t *ppos)
3480{
3481 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3482
3483 neigh_proc_update(ctl, write);
3484 return ret;
3485}
3486EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3487
3488static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3489 void *buffer, size_t *lenp,
3490 loff_t *ppos)
3491{
3492 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3493
3494 neigh_proc_update(ctl, write);
3495 return ret;
3496}
3497
3498int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3499 void *buffer, size_t *lenp, loff_t *ppos)
3500{
3501 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3502
3503 neigh_proc_update(ctl, write);
3504 return ret;
3505}
3506EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3507
3508static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3509 void *buffer, size_t *lenp,
3510 loff_t *ppos)
3511{
3512 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3513
3514 neigh_proc_update(ctl, write);
3515 return ret;
3516}
3517
3518static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3519 void *buffer, size_t *lenp,
3520 loff_t *ppos)
3521{
3522 struct neigh_parms *p = ctl->extra2;
3523 int ret;
3524
3525 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3526 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3527 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3528 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3529 else
3530 ret = -1;
3531
3532 if (write && ret == 0) {
3533 /* update reachable_time as well, otherwise, the change will
3534 * only be effective after the next time neigh_periodic_work
3535 * decides to recompute it
3536 */
3537 p->reachable_time =
3538 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3539 }
3540 return ret;
3541}
3542
3543#define NEIGH_PARMS_DATA_OFFSET(index) \
3544 (&((struct neigh_parms *) 0)->data[index])
3545
3546#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3547 [NEIGH_VAR_ ## attr] = { \
3548 .procname = name, \
3549 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3550 .maxlen = sizeof(int), \
3551 .mode = mval, \
3552 .proc_handler = proc, \
3553 }
3554
3555#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3556 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3557
3558#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3559 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3560
3561#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3562 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3563
3564#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3565 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3566
3567#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3568 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3569
3570static struct neigh_sysctl_table {
3571 struct ctl_table_header *sysctl_header;
3572 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3573} neigh_sysctl_template __read_mostly = {
3574 .neigh_vars = {
3575 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3576 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3577 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3578 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3579 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3580 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3581 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3582 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3583 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3584 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3585 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3586 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3587 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3588 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3589 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3590 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3591 [NEIGH_VAR_GC_INTERVAL] = {
3592 .procname = "gc_interval",
3593 .maxlen = sizeof(int),
3594 .mode = 0644,
3595 .proc_handler = proc_dointvec_jiffies,
3596 },
3597 [NEIGH_VAR_GC_THRESH1] = {
3598 .procname = "gc_thresh1",
3599 .maxlen = sizeof(int),
3600 .mode = 0644,
3601 .extra1 = SYSCTL_ZERO,
3602 .extra2 = SYSCTL_INT_MAX,
3603 .proc_handler = proc_dointvec_minmax,
3604 },
3605 [NEIGH_VAR_GC_THRESH2] = {
3606 .procname = "gc_thresh2",
3607 .maxlen = sizeof(int),
3608 .mode = 0644,
3609 .extra1 = SYSCTL_ZERO,
3610 .extra2 = SYSCTL_INT_MAX,
3611 .proc_handler = proc_dointvec_minmax,
3612 },
3613 [NEIGH_VAR_GC_THRESH3] = {
3614 .procname = "gc_thresh3",
3615 .maxlen = sizeof(int),
3616 .mode = 0644,
3617 .extra1 = SYSCTL_ZERO,
3618 .extra2 = SYSCTL_INT_MAX,
3619 .proc_handler = proc_dointvec_minmax,
3620 },
3621 {},
3622 },
3623};
3624
3625int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3626 proc_handler *handler)
3627{
3628 int i;
3629 struct neigh_sysctl_table *t;
3630 const char *dev_name_source;
3631 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3632 char *p_name;
3633
3634 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3635 if (!t)
3636 goto err;
3637
3638 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3639 t->neigh_vars[i].data += (long) p;
3640 t->neigh_vars[i].extra1 = dev;
3641 t->neigh_vars[i].extra2 = p;
3642 }
3643
3644 if (dev) {
3645 dev_name_source = dev->name;
3646 /* Terminate the table early */
3647 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3648 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3649 } else {
3650 struct neigh_table *tbl = p->tbl;
3651 dev_name_source = "default";
3652 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3653 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3654 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3655 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3656 }
3657
3658 if (handler) {
3659 /* RetransTime */
3660 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3661 /* ReachableTime */
3662 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3663 /* RetransTime (in milliseconds)*/
3664 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3665 /* ReachableTime (in milliseconds) */
3666 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3667 } else {
3668 /* Those handlers will update p->reachable_time after
3669 * base_reachable_time(_ms) is set to ensure the new timer starts being
3670 * applied after the next neighbour update instead of waiting for
3671 * neigh_periodic_work to update its value (can be multiple minutes)
3672 * So any handler that replaces them should do this as well
3673 */
3674 /* ReachableTime */
3675 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3676 neigh_proc_base_reachable_time;
3677 /* ReachableTime (in milliseconds) */
3678 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3679 neigh_proc_base_reachable_time;
3680 }
3681
3682 /* Don't export sysctls to unprivileged users */
3683 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3684 t->neigh_vars[0].procname = NULL;
3685
3686 switch (neigh_parms_family(p)) {
3687 case AF_INET:
3688 p_name = "ipv4";
3689 break;
3690 case AF_INET6:
3691 p_name = "ipv6";
3692 break;
3693 default:
3694 BUG();
3695 }
3696
3697 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3698 p_name, dev_name_source);
3699 t->sysctl_header =
3700 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3701 if (!t->sysctl_header)
3702 goto free;
3703
3704 p->sysctl_table = t;
3705 return 0;
3706
3707free:
3708 kfree(t);
3709err:
3710 return -ENOBUFS;
3711}
3712EXPORT_SYMBOL(neigh_sysctl_register);
3713
3714void neigh_sysctl_unregister(struct neigh_parms *p)
3715{
3716 if (p->sysctl_table) {
3717 struct neigh_sysctl_table *t = p->sysctl_table;
3718 p->sysctl_table = NULL;
3719 unregister_net_sysctl_table(t->sysctl_header);
3720 kfree(t);
3721 }
3722}
3723EXPORT_SYMBOL(neigh_sysctl_unregister);
3724
3725#endif /* CONFIG_SYSCTL */
3726
3727static int __init neigh_init(void)
3728{
3729 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3730 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3731 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3732
3733 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3734 0);
3735 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3736
3737 return 0;
3738}
3739
3740subsys_initcall(neigh_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Generic address resolution entity
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
8 *
9 * Fixes:
10 * Vitaly E. Lavrov releasing NULL neighbor in neigh_add.
11 * Harald Welte Add neighbour cache statistics like rtstat
12 */
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
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>
24#ifdef CONFIG_SYSCTL
25#include <linux/sysctl.h>
26#endif
27#include <linux/times.h>
28#include <net/net_namespace.h>
29#include <net/neighbour.h>
30#include <net/arp.h>
31#include <net/dst.h>
32#include <net/sock.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>
41
42#include <trace/events/neigh.h>
43
44#define DEBUG
45#define NEIGH_DEBUG 1
46#define neigh_dbg(level, fmt, ...) \
47do { \
48 if (level <= NEIGH_DEBUG) \
49 pr_debug(fmt, ##__VA_ARGS__); \
50} while (0)
51
52#define PNEIGH_HASHMASK 0xF
53
54static void neigh_timer_handler(struct timer_list *t);
55static void __neigh_notify(struct neighbour *n, int type, int flags,
56 u32 pid);
57static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid);
58static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
59 struct net_device *dev);
60
61#ifdef CONFIG_PROC_FS
62static const struct seq_operations neigh_stat_seq_ops;
63#endif
64
65/*
66 Neighbour hash table buckets are protected with rwlock tbl->lock.
67
68 - All the scans/updates to hash buckets MUST be made under this lock.
69 - NOTHING clever should be made under this lock: no callbacks
70 to protocol backends, no attempts to send something to network.
71 It will result in deadlocks, if backend/driver wants to use neighbour
72 cache.
73 - If the entry requires some non-trivial actions, increase
74 its reference count and release table lock.
75
76 Neighbour entries are protected:
77 - with reference count.
78 - with rwlock neigh->lock
79
80 Reference count prevents destruction.
81
82 neigh->lock mainly serializes ll address data and its validity state.
83 However, the same lock is used to protect another entry fields:
84 - timer
85 - resolution queue
86
87 Again, nothing clever shall be made under neigh->lock,
88 the most complicated procedure, which we allow is dev->hard_header.
89 It is supposed, that dev->hard_header is simplistic and does
90 not make callbacks to neighbour tables.
91 */
92
93static int neigh_blackhole(struct neighbour *neigh, struct sk_buff *skb)
94{
95 kfree_skb(skb);
96 return -ENETDOWN;
97}
98
99static void neigh_cleanup_and_release(struct neighbour *neigh)
100{
101 if (neigh->parms->neigh_cleanup)
102 neigh->parms->neigh_cleanup(neigh);
103
104 trace_neigh_cleanup_and_release(neigh, 0);
105 __neigh_notify(neigh, RTM_DELNEIGH, 0, 0);
106 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
107 neigh_release(neigh);
108}
109
110/*
111 * It is random distribution in the interval (1/2)*base...(3/2)*base.
112 * It corresponds to default IPv6 settings and is not overridable,
113 * because it is really reasonable choice.
114 */
115
116unsigned long neigh_rand_reach_time(unsigned long base)
117{
118 return base ? (prandom_u32() % base) + (base >> 1) : 0;
119}
120EXPORT_SYMBOL(neigh_rand_reach_time);
121
122static void neigh_mark_dead(struct neighbour *n)
123{
124 n->dead = 1;
125 if (!list_empty(&n->gc_list)) {
126 list_del_init(&n->gc_list);
127 atomic_dec(&n->tbl->gc_entries);
128 }
129}
130
131static void neigh_update_gc_list(struct neighbour *n)
132{
133 bool on_gc_list, exempt_from_gc;
134
135 write_lock_bh(&n->tbl->lock);
136 write_lock(&n->lock);
137
138 /* remove from the gc list if new state is permanent or if neighbor
139 * is externally learned; otherwise entry should be on the gc list
140 */
141 exempt_from_gc = n->nud_state & NUD_PERMANENT ||
142 n->flags & NTF_EXT_LEARNED;
143 on_gc_list = !list_empty(&n->gc_list);
144
145 if (exempt_from_gc && on_gc_list) {
146 list_del_init(&n->gc_list);
147 atomic_dec(&n->tbl->gc_entries);
148 } else if (!exempt_from_gc && !on_gc_list) {
149 /* add entries to the tail; cleaning removes from the front */
150 list_add_tail(&n->gc_list, &n->tbl->gc_list);
151 atomic_inc(&n->tbl->gc_entries);
152 }
153
154 write_unlock(&n->lock);
155 write_unlock_bh(&n->tbl->lock);
156}
157
158static bool neigh_update_ext_learned(struct neighbour *neigh, u32 flags,
159 int *notify)
160{
161 bool rc = false;
162 u8 ndm_flags;
163
164 if (!(flags & NEIGH_UPDATE_F_ADMIN))
165 return rc;
166
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;
171 else
172 neigh->flags &= ~NTF_EXT_LEARNED;
173 rc = true;
174 *notify = 1;
175 }
176
177 return rc;
178}
179
180static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
181 struct neigh_table *tbl)
182{
183 bool retval = false;
184
185 write_lock(&n->lock);
186 if (refcount_read(&n->refcnt) == 1) {
187 struct neighbour *neigh;
188
189 neigh = rcu_dereference_protected(n->next,
190 lockdep_is_held(&tbl->lock));
191 rcu_assign_pointer(*np, neigh);
192 neigh_mark_dead(n);
193 retval = true;
194 }
195 write_unlock(&n->lock);
196 if (retval)
197 neigh_cleanup_and_release(n);
198 return retval;
199}
200
201bool neigh_remove_one(struct neighbour *ndel, struct neigh_table *tbl)
202{
203 struct neigh_hash_table *nht;
204 void *pkey = ndel->primary_key;
205 u32 hash_val;
206 struct neighbour *n;
207 struct neighbour __rcu **np;
208
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);
213
214 np = &nht->hash_buckets[hash_val];
215 while ((n = rcu_dereference_protected(*np,
216 lockdep_is_held(&tbl->lock)))) {
217 if (n == ndel)
218 return neigh_del(n, np, tbl);
219 np = &n->next;
220 }
221 return false;
222}
223
224static int neigh_forced_gc(struct neigh_table *tbl)
225{
226 int max_clean = atomic_read(&tbl->gc_entries) - tbl->gc_thresh2;
227 unsigned long tref = jiffies - 5 * HZ;
228 struct neighbour *n, *tmp;
229 int shrunk = 0;
230
231 NEIGH_CACHE_STAT_INC(tbl, forced_gc_runs);
232
233 write_lock_bh(&tbl->lock);
234
235 list_for_each_entry_safe(n, tmp, &tbl->gc_list, gc_list) {
236 if (refcount_read(&n->refcnt) == 1) {
237 bool remove = false;
238
239 write_lock(&n->lock);
240 if ((n->nud_state == NUD_FAILED) ||
241 time_after(tref, n->updated))
242 remove = true;
243 write_unlock(&n->lock);
244
245 if (remove && neigh_remove_one(n, tbl))
246 shrunk++;
247 if (shrunk >= max_clean)
248 break;
249 }
250 }
251
252 tbl->last_flush = jiffies;
253
254 write_unlock_bh(&tbl->lock);
255
256 return shrunk;
257}
258
259static void neigh_add_timer(struct neighbour *n, unsigned long when)
260{
261 neigh_hold(n);
262 if (unlikely(mod_timer(&n->timer, when))) {
263 printk("NEIGH: BUG, double timer add, state is %x\n",
264 n->nud_state);
265 dump_stack();
266 }
267}
268
269static int neigh_del_timer(struct neighbour *n)
270{
271 if ((n->nud_state & NUD_IN_TIMER) &&
272 del_timer(&n->timer)) {
273 neigh_release(n);
274 return 1;
275 }
276 return 0;
277}
278
279static void pneigh_queue_purge(struct sk_buff_head *list)
280{
281 struct sk_buff *skb;
282
283 while ((skb = skb_dequeue(list)) != NULL) {
284 dev_put(skb->dev);
285 kfree_skb(skb);
286 }
287}
288
289static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
290 bool skip_perm)
291{
292 int i;
293 struct neigh_hash_table *nht;
294
295 nht = rcu_dereference_protected(tbl->nht,
296 lockdep_is_held(&tbl->lock));
297
298 for (i = 0; i < (1 << nht->hash_shift); i++) {
299 struct neighbour *n;
300 struct neighbour __rcu **np = &nht->hash_buckets[i];
301
302 while ((n = rcu_dereference_protected(*np,
303 lockdep_is_held(&tbl->lock))) != NULL) {
304 if (dev && n->dev != dev) {
305 np = &n->next;
306 continue;
307 }
308 if (skip_perm && n->nud_state & NUD_PERMANENT) {
309 np = &n->next;
310 continue;
311 }
312 rcu_assign_pointer(*np,
313 rcu_dereference_protected(n->next,
314 lockdep_is_held(&tbl->lock)));
315 write_lock(&n->lock);
316 neigh_del_timer(n);
317 neigh_mark_dead(n);
318 if (refcount_read(&n->refcnt) != 1) {
319 /* The most unpleasant situation.
320 We must destroy neighbour entry,
321 but someone still uses it.
322
323 The destroy will be delayed until
324 the last user releases us, but
325 we must kill timers etc. and move
326 it to safe state.
327 */
328 __skb_queue_purge(&n->arp_queue);
329 n->arp_queue_len_bytes = 0;
330 n->output = neigh_blackhole;
331 if (n->nud_state & NUD_VALID)
332 n->nud_state = NUD_NOARP;
333 else
334 n->nud_state = NUD_NONE;
335 neigh_dbg(2, "neigh %p is stray\n", n);
336 }
337 write_unlock(&n->lock);
338 neigh_cleanup_and_release(n);
339 }
340 }
341}
342
343void neigh_changeaddr(struct neigh_table *tbl, struct net_device *dev)
344{
345 write_lock_bh(&tbl->lock);
346 neigh_flush_dev(tbl, dev, false);
347 write_unlock_bh(&tbl->lock);
348}
349EXPORT_SYMBOL(neigh_changeaddr);
350
351static int __neigh_ifdown(struct neigh_table *tbl, struct net_device *dev,
352 bool skip_perm)
353{
354 write_lock_bh(&tbl->lock);
355 neigh_flush_dev(tbl, dev, skip_perm);
356 pneigh_ifdown_and_unlock(tbl, dev);
357
358 del_timer_sync(&tbl->proxy_timer);
359 pneigh_queue_purge(&tbl->proxy_queue);
360 return 0;
361}
362
363int neigh_carrier_down(struct neigh_table *tbl, struct net_device *dev)
364{
365 __neigh_ifdown(tbl, dev, true);
366 return 0;
367}
368EXPORT_SYMBOL(neigh_carrier_down);
369
370int neigh_ifdown(struct neigh_table *tbl, struct net_device *dev)
371{
372 __neigh_ifdown(tbl, dev, false);
373 return 0;
374}
375EXPORT_SYMBOL(neigh_ifdown);
376
377static struct neighbour *neigh_alloc(struct neigh_table *tbl,
378 struct net_device *dev,
379 bool exempt_from_gc)
380{
381 struct neighbour *n = NULL;
382 unsigned long now = jiffies;
383 int entries;
384
385 if (exempt_from_gc)
386 goto do_alloc;
387
388 entries = atomic_inc_return(&tbl->gc_entries) - 1;
389 if (entries >= tbl->gc_thresh3 ||
390 (entries >= tbl->gc_thresh2 &&
391 time_after(now, tbl->last_flush + 5 * HZ))) {
392 if (!neigh_forced_gc(tbl) &&
393 entries >= tbl->gc_thresh3) {
394 net_info_ratelimited("%s: neighbor table overflow!\n",
395 tbl->id);
396 NEIGH_CACHE_STAT_INC(tbl, table_fulls);
397 goto out_entries;
398 }
399 }
400
401do_alloc:
402 n = kzalloc(tbl->entry_size + dev->neigh_priv_len, GFP_ATOMIC);
403 if (!n)
404 goto out_entries;
405
406 __skb_queue_head_init(&n->arp_queue);
407 rwlock_init(&n->lock);
408 seqlock_init(&n->ha_lock);
409 n->updated = n->used = now;
410 n->nud_state = NUD_NONE;
411 n->output = neigh_blackhole;
412 seqlock_init(&n->hh.hh_lock);
413 n->parms = neigh_parms_clone(&tbl->parms);
414 timer_setup(&n->timer, neigh_timer_handler, 0);
415
416 NEIGH_CACHE_STAT_INC(tbl, allocs);
417 n->tbl = tbl;
418 refcount_set(&n->refcnt, 1);
419 n->dead = 1;
420 INIT_LIST_HEAD(&n->gc_list);
421
422 atomic_inc(&tbl->entries);
423out:
424 return n;
425
426out_entries:
427 if (!exempt_from_gc)
428 atomic_dec(&tbl->gc_entries);
429 goto out;
430}
431
432static void neigh_get_hash_rnd(u32 *x)
433{
434 *x = get_random_u32() | 1;
435}
436
437static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
438{
439 size_t size = (1 << shift) * sizeof(struct neighbour *);
440 struct neigh_hash_table *ret;
441 struct neighbour __rcu **buckets;
442 int i;
443
444 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
445 if (!ret)
446 return NULL;
447 if (size <= PAGE_SIZE) {
448 buckets = kzalloc(size, GFP_ATOMIC);
449 } else {
450 buckets = (struct neighbour __rcu **)
451 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
452 get_order(size));
453 kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
454 }
455 if (!buckets) {
456 kfree(ret);
457 return NULL;
458 }
459 ret->hash_buckets = buckets;
460 ret->hash_shift = shift;
461 for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
462 neigh_get_hash_rnd(&ret->hash_rnd[i]);
463 return ret;
464}
465
466static void neigh_hash_free_rcu(struct rcu_head *head)
467{
468 struct neigh_hash_table *nht = container_of(head,
469 struct neigh_hash_table,
470 rcu);
471 size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
472 struct neighbour __rcu **buckets = nht->hash_buckets;
473
474 if (size <= PAGE_SIZE) {
475 kfree(buckets);
476 } else {
477 kmemleak_free(buckets);
478 free_pages((unsigned long)buckets, get_order(size));
479 }
480 kfree(nht);
481}
482
483static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
484 unsigned long new_shift)
485{
486 unsigned int i, hash;
487 struct neigh_hash_table *new_nht, *old_nht;
488
489 NEIGH_CACHE_STAT_INC(tbl, hash_grows);
490
491 old_nht = rcu_dereference_protected(tbl->nht,
492 lockdep_is_held(&tbl->lock));
493 new_nht = neigh_hash_alloc(new_shift);
494 if (!new_nht)
495 return old_nht;
496
497 for (i = 0; i < (1 << old_nht->hash_shift); i++) {
498 struct neighbour *n, *next;
499
500 for (n = rcu_dereference_protected(old_nht->hash_buckets[i],
501 lockdep_is_held(&tbl->lock));
502 n != NULL;
503 n = next) {
504 hash = tbl->hash(n->primary_key, n->dev,
505 new_nht->hash_rnd);
506
507 hash >>= (32 - new_nht->hash_shift);
508 next = rcu_dereference_protected(n->next,
509 lockdep_is_held(&tbl->lock));
510
511 rcu_assign_pointer(n->next,
512 rcu_dereference_protected(
513 new_nht->hash_buckets[hash],
514 lockdep_is_held(&tbl->lock)));
515 rcu_assign_pointer(new_nht->hash_buckets[hash], n);
516 }
517 }
518
519 rcu_assign_pointer(tbl->nht, new_nht);
520 call_rcu(&old_nht->rcu, neigh_hash_free_rcu);
521 return new_nht;
522}
523
524struct neighbour *neigh_lookup(struct neigh_table *tbl, const void *pkey,
525 struct net_device *dev)
526{
527 struct neighbour *n;
528
529 NEIGH_CACHE_STAT_INC(tbl, lookups);
530
531 rcu_read_lock_bh();
532 n = __neigh_lookup_noref(tbl, pkey, dev);
533 if (n) {
534 if (!refcount_inc_not_zero(&n->refcnt))
535 n = NULL;
536 NEIGH_CACHE_STAT_INC(tbl, hits);
537 }
538
539 rcu_read_unlock_bh();
540 return n;
541}
542EXPORT_SYMBOL(neigh_lookup);
543
544struct neighbour *neigh_lookup_nodev(struct neigh_table *tbl, struct net *net,
545 const void *pkey)
546{
547 struct neighbour *n;
548 unsigned int key_len = tbl->key_len;
549 u32 hash_val;
550 struct neigh_hash_table *nht;
551
552 NEIGH_CACHE_STAT_INC(tbl, lookups);
553
554 rcu_read_lock_bh();
555 nht = rcu_dereference_bh(tbl->nht);
556 hash_val = tbl->hash(pkey, NULL, nht->hash_rnd) >> (32 - nht->hash_shift);
557
558 for (n = rcu_dereference_bh(nht->hash_buckets[hash_val]);
559 n != NULL;
560 n = rcu_dereference_bh(n->next)) {
561 if (!memcmp(n->primary_key, pkey, key_len) &&
562 net_eq(dev_net(n->dev), net)) {
563 if (!refcount_inc_not_zero(&n->refcnt))
564 n = NULL;
565 NEIGH_CACHE_STAT_INC(tbl, hits);
566 break;
567 }
568 }
569
570 rcu_read_unlock_bh();
571 return n;
572}
573EXPORT_SYMBOL(neigh_lookup_nodev);
574
575static struct neighbour *___neigh_create(struct neigh_table *tbl,
576 const void *pkey,
577 struct net_device *dev,
578 bool exempt_from_gc, bool want_ref)
579{
580 struct neighbour *n1, *rc, *n = neigh_alloc(tbl, dev, exempt_from_gc);
581 u32 hash_val;
582 unsigned int key_len = tbl->key_len;
583 int error;
584 struct neigh_hash_table *nht;
585
586 trace_neigh_create(tbl, dev, pkey, n, exempt_from_gc);
587
588 if (!n) {
589 rc = ERR_PTR(-ENOBUFS);
590 goto out;
591 }
592
593 memcpy(n->primary_key, pkey, key_len);
594 n->dev = dev;
595 dev_hold(dev);
596
597 /* Protocol specific setup. */
598 if (tbl->constructor && (error = tbl->constructor(n)) < 0) {
599 rc = ERR_PTR(error);
600 goto out_neigh_release;
601 }
602
603 if (dev->netdev_ops->ndo_neigh_construct) {
604 error = dev->netdev_ops->ndo_neigh_construct(dev, n);
605 if (error < 0) {
606 rc = ERR_PTR(error);
607 goto out_neigh_release;
608 }
609 }
610
611 /* Device specific setup. */
612 if (n->parms->neigh_setup &&
613 (error = n->parms->neigh_setup(n)) < 0) {
614 rc = ERR_PTR(error);
615 goto out_neigh_release;
616 }
617
618 n->confirmed = jiffies - (NEIGH_VAR(n->parms, BASE_REACHABLE_TIME) << 1);
619
620 write_lock_bh(&tbl->lock);
621 nht = rcu_dereference_protected(tbl->nht,
622 lockdep_is_held(&tbl->lock));
623
624 if (atomic_read(&tbl->entries) > (1 << nht->hash_shift))
625 nht = neigh_hash_grow(tbl, nht->hash_shift + 1);
626
627 hash_val = tbl->hash(n->primary_key, dev, nht->hash_rnd) >> (32 - nht->hash_shift);
628
629 if (n->parms->dead) {
630 rc = ERR_PTR(-EINVAL);
631 goto out_tbl_unlock;
632 }
633
634 for (n1 = rcu_dereference_protected(nht->hash_buckets[hash_val],
635 lockdep_is_held(&tbl->lock));
636 n1 != NULL;
637 n1 = rcu_dereference_protected(n1->next,
638 lockdep_is_held(&tbl->lock))) {
639 if (dev == n1->dev && !memcmp(n1->primary_key, n->primary_key, key_len)) {
640 if (want_ref)
641 neigh_hold(n1);
642 rc = n1;
643 goto out_tbl_unlock;
644 }
645 }
646
647 n->dead = 0;
648 if (!exempt_from_gc)
649 list_add_tail(&n->gc_list, &n->tbl->gc_list);
650
651 if (want_ref)
652 neigh_hold(n);
653 rcu_assign_pointer(n->next,
654 rcu_dereference_protected(nht->hash_buckets[hash_val],
655 lockdep_is_held(&tbl->lock)));
656 rcu_assign_pointer(nht->hash_buckets[hash_val], n);
657 write_unlock_bh(&tbl->lock);
658 neigh_dbg(2, "neigh %p is created\n", n);
659 rc = n;
660out:
661 return rc;
662out_tbl_unlock:
663 write_unlock_bh(&tbl->lock);
664out_neigh_release:
665 if (!exempt_from_gc)
666 atomic_dec(&tbl->gc_entries);
667 neigh_release(n);
668 goto out;
669}
670
671struct neighbour *__neigh_create(struct neigh_table *tbl, const void *pkey,
672 struct net_device *dev, bool want_ref)
673{
674 return ___neigh_create(tbl, pkey, dev, false, want_ref);
675}
676EXPORT_SYMBOL(__neigh_create);
677
678static u32 pneigh_hash(const void *pkey, unsigned int key_len)
679{
680 u32 hash_val = *(u32 *)(pkey + key_len - 4);
681 hash_val ^= (hash_val >> 16);
682 hash_val ^= hash_val >> 8;
683 hash_val ^= hash_val >> 4;
684 hash_val &= PNEIGH_HASHMASK;
685 return hash_val;
686}
687
688static struct pneigh_entry *__pneigh_lookup_1(struct pneigh_entry *n,
689 struct net *net,
690 const void *pkey,
691 unsigned int key_len,
692 struct net_device *dev)
693{
694 while (n) {
695 if (!memcmp(n->key, pkey, key_len) &&
696 net_eq(pneigh_net(n), net) &&
697 (n->dev == dev || !n->dev))
698 return n;
699 n = n->next;
700 }
701 return NULL;
702}
703
704struct pneigh_entry *__pneigh_lookup(struct neigh_table *tbl,
705 struct net *net, const void *pkey, struct net_device *dev)
706{
707 unsigned int key_len = tbl->key_len;
708 u32 hash_val = pneigh_hash(pkey, key_len);
709
710 return __pneigh_lookup_1(tbl->phash_buckets[hash_val],
711 net, pkey, key_len, dev);
712}
713EXPORT_SYMBOL_GPL(__pneigh_lookup);
714
715struct pneigh_entry * pneigh_lookup(struct neigh_table *tbl,
716 struct net *net, const void *pkey,
717 struct net_device *dev, int creat)
718{
719 struct pneigh_entry *n;
720 unsigned int key_len = tbl->key_len;
721 u32 hash_val = pneigh_hash(pkey, key_len);
722
723 read_lock_bh(&tbl->lock);
724 n = __pneigh_lookup_1(tbl->phash_buckets[hash_val],
725 net, pkey, key_len, dev);
726 read_unlock_bh(&tbl->lock);
727
728 if (n || !creat)
729 goto out;
730
731 ASSERT_RTNL();
732
733 n = kmalloc(sizeof(*n) + key_len, GFP_KERNEL);
734 if (!n)
735 goto out;
736
737 n->protocol = 0;
738 write_pnet(&n->net, net);
739 memcpy(n->key, pkey, key_len);
740 n->dev = dev;
741 if (dev)
742 dev_hold(dev);
743
744 if (tbl->pconstructor && tbl->pconstructor(n)) {
745 if (dev)
746 dev_put(dev);
747 kfree(n);
748 n = NULL;
749 goto out;
750 }
751
752 write_lock_bh(&tbl->lock);
753 n->next = tbl->phash_buckets[hash_val];
754 tbl->phash_buckets[hash_val] = n;
755 write_unlock_bh(&tbl->lock);
756out:
757 return n;
758}
759EXPORT_SYMBOL(pneigh_lookup);
760
761
762int pneigh_delete(struct neigh_table *tbl, struct net *net, const void *pkey,
763 struct net_device *dev)
764{
765 struct pneigh_entry *n, **np;
766 unsigned int key_len = tbl->key_len;
767 u32 hash_val = pneigh_hash(pkey, key_len);
768
769 write_lock_bh(&tbl->lock);
770 for (np = &tbl->phash_buckets[hash_val]; (n = *np) != NULL;
771 np = &n->next) {
772 if (!memcmp(n->key, pkey, key_len) && n->dev == dev &&
773 net_eq(pneigh_net(n), net)) {
774 *np = n->next;
775 write_unlock_bh(&tbl->lock);
776 if (tbl->pdestructor)
777 tbl->pdestructor(n);
778 if (n->dev)
779 dev_put(n->dev);
780 kfree(n);
781 return 0;
782 }
783 }
784 write_unlock_bh(&tbl->lock);
785 return -ENOENT;
786}
787
788static int pneigh_ifdown_and_unlock(struct neigh_table *tbl,
789 struct net_device *dev)
790{
791 struct pneigh_entry *n, **np, *freelist = NULL;
792 u32 h;
793
794 for (h = 0; h <= PNEIGH_HASHMASK; h++) {
795 np = &tbl->phash_buckets[h];
796 while ((n = *np) != NULL) {
797 if (!dev || n->dev == dev) {
798 *np = n->next;
799 n->next = freelist;
800 freelist = n;
801 continue;
802 }
803 np = &n->next;
804 }
805 }
806 write_unlock_bh(&tbl->lock);
807 while ((n = freelist)) {
808 freelist = n->next;
809 n->next = NULL;
810 if (tbl->pdestructor)
811 tbl->pdestructor(n);
812 if (n->dev)
813 dev_put(n->dev);
814 kfree(n);
815 }
816 return -ENOENT;
817}
818
819static void neigh_parms_destroy(struct neigh_parms *parms);
820
821static inline void neigh_parms_put(struct neigh_parms *parms)
822{
823 if (refcount_dec_and_test(&parms->refcnt))
824 neigh_parms_destroy(parms);
825}
826
827/*
828 * neighbour must already be out of the table;
829 *
830 */
831void neigh_destroy(struct neighbour *neigh)
832{
833 struct net_device *dev = neigh->dev;
834
835 NEIGH_CACHE_STAT_INC(neigh->tbl, destroys);
836
837 if (!neigh->dead) {
838 pr_warn("Destroying alive neighbour %p\n", neigh);
839 dump_stack();
840 return;
841 }
842
843 if (neigh_del_timer(neigh))
844 pr_warn("Impossible event\n");
845
846 write_lock_bh(&neigh->lock);
847 __skb_queue_purge(&neigh->arp_queue);
848 write_unlock_bh(&neigh->lock);
849 neigh->arp_queue_len_bytes = 0;
850
851 if (dev->netdev_ops->ndo_neigh_destroy)
852 dev->netdev_ops->ndo_neigh_destroy(dev, neigh);
853
854 dev_put(dev);
855 neigh_parms_put(neigh->parms);
856
857 neigh_dbg(2, "neigh %p is destroyed\n", neigh);
858
859 atomic_dec(&neigh->tbl->entries);
860 kfree_rcu(neigh, rcu);
861}
862EXPORT_SYMBOL(neigh_destroy);
863
864/* Neighbour state is suspicious;
865 disable fast path.
866
867 Called with write_locked neigh.
868 */
869static void neigh_suspect(struct neighbour *neigh)
870{
871 neigh_dbg(2, "neigh %p is suspected\n", neigh);
872
873 neigh->output = neigh->ops->output;
874}
875
876/* Neighbour state is OK;
877 enable fast path.
878
879 Called with write_locked neigh.
880 */
881static void neigh_connect(struct neighbour *neigh)
882{
883 neigh_dbg(2, "neigh %p is connected\n", neigh);
884
885 neigh->output = neigh->ops->connected_output;
886}
887
888static void neigh_periodic_work(struct work_struct *work)
889{
890 struct neigh_table *tbl = container_of(work, struct neigh_table, gc_work.work);
891 struct neighbour *n;
892 struct neighbour __rcu **np;
893 unsigned int i;
894 struct neigh_hash_table *nht;
895
896 NEIGH_CACHE_STAT_INC(tbl, periodic_gc_runs);
897
898 write_lock_bh(&tbl->lock);
899 nht = rcu_dereference_protected(tbl->nht,
900 lockdep_is_held(&tbl->lock));
901
902 /*
903 * periodically recompute ReachableTime from random function
904 */
905
906 if (time_after(jiffies, tbl->last_rand + 300 * HZ)) {
907 struct neigh_parms *p;
908 tbl->last_rand = jiffies;
909 list_for_each_entry(p, &tbl->parms_list, list)
910 p->reachable_time =
911 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
912 }
913
914 if (atomic_read(&tbl->entries) < tbl->gc_thresh1)
915 goto out;
916
917 for (i = 0 ; i < (1 << nht->hash_shift); i++) {
918 np = &nht->hash_buckets[i];
919
920 while ((n = rcu_dereference_protected(*np,
921 lockdep_is_held(&tbl->lock))) != NULL) {
922 unsigned int state;
923
924 write_lock(&n->lock);
925
926 state = n->nud_state;
927 if ((state & (NUD_PERMANENT | NUD_IN_TIMER)) ||
928 (n->flags & NTF_EXT_LEARNED)) {
929 write_unlock(&n->lock);
930 goto next_elt;
931 }
932
933 if (time_before(n->used, n->confirmed))
934 n->used = n->confirmed;
935
936 if (refcount_read(&n->refcnt) == 1 &&
937 (state == NUD_FAILED ||
938 time_after(jiffies, n->used + NEIGH_VAR(n->parms, GC_STALETIME)))) {
939 *np = n->next;
940 neigh_mark_dead(n);
941 write_unlock(&n->lock);
942 neigh_cleanup_and_release(n);
943 continue;
944 }
945 write_unlock(&n->lock);
946
947next_elt:
948 np = &n->next;
949 }
950 /*
951 * It's fine to release lock here, even if hash table
952 * grows while we are preempted.
953 */
954 write_unlock_bh(&tbl->lock);
955 cond_resched();
956 write_lock_bh(&tbl->lock);
957 nht = rcu_dereference_protected(tbl->nht,
958 lockdep_is_held(&tbl->lock));
959 }
960out:
961 /* Cycle through all hash buckets every BASE_REACHABLE_TIME/2 ticks.
962 * ARP entry timeouts range from 1/2 BASE_REACHABLE_TIME to 3/2
963 * BASE_REACHABLE_TIME.
964 */
965 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
966 NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME) >> 1);
967 write_unlock_bh(&tbl->lock);
968}
969
970static __inline__ int neigh_max_probes(struct neighbour *n)
971{
972 struct neigh_parms *p = n->parms;
973 return NEIGH_VAR(p, UCAST_PROBES) + NEIGH_VAR(p, APP_PROBES) +
974 (n->nud_state & NUD_PROBE ? NEIGH_VAR(p, MCAST_REPROBES) :
975 NEIGH_VAR(p, MCAST_PROBES));
976}
977
978static void neigh_invalidate(struct neighbour *neigh)
979 __releases(neigh->lock)
980 __acquires(neigh->lock)
981{
982 struct sk_buff *skb;
983
984 NEIGH_CACHE_STAT_INC(neigh->tbl, res_failed);
985 neigh_dbg(2, "neigh %p is failed\n", neigh);
986 neigh->updated = jiffies;
987
988 /* It is very thin place. report_unreachable is very complicated
989 routine. Particularly, it can hit the same neighbour entry!
990
991 So that, we try to be accurate and avoid dead loop. --ANK
992 */
993 while (neigh->nud_state == NUD_FAILED &&
994 (skb = __skb_dequeue(&neigh->arp_queue)) != NULL) {
995 write_unlock(&neigh->lock);
996 neigh->ops->error_report(neigh, skb);
997 write_lock(&neigh->lock);
998 }
999 __skb_queue_purge(&neigh->arp_queue);
1000 neigh->arp_queue_len_bytes = 0;
1001}
1002
1003static void neigh_probe(struct neighbour *neigh)
1004 __releases(neigh->lock)
1005{
1006 struct sk_buff *skb = skb_peek_tail(&neigh->arp_queue);
1007 /* keep skb alive even if arp_queue overflows */
1008 if (skb)
1009 skb = skb_clone(skb, GFP_ATOMIC);
1010 write_unlock(&neigh->lock);
1011 if (neigh->ops->solicit)
1012 neigh->ops->solicit(neigh, skb);
1013 atomic_inc(&neigh->probes);
1014 consume_skb(skb);
1015}
1016
1017/* Called when a timer expires for a neighbour entry. */
1018
1019static void neigh_timer_handler(struct timer_list *t)
1020{
1021 unsigned long now, next;
1022 struct neighbour *neigh = from_timer(neigh, t, timer);
1023 unsigned int state;
1024 int notify = 0;
1025
1026 write_lock(&neigh->lock);
1027
1028 state = neigh->nud_state;
1029 now = jiffies;
1030 next = now + HZ;
1031
1032 if (!(state & NUD_IN_TIMER))
1033 goto out;
1034
1035 if (state & NUD_REACHABLE) {
1036 if (time_before_eq(now,
1037 neigh->confirmed + neigh->parms->reachable_time)) {
1038 neigh_dbg(2, "neigh %p is still alive\n", neigh);
1039 next = neigh->confirmed + neigh->parms->reachable_time;
1040 } else if (time_before_eq(now,
1041 neigh->used +
1042 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1043 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1044 neigh->nud_state = NUD_DELAY;
1045 neigh->updated = jiffies;
1046 neigh_suspect(neigh);
1047 next = now + NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME);
1048 } else {
1049 neigh_dbg(2, "neigh %p is suspected\n", neigh);
1050 neigh->nud_state = NUD_STALE;
1051 neigh->updated = jiffies;
1052 neigh_suspect(neigh);
1053 notify = 1;
1054 }
1055 } else if (state & NUD_DELAY) {
1056 if (time_before_eq(now,
1057 neigh->confirmed +
1058 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME))) {
1059 neigh_dbg(2, "neigh %p is now reachable\n", neigh);
1060 neigh->nud_state = NUD_REACHABLE;
1061 neigh->updated = jiffies;
1062 neigh_connect(neigh);
1063 notify = 1;
1064 next = neigh->confirmed + neigh->parms->reachable_time;
1065 } else {
1066 neigh_dbg(2, "neigh %p is probed\n", neigh);
1067 neigh->nud_state = NUD_PROBE;
1068 neigh->updated = jiffies;
1069 atomic_set(&neigh->probes, 0);
1070 notify = 1;
1071 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
1072 }
1073 } else {
1074 /* NUD_PROBE|NUD_INCOMPLETE */
1075 next = now + NEIGH_VAR(neigh->parms, RETRANS_TIME);
1076 }
1077
1078 if ((neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) &&
1079 atomic_read(&neigh->probes) >= neigh_max_probes(neigh)) {
1080 neigh->nud_state = NUD_FAILED;
1081 notify = 1;
1082 neigh_invalidate(neigh);
1083 goto out;
1084 }
1085
1086 if (neigh->nud_state & NUD_IN_TIMER) {
1087 if (time_before(next, jiffies + HZ/2))
1088 next = jiffies + HZ/2;
1089 if (!mod_timer(&neigh->timer, next))
1090 neigh_hold(neigh);
1091 }
1092 if (neigh->nud_state & (NUD_INCOMPLETE | NUD_PROBE)) {
1093 neigh_probe(neigh);
1094 } else {
1095out:
1096 write_unlock(&neigh->lock);
1097 }
1098
1099 if (notify)
1100 neigh_update_notify(neigh, 0);
1101
1102 trace_neigh_timer_handler(neigh, 0);
1103
1104 neigh_release(neigh);
1105}
1106
1107int __neigh_event_send(struct neighbour *neigh, struct sk_buff *skb)
1108{
1109 int rc;
1110 bool immediate_probe = false;
1111
1112 write_lock_bh(&neigh->lock);
1113
1114 rc = 0;
1115 if (neigh->nud_state & (NUD_CONNECTED | NUD_DELAY | NUD_PROBE))
1116 goto out_unlock_bh;
1117 if (neigh->dead)
1118 goto out_dead;
1119
1120 if (!(neigh->nud_state & (NUD_STALE | NUD_INCOMPLETE))) {
1121 if (NEIGH_VAR(neigh->parms, MCAST_PROBES) +
1122 NEIGH_VAR(neigh->parms, APP_PROBES)) {
1123 unsigned long next, now = jiffies;
1124
1125 atomic_set(&neigh->probes,
1126 NEIGH_VAR(neigh->parms, UCAST_PROBES));
1127 neigh_del_timer(neigh);
1128 neigh->nud_state = NUD_INCOMPLETE;
1129 neigh->updated = now;
1130 next = now + max(NEIGH_VAR(neigh->parms, RETRANS_TIME),
1131 HZ/2);
1132 neigh_add_timer(neigh, next);
1133 immediate_probe = true;
1134 } else {
1135 neigh->nud_state = NUD_FAILED;
1136 neigh->updated = jiffies;
1137 write_unlock_bh(&neigh->lock);
1138
1139 kfree_skb(skb);
1140 return 1;
1141 }
1142 } else if (neigh->nud_state & NUD_STALE) {
1143 neigh_dbg(2, "neigh %p is delayed\n", neigh);
1144 neigh_del_timer(neigh);
1145 neigh->nud_state = NUD_DELAY;
1146 neigh->updated = jiffies;
1147 neigh_add_timer(neigh, jiffies +
1148 NEIGH_VAR(neigh->parms, DELAY_PROBE_TIME));
1149 }
1150
1151 if (neigh->nud_state == NUD_INCOMPLETE) {
1152 if (skb) {
1153 while (neigh->arp_queue_len_bytes + skb->truesize >
1154 NEIGH_VAR(neigh->parms, QUEUE_LEN_BYTES)) {
1155 struct sk_buff *buff;
1156
1157 buff = __skb_dequeue(&neigh->arp_queue);
1158 if (!buff)
1159 break;
1160 neigh->arp_queue_len_bytes -= buff->truesize;
1161 kfree_skb(buff);
1162 NEIGH_CACHE_STAT_INC(neigh->tbl, unres_discards);
1163 }
1164 skb_dst_force(skb);
1165 __skb_queue_tail(&neigh->arp_queue, skb);
1166 neigh->arp_queue_len_bytes += skb->truesize;
1167 }
1168 rc = 1;
1169 }
1170out_unlock_bh:
1171 if (immediate_probe)
1172 neigh_probe(neigh);
1173 else
1174 write_unlock(&neigh->lock);
1175 local_bh_enable();
1176 trace_neigh_event_send_done(neigh, rc);
1177 return rc;
1178
1179out_dead:
1180 if (neigh->nud_state & NUD_STALE)
1181 goto out_unlock_bh;
1182 write_unlock_bh(&neigh->lock);
1183 kfree_skb(skb);
1184 trace_neigh_event_send_dead(neigh, 1);
1185 return 1;
1186}
1187EXPORT_SYMBOL(__neigh_event_send);
1188
1189static void neigh_update_hhs(struct neighbour *neigh)
1190{
1191 struct hh_cache *hh;
1192 void (*update)(struct hh_cache*, const struct net_device*, const unsigned char *)
1193 = NULL;
1194
1195 if (neigh->dev->header_ops)
1196 update = neigh->dev->header_ops->cache_update;
1197
1198 if (update) {
1199 hh = &neigh->hh;
1200 if (hh->hh_len) {
1201 write_seqlock_bh(&hh->hh_lock);
1202 update(hh, neigh->dev, neigh->ha);
1203 write_sequnlock_bh(&hh->hh_lock);
1204 }
1205 }
1206}
1207
1208
1209
1210/* Generic update routine.
1211 -- lladdr is new lladdr or NULL, if it is not supplied.
1212 -- new is new state.
1213 -- flags
1214 NEIGH_UPDATE_F_OVERRIDE allows to override existing lladdr,
1215 if it is different.
1216 NEIGH_UPDATE_F_WEAK_OVERRIDE will suspect existing "connected"
1217 lladdr instead of overriding it
1218 if it is different.
1219 NEIGH_UPDATE_F_ADMIN means that the change is administrative.
1220
1221 NEIGH_UPDATE_F_OVERRIDE_ISROUTER allows to override existing
1222 NTF_ROUTER flag.
1223 NEIGH_UPDATE_F_ISROUTER indicates if the neighbour is known as
1224 a router.
1225
1226 Caller MUST hold reference count on the entry.
1227 */
1228
1229static int __neigh_update(struct neighbour *neigh, const u8 *lladdr,
1230 u8 new, u32 flags, u32 nlmsg_pid,
1231 struct netlink_ext_ack *extack)
1232{
1233 bool ext_learn_change = false;
1234 u8 old;
1235 int err;
1236 int notify = 0;
1237 struct net_device *dev;
1238 int update_isrouter = 0;
1239
1240 trace_neigh_update(neigh, lladdr, new, flags, nlmsg_pid);
1241
1242 write_lock_bh(&neigh->lock);
1243
1244 dev = neigh->dev;
1245 old = neigh->nud_state;
1246 err = -EPERM;
1247
1248 if (!(flags & NEIGH_UPDATE_F_ADMIN) &&
1249 (old & (NUD_NOARP | NUD_PERMANENT)))
1250 goto out;
1251 if (neigh->dead) {
1252 NL_SET_ERR_MSG(extack, "Neighbor entry is now dead");
1253 goto out;
1254 }
1255
1256 ext_learn_change = neigh_update_ext_learned(neigh, flags, ¬ify);
1257
1258 if (!(new & NUD_VALID)) {
1259 neigh_del_timer(neigh);
1260 if (old & NUD_CONNECTED)
1261 neigh_suspect(neigh);
1262 neigh->nud_state = new;
1263 err = 0;
1264 notify = old & NUD_VALID;
1265 if ((old & (NUD_INCOMPLETE | NUD_PROBE)) &&
1266 (new & NUD_FAILED)) {
1267 neigh_invalidate(neigh);
1268 notify = 1;
1269 }
1270 goto out;
1271 }
1272
1273 /* Compare new lladdr with cached one */
1274 if (!dev->addr_len) {
1275 /* First case: device needs no address. */
1276 lladdr = neigh->ha;
1277 } else if (lladdr) {
1278 /* The second case: if something is already cached
1279 and a new address is proposed:
1280 - compare new & old
1281 - if they are different, check override flag
1282 */
1283 if ((old & NUD_VALID) &&
1284 !memcmp(lladdr, neigh->ha, dev->addr_len))
1285 lladdr = neigh->ha;
1286 } else {
1287 /* No address is supplied; if we know something,
1288 use it, otherwise discard the request.
1289 */
1290 err = -EINVAL;
1291 if (!(old & NUD_VALID)) {
1292 NL_SET_ERR_MSG(extack, "No link layer address given");
1293 goto out;
1294 }
1295 lladdr = neigh->ha;
1296 }
1297
1298 /* Update confirmed timestamp for neighbour entry after we
1299 * received ARP packet even if it doesn't change IP to MAC binding.
1300 */
1301 if (new & NUD_CONNECTED)
1302 neigh->confirmed = jiffies;
1303
1304 /* If entry was valid and address is not changed,
1305 do not change entry state, if new one is STALE.
1306 */
1307 err = 0;
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)) {
1314 lladdr = neigh->ha;
1315 new = NUD_STALE;
1316 } else
1317 goto out;
1318 } else {
1319 if (lladdr == neigh->ha && new == NUD_STALE &&
1320 !(flags & NEIGH_UPDATE_F_ADMIN))
1321 new = old;
1322 }
1323 }
1324
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.
1328 */
1329 if (new != old || lladdr != neigh->ha)
1330 neigh->updated = jiffies;
1331
1332 if (new != old) {
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 :
1340 0)));
1341 neigh->nud_state = new;
1342 notify = 1;
1343 }
1344
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);
1353 notify = 1;
1354 }
1355 if (new == old)
1356 goto out;
1357 if (new & NUD_CONNECTED)
1358 neigh_connect(neigh);
1359 else
1360 neigh_suspect(neigh);
1361 if (!(old & NUD_VALID)) {
1362 struct sk_buff *skb;
1363
1364 /* Again: avoid dead loop if something went wrong */
1365
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);
1371
1372 rcu_read_lock();
1373
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.
1380 */
1381 n2 = NULL;
1382 if (dst) {
1383 n2 = dst_neigh_lookup_skb(dst, skb);
1384 if (n2)
1385 n1 = n2;
1386 }
1387 n1->output(n1, skb);
1388 if (n2)
1389 neigh_release(n2);
1390 rcu_read_unlock();
1391
1392 write_lock_bh(&neigh->lock);
1393 }
1394 __skb_queue_purge(&neigh->arp_queue);
1395 neigh->arp_queue_len_bytes = 0;
1396 }
1397out:
1398 if (update_isrouter)
1399 neigh_update_is_router(neigh, flags, ¬ify);
1400 write_unlock_bh(&neigh->lock);
1401
1402 if (((new ^ old) & NUD_PERMANENT) || ext_learn_change)
1403 neigh_update_gc_list(neigh);
1404
1405 if (notify)
1406 neigh_update_notify(neigh, nlmsg_pid);
1407
1408 trace_neigh_update_done(neigh, err);
1409
1410 return err;
1411}
1412
1413int neigh_update(struct neighbour *neigh, const u8 *lladdr, u8 new,
1414 u32 flags, u32 nlmsg_pid)
1415{
1416 return __neigh_update(neigh, lladdr, new, flags, nlmsg_pid, NULL);
1417}
1418EXPORT_SYMBOL(neigh_update);
1419
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.
1422 */
1423void __neigh_set_probe_once(struct neighbour *neigh)
1424{
1425 if (neigh->dead)
1426 return;
1427 neigh->updated = jiffies;
1428 if (!(neigh->nud_state & NUD_FAILED))
1429 return;
1430 neigh->nud_state = NUD_INCOMPLETE;
1431 atomic_set(&neigh->probes, neigh_max_probes(neigh));
1432 neigh_add_timer(neigh,
1433 jiffies + NEIGH_VAR(neigh->parms, RETRANS_TIME));
1434}
1435EXPORT_SYMBOL(__neigh_set_probe_once);
1436
1437struct neighbour *neigh_event_ns(struct neigh_table *tbl,
1438 u8 *lladdr, void *saddr,
1439 struct net_device *dev)
1440{
1441 struct neighbour *neigh = __neigh_lookup(tbl, saddr, dev,
1442 lladdr || !dev->addr_len);
1443 if (neigh)
1444 neigh_update(neigh, lladdr, NUD_STALE,
1445 NEIGH_UPDATE_F_OVERRIDE, 0);
1446 return neigh;
1447}
1448EXPORT_SYMBOL(neigh_event_ns);
1449
1450/* called with read_lock_bh(&n->lock); */
1451static void neigh_hh_init(struct neighbour *n)
1452{
1453 struct net_device *dev = n->dev;
1454 __be16 prot = n->tbl->protocol;
1455 struct hh_cache *hh = &n->hh;
1456
1457 write_lock_bh(&n->lock);
1458
1459 /* Only one thread can come in here and initialize the
1460 * hh_cache entry.
1461 */
1462 if (!hh->hh_len)
1463 dev->header_ops->cache(n, hh, prot);
1464
1465 write_unlock_bh(&n->lock);
1466}
1467
1468/* Slow and careful. */
1469
1470int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
1471{
1472 int rc = 0;
1473
1474 if (!neigh_event_send(neigh, skb)) {
1475 int err;
1476 struct net_device *dev = neigh->dev;
1477 unsigned int seq;
1478
1479 if (dev->header_ops->cache && !neigh->hh.hh_len)
1480 neigh_hh_init(neigh);
1481
1482 do {
1483 __skb_pull(skb, skb_network_offset(skb));
1484 seq = read_seqbegin(&neigh->ha_lock);
1485 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1486 neigh->ha, NULL, skb->len);
1487 } while (read_seqretry(&neigh->ha_lock, seq));
1488
1489 if (err >= 0)
1490 rc = dev_queue_xmit(skb);
1491 else
1492 goto out_kfree_skb;
1493 }
1494out:
1495 return rc;
1496out_kfree_skb:
1497 rc = -EINVAL;
1498 kfree_skb(skb);
1499 goto out;
1500}
1501EXPORT_SYMBOL(neigh_resolve_output);
1502
1503/* As fast as possible without hh cache */
1504
1505int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
1506{
1507 struct net_device *dev = neigh->dev;
1508 unsigned int seq;
1509 int err;
1510
1511 do {
1512 __skb_pull(skb, skb_network_offset(skb));
1513 seq = read_seqbegin(&neigh->ha_lock);
1514 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
1515 neigh->ha, NULL, skb->len);
1516 } while (read_seqretry(&neigh->ha_lock, seq));
1517
1518 if (err >= 0)
1519 err = dev_queue_xmit(skb);
1520 else {
1521 err = -EINVAL;
1522 kfree_skb(skb);
1523 }
1524 return err;
1525}
1526EXPORT_SYMBOL(neigh_connected_output);
1527
1528int neigh_direct_output(struct neighbour *neigh, struct sk_buff *skb)
1529{
1530 return dev_queue_xmit(skb);
1531}
1532EXPORT_SYMBOL(neigh_direct_output);
1533
1534static void neigh_proxy_process(struct timer_list *t)
1535{
1536 struct neigh_table *tbl = from_timer(tbl, t, proxy_timer);
1537 long sched_next = 0;
1538 unsigned long now = jiffies;
1539 struct sk_buff *skb, *n;
1540
1541 spin_lock(&tbl->proxy_queue.lock);
1542
1543 skb_queue_walk_safe(&tbl->proxy_queue, skb, n) {
1544 long tdif = NEIGH_CB(skb)->sched_next - now;
1545
1546 if (tdif <= 0) {
1547 struct net_device *dev = skb->dev;
1548
1549 __skb_unlink(skb, &tbl->proxy_queue);
1550 if (tbl->proxy_redo && netif_running(dev)) {
1551 rcu_read_lock();
1552 tbl->proxy_redo(skb);
1553 rcu_read_unlock();
1554 } else {
1555 kfree_skb(skb);
1556 }
1557
1558 dev_put(dev);
1559 } else if (!sched_next || tdif < sched_next)
1560 sched_next = tdif;
1561 }
1562 del_timer(&tbl->proxy_timer);
1563 if (sched_next)
1564 mod_timer(&tbl->proxy_timer, jiffies + sched_next);
1565 spin_unlock(&tbl->proxy_queue.lock);
1566}
1567
1568void pneigh_enqueue(struct neigh_table *tbl, struct neigh_parms *p,
1569 struct sk_buff *skb)
1570{
1571 unsigned long now = jiffies;
1572
1573 unsigned long sched_next = now + (prandom_u32() %
1574 NEIGH_VAR(p, PROXY_DELAY));
1575
1576 if (tbl->proxy_queue.qlen > NEIGH_VAR(p, PROXY_QLEN)) {
1577 kfree_skb(skb);
1578 return;
1579 }
1580
1581 NEIGH_CB(skb)->sched_next = sched_next;
1582 NEIGH_CB(skb)->flags |= LOCALLY_ENQUEUED;
1583
1584 spin_lock(&tbl->proxy_queue.lock);
1585 if (del_timer(&tbl->proxy_timer)) {
1586 if (time_before(tbl->proxy_timer.expires, sched_next))
1587 sched_next = tbl->proxy_timer.expires;
1588 }
1589 skb_dst_drop(skb);
1590 dev_hold(skb->dev);
1591 __skb_queue_tail(&tbl->proxy_queue, skb);
1592 mod_timer(&tbl->proxy_timer, sched_next);
1593 spin_unlock(&tbl->proxy_queue.lock);
1594}
1595EXPORT_SYMBOL(pneigh_enqueue);
1596
1597static inline struct neigh_parms *lookup_neigh_parms(struct neigh_table *tbl,
1598 struct net *net, int ifindex)
1599{
1600 struct neigh_parms *p;
1601
1602 list_for_each_entry(p, &tbl->parms_list, list) {
1603 if ((p->dev && p->dev->ifindex == ifindex && net_eq(neigh_parms_net(p), net)) ||
1604 (!p->dev && !ifindex && net_eq(net, &init_net)))
1605 return p;
1606 }
1607
1608 return NULL;
1609}
1610
1611struct neigh_parms *neigh_parms_alloc(struct net_device *dev,
1612 struct neigh_table *tbl)
1613{
1614 struct neigh_parms *p;
1615 struct net *net = dev_net(dev);
1616 const struct net_device_ops *ops = dev->netdev_ops;
1617
1618 p = kmemdup(&tbl->parms, sizeof(*p), GFP_KERNEL);
1619 if (p) {
1620 p->tbl = tbl;
1621 refcount_set(&p->refcnt, 1);
1622 p->reachable_time =
1623 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
1624 dev_hold(dev);
1625 p->dev = dev;
1626 write_pnet(&p->net, net);
1627 p->sysctl_table = NULL;
1628
1629 if (ops->ndo_neigh_setup && ops->ndo_neigh_setup(dev, p)) {
1630 dev_put(dev);
1631 kfree(p);
1632 return NULL;
1633 }
1634
1635 write_lock_bh(&tbl->lock);
1636 list_add(&p->list, &tbl->parms.list);
1637 write_unlock_bh(&tbl->lock);
1638
1639 neigh_parms_data_state_cleanall(p);
1640 }
1641 return p;
1642}
1643EXPORT_SYMBOL(neigh_parms_alloc);
1644
1645static void neigh_rcu_free_parms(struct rcu_head *head)
1646{
1647 struct neigh_parms *parms =
1648 container_of(head, struct neigh_parms, rcu_head);
1649
1650 neigh_parms_put(parms);
1651}
1652
1653void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
1654{
1655 if (!parms || parms == &tbl->parms)
1656 return;
1657 write_lock_bh(&tbl->lock);
1658 list_del(&parms->list);
1659 parms->dead = 1;
1660 write_unlock_bh(&tbl->lock);
1661 if (parms->dev)
1662 dev_put(parms->dev);
1663 call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
1664}
1665EXPORT_SYMBOL(neigh_parms_release);
1666
1667static void neigh_parms_destroy(struct neigh_parms *parms)
1668{
1669 kfree(parms);
1670}
1671
1672static struct lock_class_key neigh_table_proxy_queue_class;
1673
1674static struct neigh_table *neigh_tables[NEIGH_NR_TABLES] __read_mostly;
1675
1676void neigh_table_init(int index, struct neigh_table *tbl)
1677{
1678 unsigned long now = jiffies;
1679 unsigned long phsize;
1680
1681 INIT_LIST_HEAD(&tbl->parms_list);
1682 INIT_LIST_HEAD(&tbl->gc_list);
1683 list_add(&tbl->parms.list, &tbl->parms_list);
1684 write_pnet(&tbl->parms.net, &init_net);
1685 refcount_set(&tbl->parms.refcnt, 1);
1686 tbl->parms.reachable_time =
1687 neigh_rand_reach_time(NEIGH_VAR(&tbl->parms, BASE_REACHABLE_TIME));
1688
1689 tbl->stats = alloc_percpu(struct neigh_statistics);
1690 if (!tbl->stats)
1691 panic("cannot create neighbour cache statistics");
1692
1693#ifdef CONFIG_PROC_FS
1694 if (!proc_create_seq_data(tbl->id, 0, init_net.proc_net_stat,
1695 &neigh_stat_seq_ops, tbl))
1696 panic("cannot create neighbour proc dir entry");
1697#endif
1698
1699 RCU_INIT_POINTER(tbl->nht, neigh_hash_alloc(3));
1700
1701 phsize = (PNEIGH_HASHMASK + 1) * sizeof(struct pneigh_entry *);
1702 tbl->phash_buckets = kzalloc(phsize, GFP_KERNEL);
1703
1704 if (!tbl->nht || !tbl->phash_buckets)
1705 panic("cannot allocate neighbour cache hashes");
1706
1707 if (!tbl->entry_size)
1708 tbl->entry_size = ALIGN(offsetof(struct neighbour, primary_key) +
1709 tbl->key_len, NEIGH_PRIV_ALIGN);
1710 else
1711 WARN_ON(tbl->entry_size % NEIGH_PRIV_ALIGN);
1712
1713 rwlock_init(&tbl->lock);
1714 INIT_DEFERRABLE_WORK(&tbl->gc_work, neigh_periodic_work);
1715 queue_delayed_work(system_power_efficient_wq, &tbl->gc_work,
1716 tbl->parms.reachable_time);
1717 timer_setup(&tbl->proxy_timer, neigh_proxy_process, 0);
1718 skb_queue_head_init_class(&tbl->proxy_queue,
1719 &neigh_table_proxy_queue_class);
1720
1721 tbl->last_flush = now;
1722 tbl->last_rand = now + tbl->parms.reachable_time * 20;
1723
1724 neigh_tables[index] = tbl;
1725}
1726EXPORT_SYMBOL(neigh_table_init);
1727
1728int neigh_table_clear(int index, struct neigh_table *tbl)
1729{
1730 neigh_tables[index] = NULL;
1731 /* It is not clean... Fix it to unload IPv6 module safely */
1732 cancel_delayed_work_sync(&tbl->gc_work);
1733 del_timer_sync(&tbl->proxy_timer);
1734 pneigh_queue_purge(&tbl->proxy_queue);
1735 neigh_ifdown(tbl, NULL);
1736 if (atomic_read(&tbl->entries))
1737 pr_crit("neighbour leakage\n");
1738
1739 call_rcu(&rcu_dereference_protected(tbl->nht, 1)->rcu,
1740 neigh_hash_free_rcu);
1741 tbl->nht = NULL;
1742
1743 kfree(tbl->phash_buckets);
1744 tbl->phash_buckets = NULL;
1745
1746 remove_proc_entry(tbl->id, init_net.proc_net_stat);
1747
1748 free_percpu(tbl->stats);
1749 tbl->stats = NULL;
1750
1751 return 0;
1752}
1753EXPORT_SYMBOL(neigh_table_clear);
1754
1755static struct neigh_table *neigh_find_table(int family)
1756{
1757 struct neigh_table *tbl = NULL;
1758
1759 switch (family) {
1760 case AF_INET:
1761 tbl = neigh_tables[NEIGH_ARP_TABLE];
1762 break;
1763 case AF_INET6:
1764 tbl = neigh_tables[NEIGH_ND_TABLE];
1765 break;
1766 case AF_DECnet:
1767 tbl = neigh_tables[NEIGH_DN_TABLE];
1768 break;
1769 }
1770
1771 return tbl;
1772}
1773
1774const struct nla_policy nda_policy[NDA_MAX+1] = {
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};
1786
1787static int neigh_delete(struct sk_buff *skb, struct nlmsghdr *nlh,
1788 struct netlink_ext_ack *extack)
1789{
1790 struct net *net = sock_net(skb->sk);
1791 struct ndmsg *ndm;
1792 struct nlattr *dst_attr;
1793 struct neigh_table *tbl;
1794 struct neighbour *neigh;
1795 struct net_device *dev = NULL;
1796 int err = -EINVAL;
1797
1798 ASSERT_RTNL();
1799 if (nlmsg_len(nlh) < sizeof(*ndm))
1800 goto out;
1801
1802 dst_attr = nlmsg_find_attr(nlh, sizeof(*ndm), NDA_DST);
1803 if (!dst_attr) {
1804 NL_SET_ERR_MSG(extack, "Network address not specified");
1805 goto out;
1806 }
1807
1808 ndm = nlmsg_data(nlh);
1809 if (ndm->ndm_ifindex) {
1810 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1811 if (dev == NULL) {
1812 err = -ENODEV;
1813 goto out;
1814 }
1815 }
1816
1817 tbl = neigh_find_table(ndm->ndm_family);
1818 if (tbl == NULL)
1819 return -EAFNOSUPPORT;
1820
1821 if (nla_len(dst_attr) < (int)tbl->key_len) {
1822 NL_SET_ERR_MSG(extack, "Invalid network address");
1823 goto out;
1824 }
1825
1826 if (ndm->ndm_flags & NTF_PROXY) {
1827 err = pneigh_delete(tbl, net, nla_data(dst_attr), dev);
1828 goto out;
1829 }
1830
1831 if (dev == NULL)
1832 goto out;
1833
1834 neigh = neigh_lookup(tbl, nla_data(dst_attr), dev);
1835 if (neigh == NULL) {
1836 err = -ENOENT;
1837 goto out;
1838 }
1839
1840 err = __neigh_update(neigh, NULL, NUD_FAILED,
1841 NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN,
1842 NETLINK_CB(skb).portid, extack);
1843 write_lock_bh(&tbl->lock);
1844 neigh_release(neigh);
1845 neigh_remove_one(neigh, tbl);
1846 write_unlock_bh(&tbl->lock);
1847
1848out:
1849 return err;
1850}
1851
1852static int neigh_add(struct sk_buff *skb, struct nlmsghdr *nlh,
1853 struct netlink_ext_ack *extack)
1854{
1855 int flags = NEIGH_UPDATE_F_ADMIN | NEIGH_UPDATE_F_OVERRIDE |
1856 NEIGH_UPDATE_F_OVERRIDE_ISROUTER;
1857 struct net *net = sock_net(skb->sk);
1858 struct ndmsg *ndm;
1859 struct nlattr *tb[NDA_MAX+1];
1860 struct neigh_table *tbl;
1861 struct net_device *dev = NULL;
1862 struct neighbour *neigh;
1863 void *dst, *lladdr;
1864 u8 protocol = 0;
1865 int err;
1866
1867 ASSERT_RTNL();
1868 err = nlmsg_parse_deprecated(nlh, sizeof(*ndm), tb, NDA_MAX,
1869 nda_policy, extack);
1870 if (err < 0)
1871 goto out;
1872
1873 err = -EINVAL;
1874 if (!tb[NDA_DST]) {
1875 NL_SET_ERR_MSG(extack, "Network address not specified");
1876 goto out;
1877 }
1878
1879 ndm = nlmsg_data(nlh);
1880 if (ndm->ndm_ifindex) {
1881 dev = __dev_get_by_index(net, ndm->ndm_ifindex);
1882 if (dev == NULL) {
1883 err = -ENODEV;
1884 goto out;
1885 }
1886
1887 if (tb[NDA_LLADDR] && nla_len(tb[NDA_LLADDR]) < dev->addr_len) {
1888 NL_SET_ERR_MSG(extack, "Invalid link address");
1889 goto out;
1890 }
1891 }
1892
1893 tbl = neigh_find_table(ndm->ndm_family);
1894 if (tbl == NULL)
1895 return -EAFNOSUPPORT;
1896
1897 if (nla_len(tb[NDA_DST]) < (int)tbl->key_len) {
1898 NL_SET_ERR_MSG(extack, "Invalid network address");
1899 goto out;
1900 }
1901
1902 dst = nla_data(tb[NDA_DST]);
1903 lladdr = tb[NDA_LLADDR] ? nla_data(tb[NDA_LLADDR]) : NULL;
1904
1905 if (tb[NDA_PROTOCOL])
1906 protocol = nla_get_u8(tb[NDA_PROTOCOL]);
1907
1908 if (ndm->ndm_flags & NTF_PROXY) {
1909 struct pneigh_entry *pn;
1910
1911 err = -ENOBUFS;
1912 pn = pneigh_lookup(tbl, net, dst, dev, 1);
1913 if (pn) {
1914 pn->flags = ndm->ndm_flags;
1915 if (protocol)
1916 pn->protocol = protocol;
1917 err = 0;
1918 }
1919 goto out;
1920 }
1921
1922 if (!dev) {
1923 NL_SET_ERR_MSG(extack, "Device not specified");
1924 goto out;
1925 }
1926
1927 if (tbl->allow_add && !tbl->allow_add(dev, extack)) {
1928 err = -EINVAL;
1929 goto out;
1930 }
1931
1932 neigh = neigh_lookup(tbl, dst, dev);
1933 if (neigh == NULL) {
1934 bool exempt_from_gc;
1935
1936 if (!(nlh->nlmsg_flags & NLM_F_CREATE)) {
1937 err = -ENOENT;
1938 goto out;
1939 }
1940
1941 exempt_from_gc = ndm->ndm_state & NUD_PERMANENT ||
1942 ndm->ndm_flags & NTF_EXT_LEARNED;
1943 neigh = ___neigh_create(tbl, dst, dev, exempt_from_gc, true);
1944 if (IS_ERR(neigh)) {
1945 err = PTR_ERR(neigh);
1946 goto out;
1947 }
1948 } else {
1949 if (nlh->nlmsg_flags & NLM_F_EXCL) {
1950 err = -EEXIST;
1951 neigh_release(neigh);
1952 goto out;
1953 }
1954
1955 if (!(nlh->nlmsg_flags & NLM_F_REPLACE))
1956 flags &= ~(NEIGH_UPDATE_F_OVERRIDE |
1957 NEIGH_UPDATE_F_OVERRIDE_ISROUTER);
1958 }
1959
1960 if (ndm->ndm_flags & NTF_EXT_LEARNED)
1961 flags |= NEIGH_UPDATE_F_EXT_LEARNED;
1962
1963 if (ndm->ndm_flags & NTF_ROUTER)
1964 flags |= NEIGH_UPDATE_F_ISROUTER;
1965
1966 if (ndm->ndm_flags & NTF_USE) {
1967 neigh_event_send(neigh, NULL);
1968 err = 0;
1969 } else
1970 err = __neigh_update(neigh, lladdr, ndm->ndm_state, flags,
1971 NETLINK_CB(skb).portid, extack);
1972
1973 if (protocol)
1974 neigh->protocol = protocol;
1975
1976 neigh_release(neigh);
1977
1978out:
1979 return err;
1980}
1981
1982static int neightbl_fill_parms(struct sk_buff *skb, struct neigh_parms *parms)
1983{
1984 struct nlattr *nest;
1985
1986 nest = nla_nest_start_noflag(skb, NDTA_PARMS);
1987 if (nest == NULL)
1988 return -ENOBUFS;
1989
1990 if ((parms->dev &&
1991 nla_put_u32(skb, NDTPA_IFINDEX, parms->dev->ifindex)) ||
1992 nla_put_u32(skb, NDTPA_REFCNT, refcount_read(&parms->refcnt)) ||
1993 nla_put_u32(skb, NDTPA_QUEUE_LENBYTES,
1994 NEIGH_VAR(parms, QUEUE_LEN_BYTES)) ||
1995 /* approximative value for deprecated QUEUE_LEN (in packets) */
1996 nla_put_u32(skb, NDTPA_QUEUE_LEN,
1997 NEIGH_VAR(parms, QUEUE_LEN_BYTES) / SKB_TRUESIZE(ETH_FRAME_LEN)) ||
1998 nla_put_u32(skb, NDTPA_PROXY_QLEN, NEIGH_VAR(parms, PROXY_QLEN)) ||
1999 nla_put_u32(skb, NDTPA_APP_PROBES, NEIGH_VAR(parms, APP_PROBES)) ||
2000 nla_put_u32(skb, NDTPA_UCAST_PROBES,
2001 NEIGH_VAR(parms, UCAST_PROBES)) ||
2002 nla_put_u32(skb, NDTPA_MCAST_PROBES,
2003 NEIGH_VAR(parms, MCAST_PROBES)) ||
2004 nla_put_u32(skb, NDTPA_MCAST_REPROBES,
2005 NEIGH_VAR(parms, MCAST_REPROBES)) ||
2006 nla_put_msecs(skb, NDTPA_REACHABLE_TIME, parms->reachable_time,
2007 NDTPA_PAD) ||
2008 nla_put_msecs(skb, NDTPA_BASE_REACHABLE_TIME,
2009 NEIGH_VAR(parms, BASE_REACHABLE_TIME), NDTPA_PAD) ||
2010 nla_put_msecs(skb, NDTPA_GC_STALETIME,
2011 NEIGH_VAR(parms, GC_STALETIME), NDTPA_PAD) ||
2012 nla_put_msecs(skb, NDTPA_DELAY_PROBE_TIME,
2013 NEIGH_VAR(parms, DELAY_PROBE_TIME), NDTPA_PAD) ||
2014 nla_put_msecs(skb, NDTPA_RETRANS_TIME,
2015 NEIGH_VAR(parms, RETRANS_TIME), NDTPA_PAD) ||
2016 nla_put_msecs(skb, NDTPA_ANYCAST_DELAY,
2017 NEIGH_VAR(parms, ANYCAST_DELAY), NDTPA_PAD) ||
2018 nla_put_msecs(skb, NDTPA_PROXY_DELAY,
2019 NEIGH_VAR(parms, PROXY_DELAY), NDTPA_PAD) ||
2020 nla_put_msecs(skb, NDTPA_LOCKTIME,
2021 NEIGH_VAR(parms, LOCKTIME), NDTPA_PAD))
2022 goto nla_put_failure;
2023 return nla_nest_end(skb, nest);
2024
2025nla_put_failure:
2026 nla_nest_cancel(skb, nest);
2027 return -EMSGSIZE;
2028}
2029
2030static int neightbl_fill_info(struct sk_buff *skb, struct neigh_table *tbl,
2031 u32 pid, u32 seq, int type, int flags)
2032{
2033 struct nlmsghdr *nlh;
2034 struct ndtmsg *ndtmsg;
2035
2036 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2037 if (nlh == NULL)
2038 return -EMSGSIZE;
2039
2040 ndtmsg = nlmsg_data(nlh);
2041
2042 read_lock_bh(&tbl->lock);
2043 ndtmsg->ndtm_family = tbl->family;
2044 ndtmsg->ndtm_pad1 = 0;
2045 ndtmsg->ndtm_pad2 = 0;
2046
2047 if (nla_put_string(skb, NDTA_NAME, tbl->id) ||
2048 nla_put_msecs(skb, NDTA_GC_INTERVAL, tbl->gc_interval, NDTA_PAD) ||
2049 nla_put_u32(skb, NDTA_THRESH1, tbl->gc_thresh1) ||
2050 nla_put_u32(skb, NDTA_THRESH2, tbl->gc_thresh2) ||
2051 nla_put_u32(skb, NDTA_THRESH3, tbl->gc_thresh3))
2052 goto nla_put_failure;
2053 {
2054 unsigned long now = jiffies;
2055 unsigned int flush_delta = now - tbl->last_flush;
2056 unsigned int rand_delta = now - tbl->last_rand;
2057 struct neigh_hash_table *nht;
2058 struct ndt_config ndc = {
2059 .ndtc_key_len = tbl->key_len,
2060 .ndtc_entry_size = tbl->entry_size,
2061 .ndtc_entries = atomic_read(&tbl->entries),
2062 .ndtc_last_flush = jiffies_to_msecs(flush_delta),
2063 .ndtc_last_rand = jiffies_to_msecs(rand_delta),
2064 .ndtc_proxy_qlen = tbl->proxy_queue.qlen,
2065 };
2066
2067 rcu_read_lock_bh();
2068 nht = rcu_dereference_bh(tbl->nht);
2069 ndc.ndtc_hash_rnd = nht->hash_rnd[0];
2070 ndc.ndtc_hash_mask = ((1 << nht->hash_shift) - 1);
2071 rcu_read_unlock_bh();
2072
2073 if (nla_put(skb, NDTA_CONFIG, sizeof(ndc), &ndc))
2074 goto nla_put_failure;
2075 }
2076
2077 {
2078 int cpu;
2079 struct ndt_stats ndst;
2080
2081 memset(&ndst, 0, sizeof(ndst));
2082
2083 for_each_possible_cpu(cpu) {
2084 struct neigh_statistics *st;
2085
2086 st = per_cpu_ptr(tbl->stats, cpu);
2087 ndst.ndts_allocs += st->allocs;
2088 ndst.ndts_destroys += st->destroys;
2089 ndst.ndts_hash_grows += st->hash_grows;
2090 ndst.ndts_res_failed += st->res_failed;
2091 ndst.ndts_lookups += st->lookups;
2092 ndst.ndts_hits += st->hits;
2093 ndst.ndts_rcv_probes_mcast += st->rcv_probes_mcast;
2094 ndst.ndts_rcv_probes_ucast += st->rcv_probes_ucast;
2095 ndst.ndts_periodic_gc_runs += st->periodic_gc_runs;
2096 ndst.ndts_forced_gc_runs += st->forced_gc_runs;
2097 ndst.ndts_table_fulls += st->table_fulls;
2098 }
2099
2100 if (nla_put_64bit(skb, NDTA_STATS, sizeof(ndst), &ndst,
2101 NDTA_PAD))
2102 goto nla_put_failure;
2103 }
2104
2105 BUG_ON(tbl->parms.dev);
2106 if (neightbl_fill_parms(skb, &tbl->parms) < 0)
2107 goto nla_put_failure;
2108
2109 read_unlock_bh(&tbl->lock);
2110 nlmsg_end(skb, nlh);
2111 return 0;
2112
2113nla_put_failure:
2114 read_unlock_bh(&tbl->lock);
2115 nlmsg_cancel(skb, nlh);
2116 return -EMSGSIZE;
2117}
2118
2119static int neightbl_fill_param_info(struct sk_buff *skb,
2120 struct neigh_table *tbl,
2121 struct neigh_parms *parms,
2122 u32 pid, u32 seq, int type,
2123 unsigned int flags)
2124{
2125 struct ndtmsg *ndtmsg;
2126 struct nlmsghdr *nlh;
2127
2128 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndtmsg), flags);
2129 if (nlh == NULL)
2130 return -EMSGSIZE;
2131
2132 ndtmsg = nlmsg_data(nlh);
2133
2134 read_lock_bh(&tbl->lock);
2135 ndtmsg->ndtm_family = tbl->family;
2136 ndtmsg->ndtm_pad1 = 0;
2137 ndtmsg->ndtm_pad2 = 0;
2138
2139 if (nla_put_string(skb, NDTA_NAME, tbl->id) < 0 ||
2140 neightbl_fill_parms(skb, parms) < 0)
2141 goto errout;
2142
2143 read_unlock_bh(&tbl->lock);
2144 nlmsg_end(skb, nlh);
2145 return 0;
2146errout:
2147 read_unlock_bh(&tbl->lock);
2148 nlmsg_cancel(skb, nlh);
2149 return -EMSGSIZE;
2150}
2151
2152static const struct nla_policy nl_neightbl_policy[NDTA_MAX+1] = {
2153 [NDTA_NAME] = { .type = NLA_STRING },
2154 [NDTA_THRESH1] = { .type = NLA_U32 },
2155 [NDTA_THRESH2] = { .type = NLA_U32 },
2156 [NDTA_THRESH3] = { .type = NLA_U32 },
2157 [NDTA_GC_INTERVAL] = { .type = NLA_U64 },
2158 [NDTA_PARMS] = { .type = NLA_NESTED },
2159};
2160
2161static const struct nla_policy nl_ntbl_parm_policy[NDTPA_MAX+1] = {
2162 [NDTPA_IFINDEX] = { .type = NLA_U32 },
2163 [NDTPA_QUEUE_LEN] = { .type = NLA_U32 },
2164 [NDTPA_PROXY_QLEN] = { .type = NLA_U32 },
2165 [NDTPA_APP_PROBES] = { .type = NLA_U32 },
2166 [NDTPA_UCAST_PROBES] = { .type = NLA_U32 },
2167 [NDTPA_MCAST_PROBES] = { .type = NLA_U32 },
2168 [NDTPA_MCAST_REPROBES] = { .type = NLA_U32 },
2169 [NDTPA_BASE_REACHABLE_TIME] = { .type = NLA_U64 },
2170 [NDTPA_GC_STALETIME] = { .type = NLA_U64 },
2171 [NDTPA_DELAY_PROBE_TIME] = { .type = NLA_U64 },
2172 [NDTPA_RETRANS_TIME] = { .type = NLA_U64 },
2173 [NDTPA_ANYCAST_DELAY] = { .type = NLA_U64 },
2174 [NDTPA_PROXY_DELAY] = { .type = NLA_U64 },
2175 [NDTPA_LOCKTIME] = { .type = NLA_U64 },
2176};
2177
2178static int neightbl_set(struct sk_buff *skb, struct nlmsghdr *nlh,
2179 struct netlink_ext_ack *extack)
2180{
2181 struct net *net = sock_net(skb->sk);
2182 struct neigh_table *tbl;
2183 struct ndtmsg *ndtmsg;
2184 struct nlattr *tb[NDTA_MAX+1];
2185 bool found = false;
2186 int err, tidx;
2187
2188 err = nlmsg_parse_deprecated(nlh, sizeof(*ndtmsg), tb, NDTA_MAX,
2189 nl_neightbl_policy, extack);
2190 if (err < 0)
2191 goto errout;
2192
2193 if (tb[NDTA_NAME] == NULL) {
2194 err = -EINVAL;
2195 goto errout;
2196 }
2197
2198 ndtmsg = nlmsg_data(nlh);
2199
2200 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2201 tbl = neigh_tables[tidx];
2202 if (!tbl)
2203 continue;
2204 if (ndtmsg->ndtm_family && tbl->family != ndtmsg->ndtm_family)
2205 continue;
2206 if (nla_strcmp(tb[NDTA_NAME], tbl->id) == 0) {
2207 found = true;
2208 break;
2209 }
2210 }
2211
2212 if (!found)
2213 return -ENOENT;
2214
2215 /*
2216 * We acquire tbl->lock to be nice to the periodic timers and
2217 * make sure they always see a consistent set of values.
2218 */
2219 write_lock_bh(&tbl->lock);
2220
2221 if (tb[NDTA_PARMS]) {
2222 struct nlattr *tbp[NDTPA_MAX+1];
2223 struct neigh_parms *p;
2224 int i, ifindex = 0;
2225
2226 err = nla_parse_nested_deprecated(tbp, NDTPA_MAX,
2227 tb[NDTA_PARMS],
2228 nl_ntbl_parm_policy, extack);
2229 if (err < 0)
2230 goto errout_tbl_lock;
2231
2232 if (tbp[NDTPA_IFINDEX])
2233 ifindex = nla_get_u32(tbp[NDTPA_IFINDEX]);
2234
2235 p = lookup_neigh_parms(tbl, net, ifindex);
2236 if (p == NULL) {
2237 err = -ENOENT;
2238 goto errout_tbl_lock;
2239 }
2240
2241 for (i = 1; i <= NDTPA_MAX; i++) {
2242 if (tbp[i] == NULL)
2243 continue;
2244
2245 switch (i) {
2246 case NDTPA_QUEUE_LEN:
2247 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2248 nla_get_u32(tbp[i]) *
2249 SKB_TRUESIZE(ETH_FRAME_LEN));
2250 break;
2251 case NDTPA_QUEUE_LENBYTES:
2252 NEIGH_VAR_SET(p, QUEUE_LEN_BYTES,
2253 nla_get_u32(tbp[i]));
2254 break;
2255 case NDTPA_PROXY_QLEN:
2256 NEIGH_VAR_SET(p, PROXY_QLEN,
2257 nla_get_u32(tbp[i]));
2258 break;
2259 case NDTPA_APP_PROBES:
2260 NEIGH_VAR_SET(p, APP_PROBES,
2261 nla_get_u32(tbp[i]));
2262 break;
2263 case NDTPA_UCAST_PROBES:
2264 NEIGH_VAR_SET(p, UCAST_PROBES,
2265 nla_get_u32(tbp[i]));
2266 break;
2267 case NDTPA_MCAST_PROBES:
2268 NEIGH_VAR_SET(p, MCAST_PROBES,
2269 nla_get_u32(tbp[i]));
2270 break;
2271 case NDTPA_MCAST_REPROBES:
2272 NEIGH_VAR_SET(p, MCAST_REPROBES,
2273 nla_get_u32(tbp[i]));
2274 break;
2275 case NDTPA_BASE_REACHABLE_TIME:
2276 NEIGH_VAR_SET(p, BASE_REACHABLE_TIME,
2277 nla_get_msecs(tbp[i]));
2278 /* update reachable_time as well, otherwise, the change will
2279 * only be effective after the next time neigh_periodic_work
2280 * decides to recompute it (can be multiple minutes)
2281 */
2282 p->reachable_time =
2283 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
2284 break;
2285 case NDTPA_GC_STALETIME:
2286 NEIGH_VAR_SET(p, GC_STALETIME,
2287 nla_get_msecs(tbp[i]));
2288 break;
2289 case NDTPA_DELAY_PROBE_TIME:
2290 NEIGH_VAR_SET(p, DELAY_PROBE_TIME,
2291 nla_get_msecs(tbp[i]));
2292 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
2293 break;
2294 case NDTPA_RETRANS_TIME:
2295 NEIGH_VAR_SET(p, RETRANS_TIME,
2296 nla_get_msecs(tbp[i]));
2297 break;
2298 case NDTPA_ANYCAST_DELAY:
2299 NEIGH_VAR_SET(p, ANYCAST_DELAY,
2300 nla_get_msecs(tbp[i]));
2301 break;
2302 case NDTPA_PROXY_DELAY:
2303 NEIGH_VAR_SET(p, PROXY_DELAY,
2304 nla_get_msecs(tbp[i]));
2305 break;
2306 case NDTPA_LOCKTIME:
2307 NEIGH_VAR_SET(p, LOCKTIME,
2308 nla_get_msecs(tbp[i]));
2309 break;
2310 }
2311 }
2312 }
2313
2314 err = -ENOENT;
2315 if ((tb[NDTA_THRESH1] || tb[NDTA_THRESH2] ||
2316 tb[NDTA_THRESH3] || tb[NDTA_GC_INTERVAL]) &&
2317 !net_eq(net, &init_net))
2318 goto errout_tbl_lock;
2319
2320 if (tb[NDTA_THRESH1])
2321 tbl->gc_thresh1 = nla_get_u32(tb[NDTA_THRESH1]);
2322
2323 if (tb[NDTA_THRESH2])
2324 tbl->gc_thresh2 = nla_get_u32(tb[NDTA_THRESH2]);
2325
2326 if (tb[NDTA_THRESH3])
2327 tbl->gc_thresh3 = nla_get_u32(tb[NDTA_THRESH3]);
2328
2329 if (tb[NDTA_GC_INTERVAL])
2330 tbl->gc_interval = nla_get_msecs(tb[NDTA_GC_INTERVAL]);
2331
2332 err = 0;
2333
2334errout_tbl_lock:
2335 write_unlock_bh(&tbl->lock);
2336errout:
2337 return err;
2338}
2339
2340static int neightbl_valid_dump_info(const struct nlmsghdr *nlh,
2341 struct netlink_ext_ack *extack)
2342{
2343 struct ndtmsg *ndtm;
2344
2345 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndtm))) {
2346 NL_SET_ERR_MSG(extack, "Invalid header for neighbor table dump request");
2347 return -EINVAL;
2348 }
2349
2350 ndtm = nlmsg_data(nlh);
2351 if (ndtm->ndtm_pad1 || ndtm->ndtm_pad2) {
2352 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor table dump request");
2353 return -EINVAL;
2354 }
2355
2356 if (nlmsg_attrlen(nlh, sizeof(*ndtm))) {
2357 NL_SET_ERR_MSG(extack, "Invalid data after header in neighbor table dump request");
2358 return -EINVAL;
2359 }
2360
2361 return 0;
2362}
2363
2364static int neightbl_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2365{
2366 const struct nlmsghdr *nlh = cb->nlh;
2367 struct net *net = sock_net(skb->sk);
2368 int family, tidx, nidx = 0;
2369 int tbl_skip = cb->args[0];
2370 int neigh_skip = cb->args[1];
2371 struct neigh_table *tbl;
2372
2373 if (cb->strict_check) {
2374 int err = neightbl_valid_dump_info(nlh, cb->extack);
2375
2376 if (err < 0)
2377 return err;
2378 }
2379
2380 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2381
2382 for (tidx = 0; tidx < NEIGH_NR_TABLES; tidx++) {
2383 struct neigh_parms *p;
2384
2385 tbl = neigh_tables[tidx];
2386 if (!tbl)
2387 continue;
2388
2389 if (tidx < tbl_skip || (family && tbl->family != family))
2390 continue;
2391
2392 if (neightbl_fill_info(skb, tbl, NETLINK_CB(cb->skb).portid,
2393 nlh->nlmsg_seq, RTM_NEWNEIGHTBL,
2394 NLM_F_MULTI) < 0)
2395 break;
2396
2397 nidx = 0;
2398 p = list_next_entry(&tbl->parms, list);
2399 list_for_each_entry_from(p, &tbl->parms_list, list) {
2400 if (!net_eq(neigh_parms_net(p), net))
2401 continue;
2402
2403 if (nidx < neigh_skip)
2404 goto next;
2405
2406 if (neightbl_fill_param_info(skb, tbl, p,
2407 NETLINK_CB(cb->skb).portid,
2408 nlh->nlmsg_seq,
2409 RTM_NEWNEIGHTBL,
2410 NLM_F_MULTI) < 0)
2411 goto out;
2412 next:
2413 nidx++;
2414 }
2415
2416 neigh_skip = 0;
2417 }
2418out:
2419 cb->args[0] = tidx;
2420 cb->args[1] = nidx;
2421
2422 return skb->len;
2423}
2424
2425static int neigh_fill_info(struct sk_buff *skb, struct neighbour *neigh,
2426 u32 pid, u32 seq, int type, unsigned int flags)
2427{
2428 unsigned long now = jiffies;
2429 struct nda_cacheinfo ci;
2430 struct nlmsghdr *nlh;
2431 struct ndmsg *ndm;
2432
2433 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2434 if (nlh == NULL)
2435 return -EMSGSIZE;
2436
2437 ndm = nlmsg_data(nlh);
2438 ndm->ndm_family = neigh->ops->family;
2439 ndm->ndm_pad1 = 0;
2440 ndm->ndm_pad2 = 0;
2441 ndm->ndm_flags = neigh->flags;
2442 ndm->ndm_type = neigh->type;
2443 ndm->ndm_ifindex = neigh->dev->ifindex;
2444
2445 if (nla_put(skb, NDA_DST, neigh->tbl->key_len, neigh->primary_key))
2446 goto nla_put_failure;
2447
2448 read_lock_bh(&neigh->lock);
2449 ndm->ndm_state = neigh->nud_state;
2450 if (neigh->nud_state & NUD_VALID) {
2451 char haddr[MAX_ADDR_LEN];
2452
2453 neigh_ha_snapshot(haddr, neigh, neigh->dev);
2454 if (nla_put(skb, NDA_LLADDR, neigh->dev->addr_len, haddr) < 0) {
2455 read_unlock_bh(&neigh->lock);
2456 goto nla_put_failure;
2457 }
2458 }
2459
2460 ci.ndm_used = jiffies_to_clock_t(now - neigh->used);
2461 ci.ndm_confirmed = jiffies_to_clock_t(now - neigh->confirmed);
2462 ci.ndm_updated = jiffies_to_clock_t(now - neigh->updated);
2463 ci.ndm_refcnt = refcount_read(&neigh->refcnt) - 1;
2464 read_unlock_bh(&neigh->lock);
2465
2466 if (nla_put_u32(skb, NDA_PROBES, atomic_read(&neigh->probes)) ||
2467 nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
2468 goto nla_put_failure;
2469
2470 if (neigh->protocol && nla_put_u8(skb, NDA_PROTOCOL, neigh->protocol))
2471 goto nla_put_failure;
2472
2473 nlmsg_end(skb, nlh);
2474 return 0;
2475
2476nla_put_failure:
2477 nlmsg_cancel(skb, nlh);
2478 return -EMSGSIZE;
2479}
2480
2481static int pneigh_fill_info(struct sk_buff *skb, struct pneigh_entry *pn,
2482 u32 pid, u32 seq, int type, unsigned int flags,
2483 struct neigh_table *tbl)
2484{
2485 struct nlmsghdr *nlh;
2486 struct ndmsg *ndm;
2487
2488 nlh = nlmsg_put(skb, pid, seq, type, sizeof(*ndm), flags);
2489 if (nlh == NULL)
2490 return -EMSGSIZE;
2491
2492 ndm = nlmsg_data(nlh);
2493 ndm->ndm_family = tbl->family;
2494 ndm->ndm_pad1 = 0;
2495 ndm->ndm_pad2 = 0;
2496 ndm->ndm_flags = pn->flags | NTF_PROXY;
2497 ndm->ndm_type = RTN_UNICAST;
2498 ndm->ndm_ifindex = pn->dev ? pn->dev->ifindex : 0;
2499 ndm->ndm_state = NUD_NONE;
2500
2501 if (nla_put(skb, NDA_DST, tbl->key_len, pn->key))
2502 goto nla_put_failure;
2503
2504 if (pn->protocol && nla_put_u8(skb, NDA_PROTOCOL, pn->protocol))
2505 goto nla_put_failure;
2506
2507 nlmsg_end(skb, nlh);
2508 return 0;
2509
2510nla_put_failure:
2511 nlmsg_cancel(skb, nlh);
2512 return -EMSGSIZE;
2513}
2514
2515static void neigh_update_notify(struct neighbour *neigh, u32 nlmsg_pid)
2516{
2517 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, neigh);
2518 __neigh_notify(neigh, RTM_NEWNEIGH, 0, nlmsg_pid);
2519}
2520
2521static bool neigh_master_filtered(struct net_device *dev, int master_idx)
2522{
2523 struct net_device *master;
2524
2525 if (!master_idx)
2526 return false;
2527
2528 master = dev ? netdev_master_upper_dev_get(dev) : NULL;
2529 if (!master || master->ifindex != master_idx)
2530 return true;
2531
2532 return false;
2533}
2534
2535static bool neigh_ifindex_filtered(struct net_device *dev, int filter_idx)
2536{
2537 if (filter_idx && (!dev || dev->ifindex != filter_idx))
2538 return true;
2539
2540 return false;
2541}
2542
2543struct neigh_dump_filter {
2544 int master_idx;
2545 int dev_idx;
2546};
2547
2548static int neigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2549 struct netlink_callback *cb,
2550 struct neigh_dump_filter *filter)
2551{
2552 struct net *net = sock_net(skb->sk);
2553 struct neighbour *n;
2554 int rc, h, s_h = cb->args[1];
2555 int idx, s_idx = idx = cb->args[2];
2556 struct neigh_hash_table *nht;
2557 unsigned int flags = NLM_F_MULTI;
2558
2559 if (filter->dev_idx || filter->master_idx)
2560 flags |= NLM_F_DUMP_FILTERED;
2561
2562 rcu_read_lock_bh();
2563 nht = rcu_dereference_bh(tbl->nht);
2564
2565 for (h = s_h; h < (1 << nht->hash_shift); h++) {
2566 if (h > s_h)
2567 s_idx = 0;
2568 for (n = rcu_dereference_bh(nht->hash_buckets[h]), idx = 0;
2569 n != NULL;
2570 n = rcu_dereference_bh(n->next)) {
2571 if (idx < s_idx || !net_eq(dev_net(n->dev), net))
2572 goto next;
2573 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2574 neigh_master_filtered(n->dev, filter->master_idx))
2575 goto next;
2576 if (neigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2577 cb->nlh->nlmsg_seq,
2578 RTM_NEWNEIGH,
2579 flags) < 0) {
2580 rc = -1;
2581 goto out;
2582 }
2583next:
2584 idx++;
2585 }
2586 }
2587 rc = skb->len;
2588out:
2589 rcu_read_unlock_bh();
2590 cb->args[1] = h;
2591 cb->args[2] = idx;
2592 return rc;
2593}
2594
2595static int pneigh_dump_table(struct neigh_table *tbl, struct sk_buff *skb,
2596 struct netlink_callback *cb,
2597 struct neigh_dump_filter *filter)
2598{
2599 struct pneigh_entry *n;
2600 struct net *net = sock_net(skb->sk);
2601 int rc, h, s_h = cb->args[3];
2602 int idx, s_idx = idx = cb->args[4];
2603 unsigned int flags = NLM_F_MULTI;
2604
2605 if (filter->dev_idx || filter->master_idx)
2606 flags |= NLM_F_DUMP_FILTERED;
2607
2608 read_lock_bh(&tbl->lock);
2609
2610 for (h = s_h; h <= PNEIGH_HASHMASK; h++) {
2611 if (h > s_h)
2612 s_idx = 0;
2613 for (n = tbl->phash_buckets[h], idx = 0; n; n = n->next) {
2614 if (idx < s_idx || pneigh_net(n) != net)
2615 goto next;
2616 if (neigh_ifindex_filtered(n->dev, filter->dev_idx) ||
2617 neigh_master_filtered(n->dev, filter->master_idx))
2618 goto next;
2619 if (pneigh_fill_info(skb, n, NETLINK_CB(cb->skb).portid,
2620 cb->nlh->nlmsg_seq,
2621 RTM_NEWNEIGH, flags, tbl) < 0) {
2622 read_unlock_bh(&tbl->lock);
2623 rc = -1;
2624 goto out;
2625 }
2626 next:
2627 idx++;
2628 }
2629 }
2630
2631 read_unlock_bh(&tbl->lock);
2632 rc = skb->len;
2633out:
2634 cb->args[3] = h;
2635 cb->args[4] = idx;
2636 return rc;
2637
2638}
2639
2640static int neigh_valid_dump_req(const struct nlmsghdr *nlh,
2641 bool strict_check,
2642 struct neigh_dump_filter *filter,
2643 struct netlink_ext_ack *extack)
2644{
2645 struct nlattr *tb[NDA_MAX + 1];
2646 int err, i;
2647
2648 if (strict_check) {
2649 struct ndmsg *ndm;
2650
2651 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2652 NL_SET_ERR_MSG(extack, "Invalid header for neighbor dump request");
2653 return -EINVAL;
2654 }
2655
2656 ndm = nlmsg_data(nlh);
2657 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_ifindex ||
2658 ndm->ndm_state || ndm->ndm_type) {
2659 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor dump request");
2660 return -EINVAL;
2661 }
2662
2663 if (ndm->ndm_flags & ~NTF_PROXY) {
2664 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor dump request");
2665 return -EINVAL;
2666 }
2667
2668 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg),
2669 tb, NDA_MAX, nda_policy,
2670 extack);
2671 } else {
2672 err = nlmsg_parse_deprecated(nlh, sizeof(struct ndmsg), tb,
2673 NDA_MAX, nda_policy, extack);
2674 }
2675 if (err < 0)
2676 return err;
2677
2678 for (i = 0; i <= NDA_MAX; ++i) {
2679 if (!tb[i])
2680 continue;
2681
2682 /* all new attributes should require strict_check */
2683 switch (i) {
2684 case NDA_IFINDEX:
2685 filter->dev_idx = nla_get_u32(tb[i]);
2686 break;
2687 case NDA_MASTER:
2688 filter->master_idx = nla_get_u32(tb[i]);
2689 break;
2690 default:
2691 if (strict_check) {
2692 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor dump request");
2693 return -EINVAL;
2694 }
2695 }
2696 }
2697
2698 return 0;
2699}
2700
2701static int neigh_dump_info(struct sk_buff *skb, struct netlink_callback *cb)
2702{
2703 const struct nlmsghdr *nlh = cb->nlh;
2704 struct neigh_dump_filter filter = {};
2705 struct neigh_table *tbl;
2706 int t, family, s_t;
2707 int proxy = 0;
2708 int err;
2709
2710 family = ((struct rtgenmsg *)nlmsg_data(nlh))->rtgen_family;
2711
2712 /* check for full ndmsg structure presence, family member is
2713 * the same for both structures
2714 */
2715 if (nlmsg_len(nlh) >= sizeof(struct ndmsg) &&
2716 ((struct ndmsg *)nlmsg_data(nlh))->ndm_flags == NTF_PROXY)
2717 proxy = 1;
2718
2719 err = neigh_valid_dump_req(nlh, cb->strict_check, &filter, cb->extack);
2720 if (err < 0 && cb->strict_check)
2721 return err;
2722
2723 s_t = cb->args[0];
2724
2725 for (t = 0; t < NEIGH_NR_TABLES; t++) {
2726 tbl = neigh_tables[t];
2727
2728 if (!tbl)
2729 continue;
2730 if (t < s_t || (family && tbl->family != family))
2731 continue;
2732 if (t > s_t)
2733 memset(&cb->args[1], 0, sizeof(cb->args) -
2734 sizeof(cb->args[0]));
2735 if (proxy)
2736 err = pneigh_dump_table(tbl, skb, cb, &filter);
2737 else
2738 err = neigh_dump_table(tbl, skb, cb, &filter);
2739 if (err < 0)
2740 break;
2741 }
2742
2743 cb->args[0] = t;
2744 return skb->len;
2745}
2746
2747static int neigh_valid_get_req(const struct nlmsghdr *nlh,
2748 struct neigh_table **tbl,
2749 void **dst, int *dev_idx, u8 *ndm_flags,
2750 struct netlink_ext_ack *extack)
2751{
2752 struct nlattr *tb[NDA_MAX + 1];
2753 struct ndmsg *ndm;
2754 int err, i;
2755
2756 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*ndm))) {
2757 NL_SET_ERR_MSG(extack, "Invalid header for neighbor get request");
2758 return -EINVAL;
2759 }
2760
2761 ndm = nlmsg_data(nlh);
2762 if (ndm->ndm_pad1 || ndm->ndm_pad2 || ndm->ndm_state ||
2763 ndm->ndm_type) {
2764 NL_SET_ERR_MSG(extack, "Invalid values in header for neighbor get request");
2765 return -EINVAL;
2766 }
2767
2768 if (ndm->ndm_flags & ~NTF_PROXY) {
2769 NL_SET_ERR_MSG(extack, "Invalid flags in header for neighbor get request");
2770 return -EINVAL;
2771 }
2772
2773 err = nlmsg_parse_deprecated_strict(nlh, sizeof(struct ndmsg), tb,
2774 NDA_MAX, nda_policy, extack);
2775 if (err < 0)
2776 return err;
2777
2778 *ndm_flags = ndm->ndm_flags;
2779 *dev_idx = ndm->ndm_ifindex;
2780 *tbl = neigh_find_table(ndm->ndm_family);
2781 if (*tbl == NULL) {
2782 NL_SET_ERR_MSG(extack, "Unsupported family in header for neighbor get request");
2783 return -EAFNOSUPPORT;
2784 }
2785
2786 for (i = 0; i <= NDA_MAX; ++i) {
2787 if (!tb[i])
2788 continue;
2789
2790 switch (i) {
2791 case NDA_DST:
2792 if (nla_len(tb[i]) != (int)(*tbl)->key_len) {
2793 NL_SET_ERR_MSG(extack, "Invalid network address in neighbor get request");
2794 return -EINVAL;
2795 }
2796 *dst = nla_data(tb[i]);
2797 break;
2798 default:
2799 NL_SET_ERR_MSG(extack, "Unsupported attribute in neighbor get request");
2800 return -EINVAL;
2801 }
2802 }
2803
2804 return 0;
2805}
2806
2807static inline size_t neigh_nlmsg_size(void)
2808{
2809 return NLMSG_ALIGN(sizeof(struct ndmsg))
2810 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2811 + nla_total_size(MAX_ADDR_LEN) /* NDA_LLADDR */
2812 + nla_total_size(sizeof(struct nda_cacheinfo))
2813 + nla_total_size(4) /* NDA_PROBES */
2814 + nla_total_size(1); /* NDA_PROTOCOL */
2815}
2816
2817static int neigh_get_reply(struct net *net, struct neighbour *neigh,
2818 u32 pid, u32 seq)
2819{
2820 struct sk_buff *skb;
2821 int err = 0;
2822
2823 skb = nlmsg_new(neigh_nlmsg_size(), GFP_KERNEL);
2824 if (!skb)
2825 return -ENOBUFS;
2826
2827 err = neigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0);
2828 if (err) {
2829 kfree_skb(skb);
2830 goto errout;
2831 }
2832
2833 err = rtnl_unicast(skb, net, pid);
2834errout:
2835 return err;
2836}
2837
2838static inline size_t pneigh_nlmsg_size(void)
2839{
2840 return NLMSG_ALIGN(sizeof(struct ndmsg))
2841 + nla_total_size(MAX_ADDR_LEN) /* NDA_DST */
2842 + nla_total_size(1); /* NDA_PROTOCOL */
2843}
2844
2845static int pneigh_get_reply(struct net *net, struct pneigh_entry *neigh,
2846 u32 pid, u32 seq, struct neigh_table *tbl)
2847{
2848 struct sk_buff *skb;
2849 int err = 0;
2850
2851 skb = nlmsg_new(pneigh_nlmsg_size(), GFP_KERNEL);
2852 if (!skb)
2853 return -ENOBUFS;
2854
2855 err = pneigh_fill_info(skb, neigh, pid, seq, RTM_NEWNEIGH, 0, tbl);
2856 if (err) {
2857 kfree_skb(skb);
2858 goto errout;
2859 }
2860
2861 err = rtnl_unicast(skb, net, pid);
2862errout:
2863 return err;
2864}
2865
2866static int neigh_get(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2867 struct netlink_ext_ack *extack)
2868{
2869 struct net *net = sock_net(in_skb->sk);
2870 struct net_device *dev = NULL;
2871 struct neigh_table *tbl = NULL;
2872 struct neighbour *neigh;
2873 void *dst = NULL;
2874 u8 ndm_flags = 0;
2875 int dev_idx = 0;
2876 int err;
2877
2878 err = neigh_valid_get_req(nlh, &tbl, &dst, &dev_idx, &ndm_flags,
2879 extack);
2880 if (err < 0)
2881 return err;
2882
2883 if (dev_idx) {
2884 dev = __dev_get_by_index(net, dev_idx);
2885 if (!dev) {
2886 NL_SET_ERR_MSG(extack, "Unknown device ifindex");
2887 return -ENODEV;
2888 }
2889 }
2890
2891 if (!dst) {
2892 NL_SET_ERR_MSG(extack, "Network address not specified");
2893 return -EINVAL;
2894 }
2895
2896 if (ndm_flags & NTF_PROXY) {
2897 struct pneigh_entry *pn;
2898
2899 pn = pneigh_lookup(tbl, net, dst, dev, 0);
2900 if (!pn) {
2901 NL_SET_ERR_MSG(extack, "Proxy neighbour entry not found");
2902 return -ENOENT;
2903 }
2904 return pneigh_get_reply(net, pn, NETLINK_CB(in_skb).portid,
2905 nlh->nlmsg_seq, tbl);
2906 }
2907
2908 if (!dev) {
2909 NL_SET_ERR_MSG(extack, "No device specified");
2910 return -EINVAL;
2911 }
2912
2913 neigh = neigh_lookup(tbl, dst, dev);
2914 if (!neigh) {
2915 NL_SET_ERR_MSG(extack, "Neighbour entry not found");
2916 return -ENOENT;
2917 }
2918
2919 err = neigh_get_reply(net, neigh, NETLINK_CB(in_skb).portid,
2920 nlh->nlmsg_seq);
2921
2922 neigh_release(neigh);
2923
2924 return err;
2925}
2926
2927void neigh_for_each(struct neigh_table *tbl, void (*cb)(struct neighbour *, void *), void *cookie)
2928{
2929 int chain;
2930 struct neigh_hash_table *nht;
2931
2932 rcu_read_lock_bh();
2933 nht = rcu_dereference_bh(tbl->nht);
2934
2935 read_lock(&tbl->lock); /* avoid resizes */
2936 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2937 struct neighbour *n;
2938
2939 for (n = rcu_dereference_bh(nht->hash_buckets[chain]);
2940 n != NULL;
2941 n = rcu_dereference_bh(n->next))
2942 cb(n, cookie);
2943 }
2944 read_unlock(&tbl->lock);
2945 rcu_read_unlock_bh();
2946}
2947EXPORT_SYMBOL(neigh_for_each);
2948
2949/* The tbl->lock must be held as a writer and BH disabled. */
2950void __neigh_for_each_release(struct neigh_table *tbl,
2951 int (*cb)(struct neighbour *))
2952{
2953 int chain;
2954 struct neigh_hash_table *nht;
2955
2956 nht = rcu_dereference_protected(tbl->nht,
2957 lockdep_is_held(&tbl->lock));
2958 for (chain = 0; chain < (1 << nht->hash_shift); chain++) {
2959 struct neighbour *n;
2960 struct neighbour __rcu **np;
2961
2962 np = &nht->hash_buckets[chain];
2963 while ((n = rcu_dereference_protected(*np,
2964 lockdep_is_held(&tbl->lock))) != NULL) {
2965 int release;
2966
2967 write_lock(&n->lock);
2968 release = cb(n);
2969 if (release) {
2970 rcu_assign_pointer(*np,
2971 rcu_dereference_protected(n->next,
2972 lockdep_is_held(&tbl->lock)));
2973 neigh_mark_dead(n);
2974 } else
2975 np = &n->next;
2976 write_unlock(&n->lock);
2977 if (release)
2978 neigh_cleanup_and_release(n);
2979 }
2980 }
2981}
2982EXPORT_SYMBOL(__neigh_for_each_release);
2983
2984int neigh_xmit(int index, struct net_device *dev,
2985 const void *addr, struct sk_buff *skb)
2986{
2987 int err = -EAFNOSUPPORT;
2988 if (likely(index < NEIGH_NR_TABLES)) {
2989 struct neigh_table *tbl;
2990 struct neighbour *neigh;
2991
2992 tbl = neigh_tables[index];
2993 if (!tbl)
2994 goto out;
2995 rcu_read_lock_bh();
2996 if (index == NEIGH_ARP_TABLE) {
2997 u32 key = *((u32 *)addr);
2998
2999 neigh = __ipv4_neigh_lookup_noref(dev, key);
3000 } else {
3001 neigh = __neigh_lookup_noref(tbl, addr, dev);
3002 }
3003 if (!neigh)
3004 neigh = __neigh_create(tbl, addr, dev, false);
3005 err = PTR_ERR(neigh);
3006 if (IS_ERR(neigh)) {
3007 rcu_read_unlock_bh();
3008 goto out_kfree_skb;
3009 }
3010 err = neigh->output(neigh, skb);
3011 rcu_read_unlock_bh();
3012 }
3013 else if (index == NEIGH_LINK_TABLE) {
3014 err = dev_hard_header(skb, dev, ntohs(skb->protocol),
3015 addr, NULL, skb->len);
3016 if (err < 0)
3017 goto out_kfree_skb;
3018 err = dev_queue_xmit(skb);
3019 }
3020out:
3021 return err;
3022out_kfree_skb:
3023 kfree_skb(skb);
3024 goto out;
3025}
3026EXPORT_SYMBOL(neigh_xmit);
3027
3028#ifdef CONFIG_PROC_FS
3029
3030static struct neighbour *neigh_get_first(struct seq_file *seq)
3031{
3032 struct neigh_seq_state *state = seq->private;
3033 struct net *net = seq_file_net(seq);
3034 struct neigh_hash_table *nht = state->nht;
3035 struct neighbour *n = NULL;
3036 int bucket;
3037
3038 state->flags &= ~NEIGH_SEQ_IS_PNEIGH;
3039 for (bucket = 0; bucket < (1 << nht->hash_shift); bucket++) {
3040 n = rcu_dereference_bh(nht->hash_buckets[bucket]);
3041
3042 while (n) {
3043 if (!net_eq(dev_net(n->dev), net))
3044 goto next;
3045 if (state->neigh_sub_iter) {
3046 loff_t fakep = 0;
3047 void *v;
3048
3049 v = state->neigh_sub_iter(state, n, &fakep);
3050 if (!v)
3051 goto next;
3052 }
3053 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3054 break;
3055 if (n->nud_state & ~NUD_NOARP)
3056 break;
3057next:
3058 n = rcu_dereference_bh(n->next);
3059 }
3060
3061 if (n)
3062 break;
3063 }
3064 state->bucket = bucket;
3065
3066 return n;
3067}
3068
3069static struct neighbour *neigh_get_next(struct seq_file *seq,
3070 struct neighbour *n,
3071 loff_t *pos)
3072{
3073 struct neigh_seq_state *state = seq->private;
3074 struct net *net = seq_file_net(seq);
3075 struct neigh_hash_table *nht = state->nht;
3076
3077 if (state->neigh_sub_iter) {
3078 void *v = state->neigh_sub_iter(state, n, pos);
3079 if (v)
3080 return n;
3081 }
3082 n = rcu_dereference_bh(n->next);
3083
3084 while (1) {
3085 while (n) {
3086 if (!net_eq(dev_net(n->dev), net))
3087 goto next;
3088 if (state->neigh_sub_iter) {
3089 void *v = state->neigh_sub_iter(state, n, pos);
3090 if (v)
3091 return n;
3092 goto next;
3093 }
3094 if (!(state->flags & NEIGH_SEQ_SKIP_NOARP))
3095 break;
3096
3097 if (n->nud_state & ~NUD_NOARP)
3098 break;
3099next:
3100 n = rcu_dereference_bh(n->next);
3101 }
3102
3103 if (n)
3104 break;
3105
3106 if (++state->bucket >= (1 << nht->hash_shift))
3107 break;
3108
3109 n = rcu_dereference_bh(nht->hash_buckets[state->bucket]);
3110 }
3111
3112 if (n && pos)
3113 --(*pos);
3114 return n;
3115}
3116
3117static struct neighbour *neigh_get_idx(struct seq_file *seq, loff_t *pos)
3118{
3119 struct neighbour *n = neigh_get_first(seq);
3120
3121 if (n) {
3122 --(*pos);
3123 while (*pos) {
3124 n = neigh_get_next(seq, n, pos);
3125 if (!n)
3126 break;
3127 }
3128 }
3129 return *pos ? NULL : n;
3130}
3131
3132static struct pneigh_entry *pneigh_get_first(struct seq_file *seq)
3133{
3134 struct neigh_seq_state *state = seq->private;
3135 struct net *net = seq_file_net(seq);
3136 struct neigh_table *tbl = state->tbl;
3137 struct pneigh_entry *pn = NULL;
3138 int bucket = state->bucket;
3139
3140 state->flags |= NEIGH_SEQ_IS_PNEIGH;
3141 for (bucket = 0; bucket <= PNEIGH_HASHMASK; bucket++) {
3142 pn = tbl->phash_buckets[bucket];
3143 while (pn && !net_eq(pneigh_net(pn), net))
3144 pn = pn->next;
3145 if (pn)
3146 break;
3147 }
3148 state->bucket = bucket;
3149
3150 return pn;
3151}
3152
3153static struct pneigh_entry *pneigh_get_next(struct seq_file *seq,
3154 struct pneigh_entry *pn,
3155 loff_t *pos)
3156{
3157 struct neigh_seq_state *state = seq->private;
3158 struct net *net = seq_file_net(seq);
3159 struct neigh_table *tbl = state->tbl;
3160
3161 do {
3162 pn = pn->next;
3163 } while (pn && !net_eq(pneigh_net(pn), net));
3164
3165 while (!pn) {
3166 if (++state->bucket > PNEIGH_HASHMASK)
3167 break;
3168 pn = tbl->phash_buckets[state->bucket];
3169 while (pn && !net_eq(pneigh_net(pn), net))
3170 pn = pn->next;
3171 if (pn)
3172 break;
3173 }
3174
3175 if (pn && pos)
3176 --(*pos);
3177
3178 return pn;
3179}
3180
3181static struct pneigh_entry *pneigh_get_idx(struct seq_file *seq, loff_t *pos)
3182{
3183 struct pneigh_entry *pn = pneigh_get_first(seq);
3184
3185 if (pn) {
3186 --(*pos);
3187 while (*pos) {
3188 pn = pneigh_get_next(seq, pn, pos);
3189 if (!pn)
3190 break;
3191 }
3192 }
3193 return *pos ? NULL : pn;
3194}
3195
3196static void *neigh_get_idx_any(struct seq_file *seq, loff_t *pos)
3197{
3198 struct neigh_seq_state *state = seq->private;
3199 void *rc;
3200 loff_t idxpos = *pos;
3201
3202 rc = neigh_get_idx(seq, &idxpos);
3203 if (!rc && !(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3204 rc = pneigh_get_idx(seq, &idxpos);
3205
3206 return rc;
3207}
3208
3209void *neigh_seq_start(struct seq_file *seq, loff_t *pos, struct neigh_table *tbl, unsigned int neigh_seq_flags)
3210 __acquires(tbl->lock)
3211 __acquires(rcu_bh)
3212{
3213 struct neigh_seq_state *state = seq->private;
3214
3215 state->tbl = tbl;
3216 state->bucket = 0;
3217 state->flags = (neigh_seq_flags & ~NEIGH_SEQ_IS_PNEIGH);
3218
3219 rcu_read_lock_bh();
3220 state->nht = rcu_dereference_bh(tbl->nht);
3221 read_lock(&tbl->lock);
3222
3223 return *pos ? neigh_get_idx_any(seq, pos) : SEQ_START_TOKEN;
3224}
3225EXPORT_SYMBOL(neigh_seq_start);
3226
3227void *neigh_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3228{
3229 struct neigh_seq_state *state;
3230 void *rc;
3231
3232 if (v == SEQ_START_TOKEN) {
3233 rc = neigh_get_first(seq);
3234 goto out;
3235 }
3236
3237 state = seq->private;
3238 if (!(state->flags & NEIGH_SEQ_IS_PNEIGH)) {
3239 rc = neigh_get_next(seq, v, NULL);
3240 if (rc)
3241 goto out;
3242 if (!(state->flags & NEIGH_SEQ_NEIGH_ONLY))
3243 rc = pneigh_get_first(seq);
3244 } else {
3245 BUG_ON(state->flags & NEIGH_SEQ_NEIGH_ONLY);
3246 rc = pneigh_get_next(seq, v, NULL);
3247 }
3248out:
3249 ++(*pos);
3250 return rc;
3251}
3252EXPORT_SYMBOL(neigh_seq_next);
3253
3254void neigh_seq_stop(struct seq_file *seq, void *v)
3255 __releases(tbl->lock)
3256 __releases(rcu_bh)
3257{
3258 struct neigh_seq_state *state = seq->private;
3259 struct neigh_table *tbl = state->tbl;
3260
3261 read_unlock(&tbl->lock);
3262 rcu_read_unlock_bh();
3263}
3264EXPORT_SYMBOL(neigh_seq_stop);
3265
3266/* statistics via seq_file */
3267
3268static void *neigh_stat_seq_start(struct seq_file *seq, loff_t *pos)
3269{
3270 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3271 int cpu;
3272
3273 if (*pos == 0)
3274 return SEQ_START_TOKEN;
3275
3276 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
3277 if (!cpu_possible(cpu))
3278 continue;
3279 *pos = cpu+1;
3280 return per_cpu_ptr(tbl->stats, cpu);
3281 }
3282 return NULL;
3283}
3284
3285static void *neigh_stat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3286{
3287 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3288 int cpu;
3289
3290 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
3291 if (!cpu_possible(cpu))
3292 continue;
3293 *pos = cpu+1;
3294 return per_cpu_ptr(tbl->stats, cpu);
3295 }
3296 return NULL;
3297}
3298
3299static void neigh_stat_seq_stop(struct seq_file *seq, void *v)
3300{
3301
3302}
3303
3304static int neigh_stat_seq_show(struct seq_file *seq, void *v)
3305{
3306 struct neigh_table *tbl = PDE_DATA(file_inode(seq->file));
3307 struct neigh_statistics *st = v;
3308
3309 if (v == SEQ_START_TOKEN) {
3310 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");
3311 return 0;
3312 }
3313
3314 seq_printf(seq, "%08x %08lx %08lx %08lx %08lx %08lx %08lx "
3315 "%08lx %08lx %08lx %08lx %08lx %08lx\n",
3316 atomic_read(&tbl->entries),
3317
3318 st->allocs,
3319 st->destroys,
3320 st->hash_grows,
3321
3322 st->lookups,
3323 st->hits,
3324
3325 st->res_failed,
3326
3327 st->rcv_probes_mcast,
3328 st->rcv_probes_ucast,
3329
3330 st->periodic_gc_runs,
3331 st->forced_gc_runs,
3332 st->unres_discards,
3333 st->table_fulls
3334 );
3335
3336 return 0;
3337}
3338
3339static const struct seq_operations neigh_stat_seq_ops = {
3340 .start = neigh_stat_seq_start,
3341 .next = neigh_stat_seq_next,
3342 .stop = neigh_stat_seq_stop,
3343 .show = neigh_stat_seq_show,
3344};
3345#endif /* CONFIG_PROC_FS */
3346
3347static void __neigh_notify(struct neighbour *n, int type, int flags,
3348 u32 pid)
3349{
3350 struct net *net = dev_net(n->dev);
3351 struct sk_buff *skb;
3352 int err = -ENOBUFS;
3353
3354 skb = nlmsg_new(neigh_nlmsg_size(), GFP_ATOMIC);
3355 if (skb == NULL)
3356 goto errout;
3357
3358 err = neigh_fill_info(skb, n, pid, 0, type, flags);
3359 if (err < 0) {
3360 /* -EMSGSIZE implies BUG in neigh_nlmsg_size() */
3361 WARN_ON(err == -EMSGSIZE);
3362 kfree_skb(skb);
3363 goto errout;
3364 }
3365 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
3366 return;
3367errout:
3368 if (err < 0)
3369 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
3370}
3371
3372void neigh_app_ns(struct neighbour *n)
3373{
3374 __neigh_notify(n, RTM_GETNEIGH, NLM_F_REQUEST, 0);
3375}
3376EXPORT_SYMBOL(neigh_app_ns);
3377
3378#ifdef CONFIG_SYSCTL
3379static int unres_qlen_max = INT_MAX / SKB_TRUESIZE(ETH_FRAME_LEN);
3380
3381static int proc_unres_qlen(struct ctl_table *ctl, int write,
3382 void __user *buffer, size_t *lenp, loff_t *ppos)
3383{
3384 int size, ret;
3385 struct ctl_table tmp = *ctl;
3386
3387 tmp.extra1 = SYSCTL_ZERO;
3388 tmp.extra2 = &unres_qlen_max;
3389 tmp.data = &size;
3390
3391 size = *(int *)ctl->data / SKB_TRUESIZE(ETH_FRAME_LEN);
3392 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3393
3394 if (write && !ret)
3395 *(int *)ctl->data = size * SKB_TRUESIZE(ETH_FRAME_LEN);
3396 return ret;
3397}
3398
3399static struct neigh_parms *neigh_get_dev_parms_rcu(struct net_device *dev,
3400 int family)
3401{
3402 switch (family) {
3403 case AF_INET:
3404 return __in_dev_arp_parms_get_rcu(dev);
3405 case AF_INET6:
3406 return __in6_dev_nd_parms_get_rcu(dev);
3407 }
3408 return NULL;
3409}
3410
3411static void neigh_copy_dflt_parms(struct net *net, struct neigh_parms *p,
3412 int index)
3413{
3414 struct net_device *dev;
3415 int family = neigh_parms_family(p);
3416
3417 rcu_read_lock();
3418 for_each_netdev_rcu(net, dev) {
3419 struct neigh_parms *dst_p =
3420 neigh_get_dev_parms_rcu(dev, family);
3421
3422 if (dst_p && !test_bit(index, dst_p->data_state))
3423 dst_p->data[index] = p->data[index];
3424 }
3425 rcu_read_unlock();
3426}
3427
3428static void neigh_proc_update(struct ctl_table *ctl, int write)
3429{
3430 struct net_device *dev = ctl->extra1;
3431 struct neigh_parms *p = ctl->extra2;
3432 struct net *net = neigh_parms_net(p);
3433 int index = (int *) ctl->data - p->data;
3434
3435 if (!write)
3436 return;
3437
3438 set_bit(index, p->data_state);
3439 if (index == NEIGH_VAR_DELAY_PROBE_TIME)
3440 call_netevent_notifiers(NETEVENT_DELAY_PROBE_TIME_UPDATE, p);
3441 if (!dev) /* NULL dev means this is default value */
3442 neigh_copy_dflt_parms(net, p, index);
3443}
3444
3445static int neigh_proc_dointvec_zero_intmax(struct ctl_table *ctl, int write,
3446 void __user *buffer,
3447 size_t *lenp, loff_t *ppos)
3448{
3449 struct ctl_table tmp = *ctl;
3450 int ret;
3451
3452 tmp.extra1 = SYSCTL_ZERO;
3453 tmp.extra2 = SYSCTL_INT_MAX;
3454
3455 ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
3456 neigh_proc_update(ctl, write);
3457 return ret;
3458}
3459
3460int neigh_proc_dointvec(struct ctl_table *ctl, int write,
3461 void __user *buffer, size_t *lenp, loff_t *ppos)
3462{
3463 int ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
3464
3465 neigh_proc_update(ctl, write);
3466 return ret;
3467}
3468EXPORT_SYMBOL(neigh_proc_dointvec);
3469
3470int neigh_proc_dointvec_jiffies(struct ctl_table *ctl, int write,
3471 void __user *buffer,
3472 size_t *lenp, loff_t *ppos)
3473{
3474 int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3475
3476 neigh_proc_update(ctl, write);
3477 return ret;
3478}
3479EXPORT_SYMBOL(neigh_proc_dointvec_jiffies);
3480
3481static int neigh_proc_dointvec_userhz_jiffies(struct ctl_table *ctl, int write,
3482 void __user *buffer,
3483 size_t *lenp, loff_t *ppos)
3484{
3485 int ret = proc_dointvec_userhz_jiffies(ctl, write, buffer, lenp, ppos);
3486
3487 neigh_proc_update(ctl, write);
3488 return ret;
3489}
3490
3491int neigh_proc_dointvec_ms_jiffies(struct ctl_table *ctl, int write,
3492 void __user *buffer,
3493 size_t *lenp, loff_t *ppos)
3494{
3495 int ret = proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3496
3497 neigh_proc_update(ctl, write);
3498 return ret;
3499}
3500EXPORT_SYMBOL(neigh_proc_dointvec_ms_jiffies);
3501
3502static int neigh_proc_dointvec_unres_qlen(struct ctl_table *ctl, int write,
3503 void __user *buffer,
3504 size_t *lenp, loff_t *ppos)
3505{
3506 int ret = proc_unres_qlen(ctl, write, buffer, lenp, ppos);
3507
3508 neigh_proc_update(ctl, write);
3509 return ret;
3510}
3511
3512static int neigh_proc_base_reachable_time(struct ctl_table *ctl, int write,
3513 void __user *buffer,
3514 size_t *lenp, loff_t *ppos)
3515{
3516 struct neigh_parms *p = ctl->extra2;
3517 int ret;
3518
3519 if (strcmp(ctl->procname, "base_reachable_time") == 0)
3520 ret = neigh_proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
3521 else if (strcmp(ctl->procname, "base_reachable_time_ms") == 0)
3522 ret = neigh_proc_dointvec_ms_jiffies(ctl, write, buffer, lenp, ppos);
3523 else
3524 ret = -1;
3525
3526 if (write && ret == 0) {
3527 /* update reachable_time as well, otherwise, the change will
3528 * only be effective after the next time neigh_periodic_work
3529 * decides to recompute it
3530 */
3531 p->reachable_time =
3532 neigh_rand_reach_time(NEIGH_VAR(p, BASE_REACHABLE_TIME));
3533 }
3534 return ret;
3535}
3536
3537#define NEIGH_PARMS_DATA_OFFSET(index) \
3538 (&((struct neigh_parms *) 0)->data[index])
3539
3540#define NEIGH_SYSCTL_ENTRY(attr, data_attr, name, mval, proc) \
3541 [NEIGH_VAR_ ## attr] = { \
3542 .procname = name, \
3543 .data = NEIGH_PARMS_DATA_OFFSET(NEIGH_VAR_ ## data_attr), \
3544 .maxlen = sizeof(int), \
3545 .mode = mval, \
3546 .proc_handler = proc, \
3547 }
3548
3549#define NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(attr, name) \
3550 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_zero_intmax)
3551
3552#define NEIGH_SYSCTL_JIFFIES_ENTRY(attr, name) \
3553 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_jiffies)
3554
3555#define NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(attr, name) \
3556 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_userhz_jiffies)
3557
3558#define NEIGH_SYSCTL_MS_JIFFIES_ENTRY(attr, name) \
3559 NEIGH_SYSCTL_ENTRY(attr, attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3560
3561#define NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(attr, data_attr, name) \
3562 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_ms_jiffies)
3563
3564#define NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(attr, data_attr, name) \
3565 NEIGH_SYSCTL_ENTRY(attr, data_attr, name, 0644, neigh_proc_dointvec_unres_qlen)
3566
3567static struct neigh_sysctl_table {
3568 struct ctl_table_header *sysctl_header;
3569 struct ctl_table neigh_vars[NEIGH_VAR_MAX + 1];
3570} neigh_sysctl_template __read_mostly = {
3571 .neigh_vars = {
3572 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_PROBES, "mcast_solicit"),
3573 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(UCAST_PROBES, "ucast_solicit"),
3574 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(APP_PROBES, "app_solicit"),
3575 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(MCAST_REPROBES, "mcast_resolicit"),
3576 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(RETRANS_TIME, "retrans_time"),
3577 NEIGH_SYSCTL_JIFFIES_ENTRY(BASE_REACHABLE_TIME, "base_reachable_time"),
3578 NEIGH_SYSCTL_JIFFIES_ENTRY(DELAY_PROBE_TIME, "delay_first_probe_time"),
3579 NEIGH_SYSCTL_JIFFIES_ENTRY(GC_STALETIME, "gc_stale_time"),
3580 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(QUEUE_LEN_BYTES, "unres_qlen_bytes"),
3581 NEIGH_SYSCTL_ZERO_INTMAX_ENTRY(PROXY_QLEN, "proxy_qlen"),
3582 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(ANYCAST_DELAY, "anycast_delay"),
3583 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(PROXY_DELAY, "proxy_delay"),
3584 NEIGH_SYSCTL_USERHZ_JIFFIES_ENTRY(LOCKTIME, "locktime"),
3585 NEIGH_SYSCTL_UNRES_QLEN_REUSED_ENTRY(QUEUE_LEN, QUEUE_LEN_BYTES, "unres_qlen"),
3586 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(RETRANS_TIME_MS, RETRANS_TIME, "retrans_time_ms"),
3587 NEIGH_SYSCTL_MS_JIFFIES_REUSED_ENTRY(BASE_REACHABLE_TIME_MS, BASE_REACHABLE_TIME, "base_reachable_time_ms"),
3588 [NEIGH_VAR_GC_INTERVAL] = {
3589 .procname = "gc_interval",
3590 .maxlen = sizeof(int),
3591 .mode = 0644,
3592 .proc_handler = proc_dointvec_jiffies,
3593 },
3594 [NEIGH_VAR_GC_THRESH1] = {
3595 .procname = "gc_thresh1",
3596 .maxlen = sizeof(int),
3597 .mode = 0644,
3598 .extra1 = SYSCTL_ZERO,
3599 .extra2 = SYSCTL_INT_MAX,
3600 .proc_handler = proc_dointvec_minmax,
3601 },
3602 [NEIGH_VAR_GC_THRESH2] = {
3603 .procname = "gc_thresh2",
3604 .maxlen = sizeof(int),
3605 .mode = 0644,
3606 .extra1 = SYSCTL_ZERO,
3607 .extra2 = SYSCTL_INT_MAX,
3608 .proc_handler = proc_dointvec_minmax,
3609 },
3610 [NEIGH_VAR_GC_THRESH3] = {
3611 .procname = "gc_thresh3",
3612 .maxlen = sizeof(int),
3613 .mode = 0644,
3614 .extra1 = SYSCTL_ZERO,
3615 .extra2 = SYSCTL_INT_MAX,
3616 .proc_handler = proc_dointvec_minmax,
3617 },
3618 {},
3619 },
3620};
3621
3622int neigh_sysctl_register(struct net_device *dev, struct neigh_parms *p,
3623 proc_handler *handler)
3624{
3625 int i;
3626 struct neigh_sysctl_table *t;
3627 const char *dev_name_source;
3628 char neigh_path[ sizeof("net//neigh/") + IFNAMSIZ + IFNAMSIZ ];
3629 char *p_name;
3630
3631 t = kmemdup(&neigh_sysctl_template, sizeof(*t), GFP_KERNEL);
3632 if (!t)
3633 goto err;
3634
3635 for (i = 0; i < NEIGH_VAR_GC_INTERVAL; i++) {
3636 t->neigh_vars[i].data += (long) p;
3637 t->neigh_vars[i].extra1 = dev;
3638 t->neigh_vars[i].extra2 = p;
3639 }
3640
3641 if (dev) {
3642 dev_name_source = dev->name;
3643 /* Terminate the table early */
3644 memset(&t->neigh_vars[NEIGH_VAR_GC_INTERVAL], 0,
3645 sizeof(t->neigh_vars[NEIGH_VAR_GC_INTERVAL]));
3646 } else {
3647 struct neigh_table *tbl = p->tbl;
3648 dev_name_source = "default";
3649 t->neigh_vars[NEIGH_VAR_GC_INTERVAL].data = &tbl->gc_interval;
3650 t->neigh_vars[NEIGH_VAR_GC_THRESH1].data = &tbl->gc_thresh1;
3651 t->neigh_vars[NEIGH_VAR_GC_THRESH2].data = &tbl->gc_thresh2;
3652 t->neigh_vars[NEIGH_VAR_GC_THRESH3].data = &tbl->gc_thresh3;
3653 }
3654
3655 if (handler) {
3656 /* RetransTime */
3657 t->neigh_vars[NEIGH_VAR_RETRANS_TIME].proc_handler = handler;
3658 /* ReachableTime */
3659 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler = handler;
3660 /* RetransTime (in milliseconds)*/
3661 t->neigh_vars[NEIGH_VAR_RETRANS_TIME_MS].proc_handler = handler;
3662 /* ReachableTime (in milliseconds) */
3663 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler = handler;
3664 } else {
3665 /* Those handlers will update p->reachable_time after
3666 * base_reachable_time(_ms) is set to ensure the new timer starts being
3667 * applied after the next neighbour update instead of waiting for
3668 * neigh_periodic_work to update its value (can be multiple minutes)
3669 * So any handler that replaces them should do this as well
3670 */
3671 /* ReachableTime */
3672 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME].proc_handler =
3673 neigh_proc_base_reachable_time;
3674 /* ReachableTime (in milliseconds) */
3675 t->neigh_vars[NEIGH_VAR_BASE_REACHABLE_TIME_MS].proc_handler =
3676 neigh_proc_base_reachable_time;
3677 }
3678
3679 /* Don't export sysctls to unprivileged users */
3680 if (neigh_parms_net(p)->user_ns != &init_user_ns)
3681 t->neigh_vars[0].procname = NULL;
3682
3683 switch (neigh_parms_family(p)) {
3684 case AF_INET:
3685 p_name = "ipv4";
3686 break;
3687 case AF_INET6:
3688 p_name = "ipv6";
3689 break;
3690 default:
3691 BUG();
3692 }
3693
3694 snprintf(neigh_path, sizeof(neigh_path), "net/%s/neigh/%s",
3695 p_name, dev_name_source);
3696 t->sysctl_header =
3697 register_net_sysctl(neigh_parms_net(p), neigh_path, t->neigh_vars);
3698 if (!t->sysctl_header)
3699 goto free;
3700
3701 p->sysctl_table = t;
3702 return 0;
3703
3704free:
3705 kfree(t);
3706err:
3707 return -ENOBUFS;
3708}
3709EXPORT_SYMBOL(neigh_sysctl_register);
3710
3711void neigh_sysctl_unregister(struct neigh_parms *p)
3712{
3713 if (p->sysctl_table) {
3714 struct neigh_sysctl_table *t = p->sysctl_table;
3715 p->sysctl_table = NULL;
3716 unregister_net_sysctl_table(t->sysctl_header);
3717 kfree(t);
3718 }
3719}
3720EXPORT_SYMBOL(neigh_sysctl_unregister);
3721
3722#endif /* CONFIG_SYSCTL */
3723
3724static int __init neigh_init(void)
3725{
3726 rtnl_register(PF_UNSPEC, RTM_NEWNEIGH, neigh_add, NULL, 0);
3727 rtnl_register(PF_UNSPEC, RTM_DELNEIGH, neigh_delete, NULL, 0);
3728 rtnl_register(PF_UNSPEC, RTM_GETNEIGH, neigh_get, neigh_dump_info, 0);
3729
3730 rtnl_register(PF_UNSPEC, RTM_GETNEIGHTBL, NULL, neightbl_dump_info,
3731 0);
3732 rtnl_register(PF_UNSPEC, RTM_SETNEIGHTBL, neightbl_set, NULL, 0);
3733
3734 return 0;
3735}
3736
3737subsys_initcall(neigh_init);